In late 1994, we decided to learn and investigate Object Oriented programming and C++ to better judge the suitability of these relatively new techniques for scientific programming. We knew that there is no better way to learn a new programming environment than to use it to write a program that can solve a real problem. After a few weeks, we had our first histogramming package in C++. A few weeks later we had a rewrite of the same package using the, at that time, very new template features of C++. Again, a few weeks later we had another rewrite of the package without templates since we could only compile the version with templates on one single platform using a specific compiler. Finally, after about four months we had a histogramming package that was faster and more efficient than the well-known FORTRAN based HBOOK histogramming package. This gave us enough confidence in the new technologies to decide to continue the development. Thus was born ROOT. Since its first public release at the end of 1995, ROOT has enjoyed an ever-increasing popularity. Currently it is being used in all major High Energy and Nuclear Physics laboratories around the world to monitor, to store and to analyse data. In the other sciences as well as the medical and financial industries, many people are using ROOT. We estimate the current user base to be around several thousand people. In 1997, Eric Raymond analysed in his paper “The Cathedral and the Bazaar” the development method that makes Linux such a success. The essence of that method is: “release early, release often and listen to your customers”. This is precisely how ROOT is being developed. Over the last five years, many of our “customers” became co-developers. Here we would like to thank our main co-developers and contributors:

Masaharu Goto wrote the CINT C++ interpreter that became an essential part of ROOT. Despite being 8 time zones ahead of us, we have the feeling he has been sitting in the room next door since 1995.

Andrei and Mihaela Gheata (Alice collaboration) are co-authors of the ROOT geometry classes and Virtual Monte-Carlo. They have been working with the ROOT team since 2000.

Olivier Couet, who after a successful development and maintenance of PAW, has joined the ROOT team in 2000 and has been working on the graphics sub-system.

Ilka Antcheva has been working on the Graphical User Interface classes. She is also responsible for this latest edition of the Users Guide with a better style, improved index and several new chapters (since 2002).

Bertrand Bellenot has been developing and maintaining the Win32GDK version of ROOT. Bertrand has also many other contributions like the nice RootShower example (since 2001).

Valeriy Onoutchin has been working on several ROOT packages, in particular the graphics sub-system for Windows and the GUI Builder (since 2000).

Gerri Ganis has been working on the authentication procedures to be used by the root daemons and the PROOF system (since 2002).

Maarten Ballintijn (MIT) is one of the main developers of the PROOF sub-system (since 1995).

Valeri Fine (now at BNL) ported ROOT to Windows and contributed largely to the 3-D graphics. He is currently working on the Qt layer of ROOT (since 1995).

Victor Perevoztchikov (BNL) worked on key elements of the I/O system, in particular the improved support for STL collections (1997-2001).

Nenad Buncic developed the HTML documentation generation system and integrated the X3D viewer inside ROOT (1995-1997).

Suzanne Panacek was the author of the first version of this User’s Guide and very active in preparing tutorials and giving lectures about ROOT (1999-2002).

Axel Naumann has been developing further the HTML Reference Guide and helps in porting ROOT under Windows (cygwin/gcc implementation) (since 2000).

Anna Kreshuk has developed the Linear Fitter and Robust Fitter classes as well as many functions in TMath, TF1, TGraph (since 2005).

Richard Maunder has contributed to the GL viewer classes (since 2004).

Timur Pocheptsov has contributed to the GL viewer classes and GL in pad classes (since 2004).

Sergei Linev has developed the XML driver and the TSQLFile classes (since 2003).

Stefan Roiser has been contributing to the reflex and cintex packages (since 2005).

Lorenzo Moneta has been contributing the MathCore, MathMore, Smatrix & Minuit2 packages (since 2005).

Wim Lavrijsen is the author of the PyRoot package (since 2004).

Further we would like to thank all the people mentioned in the $ROOTSYS/README/CREDITS file for their contributions, and finally, everybody who gave comments, reported bugs and provided fixes.

Happy ROOTing!

Rene Brun & Fons Rademakers

Geneva, July 2007

1 Introduction

In the mid 1990’s, René Brun and Fons Rademakers had many years of experience developing interactive tools and simulation packages. They had lead successful projects such as PAW, PIAF, and GEANT, and they knew PAW the twenty-year-old FORTRAN libraries had reached their limits. Although still very popular, these tools could not scale up to the challenges offered by the Large Hadron Collider, where the data is a few orders of magnitude larger than anything seen before.

At the same time, computer science had made leaps of progress especially in the area of Object Oriented Design, and René and Fons were ready to take advantage of it.

ROOT was developed in the context of the NA49 experiment at CERN. NA49 has generated an impressive amount of data, around 10 Terabytes per run. This rate provided the ideal environment to develop and test the next generation data analysis.

One cannot mention ROOT without mentioning CINT, its C++ interpreter. CINT was created by Masa Goto in Japan. It is an independent product, which ROOT is using for the command line and script processor.

ROOT was, and still is, developed in the “Bazaar style”, a term from the book “The Cathedral and the Bazaar” by Eric S. Raymond. It means a liberal, informal development style that heavily relies on the diverse and deep talent of the user community. The result is that physicists developed ROOT for themselves; this made it specific, appropriate, useful, and over time refined and very powerful. The development of ROOT is a continuous conversation between users and developers with the line between the two blurring at times and the users becoming co-developers.

When it comes to storing and mining large amount of data, physics plows the way with its Terabytes, but other fields and industry follow close behind as they acquiring more and more data over time. They are ready to use the true and tested technologies physics has invented. In this way, other fields and industries have found ROOT useful and they have started to use it also.

In the bazaar view, software is released early and frequently to expose it to thousands of eager co-developers to pound on, report bugs, and contribute possible fixes. More users find more bugs, because they stress the program in different ways. By now, after ten years, the age of ROOT is quite mature. Most likely, you will find the features you are looking for, and if you have found a hole, you are encouraged to participate in the dialog and post your suggestion or even implementation on roottalk, the ROOT mailing list.

1.1 The ROOT Mailing Lists

The roottalk was the very first active ROOT mailing list. mailing list People can subscribe to it by registering at the ROOT web site: The RootTalk Forum has been gradually replaced this mailing list since September 2003. The RootTalk Forum is a web-based news group with about 10 discussion sub-units.

If you have a question, it is likely that it has been asked, answered, and stored in the roottalk or RootTalk Forum archives. Please use the search engine to see if your question has already been answered before sending a mail to the roottalk list or post a topic in the Forum.

You can browse the roottalk archives at: You can send your question without subscribing to:

1.2 Contact Information

Several authors wrote this book and you may see a “change of voice” from one chapter to the next. We felt we could accept this in order to have the expert explain what they know best. If you would like to contribute a chapter or add to a section, please contact . We count on you to send us suggestions on additional topics or on the topics that need more documentation. Please send your comments, corrections, questions, and suggestions to the rootdoc list:

We attempt to give the user insight into the many capabilities of ROOT. The book begins with the elementary functionality and progresses in complexity reaching the specialized topics at the end. The experienced user looking for special topics may find these chapters useful: see “Networking”, “Writing a Graphical User Interface”, “Threads”, and “PROOF: Parallel Processing”.

1.3 Conventions Used in This Book

We tried to follow a style convention for the sake of clarity. The styles in used are described below.

To show source code in scripts or source files:

   cout << " Hello" << endl;
   float x = 3.;
   float y = 5.;
   int   i = 101;
   cout <<" x = "<<x<<" y = "<<y<<" i = "<<i<< endl;

To show the ROOT command line, we show the ROOT prompt without numbers. In the interactive system, the ROOT prompt has a line number (root[12]); for the sake of simplicity, the line numbers are left off.

root[] TLine l
root[] l.Print()
TLine  X1=0.000000 Y1=0.000000 X2=0.000000 Y2=0.000000

Italic bold monotype font indicates a global variable, for example gDirectory.

When a variable term is used, it is shown between angled brackets. In the example below the variable term <library> can be replaced with any library in the $ROOTSYS directory: $ROOTSYS/<library>/inc.

1.4 The Framework

ROOT is an object-oriented framework aimed at solving the data analysis challenges of high-energy physics. There are two key words in this definition, object oriented and framework. First, we explain what we mean by a framework and then why it is an object-oriented framework.

1.4.1 What Is a Framework?

Programming inside a framework is a little like living in a city. Plumbing, electricity, telephone, and transportation are services provided by the city. In your house, you have interfaces to the services such as light switches, electrical outlets, and telephones. The details, for example, the routing algorithm of the phone switching system, are transparent to you as the user. You do not care; you are only interested in using the phone to communicate with your collaborators to solve your domain specific problems.

Programming outside of a framework may be compared to living in the country. In order to have transportation and water, you will have to build a road and dig a well. To have services like telephone and electricity you will need to route the wires to your home. In addition, you cannot build some things yourself. For example, you cannot build a commercial airport on your patch of land. From a global perspective, it would make no sense for everyone to build his or her own airport. You see you will be very busy building the infrastructure (or framework) before you can use the phone to communicate with your collaborators and have a drink of water at the same time. In software engineering, it is much the same way. In a framework, the basic utilities and services, such as I/O and graphics, are provided. In addition, ROOT being a HEP analysis framework, it provides a large selection of HEP specific utilities such as histograms and fitting. The drawback of a framework is that you are constrained to it, as you are constraint to use the routing algorithm provided by your telephone service. You also have to learn the framework interfaces, which in this analogy is the same as learning how to use a telephone.

If you are interested in doing physics, a good HEP framework will save you much work. Next is a list of the more commonly used components of ROOT: Command Line Interpreter, Histograms and Fitting, Writing a Graphical User Interface, 2D Graphics, Input/Output , Collection Classes, Script Processor.

There are also less commonly used components, as: 3D Graphics, Parallel Processing (PROOF), Run Time Type Identification (RTTI), Socket and Network Communication, Threads. Advantages of Frameworks

The benefits of frameworks can be summarized as follows:

1.4.2 Why Object-Oriented?

Object-Oriented Programming offers considerable benefits compared to Procedure-Oriented Programming:

1.5 Installing ROOT

To install ROOT you will need to go to the ROOT website at: You have a choice to download the binaries or the source. The source is quicker to transfer since it is only ~22 MB, but you will need to compile and link it. The binaries compiled with no degug information range from ~35 MB to ~45 MB depending on the target platform.

The installation and building of ROOT is described in Appendix A: Install and Build ROOT. You can download the binaries, or the source. The GNU g++ compiler on most UNIX platforms can compile ROOT.

Before downloading a binary version make sure your machine contains the right run-time environment. In most cases it is not possible to run a version compiled with, e.g., gcc4.0 on a platform where only gcc 3.2 is installed. In such cases you’ll have to install ROOT from source.

ROOT is currently running on the following platforms: supported platforms

1.6 The Organization of the ROOT Framework

Now after we know in abstract terms what the ROOT framework is, let us look at the physical directories and files that come with the ROOT installation. You may work on a platform where your system administrator has already installed ROOT. You will need to follow the specific development environment for your setup and you may not have write access to the directories. In any case, you will need an environment variable called ROOTSYS, which holds the path of the top ROOT directory.

> echo $ROOTSYS

In the ROOTSYS directory are examples, executables, tutorials, header tutorials files, and, if you opted to download it, the source is here. The directories of special interest to us are bin, tutorials, lib, test, andinclude. The next figure shows the contents of these directories.

ROOT framework directories

ROOT framework directories

1.6.1 $ROOTSYS/bin

The bin directory contains several executables.


shows the ROOT splash screen and calls root.exe


the executable that root calls, if you use a debugger such as gdb, you will need to run root.exe directly CINTdebugger


is the utility ROOT uses to create a class dictionary for CINT


a modified version of makedepend that is used by the ROOT build system


a script returning the needed compile flags and libraries for projects that compile and link with ROOT


the C++ interpreter executable that is independent of ROOT


the pure CINT version of rootcint, used to generate a dictionary; It is used by some of CINT install scripts to generate dictionaries for external system libraries


a small daemon used to authenticate a user of ROOT parallel processing capability (PROOF)


the actual PROOF process, which is started by proofd after a user, has successfully been authenticated


is the daemon for remote ROOT file access (see the TNetFile)

1.6.2 $ROOTSYS/lib

There are several ways to use ROOT, one way is to run the executable by typing root at the system prompt another way is to link with the ROOT libraries and make the ROOT classes available in your own program.

Here is a short description of the most relevant libraries, the ones marked with a * are only installed when the options specified them. Library Dependencies

ROOT libraries dependencies

ROOT libraries dependencies

The libraries are designed and organized to minimize dependencies, such that you can load just enough code for the task at hand rather than having to load all libraries or one monolithic chunk. The core library ( contains the essentials; it is a part of all ROOT applications. In the Figure 1-2 you see that is made up of base classes, container classes, meta information classes, operating system specific classes, and the ZIP algorithm used for compression of the ROOT files.

The CINT library ( is also needed in all ROOT applications, and even by libCore. It can be used independently of libCore, in case you only need the C++ interpreter and not ROOT. A program referencing only TObject only needs libCore and libCint. To add the ability to read and write ROOT objects one also has to load libRIO. As one would expect, none of that depends on graphics or the GUI.

Library dependencies have different consequences; depending on whether you try to build a binary, or you just try to access a class that is defined in a library. Linktime Library Dependencies

When building your own executable you will have to link against the libraries that contain the classes you use. The ROOT reference guide states the library a class is reference guide defined in. Almost all relevant classes can be found in libraries returned by root-config -glibs; the graphics libraries are retuned by root-config --libs. These commands are commonly used in Makefiles. Using root-config instead of enumerating the libraries by hand allows you to link them in a platform independent way. Also, if ROOT library names change you will not need to change your Makefile.

A batch program that does not have a graphic display, which creates, fills, and saves histograms and trees, only needs to link the core libraries (libCore, libCint, libRIO), libHist and libTree. If ROOT needs access to other libraries, it loads them dynamically. For example, if the TreeViewer is used, libTreePlayer and all libraries libTreePlayer depends on are loaded also. The dependent libraries are shown in the ROOT reference guide’s library dependency graph. The difference between reference guide libHist and libHistPainter is that the former needs to be explicitly linked and the latter will be loaded automatically at runtime when ROOT needs it, by means of the Plugin Manager. plugin manager

In the Figure 1-2, the libraries represented by green boxes outside of the core are loaded via the plugin manager plugin manager or equivalent techniques, while the white ones are not. Of course, if one wants to access a plugin library directly, it has to be explicitly linked. An example of a plugin library is libMinuit. To create and fill histograms you need to link If the code has a call to fit the histogram, the “fitter” will dynamically load libMinuit if it is not yet loaded. Plugins: Runtime Library Dependencies for Linking

plugin manager The Plugin Manager TPluginManager allows postponing library dependencies to runtime: a plugin library will only be loaded when it is needed. Non-plugins will need to be linked, and are thus loaded at start-up. Plugins are defined by a base class (e.g. TFile) that will be implemented in a plugin, a tag used to identify the plugin (e.g. ^rfio: as part of the protocol string), the plugin class of which an object will be created (e.g. TRFIOFile), the library to be loaded (in short to RFIO), and the constructor to be called (e.g. “TRFIOFile()”). This can be specified in the .rootrc which already contains many plugin definitions, or by calls to gROOT->GetPluginManager()->AddHandler(). Library Autoloading

When using a class in CINT, e.g. in an interpreted source file, ROOT will automatically load the library that defines this class. On start-up, ROOT parses all files ending on .rootmap rootmap that are in one of the $LD_LIBRARY_PATH (or $DYLD_LIBRARY_PATH for MacOS, or $PATH for Windows). They contain class names and the library names that the class depends on. After reading them, ROOT knows which classes are available, and which libraries to load for them.

When TSystem::Load("ALib") is called, ROOT uses this information to determine which libraries depends on. It will load these libraries first. Otherwise, loading the requested library could cause a system (dynamic loader) error due to unresolved symbols.

1.6.3 $ROOTSYS/tutorials

tutorials The tutorials directory contains many example example scripts. They assume some basic knowledge of ROOT, and for the new user we recommend reading the chapters: “Histograms” and “Input/Output” before trying the examples. The more experienced user can jump to chapter “The Tutorials and Tests” to find more explicit and specific information about how to build and run the examples.

The $ROOTSYS/tutorials/ directory include the following sub-directories:

fft: Fast Fourier Transform with the fftw package fit: Several examples illustrating minimization/fitting foam: Random generator in multidimensional space geom: Examples of use of the geometry package (TGeo classes) gl: Visualisation with OpenGL graphics: Basic graphics graphs: Use of TGraph, TGraphErrors, etc. gui: Scripts to create Graphical User Interface hist: Histograming image: Image Processing io: Input/Output math: Maths and Statistics functions matrix: Matrices (TMatrix) examples mlp: Neural networks with TMultiLayerPerceptron net: Network classes (client/server examples) physics: LorentzVectors, phase space pyroot: Python tutorials pythia: Example with pythia6 quadp: Quadratic Programming ruby: ruby tutorials smatrix: Matrices with a templated package spectrum: Peak finder, background, deconvolutions splot: Example of the TSplot class (signal/background estimator) sql: Interfaces to SQL (mysql, oracle, etc) thread: Using Threads tmva: Examples of the MultiVariate Analysis classes tree: Creating Trees, Playing with Trees unuran: Interface with the unuram random generator library xml: Writing/Reading xml files

You can execute the scripts in $ROOTSYS/tutorials (or sub-directories) by setting your current directory in the script directory or from any user directory with write access. Several tutorials create new files. If you have write access to the tutorials directory, the new files will be created in the tutorials directory, otherwise they will be created in the user directory.

1.6.4 $ROOTSYS/test

The test directory contains a set of examples example that represent all areas of the framework. When a new release is cut, the examples in this directory are compiled and run to test the new release’s backward compatibility. The list of source files is described in chapter “The Tutorials and Tests”.

The $ROOTSYS/test directory is a gold mine of ROOT-wisdom nuggets, and we encourage you to explore and exploit it. We recommend the new users to read the chapter “Getting Started”. The chapter “The Tutorials and Tests” has instructions on how to build all the programs and it goes over the examples Event and stress.

1.6.5 $ROOTSYS/include

The include directory contains all header files. It is especially important because the header files contain the class definitions.

1.6.6 $ROOTSYS/<library>

The directories we explored above are available when downloading the binaries. When downloading the source you also get a directory for each library with the corresponding header and source files, located in the inc and src subdirectories. To see what classes are in a library, you can check the <library>/inc directory for the list of class definitions. For example, the physics library contains these class definitions:

> ls -m $ROOTSYS/math/physics/inc/
LinkDef.h, TFeldmanCousins.h, TGenPhaseSpace.h, TLorentzRotation.h,
TLorentzVector.h, TQuaternion.h, TRobustEstimator.h, TRolke.h,
TRotation.h, TVector2.h, TVector3.h

1.7 How to Find More Information

website The ROOT web site has up to date documentation. The ROOT source code automatically generates this documentation, so each class is explicitly documented on its own web page, which is always up to date with the latest official release of ROOT.

The ROOT Reference Guide web pages can be found at class index reference guide Each page contains a class description, and an explanation of each method. It shows the class inheritance tree and lets you jump to the parent class page by clicking on the class name. If you want more details, you can even see the source. There is a help page available in the little box on the upper right hand side of each class documentation page. You can see on the next page what a typical class documentation web page looks like. The ROOT web site also contains in addition to this Reference Guide, “How To’s”, a list of publications and example applications.

1.7.1 Class Reference Guide

The top of any class reference page lets you jump to different parts of the documentation. The first line links to the class index and the index for the current module (a group of classes, often a library). The second line links to the ROOT homepage and the class overviews. The third line links the source information - a HTML version of the source and header file as well as the CVS (the source management system used for the ROOT development) information of the files. The last line links the different parts of the current pages.

Example of function documentation, with automatically generated LaTeX-like graphics

Example of function documentation, with automatically generated LaTeX-like graphics

Inheritance tree, showing what the current class derives from, and which classes inherit from it

Inheritance tree, showing what the current class derives from, and which classes inherit from it

HTML version of the source file linking all types and most functions

HTML version of the source file linking all types and most functions

2 Getting Started

We begin by showing you how to use ROOT interactively. There are two examples to click through and learn how to use the GUI. We continue by using the command line, and explaining the coding conventions, global variables and the environment setup. If you have not installed ROOT, you can do so by following the instructions in the appendix, or on the ROOT web site:

2.1 Setting the Environment Variables

Before you can run ROOT you need to set the environment variable ROOTSYS and change your path to include root/bin and library path variables to include root/lib. Please note: the syntax is for bash, if you are running tcsh you will have to use setenv instead of export.

  1. Define the variable $ROOTSYS to the directory where you unpacked the ROOT:
$ export ROOTSYS=$HOME/root
  1. Add ROOTSYS/bin to your PATH:
$ export PATH=$PATH:$ROOTSYS/bin
  1. Setting the Library Path

On HP-UX, before executing the interactive module, you must set the library path:


On AIX, before executing the interactive module, you must set the library path:

$ [ -z "$LIBPATH" ] && export LIBPATH=/lib:/usr/lib

On Linux, Solaris, Alpha OSF and SGI, before executing the interactive module, you must set the library path:


On Solaris, in case your LD_LIBRARY_PATH is empty, you should set it:

$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ROOTSYS/lib:/usr/dt/lib

If you use the afs version you should set (vers = version number, arch = architecture):

$ export ROOTSYS=/afs/

If ROOT was installed in $HOME/myroot directory on a local machine, one can do:

cd $HOME/myroot
. bin/       // or source bin/

The new $ROOTSYS/bin/thisroot.[c]sh scripts will set correctly the ROOTSYS, LD_LIBRARY_PATH or other paths depending on the platform and the MANPATH. To run the program just type: root.

2.2 Start and Quit a ROOT Session

% root
  *                                         *
  *        W E L C O M E  to  R O O T       *
  *                                         *
  *   Version   5.34/07     26 April 2013   *
  *                                         *
  *  You are welcome to visit our Web site  *
  *            *
  *                                         *

ROOT 5.34/07 (v5-34-07@c1f030b, May 13 2013, 16:42:38 on macosx64)

CINT/ROOT C/C++ Interpreter version 5.18.00, July 2, 2010
Type ? for help. Commands must be C++ statements.
Enclose multiple statements between { }.
root [0]

To start ROOT you can type root at the system prompt. This starts up CINT, the ROOT command line C/C++ interpreter, and it gives you the ROOT prompt (root[0]).

It is possible to launch ROOT with some command line options, as shown below:

%  root -?
Usage: root [-l] [-b] [-n] [-q] [dir] [[file:]data.root]
                                                [file1.C ... fileN.C]
  -b : run in batch mode without graphics
  -n : do not execute logon and logoff macros as specified in .rootrc
  -q : exit after processing command line macro files
  -l : do not show splash screen
  -x : exit on exception
 dir : if dir is a valid directory cd to it before executing

  -?       : print usage
  -h       : print usage
  --help   : print usage
  -config  : print ./configure options
  -memstat : run with memory usage monitoring

For example if you would like to run a script myMacro.C in the background, redirect the output into a file myMacro.log, and exit after the script execution, use the following syntax:

root -b -q myMacro.C > myMacro.log

If you need to pass a parameter to the script use:

root -b -q 'myMacro.C(3)' > myMacro.log

Be mindful of the quotes, i.e. if you need to pass a string as a parameter, the syntax is:

root -b -q 'myMacro.C("text")' > myMacro.log

You can build a shared library with ACLiC and then use this shared library on the command line for a quicker execution (i.e. the compiled speed rather than the interpreted speed). See also “CINT the C++ Interpreter”.

root -b -q > myMacro.log

ROOT has a powerful C/C++ interpreter giving you access to all available ROOT classes, global variables, and functions via the command line. By typing C++ statements at the prompt, you can create objects, call functions, execute scripts, etc. For example:

root[] 1+sqrt(9)
(const double)4.00000000000000000e+00
root[] for (int i = 0; i<4; i++) cout << "Hello" << i << endl
Hello 0
Hello 1
Hello 2
Hello 3
root[] .q

To exit the ROOT session, type .q.

root[] .q

2.3 Using the GUI

The basic whiteboard on which an object is drawn in ROOT is called a canvas (defined by the class TCanvas). Every object in the canvas is a graphical object in the sense that you can grab it, resize it, and change some characteristics using the mouse. The canvas area can be divided in several sub areas, so-called pads (the class TPad). A pad is a canvas sub area that can contain other pads or graphical objects. At any one time, just one pad is the so-called active pad. Any object at the moment of drawing will be drawn in the active pad. The obvious question is: what is the relation between a canvas and a pad? In fact, a canvas is a pad that spans through an entire window. This is nothing else than the notion of inheritance. The TPad class is the parent of the TCanvas class. In ROOT, most objects derive from a base class TObject. This class has a virtual method Draw() such as all objects are supposed to be able to be “drawn”. If several canvases are defined, there is only one active at a time. One draws an object in the active canvas by using the statement:


This instructs the object “object” to draw itself. If no canvas is opened, a default one (named “c1”) is created. In the next example, the first statement defines a function and the second one draws it. A default canvas is created since there was no opened one. You should see the picture as shown in the next figure.

root[] TF1 f1("func1","sin(x)/x",0,10)
root[] f1.Draw()
<TCanvas::MakeDefCanvas>: created default TCanvas with name c1
A canvas with drawing

A canvas with drawing

The following components comprise the canvas window:

At the top of the canvas window are File, Edit, View, Options, Inspect, Classes and Help menus. File Menu Edit Menu

There is only one active menu entry in the Edit menu. The others menu entries will be implemented and will become active in the near future. View Menu Options Menu Inspect Menu Help Menu Classes Menu Toolbar

The following menu shortcuts and utilities are available from the toolbar:

Create a new canvas window.

Popup the Open File dialog.

Popup the Save As… dialog.

Popup the Print dialog.

Interrupts the current drawing process.

Redraw the canvas.

Inspect the gROOT object.

Create a new objects’ browser.

You can create the following graphical objects using the toolbar buttons for primitive drawing. Tool tips are provided for helping your choice.

An Arc or circle: Click on the center of the arc, and then move the mouse. A rubber band circle is shown. Click again with the left button to freeze the arc.

A Line: Click with the left button at the point where you want to start the line, then move the mouse and click again with the left button to freeze the line.

An Arrow:Click with the left button at the point where you want to start the arrow, then move the mouse and click again with the left button to freeze the arrow.

A Diamond: Click with the left button and freeze again with the left button. The editor draws a rubber band box to suggest the outline of the diamond.

An Ellipse: Proceed like for an arc. You can grow/shrink the ellipse by pointing to the sensitive points. They are highlighted. You can move the ellipse by clicking on the ellipse, but not on the sensitive points. If, with the ellipse context menu, you have selected a fill area color, you can move a filled-ellipse by pointing inside the ellipse and dragging it to its new position.

A Pad: Click with the left button and freeze again with the left button. The editor draws a rubber band box to suggest the outline of the pad.

A PaveLabel: Proceed like for a pad. Type the text of label and finish with a carriage return. The text will appear in the box.

A Pave Text: Proceed like for a pad. You can then click on the TPaveText object with the right mouse button and select the option InsertText.

Paves Text: Proceed like for a TPaveText.

A Poly Line: Click with the left button for the first point, move the moose, click again with the left button for a new point. Close the poly-line with a double click. To edit one vertex point, pick it with the left button and drag to the new point position.

A Curly Line: Proceed as for the arrow or line. Once done, click with the third button to change the characteristics of the curly line, like transform it to wave, change the wavelength, etc.

A Curly Arc: Proceed like for an ellipse. The first click is located at the position of the center, the second click at the position of the arc beginning. Once done, one obtains a curly ellipse, for which one can click with the third button to change the characteristics, like transform it to wavy, change the wavelength, set the minimum and maximum angle to make an arc that is not closed, etc.

A Text/Latex string: Click with the left button where you want to draw the text and then type in the text terminated by carriage return. All TLatex expressions are valid. To move the text or formula, point on it keeping the left mouse button pressed and drag the text to its new position. You can grow/shrink the text if you position the mouse to the first top-third part of the string, then move the mouse up or down to grow or shrink the text respectively. If you position the mouse near the bottom-end of the text, you can rotate it.

A Marker: Click with the left button where to place the marker. The marker can be modified by using the method SetMarkerStyle() of TSystem.

A Graphical Cut: Click with the left button on each point of a polygon delimiting the selected area. Close the cut by double clicking on the last point. A TCutG object is created. It can be used as a selection for a TTree::Draw. You can get a pointer to this object with:

TCutG cut = (TCutG*)gPad->GetPrimitive("CUTG")

Once you are happy with your picture, you can select the Save as canvas.C item in the canvas File menu. This will automatically generate a script with the C++ statements corresponding to the picture. This facility also works if you have other objects not drawn with the graphics editor (histograms for example).

2.3.2 The Editor Frame

The ROOT graphics editor loads the corresponding object editor objEditor according to the selected object obj in the canvas respecting the class inheritance. An object in the canvas is selected after the left mouse click on it. For example, if the selected object is TAxis, the TAxisEditor will shows up in the editor frame giving the possibility for changing different axis attributes. The graphics editor can be:

Embedded - connected only with the canvas in the application window that appears on the left of the canvas window after been activated via View menu / Editor. It appears on the left side if the canvas window allowing users to edit the attributes of the selected object via provided user interface. The name of the selected object is displayed on the top of the editor frame in red color. If the user interface needs more space then the height of the canvas window, a vertical scroll bar appears for easer navigation.

Global - has own application window and can be connected to any created canvas in a ROOT session. It can be activated via the context menu entries for setting line, fill, text and marker attributes for backward compatibility, but there will be a unique entry in the near future.

The user interface for the following classes is available since ROOT v.4.04: TAttLine, TAttFill, TAttMarker, TAttText, TArrow, TAxis, TCurlyArc, TCurlyLine, TFrame, TH1, TH2, TGraph, TPad, TCanvas, TPaveStats. For more details, see “The Graphics Editor”, “The User Interface for Histograms”, “The User Interface for Graphs”.

2.3.3 Classes, Methods and Constructors

Object oriented programming introduces objects, which have data members and methods. The next line creates an object named f1 of the class TF1 that is a one-dimensional function. The type of an object is called a class. The object itself is called an instance of a class. When a method builds an object, it is called a constructor.

TF1 f1("func1","sin(x)/x",0,10)

In our constructor the function sin(x)/x is defined for use, and 0 and 10 are the limits. The first parameter, func1 is the name of the object f1. Most objects in ROOT have a name. ROOT maintains a list of objects that can be searched to find any object by its given name (in our example func1).

The syntax to call an object’s method, or if one prefers, to make an object to do something is:


The dot can be replaced by “->” if object is a pointer. In compiled code, the dot MUST be replaced by a “->” if object is a pointer.


So now, we understand the two lines of code that allowed us to draw our function. f1.Draw() stands for “call the method Draw() associated with the object f1 of the class TF1”. Other methods can be applied to the object f1 of the class TF1. For example, the evaluating and calculating the derivative and the integral are what one would expect from a function.

root[] f1.Eval(3)
root[] f1.Derivative(3)
root[] f1.Integral(0,3)
root[] f1.Draw()

By default the method TF1::Paint(), that draws the function, computes 100 equidistant points to draw it. The number of points can be set to a higher value with:

root[] f1.SetNpx(2000);

Note that while the ROOT framework is an object-oriented framework, this does not prevent the user from calling plain functions.

2.3.4 User Interaction

Now we will look at some interactive capabilities. Try to draw the function sin(x)/x again. Every object in a window (which is called a canvas) is, in fact, a graphical object in the sense that you can grab it, resize it, and change its characteristics with a mouse click. For example, bring the cursor over the x-axis. The cursor changes to a hand with a pointing finger when it is over the axis. Now, left click and drag the mouse along the axis to the right. You have a very simple zoom.

When you move the mouse over any object, you can get access to selected methods by pressing the right mouse button and obtaining a context menu. If you try this on the function TF1, you will get a menu showing available methods. The other objects on this canvas are the title, a TPaveText object; the x and y-axis, TAxis objects, the frame, a TFrame object, and the canvas a TCanvas object. Try clicking on these and observe the context menu with their methods.

A context menu

A context menu

For example try selecting the SetRange() method and putting -10, 10 in the dialog box fields. This is equivalent to executing f1.SetRange(-10,10) from the command line, followed by f1.Draw(). Here are some other options you can try.

Once the picture suits your wishes, you may want to see the code you should put in a script to obtain the same result. To do that, choose Save / canvas.C entry of the File menu. This will generate a script showing the options set in the current canvas. Notice that you can also save the picture into various file formats such as PostScript, GIF, etc. Another interesting possibility is to save your canvas into the native ROOT format (.rootfile). This will enable you to open it again and to change whatever you like. All objects associated to the canvas (histograms, graphs) are saved at the same time.

2.3.5 Building a Multi-pad Canvas

Let us now try to build a canvas with several pads.

root[] TCanvas *MyC = new TCanvas("MyC","Test canvas",1)
root[] MyC->Divide(2,2)

Once again, we call the constructor of a class, this time the class TCanvas. The difference between this and the previous constructor call (TF1) is that here we are creating a pointer to an object. Next, we call the method Divide() of the TCanvas class (that is TCanvas::Divide()), which divides the canvas into four zones and sets up a pad in each of them. We set the first pad as the active one and than draw the functionf1there.

root[] MyC->cd(1)
root[] f1->Draw()

All objects will be drawn in that pad because it is the active one. The ways for changing the active pad are:

root[] MyC->cd(3)

Pads are numbered from left to right and from top to bottom. Each new pad created by TCanvas::Divide() has a name, which is the name of the canvas followed by _1, _2, etc. To apply the method cd() to the third pad, you would write:

root[] MyC_3->cd()

2.3.6 Saving the Canvas

The SaveAs… dialog

The SaveAs… dialog

Using the File menu / Save cascade menu users can save the canvas as one of the files from the list. Please note that saving the canvas this way will overwrite the file with the same name without a warning.

All supported file types can be saved via File menu / SaveAs… This dialog gives a choice to show or suppress the confirmation message for overwriting an existing file.

If the Ovewrite check box is not selected, a message dialog appears asking the user to overwrite the file (Yes/No). The user choice is saved for the next time the Save As… dialog shows up.

2.3.7 Printing the Canvas

The Print command in the canvas File menu pops-up a print dialog where the user can specify a preferred print command and the printer name.

Both print parameters can be set via the new Print.Command and Print.Printer rootrc resources as follows:

# Printer settings.
WinNT.*.Print.Command:    AcroRd32.exe
Unix.*.Print.Command:     xprint -P%p %f
Print.Printer:            32-rb205-hp
Print.Directory:          .

If the %p and %f are specified as a part of the print command, they will be replaced by the specified printer name and the file name. All other parameters will be kept as they are written. A print button is available in the canvas toolbar (activated via View menu/Toolbar).

2.4 The ROOT Command Line

We have briefly touched on how to use the command line. There are different types of commands.

  1. CINT commands start with “.
root[] .?  //this command will list all the CINT commands
root[] .L <filename>  //load [filename]
root[] .x <filename>  //load and execute [filename]
  1. SHELL commands start with “.!” for example:
root[] .! ls
  1. C++ commands follow C++ syntax (almost)
root[] TBrowser *b = new TBrowser()

2.4.1 Multi-line Commands

You can use the command line to execute multi-line commands. To begin a multi-line command you must type a single left curly bracket {, and to end it you must type a single right curly bracket }. For example:

root[] {
end with '}'> Int_t j = 0;
end with '}'> for (Int_t i = 0; i < 3; i++)
end with '}'> {
end with '}'> j= j + i;
end with '}'> cout << "i = " << i << ", j = " << j << endl;
end with '}'> }
end with '}'> }
i = 0, j = 0
i = 1, j = 1
i = 2, j = 3

It is more convenient to edit a script than the command line, and if your multi line commands are getting unmanageable, you may want to start with a script instead.

2.4.2 CINT Extensions

We should say that some things are not standard C++. The CINT interpreter has several extensions. See “ROOT/CINT Extensions to C++”.

2.4.3 Helpful Hints for Command Line Typing

The interpreter knows all the classes, functions, variables, and user defined types. This enables ROOT to help users to complete the command line. For example, if we do not know anything about the TLine class, the Tab feature helps us to get a list of all classes starting with TL(where <TAB> means type the Tab key).

root[] l = new TLi<TAB>

To list the different constructors and parameters for TLine use the <TAB> key as follows:

root[] l = new TLine(<TAB>
TLine TLine()
TLine TLine(Double_t x1,Double_t y1,Double_t x2,Double_t y2)
TLine TLine(const TLine& line)

2.4.4 Regular Expression

The meta-characters below can be used in a regular expression:

When using wildcards the regular expression is assumed to be preceded by a ‘^’ (BOL) and terminated by ‘$’ (EOL). All ‘*’ (closures) are assumed to be preceded by a ‘.’, i.e. any character, except slash _/_. Its special treatment allows the easy matching of pathnames. For example, _*.root_ will match _aap.root_, but not _pipo/aap.root_.

The escape characters are:

The class TRegexp can be used to create a regular expression from an input string. If wildcard is true then the input string contains a wildcard expression.

TRegexp(const char *re, Bool_t wildcard)

Regular expression and wildcards can be easily used in methods like:

Ssiz_t Index(const TString& string,Ssiz_t* len,Ssiz_t i) const

The method finds the first occurrence of the regular expression in the string and returns its position.

2.5 Conventions

In this paragraph, we will explain some of the conventions used in ROOT source and examples.

2.5.1 Coding Conventions

From the first days of ROOT development, it was decided to use a set of coding conventions. This allows a consistency throughout the source code. Learning these will help you identify what type of information you are dealing with and enable you to understand the code better and quicker. Of course, you can use whatever convention you want but if you are going to submit some code for inclusion into the ROOT sources, you will need to use these.

These are the coding conventions:

2.5.2 Machine Independent Types

Different machines may have different lengths for the same type. The most famous example is the int type. It may be 16 bits on some old machines and 32 bits on some newer ones. To ensure the size of your variables, use these pre defined types in ROOT:

If you do not want to save a variable on disk, you can use int or Int_t, the result will be the same and the interpreter or the compiler will treat them in exactly the same way.

2.5.3 TObject

In ROOT, almost all classes inherit from a common base class called TObject. This kind of architecture is also used in the Java language. The TObject class provides default behavior and protocol for all objects in the ROOT system. The main advantage of this approach is that it enforces the common behavior of the derived classes and consequently it ensures the consistency of the whole system. See “The Role of TObject”.

TObject provides protocol, i.e. (abstract) member functions, for:

2.6 Global Variables

ROOT has a set of global variables that apply to the session. For example, gDirectory always holds the current directory, and gStyle holds the current style.

All global variables begin with “g” followed by a capital letter.

2.6.1 gROOT

The single instance of TROOT is accessible via the global gROOT and holds information relative to the current session. By using the gROOT pointer, you get the access to every object created in a ROOT program. The TROOT object has several lists pointing to the main ROOT objects. During a ROOT session, the gROOT keeps a series of collections to manage objects. They can be accessed via gROOT::GetListOf... methods.


These methods return a TSeqCollection, meaning a collection of objects, and they can be used to do list operations such as finding an object, or traversing the list and calling a method for each of the members. See the TCollection class description for the full set of methods supported for a collection. For example, to find a canvas called c1you can do:

root[] gROOT->GetListOfCanvases()->FindObject("c1")

This returns a pointer to a TObject, and before you can use it as a canvas you need to cast it to a TCanvas*.

2.6.2 gFile

gFile is the pointer to the current opened file in the ROOT session.

2.6.3 gDirectory

gDirectory is a pointer to the current directory. The concept and role of a directory is explained in the chapter “Input/Output”.

2.6.4 gPad

A graphic object is always drawn on the active pad. It is convenient to access the active pad, no matter what it is. For that, we have gPad that is always pointing to the active pad. For example, if you want to change the fill color of the active pad to blue, but you do not know its name, you can use gPad.

root[] gPad->SetFillColor(38)

To get the list of colors, if you have an open canvas, click in the “View” menu, selecting the “Colors” entry.

2.6.5 gRandom

gRandom is a pointer to the current random number generator. By default, it points to a TRandom3 object, based on the “Mersenne-Twister” generator. This generator is very fast and has very good random proprieties (a very long period of 10600). Setting the seed to 0 implies that the seed will be uniquely generated using the TUUID. Any other value will be used as a constant. The following basic random distributions are provided: Rndm() or Uniform(min,max), Gaus(mean,sigma), Exp(tau), BreitWigner(mean,sigma), Landau(mean,sigma), Poisson(mean), Binomial(ntot,prob). You can customize your ROOT session by replacing the random number generator. You can delete gRandom and recreate it with your own. For example:

root[] delete gRandom;
root[] gRandom = new TRandom2(0); //seed=0

TRandom2 is another generator, which is also very fast and uses only three words for its state.

2.6.6 gEnv

gEnv is the global variable (of type TEnv) with all the environment settings for the current session. This variable is set by reading the contents of a .rootrc file (or $ROOTSYS/etc/system.rootrc) at the beginning of the root session. See Environment Setup below for more information.

2.7 Environment Setup

The behavior of a ROOT session can be tailored with the options in the .rootrc file. At start-up, ROOT looks for a .rootrc file in the following order:

If more than one .rootrc files are found in the search paths above, the options are merged, with precedence local, user, global. While in a session, to see current settings, you can do:

root[] gEnv->Print()

The rootrc file typically looks like:

# Path used by dynamic loader to find shared libraries
Unix.*.Root.DynamicPath:  .:~/rootlibs:$(ROOTSYS)/lib
Unix.*.Root.MacroPath:    .:~/rootmacros:$(ROOTSYS)/macros

# Path where to look for TrueType fonts
Unix.*.Root.UseTTFonts:     true
# Activate memory statistics
Rint.Root.MemStat:       1
Rint.Load:               rootalias.C
Rint.Logon:              rootlogon.C
Rint.Logoff:             rootlogoff.C
Rint.Canvas.MoveOpaque:  false
Rint.Canvas.HighLightColor: 5

The various options are explained in $ROOTSYS/etc/system.rootrc. The .rootrc file contents are combined. For example, if the flag to use true type fonts is set to true in the system.rootrc file, you have to set explicitly it false in your local .rootrc file if you do not want to use true type fonts. Removing the UseTTFontsstatement in the local .rootrc file will not disable true fonts. The value of the environment variable ROOTDEBUG overrides the value in the .rootrc file at startup. Its value is used to set gDebug and helps for quick turn on debug mode in TROOT startup.

ROOT looks for scripts in the path specified in the .rootrc file in the Root.Macro.Path variable. You can expand this path to hold your own directories.

2.7.1 Logon and Logoff Scripts

The rootlogon.C and rootlogoff.C files are scripts loaded and executed at start-up and shutdown. The rootalias.C file is loaded but not executed. It typically contains small utility functions. For example, the rootalias.C script that comes with the ROOT distributions (located in $ROOTSYS/tutorials) defines the function edit(char *file). This allows the user to call the editor from the command line. This particular function will start the VI editor if the environment variable EDITOR is not set.

root[0] edit("c1.C")

For more details, see $ROOTSYS/tutorials/rootalias.C.

2.7.2 History File

You can use the up and down arrow at the command line, to access the previous and next command. The commands are recorded in the history file $HOME/.root_hist. It is a text file, and you can edit, cut, and paste from it. You can specify the history file in the system.rootrc file, by setting the Rint.Historyoption. You can also turn off the command logging in the system.rootrc file with the option: Rint.History: -

The number of history lines to be kept can be set also in .rootrc by:

Rint.HistSize:         500
Rint.HistSave:         400

The first value defines the maximum of lines kept; once it is reached all, the last HistSave lines will be removed. One can set HistSize to 0 to disable history line management. There is also implemented an environment variable called ROOT_HIST. By setting ROOT_HIST=300:200 the above values can be overriden - the first value corresponds to HistSize, the (optional) second one to HistSave. You can set ROOT_HIST=0 to disable the history.

2.7.3 Tracking Memory Leaks

You can track memory usage and detect leaks by monitoring the number of objects that are created and deleted (see TObjectTable). To use this facility, edit the file $ROOTSYS/etc/system.rootrc or .rootrc if you have this file and add the two following lines:

Root.MemStat:            1
Root.ObjectStat:         1

In your code or on the command line you can type the line:


This line will print the list of all active classes and the number of instances for each class. By comparing consecutive print outs, you can see objects that you forgot to delete. Note that this method cannot show leaks coming from the allocation of non-objects or classes unknown to ROOT.

2.7.4 Memory Checker

A memory checking system was developed by D.Bertini and M.Ivanov and added in ROOT version 3.02.07. To activate the memory checker you can set the resource Root.MemCheck to 1 (e.g.: Root.MemCheck: 1 in the .rootrc file). You also have to link with (e.g. use root-config --new --libs) or to use rootn.exe. When these settings are in place, you will find a file “memcheck.out” in the directory where you started your ROOT program after the completion of the program execution. You can also set the resource Root.MemCheckFile to the name of a file. The memory information will be written to that file. The contents of this memcheck.out can be analyzed and transformed into printable text via the memprobe program (in $ROOTSYS/bin).

2.8 Converting from PAW to ROOT

The web page at: gives the “translation” table of some commonly used PAW commands into ROOT. If you move the mouse cursor over the picture at:, you will get the corresponding ROOT commands as tooltips.

2.8.1 Converting HBOOK/PAW Files

ROOT has a utility called h2root that you can use to convert your HBOOK/PAW histograms or ntuple files into ROOT files. To use this program, you type the shell script command:

h2root  <hbookfile>  <rootfile>

If you do not specify the second parameter, a file name is automatically generated for you. If hbookfile is of the form file.hbook, then the ROOT file will be called file.root. This utility converts HBOOK histograms into ROOT histograms of the class TH1F. HBOOK profile histograms are converted into ROOT profile histograms (see class TProfile). HBOOK row-wise and column-wise ntuples are automatically converted to ROOT Trees. See “Trees”. Some HBOOK column-wise ntuples may not be fully converted if the columns are an array of fixed dimension (e.g. var[6]) or if they are a multi-dimensional array.

HBOOK integer identifiers are converted into ROOT named objects by prefixing the integer identifier with the letter “h” if the identifier is a positive integer and by "h_" if it is a negative integer identifier. In case of row-wise or column-wise ntuples, each column is converted to a branch of a tree. Note that h2root is able to convert HBOOK files containing several levels of sub-directories. Once you have converted your file, you can look at it and draw histograms or process ntuples using the ROOT command line. An example of session is shown below:

// this connects the file hbookconverted.root
root[] TFile f("hbookconverted.root");

// display histogram named h10 (was HBBOK id 10)
root[] h10.Draw();

// display column "var" from ntuple h30
root[] h30.Draw("var");

You can also use the ROOT browser (see TBrowser) to inspect this file.

The chapter on trees explains how to read a tree. ROOT includes a function TTree::MakeClass to generate automatically the code for a skeleton analysis function. See “Example Analysis”.

In case one of the ntuple columns has a variable length (e.g. px(ntrack)), h.Draw("px") will histogram the px column for all tracks in the same histogram. Use the script quoted above to generate the skeleton function and create/fill the relevant histogram yourself.

3 Histograms

This chapter covers the functionality of the histogram classes. We begin with an overview of the histogram classes, after which we provide instructions and examples on the histogram features.

We have put this chapter ahead of the graphics chapter so that you can begin working with histograms as soon as possible. Some of the examples have graphics commands that may look unfamiliar to you. These are covered in the chapter “Input/Output”.

3.1 The Histogram Classes

ROOT supports histograms up to three dimensions. Separate concrete classes are provided for one-dimensional, two-dimensional and three-dimensional classes. The histogram classes are split into further categories, depending on the set of possible bin values:

ROOT also supports profile histograms, which constitute an elegant replacement of two-dimensional histograms in many cases. The inter-relation of two measured quantities X and Y can always be visualized with a two-dimensional histogram or scatter-plot. Profile histograms, on the other hand, are used to display the mean value of Y and its RMS for each bin in X. If Y is an unknown but single-valued approximate function of X, it will have greater precision in a profile histogram than in a scatter plot.

The class hierarchy of histogram classes

The class hierarchy of histogram classes

All ROOT histogram classes are derived from the base class TH1 (see figure above). This means that two-dimensional and three-dimensional histograms are seen as a type of a one-dimensional histogram, in the same way in which multidimensional C arrays are just an abstraction of a one-dimensional contiguous block of memory.

3.2 Creating Histograms

There are several ways in which you can create a histogram object in ROOT. The straightforward method is to use one of the several constructors provided for each concrete class in the histogram hierarchy. For more details on the constructor parameters, see the subsection “Constant or Variable Bin Width” below. Histograms may also be created by:

   // using various constructors
   TH1* h1 = new TH1I("h1", "h1 title", 100, 0.0, 4.0);
   TH2* h2 = new TH2F("h2", "h2 title", 40, 0.0, 2.0, 30, -1.5, 3.5);
   TH3* h3 = new TH3D("h3", "h3 title", 80, 0.0, 1.0, 100, -2.0, 2.0,
                       50, 0.0, 3.0);

   // cloning a histogram
   TH1* hc = (TH1*)h1->Clone();

   // projecting histograms
   // the projections always contain double values !
   TH1* hx = h2->ProjectionX(); // ! TH1D, not TH1F
   TH1* hy = h2->ProjectionY(); // ! TH1D, not TH1F

3.2.1 Constant or Variable Bin Width

The histogram classes provide a variety of ways to construct a histogram, but the most common way is to provide the name and title of histogram and for each dimension: the number of bins, the minimum x (lower edge of the first bin) and the maximum x (upper edge of the last bin).

   TH2* h = new TH2D(
      /* name */ "h2",
      /* title */ "Hist with constant bin width",
      /* X-dimension */ 100, 0.0, 4.0,
      /* Y-dimension */ 200, -3.0, 1.5);

When employing his constructor, you will create a histogram with constant (fixed) bin width on each axis. For the example above, the interval [0.0, 4.0] is divided into 100 bins of the same width w X = 4.0 - 0.0 100 = 0.04 for the X axis (dimension). Likewise, for the Y axis (dimension), we have bins of equal width w Y = 1.5 - (-3.0) 200 = 0.0225.

If you want to create histograms with variable bin widths, ROOT provides another constructor suited for this purpose. Instead of passing the data interval and the number of bins, you have to pass an array (single or double precision) of bin edges. When the histogram has n bins, then there are n+1 distinct edges, so the array you pass must be of size n+1.

   const Int_t NBINS = 5;
   Double_t edges[NBINS + 1] = {0.0, 0.2, 0.3, 0.6, 0.8, 1.0};
   // Bin 1 corresponds to range [0.0, 0.2]
   // Bin 2 corresponds to range [0.2, 0.3] etc...

   TH1* h = new TH1D(
      /* name */ "h1",
      /* title */ "Hist with variable bin width",
      /* number of bins */ NBINS,
      /* edge array */ edges

Each histogram object contains three TAxis objects: fXaxis , fYaxis, and fZaxis, but for one-dimensional histograms only the X-axis is relevant, while for two-dimensional histograms the X-axis and Y-axis are relevant. See the class TAxis for a description of all the access methods. The bin edges are always stored internally in double precision.

You can examine the actual edges / limits of the histogram bins by accessing the axis parameters, like in the example below:

   const Int_t XBINS = 5; const Int_t YBINS = 5;
   Double_t xEdges[XBINS + 1] = {0.0, 0.2, 0.3, 0.6, 0.8, 1.0};
   Double_t yEdges[YBINS + 1] = {-1.0, -0.4, -0.2, 0.5, 0.7, 1.0};

   TH2* h = new TH2D("h2", "h2", XBINS, xEdges, YBINS, yEdges);
   TAxis* xAxis = h->GetXaxis(); TAxis* yAxis = h->GetYaxis();

   cout << "Third bin on Y-dimension: " << endl; // corresponds to
                                                 // [-0.2, 0.5]
   cout << "\tLower edge: " << yAxis->GetBinLowEdge(3) << endl;
   cout << "\tCenter: " << yAxis->GetBinCenter(3) << endl;
   cout << "\tUpper edge: " << yAxis->GetBinUpEdge(3) << endl;

3.3 Bin Numbering

All histogram types support fixed or variable bin sizes. 2-D histograms may have fixed size bins along X and variable size bins along Y or vice-versa. The functions to fill, manipulate, draw, or access histograms are identical in both cases.

3.3.1 Convention

For all histogram types: nbins , xlow , xup

Bin# 0 contains the underflow.

Bin# 1 contains the first bin with low-edge ( xlow INCLUDED).

The second to last bin (bin# nbins) contains the upper-edge (xup EXCLUDED).

The Last bin (bin# nbins+1) contains the overflow.

In case of 2-D or 3-D histograms, a “global bin” number is defined. For example, assuming a 3-D histogram h with binx, biny, binz, the function returns a global/linear bin number.

   Int_t bin = h->GetBin(binx, biny, binz);

This global bin is useful to access the bin information independently of the dimension.

3.3.2 Re-binning

At any time, a histogram can be re-binned via the TH1::Rebin() method. It returns a new histogram with the re-binned contents. If bin errors were stored, they are recomputed during the re-binning.

3.4 Filling Histograms

A histogram is typically filled with statements like:

   h1->Fill(x,w); // with weight

The Fill method computes the bin number corresponding to the given x, y or z argument and increments this bin by the given weight. The Fill() method returns the bin number for 1-D histograms or global bin number for 2-D and 3-D histograms. If TH1::Sumw2() has been called before filling, the sum of squares is also stored. One can increment a bin number directly by calling TH1::AddBinContent(), replace the existing content via TH1::SetBinContent() , and access the bin content of a given bin via TH1::GetBinContent() .

   Double_t binContent = h->GetBinContent(bin);

3.4.1 Automatic Re-binning Option

By default, the number of bins is computed using the range of the axis. You can change this to re-bin automatically by setting the automatic re-binning option:


Once this is set, the Fill() method will automatically extend the axis range to accommodate the new value specified in the Fill() argument. The used method is to double the bin size until the new value fits in the range, merging bins two by two. The TTree::Draw() method extensively uses this automatic binning option when drawing histograms of variables in TTree with an unknown range. The automatic binning option is supported for 1-D, 2-D and 3-D histograms. During filling, some statistics parameters are incremented to compute the mean value and root mean square with the maximum precision. In case of histograms of type TH1C, TH1S, TH2C, TH2S, TH3C, TH3S a check is made that the bin contents do not exceed the maximum positive capacity (127 or 65 535). Histograms of all types may have positive or/and negative bin contents.

3.5 Random Numbers and Histograms

TH1::FillRandom() can be used to randomly fill a histogram using the contents of an existing TF1 function or another TH1 histogram (for all dimensions). For example, the following two statements create and fill a histogram 10 000 times with a default Gaussian distribution of mean 0 and sigma 1 :

root[] TH1F h1("h1","Histo from a Gaussian",100,-3,3);
root[] h1.FillRandom("gaus",10000);

TH1::GetRandom() can be used to get a random number distributed according the contents of a histogram. To fill a histogram following the distribution in an existing histogram you can use the second signature of TH1::FillRandom(). Next code snipped assumes that h is an existing histogram (TH1 ).

root[] TH1F h2("h2","Histo from existing histo",100,-3,3);
root[] h2.FillRandom(&h1, 1000);

The distribution contained in the histogram h1 ( TH1 ) is integrated over the channel contents. It is normalized to one. The second parameter (1000) indicates how many random numbers are generated.

Getting 1 random number implies:

You can see below an example of the TH1::GetRandom() method which can be used to get a random number distributed according the contents of a histogram.

void getrandomh() {
   TH1F *source = new TH1F("source","source hist",100,-3,3);
   TH1F *final = new TH1F("final","final hist",100,-3,3);

             // continued...

   for (Int_t i=0;i<10000;i++) {
   TCanvas *c1 = new TCanvas("c1","c1",800,1000);

3.6 Adding, Dividing, and Multiplying

Many types of operations are supported on histograms or between histograms:

Histograms objects (not pointers) TH1F h1 can be multiplied by a constant using:


A new histogram can be created without changing the original one by doing:

   TH1F h3 = 8*h1;

To multiply two histogram objects and put the result in a 3rd one do:

   TH1F h3 = h1*h2;

The same operations can be done with histogram pointers TH1F *h1, *h2 following way:

   h1->Scale(const) TH1F h3 = 8*(*h1); TH1F h3 = (*h1)*(*h2);

Of course, the TH1 methods Add , Multiply and Divide can be used instead of these operators.

If a histogram has associated error bars ( TH1::Sumw2() has been called), the resulting error bars are also computed assuming independent histograms. In case of divisions, binomial errors are also supported.

3.7 Projections

One can make:

These projections can be fit via: TH2::FitSlicesX, TH2::FitSlicesY, TH3::FitSlicesZ.

3.8 Drawing Histograms

When you call the Draw method of a histogram ( TH1::Draw ) for the first time, it creates a THistPainter object and saves a pointer to painter as a data member of the histogram. The THistPainter class specializes in the drawing of histograms. It allows logarithmic axes (x, y, z) when the CONT drawing option is using. The THistPainter class is separated from the histogram so that one can have histograms without the graphics overhead, for example in a batch program. The choice to give each histogram has its own painter rather than a central singleton painter, allows two histograms to be drawn in two threads without overwriting the painter’s values. When a displayed histogram is filled again, you do not have to call the Draw method again. The image is refreshed the next time the pad is updated. A pad is updated after one of these three actions:

By default, the TH1::Draw clears the pad before drawing the new image of the histogram. You can use the "SAME" option to leave thevprevious display in tact and superimpose the new histogram. The same histogram can be drawn with different graphics options in different pads. When a displayed histogram is deleted, its image is automatically removed from the pad. To create a copy of the histogram when drawing it, you can use TH1::DrawClone(). This will clone the histogram and allow you to change and delete the original one without affecting the clone. You can use TH1::DrawNormalized() to draw a normalized copy of a histogram.

TH1 *TH1::DrawNormalized(Option_t *option,Double_t norm) const

A clone of this histogram is normalized to norm and drawn with option. A pointer to the normalized histogram is returned. The contents of the histogram copy are scaled such that the new sum of weights (excluding under and overflow) is equal to norm .

Note that the returned normalized histogram is not added to the list of histograms in the current directory in memory. It is the user’s responsibility to delete this histogram. The kCanDelete bit is set for the returned object. If a pad containing this copy is cleared, the histogram will be automatically deleted. See “Draw Options” for the list of options.

3.8.1 Setting the Style

Histograms use the current style gStyle, which is the global object of class TStyle. To change the current style for histograms, the TStyle class provides a multitude of methods ranging from setting the fill color to the axis tick marks. Here are a few examples:

   void SetHistFillColor(Color_t color = 1)
   void SetHistFillStyle(Style_t styl = 0)
   void SetHistLineColor(Color_t color = 1)
   void SetHistLineStyle(Style_t styl = 0)
   void SetHistLineWidth(Width_t width = 1)

When you change the current style and would like to propagate the change to a previously created histogram you can call TH1::UseCurrentStyle(). You will need to call UseCurrentStyle() on each histogram. When reading many histograms from a file and you wish to update them to the current style, you can use gROOT::ForceStyle and all histograms read after this call will be updated to use the current style. See “Graphics and the Graphical User Interface”. When a histogram is automatically created as a result of a TTree::Draw , the style of the histogram is inherited from the tree attributes and the current style is ignored. The tree attributes are the ones set in the current TStyle at the time the tree was created. You can change the existing tree to use the current style, by calling TTree::UseCurrentStyle() .

3.8.2 Draw Options

The following draw options are supported on all histogram classes:

The following options are supported for 1-D histogram classes:

The following options are supported for 2-D histogram classes:

The following options are supported for 3-D histogram classes:

Most options can be concatenated without spaces or commas, for example, if h is a histogram pointer:


The options are not case sensitive. The options BOX , COL and COLZ use the color palette defined in the current style (see TStyle::SetPalette). The options CONT , SURF , and LEGO have by default 20 equidistant contour levels, you can change the number of levels with TH1::SetContour. You can also set the default drawing option with TH1::SetOption . To see the current option use TH1::GetOption . For example:

   h->Draw(); // will use the lego option
   h->Draw("scat") // will use the scatter plot option The SCATter Plot Option

By default, 2D histograms are drawn as scatter plots. For each cell (i,j) a number of points proportional to the cell content are drawn. A maximum of 500 points per cell are drawn. If the maximum is above 500 contents are normalized to 500. The ARRow Option

The ARR option shows the gradient between adjacent cells. For each cell (i,j) an arrow is drawn. The orientation of the arrow follows the cell gradient. The BOX Option

For each cell (i,j) a box is drawn with surface proportional to contents. The size of the box is proportional to the absolute value of the cell contents. The cells with negative contents are drawn with an X on top of the boxes. With option BOX1 a button is drawn for each cell with surface proportional to contents’ absolute value. A sunken button is drawn for negative values, a raised one for positive values. The ERRor Bars Options

The E1 bars’ option

The “E1” bars’ option

Note that for all options, the line and fill attributes of the histogram are used for the errors or errors contours. Use gStyle->SetErrorX(dx) to control the size of the error along x. The parameter dx is a percentage of bin width for errors along X. Set dx=0 to suppress the error along X. Use gStyle->SetEndErrorSize(np) to control the size of the lines at the end of the error bars (when option 1 is used). By default np=1 (np represents the number of pixels). The Color Option

For each cell (i,j) a box is drawn with a color proportional to the cell content. The color table used is defined in the current style (gStyle ). The color palette in TStyle can be modified with TStyle::SetPalette .

Different draw options

Different draw options The TEXT Option

For each cell (i,j) the cell content is printed. The text attributes are:

The TEXT option

The TEXT option The CONTour Options

The following contour options are supported:

Different contour options

Different contour options

The default number of contour levels is 20 equidistant levels. It can be changed with TH1::SetContour. When option “LIST” is specified together with option “CONT”, all points used for contour drawing, are saved in the TGraph object and are accessible in the following way:

   TObjArray *contours =
   Int_t ncontours = contours->GetSize(); TList *list =

Where “i” is a contour number and list contains a list of TGraph objects. For one given contour, more than one disjoint poly-line may be generated. The TGraph numbers per contour are given by list->GetSize(). Here we show how to access the first graph in the list.

   TGraph *gr1 = (TGraph*)list->First();

The tutorial macro earth.C uses these four options and produces the following picture:

The earth.C macro output

The earth.C macro output The LEGO Options

In a lego plot, the cell contents are drawn as 3D boxes, with the height of the box proportional to the cell content.

LEGO and SURF options

“LEGO” and “SURF” options

A lego plot can be represented in several coordinate systems; the default system is Cartesian coordinates. Other possible coordinate systems are CYL , POL , SPH , and PSR .

With TStyle::SetPalette the color palette can be changed. We suggest you use palette 1 with the call:

   gStyle->SetPalette(1); The SURFace Options

In a surface plot, cell contents are represented as a mesh. The height of the mesh is proportional to the cell content. A surface plot can be represented in several coordinate systems. The default is Cartesian coordinates, and the other possible systems are CYL, POL, SPH, and PSR . The following picture uses SURF1 . With TStyle::SetPalette the color palette can be changed. We suggest you use palette 1 with the call:

Different surface options

Different surface options The BAR Options

When the option “bar” or “hbar” is specified, a bar chart is drawn.

The options for vertical bar chart are “bar”, “bar0”, “bar1”, “bar2”, “bar3”, “bar4”.

Vertical bar charts

Vertical bar charts

Use TH1::SetBarWidth() to control the bar width (default is the bin width). Use TH1::SetBarOffset to control the bar offset (default is 0). See the example $ROOTSYS/tutorials/hist/hbars.C

The options for the horizontal bar chart are “hbar”, “hbar0”, “hbar1”, “hbar2”, “hbar3”, and “hbar4”.

Use TH1::SetBarWidth to control the bar width (default is the bin width). Use TH1::SetBarOffset to control the bar offset (default is 0). See the example $ROOTSYS/tutorials/hist/hbars.C

Horizontal bar charts

Horizontal bar charts The Z Option: Display the Color Palette on the Pad

The “Z” option can be specified with the options: COL, CONT, SURF, and LEGO to display the color palette with an axis indicating the value of the corresponding color on the right side of the picture. If there is not enough space on the right side, you can increase the size of the right margin by calling TPad::SetRightMargin(). The attributes used to display the palette axis values are taken from the Z axis of the object. For example, you can set the labels size on the palette axis with:

   hist->GetZaxis()->SetLabelSize(); Setting the Color Palette

You can set the color palette with TStyle::SetPalette , e.g.


For example, the option COL draws a 2-D histogram with cells represented by a box filled with a color index, which is a function of the cell content. If the cell content is N, the color index used will be the color number in colors[N] . If the maximum cell content is greater than ncolors , all cell contents are scaled to ncolors. If ncolors<=0, a default palette of 50 colors is defined. This palette is recommended for pads, labels. It defines:

The color numbers specified in this palette can be viewed by selecting the menu entry Colors in the View menu of the canvas menu bar. The color’s red, green, and blue values can be changed via TColor::SetRGB.

If ncolors == 1 && colors == 0, then a Pretty Palette with a spectrum violet to red is created with 50 colors. That’s the default rain bow palette.

Other prefined palettes with 255 colors are available when colors == 0. The following value of ncolors (with colors = 0) give access to:

The color numbers specified in the palette can be viewed by selecting the item “colors” in the “VIEW” menu of the canvas toolbar. The color parameters can be changed via TColor::SetRGB.

Note that when drawing a 2D histogram h2 with the option “COL” or “COLZ” or with any “CONT” options using the color map, the number of colors used is defined by the number of contours n specified with: h2->SetContour(n) TPaletteAxis

A TPaletteAxisobject is used to display the color palette when drawing 2D histograms. The object is automatically created when drawing a 2D histogram when the option “z” is specified. It is added to the histogram list of functions. It can be retrieved and its attributes can be changed with:

   TPaletteAxis *palette=(TPaletteAxis*)h->FindObject("palette");

The palette can be interactively moved and resized. The context menu can be used to set the axis attributes. It is possible to select a range on the axis, to set the min/max in z. The SPEC Option

The “SPEC” option offers a large set of options/attributes to visualize 2D histograms thanks to “operators” following the “SPEC” keyword. For example, to draw the 2-D histogram h2 using all default attributes except the viewing angles, one can do:

   h2->Draw("SPEC a(30,30,0)");

The operators’ names are case unsensitive (i.e. one can use “a” or “A”) and their parameters are seperated by coma “,”. Operators can be put in any order in the option and must be separated by a space " “. No space characters should be put in an operator. All the available operators are described below.

The way how a 2D histogram will be painted is controled by two parameters: the “Display modes groups” and the “Display Modes”. “Display modes groups” can take the following values:

“Display modes” can take the following values:

These parameters can be set by using the “dm” operator in the option.

   h2->Draw("SPEC dm(1,2)");

The above example draws the histogram using the “Light Display mode group” and the “Grid Display mode”. The following tables summarize all the possible combinations of both groups:




















































The “Pen Attributes” can be changed using pa(color,style,width). Next example sets line color to 2, line type to 1 and line width to 2. Note that if pa() is not specified, the histogram line attributes are used:

   h2->Draw("SPEC dm(1,2) pa(2,1,2)");

The number of “Nodes” can be changed with n(nodesx,nodesy). Example:

   h2->Draw("SPEC n(40,40)");

Sometimes the displayed region is rather large. When displaying all channels the pictures become very dense and complicated. It is very difficult to understand the overall shape of data. “n(nx,ny)” allows to change the density of displayed channels. Only the channels coinciding with given nodes are displayed.

The visualization “Angles” can be changed with “a(alpha,beta,view)”: “alpha” is the angle between the bottom horizontal screen line and the displayed space on the right side of the picture and “beta” on the left side, respectively. One can rotate the 3-d space around the vertical axis using the “view” parameter. Allowed values are 0, 90, 180 and 270 degrees.

   h2->Draw("SPEC n(40,40) dm(0,1) a(30,30,0)");

The operator “zs(scale)” changes the scale of the Z-axis. The possible values are:

If gPad->SetLogz() has been set, the log scale on Z-axis is set automatically, i.e. there is no need for using the zs() operator. Note that the X and Y axis are always linear.

The operator “ci(r,g,b)” defines the colors increments (r, g and b are floats). For sophisticated shading (Light, Height and LightHeight Display Modes Groups) the color palette starts from the basic pen color (see pa() function). There is a predefined number of color levels (256). Color in every level is calculated by adding the increments of the r , g , b components to the previous level. Using this function one can change the color increments between two neighboring color levels. The function does not apply on the Simple Display Modes Group. The default values are: (1,1,1).

The operator “ca(color_algorithm)” allows to choose the Color Algorithm. To define the colors one can use one of the following color algorithms (RGB, CMY, CIE, YIQ, HVS models). When the level of a component reaches the limit value one can choose either smooth transition (by decreasing the limit value) or a sharp modulo transition (continuing with 0 value). This allows various visual effects. One can choose from the following set of the algorithms:

This function does not apply on Simple display modes group. Default value is 0. Example choosing CMY Modulo to paint the 2D histogram:

   h2->Draw("SPEC c1(3) dm(0,1) a(30,30,0)");

The operator “lp(x,y,z)” sets the light position. In Light and LightHeight display modes groups the color palette is calculated according to the fictive light source position in 3-d space. Using this function one can change the source’s position and thus achieve various graphical effects. This function does not apply for Simple and Height display modes groups. Default is: lp(1000,1000,100) .

The operator “s(shading,shadow)” allows to set the shading. The surface picture is composed of triangles. The edges of the neighboring triangles can be smoothed (shaded). The shadow can be painted as well. The function does not apply on Simple display modes group. The possible values for shading are:

The possible values for shadow are:

Default values: s(1,0) .

The operator “b(bezier)” sets the Bezier smoothing. For Simple display modes group and for Grid, LinesX and LinesY display modes one can smooth data using Bezier smoothing algorithm. The function does not apply on other display modes groups and display modes. Possible values are: 0 = No bezier smoothing, 1 = Bezier smoothing. Default value is: b(0).

The operator “cw(width)” sets the contour width. This function applies only on for the Contours display mode. One can change the width between horizontal slices and thus their density. Default value: cw(50) .

The operator “lhw(weight)” sets the light height weight. For LightHeight display modes group one can change the weight between both shading algorithms. The function does not apply on other display modes groups. Default value is lhw(0.5) .

The operator “cm(enable,color,width,height,style)” allows to draw a marker on each node. In addition to the surface drawn using any above given algorithm one can display channel marks. One can control the color as well as the width, height (in pixels) and the style of the marks. The parameter enable can be set to 0 = Channel marks are not drawn or 1 = Channel marks drawn. The possible styles are:

The operator “cg(enable,color)” channel grid. In addition to the surface drawn using any above given algorithm one can display grid using the color parameter. The parameter enable can be set to:

See the example in $ROOTSYS/tutorials/spectrum/spectrumpainter.C .

The picture produced by spectrumpainter.C macro

The picture produced by spectrumpainter.C macro 3-D Histograms

By default a 3D scatter plot is drawn. If the “BOX” option is specified, a 3D box with a volume proportional to the cell content is drawn.

3.8.3 Drawing a Sub-range of a 2-D Histogram

The picture produced by fit2a.C macro

The picture produced by fit2a.C macro

Using a TCutG object, it is possible to draw a 2D histogram sub-range. One must create a graphical cut (mouse or C++) and specify the name of the cut between “[” and “]” in the Draw option.

For example, with a TCutGnamed “cutg”, one can call:

   myhist->Draw("surf1 [cutg]");

Or, assuming two graphical cuts with name “cut1” and “cut2”, one can do:


The second Draw will superimpose on top of the first lego plot a subset of h2using the “surf” option with:

Up to 16 cuts may be specified in the cut string delimited by "[..]". Currently only the following drawing options are sensitive to the cuts option: col , box , scat , hist , lego , surf and cartesian coordinates only. See a complete example in the tutorial $ROOTSYS/tutorials/fit/fit2a.C .

3.8.4 Superimposing Histograms with Different Scales

The following script creates two histograms; the second histogram is the bins integral of the first one. It shows a procedure to draw the two histograms in the same pad and it draws the scale of the second histogram using a new vertical axis on the right side.

Superimposed histograms with different scales

Superimposed histograms with different scales

void twoscales() {
   TCanvas *c1 = new TCanvas("c1","different scales hists",600,400);
   //create, fill and draw h1
   TH1F *h1 = new TH1F("h1","my histogram",100,-3,3);
   for (Int_t i=0;i<10000;i++) h1->Fill(gRandom->Gaus(0,1));
   //create hint1 filled with the bins integral of h1
   TH1F *hint1 = new TH1F("hint1","h1 bins integral",100,-3,3);
   Float_t sum = 0;
   for (Int_t i=1;i<=100;i++) {
      sum += h1->GetBinContent(i);
   //scale hint1 to the pad coordinates
   Float_t rightmax = 1.1*hint1->GetMaximum();
   Float_t scale    = gPad->GetUymax()/rightmax;
   //draw an axis on the right side
   TGaxis*axis = new TGaxis(gPad->GetUxmax(),gPad->GetUymin(),

3.8.5 Statistics Display

By default, a histogram drawing includes the statistics box. Use TH1::SetStats(kFALSE) to eliminate the statistics box. If the statistics box is drawn, gStyle->SetOptStat(mode) allow you to select the type of displayed information . The parameter mode has up to nine digits that can be set OFF (0) or ON as follows:

mode = ksiourmen (default =000001111)

Never call SetOptStat(0001111) , but SetOptStat(1111) , because 0001111 will be taken as an octal number.

The method TStyle::SetOptStat(Option_t*option) can also be called with a character string as a parameter. The parameter option can contain:

   gStyle->SetOptStat("ne");   // prints the histogram name and number
                               // of entries
   gStyle->SetOptStat("n");    // prints the histogram name
   gStyle->SetOptStat("nemr"); // the default value

With the option "same", the statistic box is not redrawn. With the option "sames", it is re-drawn. If it hides the previous statistics box, you can change its position with the next lines (where h is the histogram pointer):

root[] TPaveStats *s =
root[] s->SetX1NDC (newx1); // new x start position
root[] s->SetX2NDC (newx2); // new x end position

3.8.6 Setting Line, Fill, Marker, and Text Attributes

The histogram classes inherit from the attribute classes: TAttLine, TAttFill, TAttMarker and TAttText. See the description of these classes for the list of options.

3.8.7 Setting Tick Marks on the Axis

The TPad::SetTicks() method specifies the type of tick marks on the axis. Let tx=gPad->GetTickx() and ty=gPad->GetTicky().

Use TPad::SetTicks(tx,ty) to set these options. See also the methods of TAxis that set specific axis attributes. If multiple color-filled histograms are drawn on the same pad, the fill area may hide the axis tick marks. One can force the axis redrawing over all the histograms by calling:


3.8.8 Giving Titles to the X, Y and Z Axis

Because the axis title is an attribute of the axis, you have to get the axis first and then call TAxis::SetTitle.

   h->GetXaxis()->SetTitle("X axis title");
   h->GetYaxis()->SetTitle("Y axis title");
   h->GetZaxis()->SetTitle("Z axis title");

The histogram title and the axis titles can be any TLatex string. The titles are part of the persistent histogram. For example if you wanted to write E with a subscript (T) you could use this:


For a complete explanation of the Latex mathematical expressions, see “Graphics and the Graphical User Interface”. It is also possible to specify the histogram title and the axis titles at creation time. These titles can be given in the “title” parameter. They must be separated by “;”:

   TH1F* h=new TH1F("h","Histogram title;X Axis;Y Axis;Z Axis",

Any title can be omitted:

   TH1F* h=new TH1F("h","Histogram title;;Y Axis",100,0,1);
   TH1F* h=new TH1F("h",";;Y Axis",100,0,1);

The method SetTitle has the same syntax:

   h->SetTitle("Histogram title;An other X title Axis");

3.9 Making a Copy of an Histogram

Like for any other ROOT object derived from TObject , the Clone method can be used. This makes an identical copy of the original histogram including all associated errors and functions:

   TH1F *hnew = (TH1F*)h->Clone(); // renaming is recommended,
   hnew->SetName("hnew");          // because otherwise you will have
                                   // two histograms with the same
                                   // name

3.10 Normalizing Histograms

You can scale a histogram ( TH1 *h ) such that the bins integral is equal to the normalization parameter norm:

   Double_t scale = norm/h->Integral();

3.11 Saving/Reading Histograms to/from a File

The following statements create a ROOT file and store a histogram on the file. Because TH1 derives from TNamed , the key identifier on the file is the histogram name:

   TFile f("histos.root","new");
   TH1F h1("hgaus","histo from a gaussian",100,-3,3);

To read this histogram in another ROOT session, do:

   TFile f("histos.root");
   TH1F *h = (TH1F*)f.Get("hgaus");

One can save all histograms in memory to the file by:


For a more detailed explanation, see “Input/Output”.

3.12 Miscellaneous Operations

   Asymmetry = (h1 - h2)/(h1 + h2); //where h1 = this
   h3 = h1->GetAsymmetry(h2);

3.13 Alphanumeric Bin Labels

By default, a histogram axis is drawn with its numeric bin labels. One can specify alphanumeric labels instead.

3.13.1 Option 1: SetBinLabel

To set an alphanumeric bin label call:


This can always be done before or after filling. Bin labels will be automatically drawn with the histogram.

Histograms with alphanumeric bin labels

Histograms with alphanumeric bin labels

See example in $ROOTSYS/tutorials/hist/hlabels1.C , hlabels2.C

3.13.2 Option 2: Fill

You can also call a Fill() function with one of the arguments being a string:


3.13.3 Option 3: TTree::Draw

You can use a char* variable type to histogram strings with TTree::Draw().

   // here "Nation" and "Division" are two char* branches of a Tree
   tree.Draw("Nation::Division", "", "text");
Using a *char variable type in TTree::Draw

Using a *char variable type in TTree::Draw

There is an example in $ROOTSYS/tutorials/tree/cernstaff.C.

If a variable is defined as char* it is drawn as a string by default. You change that and draw the value of char[0] as an integer by adding an arithmetic operation to the expression as shown below.

   // draw the integer value of MyChar[0] where "MyChar" is char[5]
   tree.Draw("MyChar + 0");

3.13.4 Sort Options

When using the options 2 or 3 above, the labels are automatically added to the list (THashList) of labels for a given axis. By default, an axis is drawn with the order of bins corresponding to the filling sequence. It is possible to reorder the axis alphabetically or by increasing or decreasing values. The reordering can be triggered via the TAxis context menu by selecting the menu item “LabelsOption” or by calling directly.


Here axis may be X, Y, or Z. The parameter option may be:

When using the option second above, new labels are added by doubling the current number of bins in case one label does not exist yet. When the filling is terminated, it is possible to trim the number of bins to match the number of active labels by calling:


Here axis may be X, Y, or Z. This operation is automatic when using TTree::Draw . Once bin labels have been created, they become persistent if the histogram is written to a file or when generating the C++ code via SavePrimitive .

3.14 Histogram Stacks

A THStack is a collection of TH1 (or derived) objects. Use THStack::Add( TH1 *h) to add a histogram to the stack. The THStack does not own the objects in the list.

Stacked histograms

Stacked histograms

By default, THStack::Draw draws the histograms stacked as shown in the left pad in the picture above. If the option "nostack" is used, the histograms are superimposed as if they were drawn one at a time using the "same" draw option . The right pad in this picture illustrates the THStack drawn with the "nostack" option.


Next is a simple example, for a more complex one see $ROOTSYS/tutorials/hist/hstack.C.

   THStack hs("hs","test stacked histograms");
   TH1F *h1 = new TH1F("h1","test hstack",100,-4,4);
   TH1F *h2 = new TH1F("h2","test hstack",100,-4,4);
   TH1F *h3 = new TH1F("h3","test hstack",100,-4,4);
   TCanvas c1("c1","stacked hists",10,10,700,900);
   c1.Divide (1,2);;

3.15 TH2Poly

TH2Poly is a 2D Histogram class allowing to define polygonal bins of arbitary shape.

Each bin in the TH2Poly histogram is a TH2PolyBin object. TH2PolyBin is a very simple class containing the vertices and contents of the polygonal bin as well as several related functions.

Bins are defined using one of the AddBin() methods. The bin definition should be done before filling.

The following very simple macro shows how to build and fill a TH2Poly:

   TH2Poly *h2p = new TH2Poly();
   Double_t x1[] = {0, 5, 5};
   Double_t y1[] = {0, 0, 5};
   Double_t x2[] = {0, -1, -1, 0};
   Double_t y2[] = {0, 0, -1, -1};
   Double_t x3[] = {4, 3, 0, 1, 2.4};
   Double_t y3[] = {4, 3.7, 1, 4.7, 3.5};
   h2p->AddBin(3, x1, y1);
   h2p->AddBin(3, x2, y2);
   h2p->AddBin(3, x3, y3);
   h2p->Fill( 3, 1, 3);      // fill bin 1
   h2p->Fill(-0.5, -0.5, 7); // fill bin 2
   h2p->Fill(-0.7, -0.5, 1); // fill bin 2
   h2p->Fill( 1, 3, 5);      // fill bin 3

More examples can bin found in $ROOTSYS/tutorials/hist/th2poly*.C

A TH2Poly histogram example

A TH2Poly histogram example

3.16 Profile Histograms

Profile histograms are in many cases an elegant replacement of two-dimensional histograms. The relationship of two quantities X and Y can be visualized by a two-dimensional histogram or a scatter-plot; its representation is not particularly satisfactory, except for sparse data. If Y is an unknown [but single-valued] function of X, it can be displayed by a profile histogram with much better precision than by a scatter-plot. Profile histograms display the mean value of Y and its RMS for each bin in X. The following shows the contents [capital letters] and the values shown in the graphics [small letters] of the elements for bin j. When you fill a profile histogram with TProfile.Fill(x,y) :

E[j] = sum Y**2
L[j] = number of entries in bin J
H[j] = sum Y
h[j] = H[j] / L[j]
s[j] = sqrt[E[j] / L[j] - h[j]**2]
e[j] = s[j] / sqrt[L[j]]

In the special case where s[j] is zero, when there is only one entry per bin, e[j] is computed from the average of the s[j] for all bins. This approximation is used to keep the bin during a fit operation. The TProfile constructor takes up to eight arguments. The first five parameters are similar to TH1D constructor.

TProfile(const char *name,const char *title,Int_t nbinsx,
         Double_t xlow, Double_t xup, Double_t ylow, Double_t yup,
         Option_t *option)

All values of y are accepted at filling time. To fill a profile histogram, you must use TProfile::Fill function. Note that when filling the profile histogram the method TProfile::Fill checks if the variable y is between fYmin and fYmax. If a minimum or maximum value is set for the Y scale before filling, then all values below ylow or above yup will be discarded. Setting the minimum or maximum value for the Y scale before filling has the same effect as calling the special TProfile constructor above where ylow and yup are specified.

3.16.1 Build Options

The last parameter is the build option. If a bin has N data points all with the same value Y, which is the case when dealing with integers, the spread in Y for that bin is zero, and the uncertainty assigned is also zero, and the bin is ignored in making subsequent fits. If SQRT(Y) was the correct error in the case above, then SQRT(Y)/SQRT(N) would be the correct error here. In fact, any bin with non-zero number of entries N but with zero spread (spread = s[j]) should have an uncertainty SQRT(Y)/SQRT(N). Now, is SQRT(Y)/SQRT(N) really the correct uncertainty ? That it is only in the case where the Y variable is some sort of counting statistics, following a Poisson distribution. This is the default case. However, Y can be any variable from an original NTUPLE, and does not necessarily follow a Poisson distribution. The computation of errors is based on Y = values of data points; N = number of data points.

The option ‘i’ is used for integer Y values with the uncertainty of \(\pm 0.5\), assuming the probability that Y takes any value between Y-0.5 and Y+0.5 is uniform (the same argument for Y uniformly distributed between Y and Y+1). An example is an ADC measurement. The ‘G’ option is useful, if all Y variables are distributed according to some known Gaussian of standard deviation Sigma. For example when all Y’s are experimental quantities measured with the same instrument with precision Sigma. The next figure shows the graphic output of this simple example of a profile histogram.

   // Create a canvas giving the coordinates and the size
   TCanvas *c1 = new TCanvas("c1", "Profile example",200,10,700,500);

   // Create a profile with the name, title, the number of bins,
   // the low and high limit of the x-axis and the low and high
   // limit of the y-axis.
   // No option is given so the default is used.
   hprof = new TProfile("hprof",
                        "Profile of pz versus px",100,-4,4,0,20);

   // Fill the profile 25000 times with random numbers
   Float_t px, py, pz;
   for ( Int_t i=0; i<25000; i++) {
      // Use the random number generator to get two numbers following
      // a gaussian distribution with mean=0 and sigma=1
      pz = px*px + py*py;
A profile histogram example

A profile histogram example

3.16.2 Drawing a Profile without Error Bars

To draw a profile histogram and not show the error bars use the “HIST” option in the TProfile::Draw method. This will draw the outline of the TProfile.

3.16.3 Create a Profile from a 2D Histogram

You can make a profile from a histogram using the methods TH2::ProfileX and TH2::ProfileY.

3.16.4 Create a Histogram from a Profile

To create a regular histogram from a profile histogram, use the method TProfile::ProjectionX .This example instantiates a TH1D object by copying the TH1D piece of TProfile.

   TH1D *sum = myProfile.ProjectionX();

You can do the same with a 2D profile using the method TProfile2D::ProjectionXY .

3.16.5 Generating a Profile from a TTree

The 'prof' and 'profs' options in the TTree::Draw method generate a profile histogram ( TProfile ), given a two dimensional expression in the tree, or a TProfile2D given a three dimensional expression. See “Trees”. Note that you can specify 'prof' or 'profs' : 'prof' generates a TProfile with error on the mean, 'profs' generates a TProfile with error on the spread.

3.16.6 2D Profiles

The class for a 2D Profile is called TProfile2D . It is in many cases an elegant replacement of a three-dimensional histogram. The relationship of three measured quantities X, Y and Z can be visualized by a three-dimensional histogram or scatter-plot; its representation is not particularly satisfactory, except for sparse data. If Z is an unknown (but single-valued) function of (X,Y), it can be displayed with a TProfile2D with better precision than by a scatter-plot. A TProfile2D displays the mean value of Z and its RMS for each cell in X, Y. The following shows the cumulated contents (capital letters) and the values displayed (small letters) of the elements for cell i,j.

When you fill a profile histogram with TProfile2D.Fill(x,y,z):

E[i,j] = sum z
L[i,j] = sum l
h[i,j] = H[i,j ] / L[i,j]
s[i,j] = sqrt[E[i,j] / L[i,j]- h[i,j]**2]
e[i,j] = s[i,j] / sqrt[L[i,j]]

In the special case where s[i,j] is zero, when there is only one entry per cell, e[i,j] is computed from the average of the s[i,j] for all cells. This approximation is used to keep the cell during a fit operation.

A TProfile2D histogram example

A TProfile2D histogram example

   // Creating a Canvas and a TProfile2D
   TCanvas *c1 = new TCanvas("c1",
                             "Profile histogram example",
                              200, 10,700,500);
   hprof2d = new TProfile2D("hprof2d",
                            "Profile of pz versus px and py",

   // Filling the TProfile2D with 25000 points
   Float_t px, py, pz;
   for (Int_t i=0; i<25000; i++) {
      pz = px*px + py*py;

3.17 Iso Surfaces

Paint one Gouraud shaded 3d iso surface though a 3d histogram at the value computed as follow: SumOfWeights/(NbinsX*NbinsY*NbinsZ).

void hist3d() {
   TH3D *h3 = new TH3D("h3", "h3", 20, -2, 2, 20, -2, 2, 20, 0, 4);
   Double_t x,y,z;
   for (Int_t i=0; i<10000; i++) {
Iso surfaces

Iso surfaces

3.18 3D Implicit Functions

   TF3 *fun3 = new TF3("fun3","sin(x*x+y*y+z*z-36)",-2,2,-2,2,-2,2);
3D implicit function

3D implicit function

3.19 TPie

The TPie class allows to create a Pie Chart representation of a one dimensional data set. The data can come from an array of Double_t (or Float_t ) or from a 1D-histogram. The possible options to draw a TPie are:

The method SetLabelFormat() is used to customize the label format. The format string must contain one of these modifiers:

   mypie->SetLabelFormat("%txt (%frac)");

See the macro $ROOTSYS/tutorials/graphics/piechart.C .

The picture generated by tutorial macro piechart.C

The picture generated by tutorial macro piechart.C

3.20 The User Interface for Histograms

The classes T H1 Editor and T H2 Editor provides the user interface for setting histogram’s attributes and rebinning interactively.

3.20.1 TH1Editor The Style Tab Title

sets the title of the histogram. Plot

draw a 2D or 3D plot; according to the dimension, different drawing possibilities can be set. Error

add different error bars to the histogram (no errors, simple, etc.). Add

further things which can be added to the histogram (None, simple/smooth line, fill area, etc.) 2-D Plot Simple Drawing

draw a simple histogram without errors (= “HIST” draw option). In combination with some other draw options an outer line is drawn on top of the histogram Show markers

draw a marker on to of each bin (=“P” draw option). Draw bar chart

draw a bar chart (=“B” draw option). Bar option

draw a bar chart (=“BAR” draw option); if selected, it will show an additional interface elements for bars: width, offset, percentage and the possibility to draw horizontal bars. 3-D Plot Add

set histogram type Lego-Plot or Surface draw (Lego, Lego1.2, Surf, Surf1…5). Coords

set the coordinate system (Cartesian, Spheric, etc.). Error

same as for 2D plot. Bar

set the bar attributes: width and offset. Horizontal Bar

draw a horizontal bar chart. The Binning tab

The binning tab has two different layouts. One is for a histogram, which is not drawn from an ntuple. The other one is available for a histogram, which is drawn from an ntuple. In this case, the rebin algorithm can create a rebinned histogram from the original data i.e. the ntuple.

To see the differences do:

   TFile f("hsimple.root");
   hpx->Draw("BAR1"); // non ntuple histogram
   ntuple->Draw("px");// ntuple histogram Non ntuple histogram

Rebin with a slider and the number of bins (shown in the field below the slider). The number of bins can be changed to any number, which divides the number of bins of the original histogram. A click on the Apply button will delete the origin histogram and will replace it by the rebinned one on the screen. A click on the Ignore button will restore the origin histogram. Histogram drawn from an ntuple Rebin

with the slider, the number of bins can be enlarged by a factor of 2, 3, 4, 5 (moving to the right) or reduced by a factor of \(\frac{1}{2}\), \(\frac{1}{3}\), \(\frac{1}{4}\), \(\frac{1}{5}\). BinOffset with a BinOffset slider

the origin of the histogram can be changed within one binwidth. Using this slider the effect of binning the data into bins can be made visible (statistical fluctuations). Axis Range

with a double slider it is possible to zoom into the specified axis range. It is also possible to set the upper and lower limit in fields below the slider. Delayed drawing

all the Binning sliders can set to delay draw mode. Then the changes on the histogram are only updated, when the Slider is released. This should be activated if the redrawing of the histogram is time consuming.

3.20.2 TH2Editor Style Tab: Title

set the title of the histogram Histogram

change the draw options of the histogram. Plot

draw a 2D or 3D plot of the histogram; according to the dimension, the drawing possibilities are different. 2-D Plot Contour

draw a contour plot (None, Cont0…5) Cont #

set the number of Contours; Arrow

set the arrow mode and shows the gradient between adjacent cells; Col

a box is drawn for each cell with a color scale varying with contents; Text

draw bin contents as text; Box

a box is drawn for each cell with surface proportional to contents; Scat

draw a scatter-plot (default); Palette

the color palette is drawn. 3-D Plot Type

set histogram type to Lego or surface plot; draw (Lego, Lego1.2, Surf, Surf1…5) Coords

set the coordinate system (Cartesian, Spheric, etc.); Cont #

set the number of Contours (for e.g. Lego2 draw option); Errors

draw errors in a Cartesian lego plot; Palette

draw the color palette; Front

draw the front box of a Cartesian lego plot; Back

draw the back box of a Cartesian lego plot; Bar

change the bar attributes: the width and offset. Rebinning Tab

The Rebinning tab has two different layouts. One is for a histogram that is not drawn from an ntuple; the other one is available for a histogram, which is drawn from an ntuple. In this case, the rebin algorithm can create a rebinned histogram from the original data i.e. the ntuple. To see the differences do for example:

   TFile f ("hsimple.root");
   hpxpy->Draw("Lego2"); // non ntuple histogram
   ntuple->Draw("px:py","","Lego2"); // ntuple histogram Non-ntuple histogram:

Rebin with sliders (one for the x, one for the y-axis) and the number of bins (shown in the field below them can be changed to any number, which divides the number of bins of the original histogram. Selecting the Apply button will delete the origin histogram and will replace it by the rebinned one on the screen. Selecting the Ignore the origin histogram will be restored. Histogram drawn from an ntuple Rebin

with the sliders the number of bins can be enlarged by a factor of 2,3,4,5 (moving to the right) or reduced by a factor of \(\frac{1}{2}\), \(\frac{1}{3}\), \(\frac{1}{4}\), \(\frac{1}{5}\). BinOffset

with the BinOffset slider the origin of the histogram can be changed within one binwidth. Using this slider the effect of binning the data into bins can be made visible (=> statistical fluctuations). Axis Range

with a double slider that gives the possibility for zooming. It is also possible to set the upper and lower limit in fields below the slider. Delayed drawing

all the binning sliders can be set to delay draw mode. Then the changes on the histogram are only updated, when the Slider is released. This should be activated if the redrawing of the histogram is too time consuming.

4 Graphs

A graph is a graphics object made of two arrays X and Y, holding the x,y coordinates of n points. There are several graph classes; they are TGraph, TGraphErrors, TGraphAsymmErrors, and TMultiGraph.

4.1 TGraph

The TGraph class supports the general case with non-equidistant points, and the special case with equidistant points. Graphs are created with the TGraph constructor. First, we define the arrays of coordinates and then create the graph. The coordinates can be arrays of doubles or floats.

   Int_t n = 20;
   Double_t x[n], y[n];
   for (Int_t i=0; i<n; i++) {
      x[i] = i*0.1;
      y[i] = 10*sin(x[i]+0.2);
   TGraph *gr1 = new TGraph (n, x, y);

An alternative constructor takes only the number of points n. It is expected that the coordinates will be set later.

TGraph *gr2 = new TGraph(n);

The default constructor can also be used. Further calls to SetPoint() will extend the internal vectors.

TGraph *gr3 = new TGraph();

4.1.1 Graph Draw Options

The various drawing options for a graph are explained in TGraph::PaintGraph. They are:

The options are not case sensitive and they can be concatenated in most cases. Let us look at some examples. Continuous Line, Axis and Stars (AC*)

A graph drawn with axis, * markers and continuous line (option AC*)

A graph drawn with axis, * markers and continuous line (option AC*)

   Int_t n = 20;
   Double_t x[n], y[n];
   for (Int_t i=0;i<n;i++) {
      x[i] = i*0.1;
      y[i] = 10*sin(x[i]+0.2);

   // create graph
   TGraph *gr  = new TGraph(n,x,y);
   TCanvas *c1 = new TCanvas("c1","Graph Draw Options",

   // draw the graph with axis, continuous line, and put
   // a * at each point
} Bar Graphs (AB)

A graph drawn with axis and bar (option AB)

A graph drawn with axis and bar (option AB)

root[] TGraph *gr1 = new TGraph(n,x,y);
root[] gr1->SetFillColor(40);
root[] gr1->Draw("AB");

This code will only work if n, x, and y is defined. The previous example defines these. You need to set the fill color, because by default the fill color is white and will not be visible on a white canvas. You also need to give it an axis, or the bar chart will not be displayed properly. Filled Graphs (AF)

A graph drawn with axis and fill (option AF)

A graph drawn with axis and fill (option AF)

root[] TGraph *gr3 = new TGraph(n,x,y);
root[] gr3->SetFillColor(45);
root[] gr3->Draw("AF")

This code will only work if n, x, yare defined. The first example defines them. You need to set the fill color, because by default the fill color is white and will not be visible on a white canvas. You also need to give it an axis, or the filled polygon will not be displayed properly. Marker Options

Graph markers created in different ways

Graph markers created in different ways

   Int_t n = 20;
   Double_t x[n], y[n];

   // build the arrays with the coordinate of points
   for (Int_t i=0; i<n; i++) {
      x[i] = i*0.1;
      y[i] = 10*sin(x[i]+0.2);

   // create graphs
   TGraph *gr3  = new TGraph(n,x,y);
   TCanvas *c1 = new TCanvas ("c1","Graph Draw Options",

   // draw the graph with the axis,contineous line, and put
   // a marker using the graph's marker style at each point

   // get the points in the graph and put them into an array
   Double_t *nx = gr3->GetX();
   Double_t *ny = gr3->GetY();

   // create markers of different colors
   for (Int_t j=2; j<n-1; j++) {
      TMarker *m = new TMarker(nx[j], 0.5*ny[j], 22);

4.2 Superimposing Two Graphs

To super impose two graphs you need to draw the axis only once, and leave out the “A” in the draw options for the second graph. Next is an example:

Superimposing two graphs

Superimposing two graphs

   Int_t n = 20;
   Double_t x[n], y[n], x1[n], y1[n];

   // create a blue graph with a cos function

   // superimpose the second graph by leaving out the axis option "A"

4.3 Graphs with Error Bars

A TGraphErrors is a TGraph with error bars. The various draw format options of TGraphErrors::Paint() are derived from TGraph.

void TGraphErrors::Paint(Option_t *option)
Graphs with different draw options of error bars

Graphs with different draw options of error bars

In addition, it can be drawn with the “Z” option to leave off the small lines at the end of the error bars. If option contains “>”, an arrow is drawn at the end of the error bars. If option contains “|>”, a full arrow is drawn at the end of the error bars. The size of the arrow is set to 2/3 of the marker size.

The option “[]” is interesting to superimpose systematic errors on top of the graph with the statistical errors. When it is specified, only the end vertical/horizontal lines of the error bars are drawn.

To control the size of the lines at the end of the error bars (when option 1 is chosen) use SetEndErrorSize(np). By default np=1; np represents the number of pixels.


The four parameters of TGraphErrors are: X, Y (as in TGraph), X-errors, and Y-errors - the size of the errors in the x and y direction. Next example is $ROOTSYS/tutorials/graphs/gerrors.C.

   c1 = new TCanvas("c1","A Simple Graph with error bars",200,10,700,500);

   // create the coordinate arrays
   Int_t n = 10;
   Float_t x[n]  = {-.22,.05,.25,.35,.5,.61,.7,.85,.89,.95};
   Float_t y[n]  = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};

   // create the error arrays
   Float_t ex[n] = {.05,.1,.07,.07,.04,.05,.06,.07,.08,.05};
   Float_t ey[n] = {.8,.7,.6,.5,.4,.4,.5,.6,.7,.8};

   // create the TGraphErrors and draw it
   gr = new TGraphErrors(n,x,y,ex,ey);
   gr->SetTitle("TGraphErrors Example");

4.4 Graphs with Asymmetric Error Bars

A graph with asymmetric error bars

A graph with asymmetric error bars

A TGraphAsymmErrors is a TGraph with asymmetric error bars. It inherits the various draw format options from TGraph. Its method Paint(Option_t *option) paints the TGraphAsymmErrors with the current attributes. You can set the following additional options for drawing:

The constructor has six arrays as parameters: X and Y as TGraph and low X-errors and high X-errors, low Y-errors and high Y-errors. The low value is the length of the error bar to the left and down, the high value is the length of the error bar to the right and up.

   c1 = new TCanvas("c1","A Simple Graph with error bars",

   // create the arrays for the points
   Int_t n = 10;
   Double_t x[n]  = {-.22,.05,.25,.35,.5, .61,.7,.85,.89,.95};
   Double_t y[n]  = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};

   // create the arrays with high and low errors
   Double_t exl[n] = {.05,.1,.07,.07,.04,.05,.06,.07,.08,.05};
   Double_t eyl[n] = {.8,.7,.6,.5,.4,.4,.5,.6,.7,.8};
   Double_t exh[n] = {.02,.08,.05,.05,.03,.03,.04,.05,.06,.03};
   Double_t eyh[n] = {.6,.5,.4,.3,.2,.2,.3,.4,.5,.6};

   // create TGraphAsymmErrors with the arrays
   gr = new TGraphAsymmErrors(n,x,y,exl,exh,eyl,eyh);
   gr->SetTitle("TGraphAsymmErrors Example");

4.5 Graphs with Asymmetric Bent Errors

A graph with asymmetric bent error bars

A graph with asymmetric bent error bars

A TGraphBentErrors is a TGraph with bent, asymmetric error bars. The various format options to draw a TGraphBentErrors are explained in TGraphBentErrors::Paint method. The TGraphBentErrors is drawn by default with error bars and small horizontal and vertical lines at the end of the error bars. If option “z” or “Z” is specified, these small lines are not drawn. If the option “X” is specified, the errors are not drawn (the TGraph::Paint method equivalent).

This figure has been generated by the following macro:

   Int_t n = 10;
   Double_t x[n] = {-0.22,0.05,0.25,0.35,0.5,0.61,0.7,0.85,0.89,0.95};
   Double_t y[n] = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};
   Double_t exl[n] = {.05,.1,.07,.07,.04,.05,.06,.07,.08,.05};
   Double_t eyl[n] = {.8,.7,.6,.5,.4,.4,.5,.6,.7,.8};
   Double_t exh[n] = {.02,.08,.05,.05,.03,.03,.04,.05,.06,.03};
   Double_t eyh[n] = {.6,.5,.4,.3,.2,.2,.3,.4,.5,.6};
   Double_t exld[n] = {.0,.0,.0,.0,.0,.0,.0,.0,.0,.0};
   Double_t eyld[n] = {.0,.0,.0,.0,.0,.0,.0,.0,.0,.0};
   Double_t exhd[n] = {.0,.0,.0,.0,.0,.0,.0,.0,.0,.0};
   Double_t eyhd[n] = {.0,.0,.0,.0,.0,.0,.0,.0,.05,.0};
   gr = new TGraphBentErrors(n,x,y,
   gr->SetTitle("TGraphBentErrors Example");

4.6 TGraphPolar

The TGraphPolar class creates a polar graph (including error bars). A TGraphPolar is a TGraphErrors represented in polar coordinates. It uses the class TGraphPolargram to draw the polar axis.

   TCanvas *CPol = new TCanvas("CPol","TGraphPolar Examples",700,700);
   Double_t rmin=0;
   Double_t rmax=TMath::Pi()*2;
   Double_t r[1000];
   Double_t theta[1000];
   TF1 * fp1 = new TF1("fplot","cos(x)",rmin,rmax);
   for (Int_t ipt = 0; ipt < 1000; ipt++) {
      r[ipt] = ipt*(rmax-rmin)/1000+rmin;
      theta[ipt] = fp1->Eval(r[ipt]);
   TGraphPolar * grP1 = new TGraphPolar(1000,r,theta);

The TGraphPolar drawing options are:

“O” Polar labels are paint orthogonally to the polargram radius.

“P” Polymarker are paint at each point position.

“E” Paint error bars.

“F” Paint fill area (closed polygon).

“A”Force axis redrawing even if a polagram already exists.

A polar graph

A polar graph

4.7 TGraph Exclusion Zone

When a graph is painted with the option “C” or “L”, it is possible to draw a filled area on one side of the line. This is useful to show exclusion zones. This drawing mode is activated when the absolute value of the graph line width (set thanks to SetLineWidth) is greater than 99. In that case the line width number is interpreted as 100*ff+ll = ffll. The two-digit numbers “ll” represent the normal line width whereas “ff” is the filled area width. The sign of “ffll” allows flipping the filled area from one side of the line to the other. The current fill area attributes are used to draw the hatched zone.

Graphs with exclusion zones

Graphs with exclusion zones

   c1 = new TCanvas("c1","Exclusion graphs examples",200,10,700,500);

   // create the multigraph
   TMultiGraph *mg = new TMultiGraph();
   mg->SetTitle("Exclusion graphs");

   // create the graphs points
   const Int_t n = 35;
   Double_t x1[n], x2[n], x3[n], y1[n], y2[n], y3[n];
   for (Int_t i=0;i<n;i++) {
      x1[i] = i*0.1; y1[i] = 10*sin(x1[i]);
      x2[i] = x1[i]; y2[i] = 10*cos(x1[i]);
      x3[i] = x1[i]+.5; y3[i] = 10*sin(x1[i])-2;

   // create the 1st TGraph
   gr1 = new TGraph(n,x1,y1);

   // create the 2nd TGraph
   gr2 = new TGraph(n,x2,y2);

   // create the 3rd TGraph
   gr3 = new TGraph(n,x3,y3);

   // put the graphs in the multigraph

   // draw the multigraph

4.8 TGraphQQ

A TGraphQQ allows drawing quantile-quantile plots. Such plots can be drawn for two datasets, or for one dataset and a theoretical distribution function.

4.8.1 Two Datasets

Examples of qq-plots of 2 datasets

Examples of qq-plots of 2 datasets

Quantile-quantile plots are used to determine whether two samples come from the same distribution. A qq-plot draws the quantiles of one dataset against the quantile of the other. The quantiles of the dataset with fewer entries are on Y-axis, with more entries - on X-axis. A straight line, going through 0.25 and 0.75 quantiles is also plotted for reference. It represents a robust linear fit, not sensitive to the extremes of the datasets. If the datasets come from the same distribution, points of the plot should fall approximately on the 45 degrees line. If they have the same distribution function, but different parameters of location or scale, they should still fall on the straight line, but not the 45 degrees one.

The greater their departure from the straight line, the more evidence there is that the datasets come from different distributions. The advantage of qq-plot is that it not only shows that the underlying distributions are different, but, unlike the analytical methods, it also gives information on the nature of this difference: heavier tails, different location/scale, different shape, etc.

4.8.2 One Dataset

Examples of qq-plots of 1 dataset

Examples of qq-plots of 1 dataset

Quantile-quantile plots are used to determine if the dataset comes from the specified theoretical distribution, such as normal. A qq-plot draws quantiles of the dataset against quantiles of the specified theoretical distribution. Note, that density, not CDF should be specified a straight line, going through 0.25 and 0.75 quantiles could also be plotted for reference. It represents a robust linear fit, not sensitive to the extremes of the dataset. As in the two datasets case, departures from straight line indicate departures from the specified distribution. Next picture shows an example of a qq-plot of a dataset from N(3, 2) distribution and TMath::Gaus(0, 1) theoretical function. Fitting parameters are estimates of the distribution mean and sigma.

4.9 TMultiGraph

A multigraph example

A multigraph example

A TMultiGraph is a collection of TGraph (or derived) objects. Use TMultiGraph::Addto add a new graph to the list. The TMultiGraph owns the objects in the list. The drawing and fitting options are the same as for TGraph.

   // create the points
   Int_t n = 10;
   Double_t x[n]  = {-.22,.05,.25,.35,.5,.61,.7,.85,.89,.95};
   Double_t y[n]  = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};
   Double_t x2[n]  = {-.12,.15,.35,.45,.6,.71,.8,.95,.99,1.05};
   Double_t y2[n]  = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};

   // create the width of errors in x and y direction
   Double_t ex[n] = {.05,.1,.07,.07,.04,.05,.06,.07,.08,.05};
   Double_t ey[n] = {.8,.7,.6,.5,.4,.4,.5,.6,.7,.8};

   // create two graphs
   TGraph *gr1 = new TGraph(n,x2,y2);
   TGraphErrors *gr2 = new TGraphErrors(n,x,y,ex,ey);

   // create a multigraph and draw it
   TMultiGraph  *mg  = new TMultiGraph();

4.10 TGraph2D

Delaunay triangles and Voronoi diagram

Delaunay triangles and Voronoi diagram

This class is a set of N points x[i], y[i], z[i] in a non-uniform grid. Several visualization techniques are implemented, including Delaunay triangulation. Delaunay triangulation is defined as follow: ‘for a set S of points in the Euclidean plane, the unique triangulation DT(S) of S such that no point in S is inside the circum-circle of any triangle in DT(S). DT(S) is the dual of the Voronoi diagram of S. If n is the number of points in S, the Voronoi diagram of S is the partitioning of the plane containing S points into n convex polygons such that each polygon contains exactly one point and every point in a given polygon is closer to its central point than to any other. A Voronoi diagram is sometimes also known as a Dirichlet tessellation.

The TGraph2D class has the following constructors:

   TGraph2D *g = new TGraph2D(n,x,y,z);
   TGraph2D *g = new TGraph2D(n);
   TGraph2D *g = new TGraph2D();
   TGraph2D *g = new TGraph2D("graph.dat");

The arrays are read from the ASCII file “graph.dat” according to a specified format. The format’s default value is “%lg %lg %lg”. Note that in any of last three cases, the SetPoint method can be used to change a data point or to add a new one. If the data point index (i) is greater than the size of the internal arrays, they are automatically extended.

Specific drawing options can be used to paint a TGraph2D:

A TGraph2D can be also drawn with ANY options valid for 2D histogram drawing. In this case, an intermediate 2D histogram is filled using the Delaunay triangles technique to interpolate the data set. TGraph2D linearly interpolate a Z value for any (X,Y) point given some existing (X,Y,Z) points. The existing (X,Y,Z) points can be randomly scattered. The algorithm works by joining the existing points to make Delaunay triangles in (X,Y). These are then used to define flat planes in (X,Y,Z) over which to interpolate. The interpolated surface thus takes the form of tessellating triangles at various angles. Output can take the form of a 2D histogram or a vector. The triangles found can be drawn in 3D. This software cannot be guaranteed to work under all circumstances. It was originally written to work with a few hundred points in anXY space with similar X and Y ranges.

Graph2D drawn with option surf1 and tri1 p0

Graph2D drawn with option “surf1” and “tri1 p0”

   TCanvas *c = new TCanvas("c","Graph2D example",0,0,700,600);
   Double_t x, y, z, P = 6.;
   Int_t np = 200;
   TGraph2D *dt = new TGraph2D();
   TRandom *r = new TRandom();

   for (Int_t N=0; N<np; N++) {
      x = 2*P*(r->Rndm(N))-P;
      y = 2*P*(r->Rndm(N))-P;
      z = (sin(x)/x)*(sin(y)/y)+0.2;
   dt->Draw("surf1");       // use "surf1" to generate the left picture
}                           // use "tri1 p0" to generate the right one

A more complete example is $ROOTSYS/tutorials/fit/graph2dfit.C that produces the next figure.

Output of macro graph2dfit.C

Output of macro graph2dfit.C

4.11 TGraph2DErrors

A TGraph2DErrors is a TGraph2D with errors. It is useful to perform fits with errors on a 2D graph. An example is the macro $ROOTSYS/tutorials/graphs/graph2derrorsfit.C.

4.12 Fitting a Graph

The graph Fit method in general works the same way as the TH1::Fit. See “Fitting Histograms”.

4.13 Setting the Graph’s Axis Title

To give the axis of a graph a title you need to draw the graph first, only then does it actually have an axis object. Once drawn, you set the title by getting the axis and calling the TAxis::SetTitle method, and if you want to center it, you can call the TAxis::CenterTitle method.

Assuming that n, x, and y are defined. Next code sets the titles of the x and y axes.

root[] gr5 = new TGraph(n,x,y)
root[] gr5->Draw()
<TCanvas::MakeDefCanvas>: created default TCanvas with name c1
root[] gr5->Draw("ALP")
root[] gr5->GetXaxis()->SetTitle("X-Axis")
root[] gr5->GetYaxis()->SetTitle("Y-Axis")
root[] gr5->GetXaxis()->CenterTitle()
root[] gr5->GetYaxis()->CenterTitle()
root[] gr5->Draw("ALP")

For more graph examples see the scripts: $ROOTSYS/tutorials directory graph.C, gerrors.C, zdemo.C, and gerrors2.C.

A graph with axis titles

A graph with axis titles

4.14 Zooming a Graph

To zoom a graph you can create a histogram with the desired axis range first. Draw the empty histogram and then draw the graph using the existing axis from the histogram.

   c1 = new TCanvas("c1","A Zoomed Graph",200,10,700,500);
   hpx = new TH2F("hpx","Zoomed Graph Example",10,0,0.5,10,1.0,8.0);
   hpx->SetStats(kFALSE);   // no statistics
   Int_t n = 10;
   Double_t x[n] = {-.22,.05,.25,.35,.5,.61,.7,.85,.89,.95};
   Double_t y[n] = {1,2.9,5.6,7.4,9,9.6,8.7,6.3,4.5,1};
   gr = new TGraph(n,x,y);
   gr->Draw("LP");// and draw it without an axis

The next example is the same graph as above with a zoom in the x and y directions.

A zoomed graph

A zoomed graph

4.15 The User Interface for Graphs

The class TGraphEditor provides the user interface for setting the following graph attributes interactively:

5 Fitting Histograms

To fit a histogram you can use the Fit Panel on a visible histogram via the context menu, or you can use the TH1::Fit method. The Fit Panel, which is limited, is best for prototyping. The histogram needs to be drawn in a pad before the Fit Panel is invoked. The method TH1::Fit is more powerful and is used in scripts and programs.

5.1 The Fit Method

To fit a histogram programmatically, you can use the TH1::Fit method. Here is the signature of TH1::Fit and an explanation of the parameters:

   void Fit(const char *fname, Option_t *option, Option_t *goption,
            Axis_t xxmin, Axis_t  xxmax)

By default, the fitting function object is added to the histogram and is drawn in the current pad.

5.2 Fit with a Predefined Function

To fit a histogram with a predefined function, simply pass the name of the function in the first parameter of TH1::Fit. For example, this line fits histogram object hist with a Gaussian.

root[] hist.Fit("gaus");

The initial parameter values for pre-defined functions are set automatically.

5.3 Fit with a User-Defined Function

You can create a TF1 object and use it in the call the TH1::Fit. The parameter in to the Fit method is the NAME of the TF1 object. There are three ways to create a TF1.

5.3.1 Creating a TF1 with a Formula

Let’s look at the first case. Here we call the TF1 constructor by giving it the formula: sin(x)/x.

root[] TF1  *f1 = new TF1("f1","sin(x)/x",0,10)

You can also use a TF1 object in the constructor of another TF1.

root[] TF1  *f2 = new TF1("f2","f1*2",0,10)

5.3.2 Creating a TF1 with Parameters

The second way to construct a TF1 is to add parameters to the expression. Here we use two parameters:

root[] TF1 *f1 = new TF1("f1","[0]*x*sin([1]*x)",-3,3);
The function x*sin(x)

The function x*sin(x)

The parameter index is enclosed in square brackets. To set the initial parameters explicitly you can use:

root[] f1->SetParameter(0,10);

This sets parameter 0 to 10. You can also use SetParameters to set multiple parameters at once.

root[] f1->SetParameters(10,5);

This sets parameter 0 to 10 and parameter 1 to 5. We can now draw the TF1:

root[] f1->Draw()

5.3.3 Creating a TF1 with a User Function

The third way to build a TF1 is to define a function yourself and then give its name to the constructor. A function for a TF1 constructor needs to have this exact signature:

Double_t fitf(Double_t *x,Double_t *par)

The two parameters are:

The following script $ROOTSYS/tutorials/fit/myfit.C illustrates how to fit a 1D histogram with a user-defined function. First we declare the function.

   // define a function with 3 parameters
   Double_t fitf(Double_t *x,Double_t *par) {
      Double_t arg = 0;
      if (par[2]!=0) arg = (x[0] - par[1])/par[2];
      Double_t fitval = par[0]*TMath::Exp(-0.5*arg*arg);
      return fitval;

Now we use the function:

   // this function used fitf to fit a histogram
   void fitexample() {

      // open a file and get a histogram
      TFile *f = new TFile("hsimple.root");
      TH1F *hpx = (TH1F*)f->Get("hpx");

      // Create a TF1 object using the function defined above.
      // The last three parameters specify the number of parameters
      // for the function.
      TF1 *func = new TF1("fit",fitf,-3,3,3);
      // set the parameters to the mean and RMS of the histogram

      // give the parameters meaningful names
      func->SetParNames ("Constant","Mean_value","Sigma");

      // call TH1::Fit with the name of the TF1 object

5.4 Fixing and Setting Parameters’ Bounds

Parameters must be initialized before invoking the Fit method. The setting of the parameter initial values is automatic for the predefined functions: poln, exp, gaus, and landau. You can fix one or more parameters by specifying the “B” option when calling the Fit method. When a function is not predefined, the fit parameters must be initialized to some value as close as possible to the expected values before calling the fit function.

To set bounds for one parameter, use TF1::SetParLimits:


When the lower and upper limits are equal, the parameter is fixed. Next two statements fix parameter 4 at 10.


However, to fix a parameter to 0, one must call the FixParameter function:


Note that you are not forced to set the limits for all parameters. For example, if you fit a function with 6 parameters, you can:


With this setup, parameters 0->2 can vary freely, parameter 3 has boundaries [-10, 4] with initial value -1.5, and parameter 4 is fixed to 0.

5.5 Fitting Sub Ranges

By default, TH1::Fit will fit the function on the defined histogram range. You can specify the option “R” in the second parameter of TH1::Fit to restrict the fit to the range specified in the TF1 constructor. In this example, the fit will be limited to -3 to 3, the range specified in the TF1 constructor.

root[] TF1 *f1 = new TF1("f1","[0]*x*sin([1]*x)",-3,3);
root[] hist->Fit("f1","R");

You can also specify a range in the call to TH1::Fit:

root[] hist->Fit("f1","","",-2,2)

See macros $ROOTSYS/tutorials/fit/myfit.C and multifit.C as more completed examples.

5.6 The Fit Panel

The Fit Panel

The Fit Panel

To display the Fit Panel right click on a histogram to pop up the context menu, and then select the menu entry Fit Panel.

The new Fit Panel GUI is available in ROOT v5.14. Its goal is to replace the old Fit Panel and to provide more user friendly way for performing, exploring and comparing fits.

By design, this user interface is planned to contain two tabs: “General” and “Minimization”. Currently, the “General” tab provides user interface elements for setting the fit function, fit method and different fit, draw, print options.

The new fit panel is a modeless dialog, i.e. when opened, it does not prevent users from interacting with other windows. Its first prototype is a singleton application. When the Fit Panel is activated, users can select an object for fitting in the usual way, i.e. by left-mouse click on it. If the selected object is suitable for fitting, the fit panel is connected with this object and users can perform fits by setting different parameters and options.

5.6.1 Function Choice and Settings

‘Predefined’ combo box - contains a list of predefined functions in ROOT. You have a choice of several polynomials, a Gaussian, a Landau, and an Exponential function. The default one is Gaussian.

‘Operation’ radio button group defines the selected operational mode between functions:

Nop - no operation (default);

Add - addition;

Conv - convolution (will be implemented in the future).

Users can enter the function expression into the text entry field below the ‘Predefined’ combo box. The entered string is checked after the Enter key was pressed and an error message shows up, if the function string is not accepted.

Set Parameters’ button opens a dialog for parameters settings, which will be explaned later.

5.6.2 Fitter Settings

‘Method’ combo box currently provides only two fit model choices: Chi-square and Binned Likelihood. The default one is Chi-square. The Binned Likelihood is recomended for bins with low statistics.

‘Linear Fit’ check button sets the use of Linear fitter when is selected. Otherwise the minimization is done by Minuit, i.e. fit option “F” is applied. The Linear fitter can be selected only for functions linears in parameters (for example - polN).

‘Robust’ number entry sets the robust value when fitting graphs.

‘No Chi-square’ check button switch On/Off the fit option “C” - do not calculate Chi-square (for Linear fitter).

‘Integral’ check button switch On/Off the option “I” - use integral of function instead of value in bin center.

‘Best Errors’ sets On/Off the option “E” - better errors estimation by using Minos technique.

‘All weights = 1’ sets On/Off the option “W”- all weights set to 1 excluding empty bins; error bars ignored.

‘Empty bins, weights=1’ sets On/Off the option “WW” - all weights equal to 1 including empty bins; error bars ignored.

‘Use range’ sets On/Off the option “R” - fit only data within the specified function range. Sliders settings are used if this option is set to On. Users can change the function range values by pressing the left mouse button near to the left/right slider edges. It is possible to change both values simultaneously by pressing the left mouse button near to the slider center and moving it to a new position.

‘Improve fit results’ sets On/Off the option “M”- after minimum is found, search for a new one.

‘Add to list’ sets On/Off the option “+”- add function to the list without deleting the previous one. When fitting a histogram, the function is attached to the histogram’s list of functions. By default, the previously fitted function is deleted and replaced with the most recent one, so the list only contains one function. Setting this option to On will add the newly fitted function to the existing list of functions for the histogram. Note that the fitted functions are saved with the histogram when it is written to a ROOT file. By default, the function is drawn on the pad displaying the histogram.

5.6.3 Draw Options

‘SAME’ sets On/Off function drawing on the same pad. When a fit is executed, the image of the function is drawn on the current pad.

‘No drawing’ sets On/Off the option “0”- do not draw the fit results.

‘Do not store/draw’ sets On/Off option “N”- do not store the function and do not draw it.

This set of options specifies the amount of feedback printed on the root command line after performed fits.

‘Verbose’ - prints fit results after each iteration.

‘Quiet’ - no fit information is printed.

‘Default’ - between Verbose and Quiet.

5.6.5 Command Buttons

Fit button - performs a fit taking different option settings via the Fit Panel interface.

Reset - sets the GUI elements and related fit settings to the default ones.

Close - closes the Fit panel window.

5.7 Fitting Multiple Sub Ranges

The script for this example is $ROOTSYS/tutorials/fit/multifit.C. It shows how to use several Gaussian functions with different parameters on separate sub ranges of the same histogram. To use a Gaussian, or any other ROOT built in function, on a sub range you need to define a new TF1. Each is ‘derived’ from the canned function gaus.

Fitting a histogram with several Gaussian functions

Fitting a histogram with several Gaussian functions

First, four TF1 objects are created - one for each sub-range:

   g1 = new TF1("m1","gaus",85,95);
   g2 = new TF1("m2","gaus",98,108);
   g3 = new TF1("m3","gaus",110,121);
   // The total is the sum of the three, each has 3 parameters
   total = new TF1("mstotal","gaus(0)+gaus(3)+gaus(6)",85,125);

Next, we fill a histogram with bins defined in the array x.

   // Create a histogram and set it's contents
   h = new TH1F("g1","Example of several fits in subranges",
   for (int i=0; i<np; i++) {
   // Define the parameter array for the total function
   Double_t par[9];

When fitting simple functions, such as a Gaussian, the initial values of the parameters are automatically computed by ROOT. In the more complicated case of the sum of 3 Gaussian functions, the initial values of parameters must be set. In this particular case, the initial values are taken from the result of the individual fits. The use of the “+” sign is explained below:

   // Fit each function and add it to the list of functions

   // Get the parameters from the fit

   // Use the parameters on the sum

5.8 Adding Functions to the List

The example $ROOTSYS/tutorials/fit/multifit.C also illustrates how to fit several functions on the same histogram. By default a Fit command deletes the previously fitted function in the histogram object. You can specify the option “+” in the second parameter to add the newly fitted function to the existing list of functions for the histogram.

root[] hist->Fit("f1","+","",-2,2)

Note that the fitted function(s) are saved with the histogram when it is written to a ROOT file.

5.9 Combining Functions

You can combine functions to fit a histogram with their sum as it is illustrated in the macro FitDemo.C ($ROOTSYS/tutorials/fit/FittingDemo.C). We have a function that is the combination of a background and Lorentzian peak. Each function contributes 3 parameters:

\[ y(E) = a_{1} + a_{2}E + a_{3}E^{2} + \frac{A_{p}(\frac{G}{2p})} {(E-m)^{2} + (\frac{G}{2})^2 } \]

BackgroundLorentzian Peak

par[0] = \(a_{1}\) par[0] = \(A_{p}\)

par[1] = \(a_{2}\) par[1] = \(G\)

par[2] = \(a_{3}\) par[2] = \(m\)

The combination function (fitFunction) has six parameters:

fitFunction = background(x,par) + LorentzianPeak(x,&par[3])

par[0]=\(a_{1}\) par[1]=\(a_{2}\) par[2]=\(a_{3}\) par[3]=\(A_{p}\) par[4]=\(G\) par[5]=\(m\)

This script creates a histogram and fits it with the combination of two functions. First we define the two functions and the combination function:

   // Quadratic background function
   Double_t background(Double_t *x, Double_t *par) {
      return par[0] + par[1]*x[0] + par[2]*x[0]*x[0];

   // Lorentzian Peak function
   Double_t lorentzianPeak(Double_t *x, Double_t *par) {
      return (0.5*par[0]*par[1]/TMath::Pi()) / TMath::Max(1.e-10,
      (x[0]-par[2])*(x[0]-par[2])+ .25*par[1]*par[1]);

   // Sum of background and peak function
   Double_t fitFunction(Double_t *x, Double_t *par) {
      return background(x,par) + lorentzianPeak(x,&par[3]);

   void FittingDemo() {
   // bevington exercise by P. Malzacher, modified by R. Brun
   const int nBins = 60;
   Stat_t data[nBins] = {  6, 1,10,12, 6,13,23,22,15,21,
   TH1F *histo = new TH1F("example_9_1",
   "Lorentzian Peak on Quadratic Background",60,0,3);

   for(int i=0; i < nBins;  i++) {
      // we use these methods to explicitly set the content
      // and error instead of using the fill method.
   // create a TF1 with the range from 0 to 3 and 6 parameters
   TF1 *fitFcn = new TF1("fitFcn",fitFunction,0,3,6);

   // first try without starting values for the parameters
   // this defaults to 1 for each param.
   // this results in an ok fit for the polynomial function however
   // the non-linear part (Lorentzian
The output of the FittingDemo() example

The output of the FittingDemo() example

5.10 Associated Function

One or more objects (typically a TF1*) can be added to the list of functions (fFunctions) associated to each histogram. A call to TH1::Fit adds the fitted function to this list. Given a histogram h, one can retrieve the associated function with:

   TF1 *myfunc = h->GetFunction("myfunc");

5.11 Access to the Fit Parameters and Results

If the histogram (or graph) is made persistent, the list of associated functions is also persistent. Retrieve a pointer to the function with the TH1::GetFunction()method. Then you can retrieve the fit parameters from the function (**TF1`**) with calls such as:

root[] TF1 *fit = hist->GetFunction(function_name);
root[] Double_t chi2 = fit->GetChisquare();
// value of the first parameter
root[] Double_t p1 = fit->GetParameter(0);
// error of the first parameter
root[] Double_t e1 = fit->GetParError(0);

5.12 Associated Errors

By default, for each bin, the sum of weights is computed at fill time. One can also call TH1::Sumw2 to force the storage and computation of the sum of the square of weights per bin. If Sumw2 has been called, the error per bin is computed as the sqrt(sum of squares of weights); otherwise, the error is set equal to the sqrt(bin content). To return the error for a given bin number, do:

   Double_t error = h->GetBinError(bin);

Empty bins are excluded in the fit when using the Chi-square fit method. When fitting the histogram with the low statistics, it is recommended to use the Log-Likelihood method (option ‘L’ or “LL”).

5.13 Fit Statistics

You can change the statistics box to display the fit parameters with the TStyle::SetOptFit(mode) method. This parameter has four digits: mode = pcev (default = 0111)

For example, to print the fit probability, parameter names/values, and errors, use:


5.14 The Minimization Package

This package was originally written in FORTRAN by Fred James and part of PACKLIB (patch D506). It has been converted to a C++ class by René Brun. The current implementation in C++ is a straightforward conversion of the original FORTRAN version. The main changes are:

   gMinuit->Command("SCAn 1");
   TGraph *gr = (TGraph*)gMinuit->GetPlot();

5.14.1 Basic Concepts of Minuit

The Minuit package acts on a multi parameter FORTRAN function to which one must give the generic name FCN. In the ROOT implementation, the function FCN is defined via the Minuit SetFCN member function when a HistogramFit command is invoked. The value of FCN will in general depend on one or more variable parameters.

To take a simple example, in case of ROOT histograms (classes TH1C, TH1S, TH1F, TH1D) the Fit function defines the Minuit fitting function as being H1FitChisquare or H1FitLikelihood depending on the options selected. H1FitChisquare calculates the chi-square between the user fitting function (Gaussian, polynomial, user defined, etc) and the data for given values of the parameters. It is the task of Minuit to find those values of the parameters which give the lowest value of chi-square.

5.14.2 The Transformation of Limited Parameters

For variable parameters with limits, Minuit uses the following transformation:

Pint = arcsin(2((Pext-a)/(b-a))-1)

Pext = a+((b-a)/(2))(sinPint+1)

so that the internal value Pint can take on any value, while the external value Pext can take on values only between the lower limit a and the ext upper limit b. Since the transformation is necessarily non-linear, it would transform a nice linear problem into a nasty non-linear one, which is the reason why limits should be avoided if not necessary. In addition, the transformation does require some computer time, so it slows down the computation a little bit, and more importantly, it introduces additional numerical inaccuracy into the problem in addition to what is introduced in the numerical calculation of the FCN value. The effects of non-linearity and numerical round off both become more important as the external value gets closer to one of the limits (expressed as the distance to nearest limit divided by distance between limits). The user must therefore be aware of the fact that, for example, if he puts limits of (0, 1010) on a parameter, then the values 0.0 and 1. 0 will be indistinguishable to the accuracy of most machines.

The transformation also affects the parameter error matrix, of course, so Minuit does a transformation of the error matrix (and the ‘’parabolic’’ parameter errors) when there are parameter limits. Users should however realize that the transformation is only a linear approximation, and that it cannot give a meaningful result if one or more parameters is very close to a limit, where \(\frac{\partial Pext}{\partial Pint} \neq 0\). Therefore, it is recommended that:

5.14.3 How to Get the Right Answer from Minuit

Minuit offers the user a choice of several minimization algorithms. The MIGRAD algorithm is in general the best minimized for nearly all functions. It is a variable-metric method with inexact line search, a stable metric updating scheme, and checks for positive-definiteness. Its main weakness is that it depends heavily on knowledge of the first derivatives, and fails miserably if they are very inaccurate.

If parameter limits are needed, in spite of the side effects, then the user should be aware of the following techniques to alleviate problems caused by limits: Getting the Right Minimum with Limits

If MIGRAD converges normally to a point where no parameter is near one of its limits, then the existence of limits has probably not prevented Minuit from finding the right minimum. On the other hand, if one or more parameters is near its limit at the minimum, this may be because the true minimum is indeed at a limit, or it may be because the minimized has become ‘’blocked’’ at a limit. This may normally happen only if the parameter is so close to a limit (internal value at an odd multiple of \(\pm \frac{\pi}{2}\) that Minuit prints a warning to this effect when it prints the parameter values. The minimized can become blocked at a limit, because at a limit the derivative seen by the minimized \(\frac{\partial F}{\partial Pint}\) is zero no matter what the real derivative \(\frac{\partial F}{\partial Pext}\) is.

\[ \left(\frac{\partial F}{\partial Pint}\right) = \left(\frac{\partial F}{\partial Pext}\right) \left(\frac{\partial Pext}{\partial Pint}\right) = \left(\frac{\partial F}{\partial Pext}\right) = 0 \] Getting the Right Parameter Errors with Limits

In the best case, where the minimum is far from any limits, Minuit will correctly transform the error matrix, and the parameter errors it reports should be accurate and very close to those you would have got without limits. In other cases (which should be more common, since otherwise you would not need limits), the very meaning of parameter errors becomes problematic. Mathematically, since the limit is an absolute constraint on the parameter, a parameter at its limit has no error, at least in one direction. The error matrix, which can assign only symmetric errors, then becomes essentially meaningless. Interpretation of Parameter Errors

There are two kinds of problems that can arise: the reliability of Minuit’s error estimates, and their statistical interpretation, assuming they are accurate. Statistical Interpretation

For discussion of basic concepts, such as the meaning of the elements of the error matrix, or setting of exact confidence levels see the articles:

5.14.4 Reliability of Minuit Error Estimates

Minuit always carries around its own current estimates of the parameter errors, which it will print out on request, no matter how accurate they are at any given point in the execution. For example, at initialization, these estimates are just the starting step sizes as specified by the user. After a HESSE step, the errors are usually quite accurate, unless there has been a problem. Minuit, when it prints out error values, also gives some indication of how reliable it thinks they are. For example, those marked CURRENT GUESS ERROR are only working values not to be believed, and APPROXIMATE ERROR means that they have been calculated but there is reason to believe that they may not be accurate.

If no mitigating adjective is given, then at least Minuit believes the errors are accurate, although there is always a small chance that Minuit has been fooled. Some visible signs that Minuit may have been fooled:

The best way to be absolutely sure of the errors is to use ‘’independent’‘calculations and compare them, or compare the calculated errors with a picture of the function. Theoretically, the covariance matrix for a’‘physical’’ function must be positive-definite at the minimum, although it may not be so for all points far away from the minimum, even for a well-determined physical problem. Therefore, if MIGRAD reports that it has found a non-positive-definite covariance matrix, this may be a sign of one or more of the following: A Non-physical Region

On its way to the minimum, MIGRAD may have traversed a region that has unphysical behavior, which is of course not a serious problem as long as it recovers and leaves such a region. An Underdetermined Problem

If the matrix is not positive-definite even at the minimum, this may mean that the solution is not well defined, for example that there are more unknowns than there are data points, or that the parameterization of the fit contains a linear dependence. If this is the case, then Minuit (or any other program) cannot solve your problem uniquely. The error matrix will necessarily be largely meaningless, so the user must remove the under determinedness by reformulating the parameterization. Minuit cannot do this itself. Numerical Inaccuracies

It is possible that the apparent lack of positive-definiteness is due to excessive round off errors in numerical calculations (in the user function), or not enough precision. This is unlikely in general, but becomes more likely if the number of free parameters is very large, or if the parameters are badly scaled (not all of the same order of magnitude), and correlations are large. In any case, whether the non-positive-definiteness is real or only numerical is largely irrelevant, since in both cases the error matrix will be unreliable and the minimum suspicious. An Ill-posed Problem

For questions of parameter dependence, see the discussion above on positive-definiteness. Possible other mathematical problems are the following:

5.15 FUMILI Minimization Package

FUMILI is used to minimize Chi-square function or to search maximum of likelihood function. Experimentally measured values \(F_{i}\) are fitted with theoretical functions \(f_{i}(\vec{x_{i}},\vec{\theta})\), where \(\vec{x_{i}}\) are coordinates, and \(\vec{\theta}\) - vector of parameters. For better convergence Chi-square function has to be the following form

\[ \frac{\chi^2}{2} = \frac{1}{2} \sum_{i=1}^{n} \left(\frac{f_{i}(\vec{x_{i}},\vec{\theta}) - F_{i}} {\sigma_{i}}\right)^{2} \]

where \(\sigma_{i}\) are errors of the measured function. The minimum condition is:

\[ \frac{\partial \chi^{2}}{\partial \theta_{i}} = \sum_{j=1}^{n} \frac{1}{\sigma_{j}^{2}} . \frac{\partial f_{i}}{\partial \theta_{i}} \left[ (\vec{x_{j}},\vec{\theta}) - F_{j}\right] = 0, i = 1 ... m \]

where \(m\) is the quantity of parameters. Expanding left part of this equation over parameter increments and retaining only linear terms one gets

\[ \left(\frac{\partial \chi^{2}}{\theta_{i}}\right) _{\theta = \vec{\theta}^{0}} + \sum_{k} \left(\frac{\partial^{2} \chi^{2}}{\partial \theta_{i} \partial \theta_{k}}\right) _{\theta = \vec{\theta}^{0}} . (\theta_{k} - \theta_{k}^{0}) = 0 \]

here \(\vec{\theta}^{0}\) is some initial value of parameters. In general case:

\[ {\partial^2\chi^2\over\partial\theta_i\partial\theta_k}= \sum^n_{j=1}{1\over\sigma^2_j} {\partial f_j\over\theta_i} {\partial f_j\over\theta_k} + \sum^n_{j=1}{(f_j - F_j)\over\sigma^2_j}\cdot {\partial^2f_j\over\partial\theta_i\partial\theta_k} \]

In FUMILI algorithm for second derivatives of Chi-square approximate expression is used when last term in previous equation is discarded. It is often done, not always wittingly, and sometimes causes troubles, for example, if user wants to limit parameters with positive values by writing down \(\theta_i^2\) instead of \(\theta_i\). FUMILI will fail if one tries minimize \(\chi^2 = g^2(\vec\theta)\) where g is arbitrary function.

Approximate value is:

\[ {\partial^2\chi^2\over\partial\theta_i\partial\theta_k}\approx Z_{ik}= \sum^n_{j=1}{1\over\sigma^2_j}{\partial f_j\over\theta_i} {\partial f_j\over\theta_k} \]

Then the equations for parameter increments are:

\[ \left(\partial\chi^2\over\partial\theta_i\right)_ {\vec\theta={\vec\theta}^0} +\sum_k Z_{ik}\cdot(\theta_k-\theta^0_k) = 0, \qquad i=1\ldots m \]

Remarkable feature of algorithm is the technique for step restriction. For an initial value of parameter \({\vec\theta}^0\) a parallelepiped \(P_0\) is built with the center at \({\vec\theta}^0\) and axes parallel to coordinate axes \(\theta_i\). The lengths of parallelepiped sides along i-th axis is \(2b_i\), where \(b_i\) is such a value that the functions \(f_j(\vec\theta)\) are quasi-linear all over the parallelepiped.

FUMILI takes into account simple linear inequalities in the form:

\[ \theta_i^{\rm min}\le\theta_i\le\theta^{\rm max}_i\]

They form parallelepiped \(P\) (\(P_0\) may be deformed by \(P\)). Very similar step formulae are used in FUMILI for negative logarithm of the likelihood function with the same idea - linearization of function argument.

5.16 Neural Networks

5.16.1 Introduction

Neural Networks are used in various fields for data analysis and classification, both for research and commercial institutions. Some randomly chosen examples are image analysis, financial movements’ predictions and analysis, or sales forecast and product shipping optimization. In particles physics neural networks are mainly used for classification tasks (signal over background discrimination). A vast majority of commonly used neural networks are multilayer perceptrons. This implementation of multilayer perceptrons is inspired from the MLPfit package, which remains one of the fastest tools for neural networks studies.

5.16.2 The MLP

The multilayer perceptron is a simple feed-forward network with the following structure showed on the left.

It is made of neurons characterized by a bias and weighted links in between - let’s call those links synapses. The input neurons receive the inputs, normalize them and forward them to the first hidden layer. Each neuron in any subsequent layer first computes a linear combination of the outputs of the previous layer. The output of the neuron is then function of that combination with f being linear for output neurons or a sigmoid for hidden layers.

Such a structure is very useful because of two theorems:

1- A linear combination of sigmoids can approximate any continuous function.

2- Trained with output=1 for the signal and 0 for the background, the approximated function of inputs X is the probability of signal, knowing X.

5.16.3 Learning Methods

The aim of all learning methods is to minimize the total error on a set of weighted examples. The error is defined as the sum in quadrate, divided by two, of the error on each individual output neuron. In all methods implemented in this library, one needs to compute the first derivative of that error with respect to the weights. Exploiting the well-known properties of the derivative, one can express this derivative as the product of the local partial derivative by the weighted sum of the outputs derivatives (for a neuron) or as the product of the input value with the local partial derivative of the output neuron (for a synapse). This computation is called “back-propagation of the errors”. Six learning methods are implemented. Stochastic Minimization

This is the most trivial learning method. The Robbins-Monro stochastic approximation is applied to multilayer perceptrons. The weights are updated after each example according to the formula:

\[ w_{ij}(t+1) = w_{ij}(t) + \Delta w_{ij}(t) \]


\[ \Delta w_{ij}(t) = - \eta \left( \frac{\partial e_p}{\partial w_{ij}} + \delta \right) + \epsilon \Delta w_{ij}(t-1) \]

The parameters for this method are Eta, EtaDecay, Delta and Epsilon. Steepest Descent With Fixed Step Size (Batch Learning)

It is the same as the stochastic minimization, but the weights are updated after considering all the examples, with the total derivative dEdw. The parameters for this method are Eta, EtaDecay, Delta and Epsilon. Steepest Descent Algorithm

Weights are set to the minimum along the line defined by the gradient. The only parameter for this method is Tau. Lower Tau = higher precision = slower search. A value Tau=3 seems reasonable. Conjugate Gradients With the Polak-Ribiere Updating Formula

Weights are set to the minimum along the line defined by the conjugate gradient. Parameters are Tau and Reset, which defines the epochs where the direction is reset to the steepest descent (estimated by using the Polak-Ribiere formula). Conjugate Gradients With the Fletcher-Reeves Updating Formula

Weights are set to the minimum along the line defined by the conjugate gradient. Parameters are Tau and Reset, which defines the epochs where the direction is reset to the steepest descent (estimated by using the Fletcher-Reeves formula). The Broyden, Fletcher, Goldfarb, Shanno (BFGS) Method

It implies the computation of a NxN matrix, but seems more powerful at least for less than 300 weights. Parameters are Tau and Reset, which defines the epochs where the direction is reset to the steepest descent.

5.16.4 Using the Network

Neural network are build from a set of “samples”. A sample is a set of values defining the inputs and the corresponding output that the network should ideally provide. In ROOT this is a TTree entry. The first thing to be decided is the network layout. This layout is described in a string where the layers are separated by semicolons. The input/output layers are defined by giving the expression for each neuron, separated by comas. Hidden layers are just described by the number of neurons.

In addition, input and output layer formulas can be preceded by ‘@’ (e.g. “@out”) if one wants to normalize the corresponding value. Also, if the string ends with ‘!’, output neurons are set up for classification, i.e. with a sigmoid (1 neuron) or softmax (more neurons) activation function.

Many questions on the good usage of neural network, including rules of dumb to determine the best network topology are addressed at

   // a simple network: 2 inputs, 10 hidden and 1 normalized
   // output neuron
   TMultiLayerPerceptron network("r,z:10:@Br",tree);

Expressions are evaluated as for TTree::Draw(). Input and outputs are taken from the TTree associated with the network. This TTree can be given as argument of the constructor or defined later with TMultiLayerPerceptron::SetData(). Events can also be weighted. The weight expression can be given in the constructor or set later with the method SetWeight() of the class TMultiLayerPerceptron. Two datasets must be defined before learning the network: a training dataset that is used when minimizing the error, and a test dataset that will avoid bias. Those two datasets can be build aside and then given to the network, or can be build from a standard expression. By default, half of the events are put in both datasets.

   // a more complex 4:8:1 network
   // the ptsumf branch is used as weigh;
   // default event lists are explicit
   TMultiLayerPerceptron  network("m,pt,acol,acopl:8:type","pt",tree,

The method TMultiLayerPerceptron::SetLearningMethod() defines the learning method. Learning methods are:

TMultiLayerPerceptron::kBFGS      // default

The training can start with TMultiLayerPerceptron::Train(Int_t nepoch,Option_t* options). The first argument is the number of epochs while option is a string that can contain “text” (simple text output), “graph” (evaluating graphical training curves), “update = X” (step for the text/graph output update) or “+” (will skip the randomization and start from the previous values). All combinations are available.

   network.Train(1000,"text,graph,update=10"); // full output every
                                               // 10 epochs
   network.Train(100,"text,+");                // 100 more epochs
   //starts with existing weights

The weights can be saved to a file (DumpWeights) and then reloaded (LoadWeights) to a new compatible network. The output can also be evaluated (Evaluate) for a given output neuron and an array of double input parameters or the network can be exported (Export) as a standalone code. Up to now, this is only as a C++ or PYTHON class, but other languages could be implemented.

5.16.5 Examples

An example of how to use TMultiLayerPerceptron is the macro mlpHiggs.C in $ROOTSYS/tutorials. Using some standard simulated information that could have been obtained at LEP, a neural network is build, which can make the difference between WW events and events containing a Higgs boson. Starting with a TFile containing two TTrees: one for the signal, the other for the background, a simple script is used:

   void mlpHiggs(Int_t ntrain=100) {
      if (!gROOT->GetClass("TMultiLayerPerceptron"))
      // prepare inputs - the 2 trees are merged into one, and a
      // "type" branch, equal to 1 for the signal and 0 for the
      // background is added
      TFile input("mlpHiggs.root");
      TTree *signal = (TTree *)input.Get("sig_filtered");
      TTree *background = (TTree *)input.Get("bg_filtered");
      TTree *simu = new TTree("MonteCarlo",
                              "Filtered Monte Carlo Events");

Since the input is a TTree and we are starting from two different TTrees (with different names), they are first merged into one, and a “type” branch is added, that says whether there is a signal or a background event. Those irrelevant details are skipped here.

      TMultiLayerPerceptron *mlp = new TMultiLayerPerceptron(
            "msumf,ptsumf, acolin, acopl:8:type","ptsumf",simu,
      mlp->Train(ntrain, "text,graph,update=10");

The neural network is instantiated and trained. “ptsumf” is used as a weight, and the standard event lists are explicit. The network that is then build has four input neurons, eight additional ones in the only hidden layer and one single output neuron.

      // Use the NN to plot the results for each sample
      TH1F *bg = new TH1F("bgh","NN output",50,-.5,1.5);
      TH1F *sig = new TH1F("sigh","NN output",50,-.5,1.5);
      Double_t params[4];
      for (i = 0; i < background->GetEntries(); i++) {
         params[0] = msumf;    params[1] = ptsumf;
         params[2] = acolin;   params[3] = acopl;
      for (i = 0; i < signal->GetEntries(); i++) {
         params[0] = msumf;
         params[1] = ptsumf;
         params[2] = acolin;
         params[3] = acopl;
      TCanvas *cv = new TCanvas("NNout_cv","Neural net output");
      TLegend *legend = new TLegend(.75,.80,.95,.95);

The neural net output is then used to display the final difference between background and signal events. The figure “The neural net output” shows this plot.

The neural net output

The neural net output

As it can be seen, this is a quite efficient technique. As mentioned earlier, neural networks are also used for fitting function. For some application with a cylindrical symmetry, a magnetic field simulation gives as output the angular component of the potential vector A, as well as the radial and z components of the B field.

One wants to fit those distributions with a function in order to plug them into the Geant simulation code. Polynomial fits could be tried, but it seems difficult to reach the desired precision over the full range. One could also use a spline interpolation between known points. In all cases, the resulting field would not be C-infinite.

An example of output (for Br) is shown. First the initial function can be seen as the target. Then, the resulting (normalized) neural net output. In order to ease the learning, the “normalize output” was used here. The initial amplitude can be recovered by multiplying by the original RMS and then shifting by the original mean.

The original and the neural net for Br

The original and the neural net for Br

6 A Little C++

This chapter introduces you to some useful insights into C++, to allow you to use of the most advanced features in ROOT. It is in no case a full course in C++.

6.1 Classes, Methods and Constructors

C++ extends C with the notion of class. If you’re used to structures in C, a class is a struct that is a group of related variables, which is extended with functions and routines specific to this structure (class). What is the interest? Consider a struct that is defined this way:

   struct Line {
      float x1;
      float y1;
      float x2;
      float y2;

This structure represents a line to be drawn in a graphical window. (x1,y1) are the coordinates of the first point, (x2,y2) the coordinates of the second point. In the standard C, if you want to draw effectively such a line, you first have to define a structure and initialize the points (you can try this):

   Line firstline;
   firstline.x1 = 0.2;
   firstline.y1 = 0.2;
   firstline.x2 = 0.8;
   firstline.y2 = 0.9;

This defines a line going from the point (0.2,0.2) to the point (0.8,0.9). To draw this line, you will have to write a function, say LineDraw(Line l) and call it with your object as argument:


In C++, we would not do that. We would instead define a class like this:

   class TLine {
      Double_t x1;
      Double_t y1;
      Double_t x2;
      Double_t y2;
      TLine(int x1, int y1, int x2, int y2);
      void Draw();

Here we added two functions, that we will call methods or member functions, to the TLine class. The first method is used for initializing the line objects we would build. It is called a constructor. The second one is the Draw method itself. Therefore, to build and draw a line, we have to do:

   TLine l(0.2,0.2,0.8,0.9);

The first line builds the object lby calling its constructor. The second line calls the TLine::Draw() method of this object. You don’t need to pass any parameters to this method since it applies to the object l, which knows the coordinates of the line. These are internal variables x1, y1, x2, y2 that were initialized by the constructor.

6.2 Inheritance and Data Encapsulation

We have defined a TLine class that contains everything necessary to draw a line. If we want to draw an arrow, is it so different from drawing a line? We just have to draw a triangle at one end. It would be very inefficient to define the class TArrow from scratch. Fortunately, inheritance allows a class to be defined from an existing class. We would write something like:

   class TArrow : public TLine {
      int ArrowHeadSize;
      void Draw();
      void SetArrowSize(int arrowsize);

The keyword “public” will be explained later. The class TArrow now contains everything that the class TLine does, and a couple of things more, the size of the arrowhead and a function that can change it. The Draw method of TArrow will draw the head and call the draw method of TLine. We just have to write the code for drawing the head!

6.2.1 Method Overriding

Giving the same name to a method (remember: method = member function of a class) in the child class (TArrow) as in the parent (TLine) does not give any problem. This is called overriding a method. Draw in TArrow overrides Draw in TLine. There is no possible ambiguity since, when one calls the Draw() method; this applies to an object which type is known. Suppose we have an object l of type TLine and an object a of type TArrow. When you want to draw the line, you do:


Draw() from TLine is called. If you do:


Draw() from TArrow is called and the arrow a is drawn.

6.2.2 Data Encapsulation

We have seen previously the keyword “public”. This keyword means that every name declared public is seen by the outside world. This is opposed to “private” that means only the class where the name was declared private could see this name. For example, suppose we declare in TArrow the variable ArrowHeadSize private.

      int ArrowHeadSize;

Then, only the methods (i.e. member functions) of TArrow will be able to access this variable. Nobody else will see it. Even the classes that we could derive from TArrow will not see it. On the other hand, if we declare the method Draw() as public, everybody will be able to see it and use it. You see that the character public or private does not depend of the type of argument. It can be a data member, a member function, or even a class. For example, in the case of TArrow, the base class TLine is declared as public:

   class TArrow : public TLine { ...

This means that all methods of TArrow will be able to access all methods of TLine, but this will be also true for anybody in the outside world. Of course, this is true if TLine accepts the outside world to see its methods/data members. If something is declared private in TLine, nobody will see it, not even TArrow members, even if TLine is declared as a public base class.

What if TLine is declared “private” instead of “public” ? Well, it will behave as any other name declared private in TArrow: only the data members and methods of TArrow will be able to access TLine, its methods and data members, nobody else. This may seem a little bit confusing and readers should read a good C++ book if they want more details. Especially since, besides public and private, a member can be protected. Usually, one puts private the methods that the class uses internally, like some utilities classes, and that the programmer does not want to be seen in the outside world.

With “good” C++ practice (which we have tried to use in ROOT), all data members of a class are private. This is called data encapsulation and is one of the strongest advantages of Object Oriented Programming (OOP). Private data members of a class are not visible, except to the class itself. So, from the outside world, if one wants to access those data members, one should use so called “getters” and “setters” methods, which are special methods used only to get or set the data members. The advantage is that if the programmers want to modify the inner workings of their classes, they can do so without changing what the user sees. The user does not even have to know that something has changed (for the better, hopefully). For example, in our TArrow class, we would have set the data member ArrowHeadSize private. The setter method is SetArrowSize(), we do not need a getter method:

   class TArrow : public TLine {
         int ArrowHeadSize;
         void Draw();
         void SetArrowSize(int arrowsize);

To define an arrow object you call the constructor. This will also call the constructor of TLine, which is the parent class of TArrow, automatically. Then we can call any of the line or arrow public methods:

root[] TArrow *myarrow = new TArrow(1,5,89,124);
root[] myarrow->SetArrowSize(10);
root[] myarrow->Draw();

6.3 Creating Objects on the Stack and Heap

To explain how objects are created on the stack and on the heap we will use the Quad class. You can find the definition in $ROOTSYS/tutorials/quadp/Quad.h and Quad.cxx. The Quad class has four methods. The constructor and destructor, Evaluate that evaluates ax**2 + bx +c, and Solve which solves the quadratic equation ax**2 + bx +c = 0.

Quad.h :

   class Quad {
         Quad(Float_t a, Float_t b, Float_t c);
         Float_t Evaluate(Float_t x) const;
         void Solve() const;
         Float_t fA;
         Float_t fB;
         Float_t fC;


   #include <iostream.h>
   #include <math.h>
   #include "Quad.h"

   Quad::Quad(Float_t a, Float_t b, Float_t c) {
      fA = a;
      fB = b;
      fC = c;
   Quad::~Quad() {
      Cout <<"deleting object with coeffts: "<< fA << "," << fB << ","
                                             << fC << endl;
   Float_t Quad::Evaluate(Float_t x) const {
      return fA*x*x + fB*x + fC;
   void Quad::Solve() const {
      Float_t temp = fB*fB - 4.*fA*fC;
      if ( temp > 0. ) {
         temp = sqrt( temp );
         cout << "There are two roots: " << ( -fB - temp ) / (2.*fA)
         << " and " << ( -fB + temp ) / (2.*fA) << endl;
      } else {
         if ( temp == 0. ) {
            cout << "There are two equal roots: " << -fB / (2.*fA)
                                                  << endl;
         } else {
            cout << "There are no roots" << endl;

Let us first look how we create an object. When we create an object by:

root[] Quad my_object(1.,2.,-3.);

We are creating an object on the stack. A FORTRAN programmer may be familiar with the idea; it is not unlike a local variable in a function or subroutine. Although there are still a few old timers who do not know it, FORTRAN is under no obligation to save local variables once the function or subroutine returns unless the SAVE statement is used. If not then it is likely that FORTRAN will place them on the stack and they will “pop off” when the RETURN statement is reached. To give an object more permanence it has to be placed on the heap.

root[] .L Quad.cxx
root[] Quad *my_objptr = new Quad(1.,2.,-3.);

The second line declares a pointer to Quad called my_objptr. From the syntax point of view, this is just like all the other declarations we have seen so far, i.e. this is a stack variable. The value of the pointer is set equal to

new Quad(1.,2.,-3.);

new, despite its looks, is an operator and creates an object or variable of the type that comes next, Quad in this case, on the heap. Just as with stack objects it has to be initialized by calling its constructor. The syntax requires that the argument list follow the type. This one statement has brought two items into existence, one on the heap and one on the stack. The heap object will live until the delete operator is applied to it.

There is no FORTRAN parallel to a heap object; variables either come or go as control passes in and out of a function or subroutine, or, like a COMMON block variables, live for the lifetime of the program. However, most people in HEP who use FORTRAN will have experience of a memory manager and the act of creating a bank is a good equivalent of a heap object. For those who know systems like ZEBRA, it will come as a relief to learn that objects do not move, C++ does not garbage collect, so there is never a danger that a pointer to an object becomes invalid for that reason. However, having created an object, it is the user’s responsibility to ensure that it is deleted when no longer needed, or to pass that responsibility onto to some other object. Failing to do that will result in a memory leak, one of the most common and most hard-to-find C++ bugs.

To send a message to an object via a pointer to it, you need to use the “->” operator e.g.:

root[] my_objptr->Solve();

Although we chose to call our pointer my_objptr, to emphasize that it is a pointer, heap objects are so common in an object-oriented program that pointer names rarely reflect the fact - you have to be careful that you know if you are dealing with an object or its pointer! Fortunately, the compiler won’t tolerate an attempt to do something like:

root[] my_objptr.Solve();

Although this is a permitted by the CINT shortcuts, it is one that you are strongly advised not to follow! As we have seen, heap objects have to be accessed via pointers, whereas stack objects can be accessed directly. They can also be accessed via pointers:

root[] Quad stack_quad(1.,2.,-3.);
root[] Quad *stack_ptr = &stack_quad;
root[] stack_ptr->Solve();

Here we have a Quad pointer that has been initialized with the address of a stack object. Be very careful if you take the address of stack objects. As we shall see soon, they are deleted automatically, which could leave you with an illegal pointer. Using it will corrupt and may well crash the program!

It is time to look at the destruction of objects. A destructor is a special C++ function that releases resources for (or destroy) an object of a class. It is opposite of a constructor that create the object of a class when is called. The compiler will provide a destructor that does nothing if none is provided. We will add one to our Quad class so that we can see when it is called. The class names the destructor but with a prefix ~ which is the C++ one’s complement i.e. bit wise complement, and hence has destruction overtones! We declare it in the .h file and define it in the .cxx file. It does not do much except print out that it has been called (still a useful debug technique despite today’s powerful debuggers!).

Now run root, load the Quad class and create a heap object:

root[] .L Quad.cxx
root[] Quad *my_objptr = new Quad(1.,2.,-3.);

To delete the object:

root[] delete my_objptr;
root[] my_objptr = 0;

You should see the print out from its destructor. Setting the pointer to zero afterwards is not strictly necessary (and CINT does it automatically), but the object is no more accessible, and any attempt to use the pointer again will, as has already been stated, cause grief. So much for heap objects, but how are stack objects deleted? In C++, a stack object is deleted as soon as control leaves the innermost compound statement that encloses it. Therefore, it is singularly futile to do something like:

root[] {  Quad my_object(1.,2.,-3.); }

CINT does not follow this rule; if you type in the above line, you will not see the destructor message. As explained in the Script lesson, you can load in compound statements, which would be a bit pointless if everything disappeared as soon as it was loaded! Instead, to reset the stack you have to type:

root[] gROOT->Reset();

This sends the Reset message via the global pointer to the ROOT object, which, amongst its many roles, acts as a resource manager. Start ROOT again and type in the following:

root[] .L Quad.cxx
root[] Quad my_object(1.,2.,-3.);
root[] Quad *my_objptr = new Quad(4.,5.,-6.);
root[] gROOT->Reset();

You will see that this deletes the first object but not the second. We have also painted ourselves into a corner, as my_objptr was also on the stack. This command will fail.

root[] my_objptr->Solve();

CINT no longer knows what my_objptr is. This is a great example of a memory leak; the heap object exists but we have lost our way to access it. In general, this is not a problem. If any object will outlive the compound statement in which it was created then a more permanent pointer will point to it, which frequently is part of another heap object. See Resetting the Interpreter Environment in the chapter “CINT the C++ Interpreter”.

7 CINT the C++ Interpreter

The subject of this chapter is CINT, ROOT command line interpreter and script processor. First, we explain what CINT is and why ROOT uses it. Then we discuss CINT as the command line interpreter, the CINT commands, and CINT extensions to C++ are discussed. CINT as the script interpreter is explained and illustrated with several examples.

7.1 What is CINT?

CINT, which is pronounced ['sint], is a C++ interpreter. An interpreter takes a program, in this case a C++ program, and carries it out by examining each instruction and in turn executing the equivalent sequence of machine language. For example, an interpreter translates and executes each statement in the body of a loop “n” times. It does not generate a machine language program. This may not be a good example, because most interpreters have become ‘smart’ about loop processing.

A compiler on the other hand, takes a program and makes a machine language executable. Once compiled the execution is very fast, which makes a compiler best suited for the case of “built once, run many times”. For example, the ROOT executable is compiled occasionally and executed many times. It takes anywhere from 1 to 45 minutes to compile ROOT for the first time (depending on the CPU). Once compiled it runs very fast. On the average, a compiled program runs roughly ten times faster than an interpreted one. Because compiling is slow, using a compiler is cumbersome for rapid prototyping when one changes and rebuilds as often as once per minute. An interpreter, on the other hand, is the perfect tool for code that changes often and runs a few times. Most of the time, interpreters are built for scripting languages, such as JavaScript, IDL, or Python. These languages are specifically designed to be interpreted rather than compiled. The advantage of using a normally compiled language is that code can be compiled once the prototype is debugged and refined. CINT is a C++ interpreter, making it a tool for rapid prototyping and scripting in C++. It is also available as a stand-alone product, see This page also has links to all the CINT documentation. The downloadable tar file contains documentation, the CINT executable, and many demo scripts that are not included in the regular ROOT distribution. Here is the list of CINT main features:

7.2 The ROOT Command Line Interface

Start up a ROOT session by typing root at the system prompt.

> root
*                                         *
*        W E L C O M E  to  R O O T       *
*                                         *
*   Version   5.16/00      27 June 2007   *
*                                         *
*  You are welcome to visit our Web site  *
*            *
*                                         *
FreeType Engine v2.1.9 used to render TrueType fonts.
Compiled on 28 June 2007 for linux with thread support.

CINT/ROOT C/C++ Interpreter version 5.16.21, June 22, 2007
Type ? for help. Commands must be C++ statements.
Enclose multiple statements between { }.

Now we create a TLine object:

root[] TLine l
root[] l.Print()
TLine  X1=0.000000 Y1=0.000000 X2=0.000000 Y2=0.000000
root[] l.SetX1(10)
root[] l.SetY1(11)
root[] l.Print()
TLine  X1=10.000000 Y1=11.000000 X2=0.000000 Y2=0.000000
root[] .g
0x4038f080 class TLine l , size=40
0x0        protected: Double_t fX1 //X of 1st point
0x0        protected: Double_t fY1 //Y of 1st point
0x0        protected: Double_t fX2 //X of 2nd point
0x0        protected: Double_t fY2 //Y of 2nd point
0x0        private: static class TClass* fgIsA

Here we note:

root[] .class TLine
class TLine //A line segment
List of base class-------------------------------
0x0        public: TObject //Basic ROOT object
0xc        public: TAttLine //Line attributes
List of member variable--------------------------
Defined in TLine
(compiled) 0x0        protected: Double_t fX1 //X of 1st point
(compiled) 0x0        protected: Double_t fY1 //Y of 1st point
(compiled) 0x0        protected: Double_t fX2 //X of 2nd point
(compiled) 0x0        protected: Double_t fY2 //Y of 2nd point
(compiled) 0x8a3a718  static const enum TLine:: kLineNDC
(compiled) 0x0        private: static TClass* fgIsA
List of member function--------------------------
filename       line:size busy function type and name  (in TLine)
(compiled) 0:0    0 public: virtual void ~TLine(void);
(compiled) 0:0    0 public: TLine TLine(void);
(compiled) 0:0    0 public: TLine TLine(Double_t x1,Double_t y1,
                                        Double_t x2,Double_t y2);
(compiled) 0:0    0 public: TLine TLine(const TLine& line);
(compiled) 0:0    0 public: virtual void Copy(TObject& line) const;
(compiled) 0:0    0 public: virtual Int_t DistancetoPrimitive(
                                                  Int_t px,Int_t py);
(compiled) 0:0    0 public: static int ImplFileLine(void);
(compiled) 0:0    0 public: static const char* ImplFileName(void);
(compiled) 0:0    0 public: static int DeclFileLine(void);
(compiled) 0:0    0 public:TLine& operator=(const TLine&);
root[] l.Print(); > test.log
root[] l.Dump(); >> test.log
root[] ?

Here we see:

Now let us execute a multi-line command:

root[] {
end with '}', '@':abort > TLine l;
end with '}', '@':abort > for (int i = 0; i < 5; i++) {
end with '}', '@':abort >    l.SetX1(i);
end with '}', '@':abort >    l.SetY1(i+1);
end with '}', '@':abort >    l.Print();
end with '}', '@':abort > }
end with '}', '@':abort > }
TLine  X1=0.000000 Y1=1.000000 X2=0.000000 Y2=0.000000
TLine  X1=1.000000 Y1=2.000000 X2=0.000000 Y2=0.000000
TLine  X1=2.000000 Y1=3.000000 X2=0.000000 Y2=0.000000
TLine  X1=3.000000 Y1=4.000000 X2=0.000000 Y2=0.000000
TLine  X1=4.000000 Y1=5.000000 X2=0.000000 Y2=0.000000
root[] .q

Here we note:

7.3 The ROOT Script Processor

ROOT script files contain pure C++ code. They can contain a simple sequence of statements like in the multi command line example given above, but also arbitrarily complex class and function definitions.

7.3.1 Un-named Scripts

Let us start with a script containing a simple list of statements (like the multi-command line example given in the previous section). This type of script must start with a { and end with a } and is called an un-named script. Assume the file is called script1.C

#include <iostream.h>
   cout << " Hello" << endl;
   float x = 3.;
   float y = 5.;
   int   i = 101;
   cout <<" x = "<<x<<" y = "<<y<<" i = "<<i<< endl;

To execute the stream of statements in script1.C do:

root[] .x script1.C

This loads the contents of file script1.C and executes all statements in the interpreter’s global scope. One can re-execute the statements by re-issuing “.x script1.C” (since there is no function entry point). Scripts are searched for in the Root.MacroPath as defined in your .rootrc file. To check which script is being executed use:

root[] .which script1.C

7.3.2 Named Scripts

Let us change the un-named script to a named script. Copy the file script1.C to script2.C and add a function statement:

#include <iostream.h>

int run()
   cout << " Hello" << endl;
   float x = 3.;
   float y = 5.;
   int   i= 101;
   cout <<" x = "<< x <<" y = "<< y <<" i = "<< i << endl;
   return 0;

Notice that no surrounding {} are required in this case. To execute function run() in script2.C do:

root[] .L script2.C             // load script in memory
root[] run()                    // execute entry point run
x = 3 y = 5 i = 101
root[] run()                    // execute run() again
x = 3 y = 5 i = 101
root[] .func                    // list all functions known by CINT
filename       line:size busy function type and name
script2.C          4:9   0 public: int run();

The last command shows that run() has been loaded from file script2.C, that the function run() starts on line 4 and is 9 lines long. Notice that once a function has been loaded it becomes part of the system just like a compiled function. Now we copy the file script2.C to the script3.C and change the function name from run() to script3(int j = 10):

#include <iostream.h>
int script3(int j = 10) {
   cout << " Hello" << endl;
   float x = 3.;
   float y = 5.;
   int   i = j;
   cout <<" x = "<< x <<", y = "<< y <<", i = "<< i << endl;
   return 0;

To execute script3() in script3.C type:

root[] .x script3.C(8)

This loads the contents of file script3.C and executes entry point script3(8). Note that the above only works when the filename (minus extension) and function entry point are both the same.

The function script3() can still be executed multiple times:

root[] script3()
x = 3, y = 5, i = 10
root[] script3(33)
x = 3, y = 5, i = 33

In a named script, the objects created on the stack are deleted when the function exits. For example, this scenario is very common. You create a histogram in a named script on the stack. You draw the histogram, but when the function exits the canvas is empty and the histogram disappeared. To avoid histogram from disappearing you can create it on the heap (by using new). This will leave the histogram object intact, but the pointer in the named script scope will be deleted. Since histograms (and trees) are added to the list of objects in the current directory, you can always retrieve them to delete them if needed.

root[] TH1F *h = (TH1F*)gDirectory->Get("myHist");              // or
root[] TH1F *h = (TH1F*)gDirectory->GetList()->FindObject("myHist");

In addition, histograms and trees are automatically deleted when the current directory is closed. This will automatically take care of the clean up. See “Input/Output”.

7.3.3 Executing a Script from a Script

You may want to execute a script conditionally inside another script. To do it you need to call the interpreter and you can do that with TROOT::ProcessLine(). The example $ROOTSYS/tutorials/tree/cernstaff.C calls a script to build the root file if it does not exist:

void cernstaff() {
   if (gSystem->AccessPathName("cernstaff.root")) {
      gROOT->ProcessLine(".x cernbuild.C");

ProcessLine takes a parameter, which is a pointer to an int or to a TInterpreter::EErrorCode to let you access the CINT error code after an attempt to interpret. This will contain the CINT error as defined in enum TInterpreter::EErrorCode.

7.4 Resetting the Interpreter Environment

Variables created on the command line and in un-named scripts are in the interpreter’s global scope, which makes the variables created in un-named scripts available on the command line event after the script is done executing. This is the opposite of a named script where the stack variables are deleted when the function in which they are defined has finished execution.

When running an un-named script over again and this is frequently the case since un-named scripts are used to prototype, one should reset the global environment to clear the variables. This is done by calling gROOT->Reset(). It is good practice, and you will see this in the examples, to begin an un-named script with gROOT->Reset(). It clears the global scope to the state just before executing the previous script (not including any logon scripts). The gROOT->Reset() calls the destructor of the objects if the object was created on the stack. If the object was created on the heap (via new) it is not deleted, but the variable is no longer associated with it. Creating variables on the heap in un-named scripts and calling gROOT->Reset() without you calling the destructor explicitly will cause a memory leak. This may be surprising, but it follows the scope rules. For example, creating an object on the heap in a function (in a named script) without explicitly deleting it will also cause a memory leak. Since when exiting the function only the stack variables are deleted. The code below shows gROOT->Reset() calling the destructor for the stack variable, but not for the heap variable. In the end, neither variable is available, but the memory for the heap variable is not released. Here is an example:

root[] gDebug = 1
(const int)1
root[] TFile stackVar("stack.root","RECREATE")
TKey Writing 86 bytes at address 64 for ID= stack.root Title=
root[] TFile *heapVar = new TFile("heap.root","RECREATE")
TKey Writing 84 bytes at address 64 for ID= heap.root Title=

We turn on Debug to see what the subsequent calls are doing. Then we create two variables, one on the stack and one on the heap.

root[] gROOT->Reset()
TKey Writing 48 bytes at address 150 for ID= stack.root Title=
TKey Writing 54 bytes at address 198 for ID= stack.root Title=
TFile dtor called for stack.root
TDirectory dtor called for stack.root

When we call gROOT->Reset(), CINT tells us that the destructor is called for the stack variable, but it does not mention the heap variable.

root[] stackVar
Error: No symbol stackVar in current scope
FILE:/var/tmp/faaa01jWe_cint LINE:1
*** Interpreter error recovered ***
root[] heapVar
Error: No symbol heapVar in current scope
FILE:/var/tmp/gaaa01jWe_cint LINE:1
*** Interpreter error recovered ***

Neither variable is available in after the call to reset.

root[] gROOT->FindObject("stack.root")
(class TObject*)0x0
root[] gROOT->FindObject("heap.root")
(class TObject*)0x106bfb30

The object on the stack is deleted and shows a null pointer when we do a FindObject. However, the heap object is still around and taking up memory.

Note gROOT->Reset() should be never called in a named script or a compiled program.

7.5 A Script Containing a Class Definition

Lets create a small class TMyClass and a derived class TChild. The virtual method TMyClass::Print()is overridden in TChild. Save this in file called script4.C.

#include <iostream.h>

class TMyClass {
      float   fX;     //x position in centimeters
      float   fY;     //y position in centimeters
      TMyClass() { fX = fY = -1; }
      virtual void Print() const;
      void SetX(float x) { fX = x; }
      void SetY(float y) { fY = y; }
void TMyClass::Print() const  // parent print method
   cout << "fX = " << fX << ", fY = " << fY << endl;
class TChild : public TMyClass {
      void Print() const;
void TChild::Print() const  // child print metod
   cout << "This is TChild::Print()" << endl;

To execute script4.C do:

root[] .L script4.C
root[] TMyClass *a = new TChild
root[] a->Print()
This is TChild::Print()
fX = -1, fY = -1
root[] a->SetX(10)
root[] a->SetY(12)
root[] a->Print()
This is TChild::Print()
fX = 10, fY = 12
root[] .class TMyClass
class TMyClass
size=0x8 FILE:script4.C LINE:3
List of base class-----------------------------------
List of member variable------------------------------
Defined in TMyClass
0x0        private: float fX
0x4        private: float fY
List of member function------------------------------
Defined in TMyClass
filename       line:size busy function type and name
script4.C         16:5    0 public: class TMyClass TMyClass(void);
script4.C         22:4    0 public: void Print(void);
script4.C         12:1    0 public: void SetX(float x);
script4.C         13:1    0 public: void SetY(float y);
root[] .q

As you can see, an interpreted class behaves just like a compiled class.

There are some limitations for a class created in a script:

See “Adding a Class” for ways how to add a class with a shared library and with ACLiC.

7.6 Debugging Scripts

A powerful feature of CINT is the ability to debug interpreted functions by means of setting breakpoints and being able to single step through the code and print variable values on the way. Assume we have script4.C still loaded, we can then do:

root[] .b TChild::Print
Break point set to line 26 script4.C
root[] a.Print()

26   TChild::Print() const
27   {
28      cout << "This is TChild::Print()" << endl;
FILE:script4.C LINE:28 cint> .s

311  operator<<(ostream& ostr,G__CINT_ENDL& i) {return(endl(ostr));
FILE:iostream.h LINE:311 cint> .s
This is TChild::Print()

29      MyClass::Print();
FILE:script4.C LINE:29 cint> .s

16   MyClass::Print() const
17   {
18      cout << "fX = " << fX << ", fY = " << fY << endl;
FILE:script4.C LINE:18 cint> .p fX
FILE:script4.C LINE:18 cint> .s

311  operator<<(ostream& ostr,G__CINT_ENDL& i) {return(endl(ostr));
FILE:iostream.h LINE:311 cint> .s
fX = 10, fY = 12

19   }

30   }

2    }
root[] .q

7.7 Inspecting Objects

An object of a class inheriting from TObject can be inspected, with the Inspect() method. The TObject::Inspect method creates a window listing the current values of the objects members. For example, the next picture is of TFile.

root[] TFile f("staff.root")
root[] f.Inspect()

You can see the pointers are in red and can be clicked on to follow the pointer to the object. If you clicked on fList, the list of objects in memory and there were none, no new canvas would be shown. On top of the page are the navigation buttons to see the previous and next screen.

ROOT object inspector of TFile

ROOT object inspector of TFile

The object inspector of fKeys, the list of keys in the memory

The object inspector of fKeys, the list of keys in the memory

7.8 ROOT/CINT Extensions to C++

In the next example, we demonstrate three of the most important extensions ROOT/CINT makes to C++. Start ROOT in the directory $ROOTSYS/tutorials (make sure to have first run ".x hsimple.C"):

root[] f = new TFile("hsimple.root")
(class TFile*)0x4045e690
TFile**         hsimple.root
TFile*         hsimple.root
KEY: TH1F     hpx;1   This is the px distribution
KEY: TH2F     hpxpy;1 py ps px
KEY: THProfile        hprof;1 Profile of pz versus px
KEY: TNtuple  ntuple;1        Demo ntuple
root[] hpx.Draw()
Warning in <MakeDefCanvas>: creating a default canvas with name c1
root[] .q

The first command shows the first extension; the declaration of f may be omitted when new is used. CINT will correctly create f as pointer to object of class TFile.

The second extension is shown in the second command. Although f is a pointer to TFile we don’t have to use the pointer de-referencing syntax “->” but can use the simple “.” notation.

The third extension is more important. In case CINT cannot find an object being referenced, it will ask ROOT to search for an object with an identical name in the search path defined by TROOT::FindObject(). If ROOT finds the object, it returns CINT a pointer to this object and a pointer to its class definition and CINT will execute the requested member function. This shortcut is quite natural for an interactive system and saves much typing. In this example, ROOT searches for hpx and finds it in simple.root.

The fourth is shown below. There is no need to put a semicolon at the end of a line. The difference between having it and leaving it off is that when you leave it off the return value of the command will be printed on the next line. For example:

root[] 23+5  // no semicolon prints the return value
root[] 23+5; // semicolon no return value is printed

Be aware that these extensions do not work when a compiler replaces the interpreter. Your code will not compile, hence when writing large scripts, it is best to stay away from these shortcuts. It will save you from having problems compiling your scripts using a real C++ compiler.

7.9 ACLiC - The Automatic Compiler of Libraries for CINT

Instead of having CINT interpret your script there is a way to have your scripts compiled, linked and dynamically loaded using the C++ compiler and linker. The advantage of this is that your scripts will run with the speed of compiled C++ and that you can use language constructs that are not fully supported by CINT. On the other hand, you cannot use any CINT shortcuts (see ROOT/CINT Extensions to C++) and for small scripts, the overhead of the compile/link cycle might be larger than just executing the script in the interpreter.

ACLiC will build a CINT dictionary and a shared library from your C++ script, using the compiler and the compiler options that were used to compile the ROOT executable. You do not have to write a makefile remembering the correct compiler options, and you do not have to exit ROOT.

7.9.1 Usage

Before you can compile your interpreted script you need to add include statements for the classes used in the script. Once you did that, you can build and load a shared library containing your script. To load it use the command .L and append the file name with a “+”.

root[] .L MyScript.C+
root[] .files

The + option generates the shared library and names it by taking the name of the file “filename” but replacing the dot before the extension by an underscore and by adding the shared library extension for the current platform. For example on most platforms, hsimple.cxx will generate If we execute a .files command we can see the newly created shared library is in the list of loaded files.

The + command rebuild the library only if the script or any of the files it includes are newer than the library. When checking the timestamp, ACLiC generates a dependency file which name is the same as the library name, just replacing the ‘so’ extension by the extension ‘d’. For example on most platforms, hsimple.cxx will generate hsimple_cxx.d.

To ensure that the shared library is rebuilt you can use the ++ syntax:

root[] .L MyScript.C++

To build, load, and execute the function with the same name as the file you can use the .x command. This is the same as executing a named script. You can have parameters and use .xor .X. The only difference is you need to append a + or a ++.

root[] .x MyScript.C+ (4000)
Creating shared library /home/./

You can select whether the script in compiled with debug symbol or with optimization by appending the letter ‘g’ or ‘O’ after the ‘+’ or ‘++’. Without the specification, the script is compiled with the same level of debugging symbol and optimization as the currently running ROOT executable. For example:

root[] .L MyScript.C++g

will compile MyScript.C with debug symbols; usually this means giving the -g option to compiler.

root[] .L MyScript.C++O

will compile MyScript.C with optimizations; usually this means giving the -O option to compiler. The syntax:

root[] .L MyScript.C++

is using the default optimization level. The initial default is to compile with the same level of optimization as the root executable itself. The default can be changed by:

root[] gSystem->SetAclicMode(TSystem::kDebug);
root[] gSystem->SetAclicMode(TSystem::kOpt);

Note that the commands:

root[] .L MyScript.C+g
root[] .L MyScript.C+O

respectively compile MyScript.C with debug and optimization if the library does not exist yet; they will not change the debug and the optimization level if the library already exist and it is up to date. To use ACLiC from compiled code or from inside another macro, we recommend using the ProcessLine() method of TROOT. For example, in one script you can use ACLiC to compile and load another script.

gROOT->ProcessLine(".L MyScript.C+")
gROOT->ProcessLine(".L MyScript.C++")

7.9.2 Setting the Include Path

You can get the include path by typing:

root[] .include

You can append to the include path by typing:

root[] .include $HOME/mypackage/include

In a script you can append to the include path:

gSystem->AddIncludePath(" -I$HOME/mypackage/include ")

You can also overwrite the existing include path:

gSystem->SetIncludePath(" -I$HOME/mypackage/include ")

The $ROOTSYS/include directory is automatically appended to the include path, so you do not have to worry about including it. To add library that should be used during linking of the shared library use something like:

gSystem->AddLinkedLibs("-L/my/path -lanylib");

This is especially useful for static libraries. For shared ones you can also simply load them before trying to compile the script:


ACLiC uses the directive fMakeSharedLibs to create the shared library. If loading the shared library fails, it tries to output a list of missing symbols by creating an executable (on some platforms like OSF, this does not HAVE to be an executable) containing the script. It uses the directive fMakeExe to do so. For both directives, before passing them to TSystem::Exec(), it expands the variables $SourceFiles, $SharedLib, $LibName, $IncludePath, $LinkedLibs, $ExeNameand$ObjectFiles. See SetMakeSharedLib() for more information on those variables. When the file being passed to ACLiC is on a read only file system, ACLiC warns the user and creates the library in a temporary directory:

root[] .L readonly/t.C++
Warning in <ACLiC>: /scratch/aclic/subs/./readonly is not writable!
Warning in <ACLiC>: Output will be written to /tmp
Info in <TUnixSystem::ACLiC>: creating shared library

To select the temporary directory ACLiC looks at $TEMP, $TEMP_DIR, $TEMPDIR, $TMP, $TMPDIR, $TMP_DIRor uses /tmp (or C:/). Also, a new interface TSystem::Get/SetBuildDir is introduced to let users select an alternative ‘root’ for building of the ACLiC libraries. For filename/full/path/name/macro.C, the library is created as fBuildDir/full/path/name/

7.9.3 Dictionary Generation

You can direct what is added to the dictionary generated by ACLiC in two ways. The simplest way is to add at the end of script (i.e. after the symbols have been defined) something like:

#if defined(__MAKECINT__)
#pragma link C++ class MyOtherClass;

You can also write this portion of code in a file name MyScript_linkdef.h where the suffix '_linkdef'is the prefix defined by the key ‘ACLiC.Linkdef‘ in the currently used resource file (usually .rootrcor$ROOTSYS/etc/system.rootrc) and the prefix is the name of your script.

In ROOT 3.05/03 and above, the default behavior of rootcint is to not link in (i.e. generate the dictionary for) any of the symbols. In particular, this means that the following lines are now, in the general case, unnecessary.

#pragma link off all globals;
#pragma link off all classes;
#pragma link off all functions;

This also means that linking the instantiation of a class template:

#pragma link C++ class mytemplate<int>;

ONLY links this specific class. In previous versions of ROOT, depending on many factors, this might also have included the linking of many other instantiation of class template used directly or indirectly by 'mytemplate'.

A typical case would have been to rely on:

#pragma link C++ class vector<MyClass>;

to also induce the generation of the iterators. You now need to request them explicitly. Another advantage of the change is that now, if you omit the ‘pragma link off’ line from your linkdef file, you can actually sprinkle the ‘pragma link C++ class’ across as many of you header as file as you need.

See the documentation of rootcint for details how pragma can be used.

NOTE: You should not call ACLiC with a script that has a function called main(). When ACLiC calls rootcint with a function called main it tries to add every symbol it finds while parsing the script and the header files to the dictionary. This includes the system header files and the ROOT header files. It will result in duplicate entries at best and crashes at worst, because some classes in ROOT need special attention before they can be added to the dictionary.

7.9.4 Intermediate Steps and Files

ACLiC executes two steps and a third one if needed. These are:

ACLiC makes a shared library with a CINT dictionary containing the classes and functions declared in the script. It also adds the classes and functions declared in included files with the same name as the script file and any of the following extensions: .h, .hh, .hpp, .hxx, .hPP, .hXX. This means that, by default, you cannot combine scripts from different files into one library by using #include statements; you will need to compile each script separately. In a future release, we plan to add the global variables declared in the script to the dictionary also. If you are curious about the specific calls, you can raise the ROOT debug level: gDebug=3 and ACLiC will print these steps. If you need to keep the intermediate files around, for example when debugging the script using gdb, use gDebug=7.

7.9.5 Moving between Interpreter and Compiler

The best way to develop portable scripts is to make sure you can always run them with both, the interpreter and with ACLiC. To do so, do not use the CINT extensions and program around the CINT limitations. When it is not possible or desirable to program around the CINT limitations, you can use the C preprocessor symbols defined for CINT and rootcint.

The preprocessor symbol __CINT__ is defined for both CINT and rootcint. The symbol __MAKECINT__ is only defined in rootcint.

Use !defined(__CINT__) || defined(__MAKECINT__) to bracket code that needs to be seen by the compiler and rootcint, but will be invisible to the interpreter.

Use !defined(__CINT__)to bracket code that should be seen only by the compiler and not by CINT or rootcint.For example, the following will hide the declaration and initialization of the array gArray from both CINT and rootcint.

#if !defined(__CINT__)
int gArray[] = { 2, 3, 4};

Because ACLiC calls rootcint to build a dictionary, the declaration of gArray will not be included in the dictionary, and consequently, gArray will not be available at the command line even if ACLiC is used. CINT and rootcint will ignore all statements between the "#if !defined (__CINT__)" and “#endif". If you want to use gArray in the same script as its declaration, you can do so. However, if you want use the script in the interpreter you have to bracket the usage of gArray between #if's, since the definition is not visible. If you add the following preprocessor statements:

#if !defined(__CINT__)
int gArray[] = { 2, 3, 4};
#elif defined(__MAKECINT__)
int gArray[];

gArray will be visible to rootcint but still not visible to CINT. If you use ACLiC, gArray will be available at the command line and be initialized properly by the compiled code.

We recommend you always write scripts with the needed include statements. In most cases, the script will still run with the interpreter. However, a few header files are not handled very well by CINT.

These types of headers can be included in interpreted and compiled mode:

A few headers will cause problems when they are included in interpreter mode, because the interpreter itself already includes them. In general, the interpreter needs to know whether to use the interpreted or compiled version. The mode of the definition needs to match the mode of the reference.

Here are the cases that need to be excluded in interpreted mode, but included for rootcint. Bracket these with: !defined(__CINT__) || defined(__MAKECINT__)

Hiding header files from rootcint that are necessary for the compiler but optional for the interpreter can lead to a subtle but fatal error. For example:

#ifndef __CINT__
#include "TTree.h"
class TTree;

class subTree : public TTree {

In this case, rootcint does not have enough information about the TTree class to produce the correct dictionary file. If you try this, rootcint and compiling will be error free, however, instantiating a subTree object from the CINT command line will cause a fatal error. In general, it is recommended to let rootcint see as many header files as possible.

7.10 Reflex

Reflection is the ability of a programming language to introspect its data structures and interact with them at runtime without prior knowledge. Reflex provides reflection capabilities for C++. With the ROOT v5.08, Reflex is an optional package. It will become a mandatory package (loaded by default) with the next ROOT versions. In order to build it you have to ./configure --enable-reflex

7.10.1 Overview

Inside ROOT Reflex is thought to replace the current reflection system, which is inherent to CINT. This is an ongoing work and not part of this release. Nevertheless, Reflex dictionaries can be used inside ROOT while populating the current CINT data structures via a special gateway called Cintex (see “Cintex”).

In order to use reflection a dictionary of the data structures involved has to be generated. Before generating the dictionaries, the source code has to be parsed and the information extracted. In the ROOT environment, there are two ways to generate dictionaries for the Reflex library.

rootcint -reflex -f module/src/G__Module.cxx -c module/inc/TMod1.h
module/inc/TMod2.h module/inc/Linkdef.h
rootcint -gccxml -f module/src/G__Module.cxx -c module/inc/TMod1.h
module/inc/TMod2.h module/inc/Linkdef.h

Note: an installation of Python and gccxml is required for using this option.

7.10.2 Selecting Types And Members

You can use selection files to tell genreflex what it should generate a dictionary for. If you do not use it, it will generate a dictionary for all types in the files passed at the command line, or when specifying --deep for all types it finds.

The selection file is passed to genreflex with the -s parameters like this:

genreflex -s selection.xml header1.h header2.h.

It is an XML file with the following structure:

<class [name="classname"] [pattern="wildname"]
[file_name="filename"] [file_pattern="wildname"]
[id="xxxx"] [type="vector"]/>
<class name="classname" >
<field name="m_transient" transient="true"/>
<field name="m_anothertransient" transient="true"/>
<properties prop1="value1" [prop2="value2"]/>
<function [name="funcname"] [pattern="wildname"]
[proto_name="name(int)"] [proto_pattern="name(int,*)"] />
<enum [name="enumname"] [patter="wildname"] />
<variable [name="varname"] [patter="wildname"] />
<class [name="classname"] [pattern="wildname"] />
<method name="unwanted" />

7.10.3 Genreflex and Templates

The program parsing the header files and providing genreflex with the information what’s in them is called GCCXML. It only sees templates if they are instantiated. See the C++ standard on when template instantiation happens. The rule of thumb is: if you design a templated class then it probably does not happen in that templated class’s header.

So you need to help GCCXML. There are two common approaches: the struct member, and the “proper” C++ way. Explicit Template Instantiation

This is the preferred method, but it is not widely used. Suppose you have a templated template class C and a templated function template T A::f(const T&) const;. You can instantiate them (say with template parameter long long) using:

#ifdef __GCCXML__
// GCCXML explicit template instantiation block
template class C<long long>;
template long long A::f(const long long&);

You can even put this into your regular header file: it is surrounded by an #ifdef __GCCXML__ and will thus be invisible to any other compiler. Template Instantiation by struct Members

Suppose you have a templated template class C and a templated function template T f(const T&) const; defined in file C.h. For the templated class you can use:

#include "C.h"
#ifdef __GCCXML__
// GCCXML explicit template instantiation block
namespace {
      C<long long> dummyMember;

Often people put these instantiations into a separate header which in turn #includes the actual header, such that the C++ sources do not see the GCCXML_DUMMY_INSTANTIATION.

7.10.4 GCCXML Installation

Gccxml is a front-end to the gcc compiler suite, which generates xml code out of parsed C++ definitions. Gccxml needs to be installed in order to use this option. Now we are using a patched version of gccxml release 0.6.0 called (0.6.0_patch3). This installation can be downloaded from

Once the dictionary sources have been generated, they can be compiled into a library and loaded via the Reflex builder system. The dictionary information can be used via the Reflex API. For this purpose, Reflex provides eight classes, which exploit the whole functionality of the system.

7.10.5 Reflex API

Reflex offers a simple yet powerful API to access Reflex reflection database. The following classes are defined in the namespace ROOT::Reflex and documented at

An object is an abstraction of a user object. It contains the information about its type and it is location in memory.

Type is an abstraction of a C++ type. Types in Reflex are:

A scope is an abstraction of a C++ type. It holds information such as its declaring scope, it is underlying scope and it is data/function members. Scopes are:

A member lives inside a scope and is of a given Type. Members can be distinguished as:

Base holds the information about the inheritance structure of classes. It contains information such as the offset to the base class and the type of the base class.

Properties are key/value pairs where the key is a string and the value an Any object (Boost::Any). Any objects can hold any type of information be it a string, int or any arbitrary object. Properties can be attached to Types, Scopes and Members and hold any kind of information that is not specific to C++. Examples for Properties would be the class author, a description of a member or the class id.

A MemberTemplate is an abstraction of a templated member. It holds the information about its template parameters and a list of its instantiations.

A TypeTemplate is an abstraction of a templated type (e.g. class). It holds the same information as the MemberTemplate (e.g. template parameters, list of instantiations)

The Reflex package lives in the namespace ROOT::Reflex. Below some examples of usage of the package are given. For further information please see the documentation of the different API classes.

The next examples will refer to the example class MyClass:

class MyClass {

   MyClass() : fMem1(47), fMem2("foo") { }
   int GetMem1() { return fMem1; }
   int GetMem1(int i) { return fMem1*i; }
   void SetMem1(int i) { fMem1 = i; }
   std::string GetMem2() { return fMem2; }
   void SetMem2(const std::string & str) { fMem2 = str; }

   int fMem1;
   std::string fMem2;

The first thing after loading a dictionary (which is done at the moment at the same time as the implemenation library), will be to look up a certain Type or Scope.

Type t1 = Type::ByName("MyClass");

Every API class provides the operator bool, which will return true if the information retrieved for this instance is valid and further actions on this instance can be taken.

if (t1) {
   if (t1.IsClass()) std::cout << "Class ";
   std::cout << t1.Name();

As a class is also a scope (as enum and union) we can now also iterate over its members. This can be done either with stl like iterators or with an iteration by number:

For (Member_Iterator mi = t1.DataMember_Begin();
     mi != DataMember_End(); ++mi) {
   std::cout << (*mi).Name(SCOPED) << " "
             << (*mi).TypeOf().Name(QUALIFIED);

Member m;
for (size_t i = 0; i < t1.FunctionMemberSize(); ++i) {
   m = t1.FunctionMemberAt(i);
   std::cout << m.Name() << " " << m.TypeOf().Name();
   for (Type_Iterator ti = m.FunctionParaeter_Begin(); ti !=
   m.FunctionParameter_End(); ++ti) {
      std::cout << (*ti).Name() << std::endl;

It is not only possible to introspect information through Reflex but also take actions. E.g. instantiate classes/structs, invoke functions, set data members, etc. The instantiation of a type which represents a class struct can be done with:

Object o1 = t1.Construct();

which will call the default constructor for this type and allocate the memory for this type inside the Object. The Object will also contain the type information constructed.

Now the object of a certain type has been constructed one may interact with it. E.g. getting the value of a data member can be done via which will return an Object of the data member in question.

Object mem_obj = o1.Get("fMem1");
int real_value = 0;
if (mem_obj.TypeOf().Name() == "int)
   int real_value = Object_Cast<int>(mem_obj);

It is also possible to invoke function members via the Object class. A function member can be looked up by name, if the member is overloaded an additional parameter which is the string representation of the type can be passed. Currently parameters for the function to invoke shall be passed as a vector of memory addresses of the parameters. This may change in the future to pass a vector of Objects.

int par1 = 2;
std::vector<void*> parVec;
int ret_val = Object_Cast<int>(
   o1.Invoke("GetMem1","int (int)",parVec));

Calling the destructor of an Object can be done via, this will call both the destructor and of the object type and deallocate the memory.


7.10.6 Cintex

Cintex is an optional package inside ROOT. In order to build it you have to

./configure --enable-cintex at the ROOT configuration step.

The purpose of the Cintex package is to bridge uni-directional information from the Reflex to the CINT dictionary system. This package will be needed as long as the unification of the Reflex and CINT dictionaries has not been completed. This unification is work ongoing. In order to use Cintex functionality it will be needed to load the Cintex library (e.g. on linux systems) and enable the Cintex gateway with


After these two steps have been taken, any Reflex dictionary information should be propagated to the CINT dictionaries and subsequently usable inside the CINT environment (e.g. from the root prompt). If wanted debugging information while loading Reflex dictionaries can be turned on with (any number greater than 0 can be used as argument but will not make any difference in the amount of debugging output for the time being).


8 Object Ownership

An object has ownership of another object if it has permission to delete it. Usually a collection or a parent object such as a pad holds ownership. To prevent memory leaks and multiple attempts to delete an object, you need to know which objects ROOT owns and which are owned by you.

The following rules apply to the ROOT classes.

If an object fits none of these cases, the user has ownership. The next paragraphs describe each rule and user ownership in more detail.

8.1 Ownership by Current Directory (gDirectory)

When a histogram, tree, or event list (TEventList) is created, it is added to the list of objects in the current directory by default. You can get the list of objects in a directory and retrieve a pointer to a specific object with the GetList method. This example retrieves a histogram.

   TH1F *h = (TH1F*)gDirectory->GetList()->FindObject("myHist");

The method TDirectory::GetList() returns a TList of objects in the directory. It looks in memory, and is implemented in all ROOT collections. You can change the directory of a histogram, tree, or event list with the SetDirectory method. Here we use a histogram for an example, but the same applies to trees and event lists.


You can also remove a histogram from a directory by using SetDirectory(0). Once a histogram is removed from the directory, it will not be deleted when the directory is closed. It is now your responsibility to delete this histogram once you have finished with it. To change the default that automatically adds the histogram to the current directory, you can call the static function:


Not all histograms created here after will be added to the current directory. In this case, you own all histogram objects and you will need to delete them and clean up the references. You can still set the directory of a histogram by calling SetDirectory once it has been created as described above.

Note that, when a file goes out of scope or is closed all objects on its object list are deleted.

8.2 Ownership by the Master TROOT Object (gROOT)

The master object gROOT, maintains several collections of objects. For example, a canvas is added to the collection of canvases and it is owned by the canvas collection.

TSeqCollection* fFiles        List of TFile
TSeqCollection* fMappedFiles  List of TMappedFile
TSeqCollection* fSockets      List of TSocket and TServerSocket
TSeqCollection* fCanvases     List of TCanvas
TSeqCollection* fStyles       List of TStyle
TSeqCollection* fFunctions    List of TF1, TF2, TF3
TSeqCollection* fTasks        List of TTask
TSeqCollection* fColors       List of TColor
TSeqCollection* fGeometries   List of geometries
TSeqCollection* fBrowsers     List of TBrowser
TSeqCollection* fSpecials     List of special objects
TSeqCollection* fCleanups     List of recursiveRemove collections

These collections are also displayed in the root folder of the Object Browser. Most of these collections are self explanatory. The special cases are the collections of specials and cleanups.

8.2.1 The Collection of Specials

This collection contains objects of the following classes: TCutG, TMultiDimFit, TPrincipal, TChains. In addition it contains the gHtml object, gMinuit objects, and the array of contours graphs (TGraph) created when calling the Draw method of a histogram with the "CONT, LIST" option.

8.2.2 Access to the Collection Contents

The current content for a collection listed above can be accessed with the corresponding gROOT->GetListOf method (for example gROOT->GetListOfCanvases). In addition, gROOT->GetListOfBrowsables returns a collection of all objects visible on the left side panel in the browser. See the image of the Object Browser in the next figure.

The ROOT Object Browser

The ROOT Object Browser

8.3 Ownership by Other Objects

When an object creates another, the creating object is the owner of the created one. For example:


The call to Fit copies the global TF1 Gaussian function and attaches the copy to the histogram. When the histogram is deleted, the copy is deleted also.

When a pad is deleted or cleared, all objects in the pad with the kCanDelete bit set are deleted automatically. Currently the objects created by the DrawCopy methods, have the kCanDelete bit set and are therefore owned by the pad.

8.4 Ownership by the User

The user owns all objects not described in one of the above cases. TObject has two bits, kCanDelete and kMustCleanup, that influence how an object is managed (in TObject::fBits). These are in an enumeration in TObject.h. To set these bits do:


The bits can be reset and tested with the TObject::ResetBit and TObject::TestBit methods.

8.4.1 The kCanDelete Bit

The gROOT collections (see above) own their members and will delete them regardless of the kCanDelete bit. In all other collections, when the collection Clear method is called (i.e. TList::Clear()), members with the kCanDelete bit set, are deleted and removed from the collection. If the kCanDelete bit is not set, the object is only removed from the collection but not deleted.

If a collection Delete (TList::Delete()) method is called, all objects in the collection are deleted without considering the kCanDelete bit. It is important to realize that deleting the collection (i.e. delete MyCollection), DOES NOT delete the members of the collection.

If the user specified MyCollection->SetOwner() the collection owns the objects and delete MyCollection will delete all its members. Otherwise, you need to:

   // delete all member objects in the collection

   // and delete the collection object
   delete MyCollection;

Note that kCanDelete is automatically set by the DrawCopy method and the user can set it for any object. For example, the user must manage all graphics primitives. If you want TCanvas to delete the primitive you created you have to set the kCanDelete bit.

The kCanDelete bit setting is displayed with TObject::ls(). The last number is either 1 or 0 and is the kCanDelete bit.

root[] TCanvas MyCanvas("MyCanvas")
root[] MyCanvas.Divide(2,1)
root[] MyCanvas->cd(MyCanvas_1)
root[] hstat.Draw()             // hstat is an existing TH1F
root[] MyCanvas->cd(MyCanvas_2)
root[] hstat.DrawCopy()         // DrawCopy sets the kCanDelete bit
(class TH1*)0x88e73f8
Canvas Name=MyCanvas ...
 TCanvas ... Name= MyCanvas ...
  TPad   ... Name= MyCanvas_1 ...
   TFrame  ...
   OBJ: TH1F    hstat   Event Histogram : 0
   TPaveText   ... title
   TPaveStats  ... stats
  TPad ... Name= MyCanvas_2 ...
   TFrame  ...
   OBJ: TH1F    hstat   Event Histogram : 1
   TPaveText   ... title
TPaveStats  ... stats

8.4.2 The kMustCleanup Bit

When the kMustCleanup bit is set, the object destructor will remove the object and its references from all collections in the clean up collection (gROOT::fCleanups). An object can be in several collections, for example if an object is in a browser and on two canvases. If the kMustCleanup bit is set, it will be removed automatically from the browser and both canvases when the destructor of the object is called.

The kMustCleanup bit is set:

The user can add his own collection to the collection of clean ups, to take advantage of the automatic garbage collection. For example:

   // create two list
   TList *myList1, *myList2;

   // add both to of clean ups

   // assuming myObject is in myList1 and myList2, when calling:
   delete myObject;

   // the object is deleted from both lists

9 Graphics and the Graphical User Interface

Graphical capabilities of ROOT range from 2D objects (lines, polygons, arrows) to various plots, histograms, and 3D graphical objects. In this chapter, we are going to focus on principals of graphics and 2D objects. Plots and histograms are discussed in a chapter of their own.

9.1 Drawing Objects

In ROOT, most objects derive from a base class TObject. This class has a virtual method Draw() so all objects are supposed to be able to be “drawn”. The basic whiteboard on which an object is drawn is called a canvas (defined by the class TCanvas). If several canvases are defined, there is only one active at a time. One draws an object in the active canvas by using the statement:


This instructs the object “object” to draw itself. If no canvas is opened, a default one (named “c1”) is instantiated and is drawn.

root[] TLine a(0.1,0.1,0.6,0.6)
root[] a.Draw()
<TCanvas::MakeDefCanvas>: created default TCanvas with name c1

The first statement defines a line and the second one draws it. A default canvas is drawn since there was no opened one.

9.2 Interacting with Graphical Objects

When an object is drawn, one can interact with it. For example, the line drawn in the previous paragraph may be moved or transformed. One very important characteristic of ROOT is that transforming an object on the screen will also transform it in memory. One actually interacts with the real object, not with a copy of it on the screen. You can try for instance to look at the starting X coordinate of the line:

root[] a.GetX1()

X1 is the x value of the starting coordinate given in the definition above. Now move it interactively by clicking with the left mouse button in the line’s middle and try to do again:

root[] a.GetX1()

You do not obtain the same result as before, the coordinates of ‘a’ have changed. As said, interacting with an object on the screen changes the object in memory.

9.2.1 Moving, Resizing and Modifying Objects

Changing the graphic objects attributes can be done with the GUI or programmatically. First, let’s see how it is done in the GUI. The Left Mouse Button

As was just seen moving or resizing an object is done with the left mouse button. The cursor changes its shape to indicate what may be done:

Point the object or one part of it:


Resize (exists also for the other directions):

Enlarge (used for text):


Here are some examples of:

Moving: Resizing:

Rotating: With C++ Statements (Programmatically)

How would one move an object in a script? Since there is a tight correspondence between what is seen on the screen and the object in memory, changing the object changes it on the screen. For example, try to do:

root[] a.SetX1(0.9)

This should change one of the coordinates of our line, but nothing happens on the screen. Why is that? In short, the canvas is not updated with each change for performance reasons. See “Updating the Pad”.

9.2.2 Selecting Objects The Middle Mouse Button

Objects in a canvas, as well as in a pad, are stacked on top of each other in the order they were drawn. Some objects may become “active” objects, which mean they are reordered to be on top of the others. To interactively make an object “active”, you can use the middle mouse button. In case of canvases or pads, the border becomes highlighted when it is active. With C++ Statements (Programmatically)

Frequently we want to draw in different canvases or pads. By default, the objects are drawn in the active canvas. To activate a canvas you can use the TPad::cd() method.

root[] c1->cd()

9.2.3 Context Menus: the Right Mouse Button

The context menus are a way to interactively call certain methods of an object. When designing a class, the programmer can add methods to the context menu of the object by making minor changes to the header file. Using Context Menus

On a ROOT canvas, you can right-click on any object and see the context menu for it. The script hsimple.C draws a histogram. The image below shows the context menus for some of the objects on the canvas. Next picture shows that drawing a simple histogram involves as many as seven objects. When selecting a method from the context menu and that method has options, the user will be asked for numerical values or strings to fill in the option. For example, TAxis::SetTitle will prompt you for a string to use for the axis title.

Context menus of different objects in a canvas

Context menus of different objects in a canvas Structure of the Context Menus

The curious reader will have noticed that each entry in the context menu corresponds to a method of the class. Look for example to the menu named TAxis::xaxis. xaxis is the name of the object and TAxis the name of its class. If we look at the list of TAxis methods, for example in, we see the methods SetTimeDisplay() andUnZoom(), which appear also in the context menu.

There are several divisions in the context menu, separated by lines. The top division is a list of the class methods; the second division is a list of the parent class methods. The subsequent divisions are the methods other parent classes in case of multiple inheritance. For example, see the TPaveText::title context menu. A TPaveText inherits from TAttLine, which has the method SetLineAttributes(). Adding Context Menus for a Class

For a method to appear in the context menu of the object it has to be marked by // *MENU* in the header file. Below is the line from TAttLine.h that adds the SetLineAttribute method to the context menu.

virtual void  SetLineAttributes(); // *MENU*

Nothing else is needed, since CINT knows the classes and their methods. It takes advantage of that to create the context menu on the fly when the object is clicking on. If you click on an axis, ROOT will ask the interpreter what are the methods of the TAxis and which ones are set for being displayed in a context menu.

Now, how does the interpreter know this? Remember, when you build a class that you want to use in the ROOT environment, you use rootcint that builds the so-called stub functions and the dictionary. These functions and the dictionary contain the knowledge of the used classes. To do this, rootcint parses all the header files. ROOT has defined some special syntax to inform CINT of certain things, this is done in the comments so that the code still compiles with a C++ compiler.

For example, you have a class with a Draw() method, which will display itself. You would like a context menu to appear when on clicks on the image of an object of this class. The recipe is the following:

For example:

class MyClass : public TObject {
   int      fV1;   // first variable
   double   fV2;   // second variable
   int    GetV1() {return fV1;}
   double GetV2() {return fV2;}
   void   SetV1(int x1) { fV1 = x1;}     // *MENU*
   void   SetV2(double d2) { fV2 = d2;}  // *MENU*
   void   SetBoth(int x1, double d2) {fV1 = x1; fV2 = d2;}

   ClassDef (MyClass,1)

To specify arguments:

void SetXXX(Int_t x1, Float_t y2); //*MENU* *ARGS={x1=>fV1}

This statement is in the comment field, after the *MENU*. If there is more than one argument, these arguments are separated by commas, where fX1 and fY2 are data fields in the same class.

void SetXXX(Int_t x1, Float_t y2); //*MENU* *ARGS={x1=>fX1,y2=>fY2}

If the arguments statement is present, the option dialog displayed when selecting SetXXX field will show the values of variables. We indicate to the system which argument corresponds to which data member of the class.

9.2.4 Executing Events when a Cursor Passes on Top of an Object

This paragraph is for class designers. When a class is designed, it is often desirable to include drawing methods for it. We will have a more extensive discussion about this, but drawing an object in a canvas or a pad consists in “attaching” the object to that pad. When one uses object.Draw(), the object is NOT painted at this moment. It is only attached to the active pad or canvas.

Another method should be provided for the object to be painted, the Paint() method. This is all explained in the next paragraph. As well as Draw() and Paint(), other methods may be provided by the designer of the class. When the mouse is moved or a button pressed/released, the TCanvas function named HandleInput() scans the list of objects in all it’s pads and for each object calls some standard methods to make the object react to the event (mouse movement, click or whatever).

The second one is DistanceToPrimitive(px,py). This function computes a “distance” to an object from the mouse position at the pixel position (px, py, see definition at the end of this paragraph) and returns this distance in pixel units. The selected object will be the one with the shortest computed distance. To see how this works, select the “Event Status” item in the canvas “Options” menu. ROOT will display one status line showing the picked object. If the picked object is, for example, a histogram, the status line indicates the name of the histogram, the position x,y in histogram coordinates, the channel number and the channel content.

It is nice for the canvas to know what the closest object from the mouse is, but it’s even nicer to be able to make this object react. The third standard method to be provided is ExecuteEvent(). This method actually does the event reaction. Its prototype is where px and py are the coordinates at which the event occurred, except if the event is a key press, in which case px contains the key code.

void ExecuteEvent(Int_t event, Int_t px, Int_t py);

Where event is the event that occurs and is one of the following (defined in Buttons.h):

kNoEvent,          kButton1Down,      kButton2Down,
kButton3Down,      kKeyDown,          kButton1Up,
kButton2Up,        kButton3Up,        kButton1Motion,
kButton2Motion,    kButton3Motion,    kKeyPress,
kButton1Locate,    kButton2Locate,    kButton3Locate,
kKeyUp,            kButton1Double,    kButton2Double,
kButton3Double,    kMouseMotion,      kMouseEnter,

We hope the names are self-explanatory.

Designing an ExecuteEvent method is not very easy, except if one wants very basic treatment. We will not go into that and let the reader refer to the sources of classes like TLine or TBox. Go and look at their ExecuteEvent method! We can nevertheless give some reference to the various actions that may be performed. For example, one often wants to change the shape of the cursor when passing on top of an object. This is done with the SetCursor method:


The argument cursor is the type of cursor. It may be:

kBottomLeft,  kBottomRight,  kTopLeft,
kTopRight,    kBottomSide,   kLeftSide,
kTopSide,     kRightSide,    kMove,
kCross,       kArrowHor,     kArrowVer,
kHand,        kRotate,       kPointer,
kArrowRight,  kCaret,        kWatch

They are defined in TVirtualX.h and again we hope the names are self-explanatory. If not, try them by designing a small class. It may derive from something already known like TLine.

Note that the ExecuteEvent() functions may in turn; invoke such functions for other objects, in case an object is drawn using other objects. You can also exploit at best the virtues of inheritance. See for example how the class TArrow (derived from TLine) use or redefine the picking functions in its base class.

The last comment is that mouse position is always given in pixel units in all these standard functions. px=0 and py=0 corresponds to the top-left corner of the canvas. Here, we have followed the standard convention in windowing systems. Note that user coordinates in a canvas (pad) have the origin at the bottom-left corner of the canvas (pad). This is all explained in the paragraph “The Coordinate Systems of a Pad”.

9.3 Graphical Containers: Canvas and Pad

We have talked a lot about canvases, which may be seen as windows. More generally, a graphical entity that contains graphical objects is called a Pad. A Canvas is a special kind of Pad. From now on, when we say something about pads, this also applies to canvases. A pad (class TPad) is a graphical container in the sense it contains other graphical objects like histograms and arrows. It may contain other pads (sub-pads) as well. A Pad is a linked list of primitives of any type (graphs, histograms, shapes, tracks, etc.). It is a kind of display list.

The pad display list

The pad display list

Drawing an object is nothing more than adding its pointer to this list. Look for example at the code of TH1::Draw(). It is merely ten lines of code. The last statement is AppendPad(). This statement calls method of TObject that just adds the pointer of the object, here a histogram, to the list of objects attached to the current pad. Since this is a TObject’s method, every object may be “drawn”, which means attached to a pad.

When is the painting done then ? The answer is: when needed. Every object that derives from TObject has a Paint() method. It may be empty, but for graphical objects, this routine contains all the instructions to paint effectively it in the active pad. Since a Pad has the list of objects it owns, it will call successively the Paint() method of each object, thus re-painting the whole pad on the screen. If the object is a sub-pad, its Paint() method will call the Paint() method of the objects attached, recursively calling Paint() for all the objects.

Pad painting

Pad painting

In some cases a pad need to be painted during a macro execution. To force the pad painting gPad->Update() (see next section) should be performed.

The list of primitives stored in the pad is also used to pick objects and to interact with them.

9.3.1 The Global Pad: gPad

When an object is drawn, it is always in the so-called active pad. For every day use, it is comfortable to be able to access the active pad, whatever it is. For that purpose, there is a global pointer, called gPad. It is always pointing to the active pad. If you want to change the fill color of the active pad to blue but you do not know its name, do this.

root[] gPad->SetFillColor(38)

To get the list of colors, go to the paragraph “Color and color palettes” or if you have an opened canvas, click on the View menu, selecting the Colors item. Finding an Object in a Pad

Now that we have a pointer to the active pad, gPad and that we know this pad contains some objects, it is sometimes interesting to access one of those objects. The method GetPrimitive() of TPad, i.e. TPad::GetPrimitive(const char* name) does exactly this. Since most of the objects that a pad contains derive from TObject, they have a name. The following statement will return a pointer to the object myobjectname and put that pointer into the variable obj. As you can see, the type of returned pointer is TObject*.

root[] obj = gPad->GetPrimitive("myobjectname")
(class TObject*)0x1063cba8

Even if your object is something more complicated, like a histogram TH1F, this is normal. A function cannot return more than one type. So the one chosen was the lowest common denominator to all possible classes, the class from which everything derives, TObject. How do we get the right pointer then? Simply do a cast of the function output that will transform the output (pointer) into the right type. For example if the object is a TPaveLabel:

root[] obj = (TPaveLabel*)(gPad->GetPrimitive("myobjectname"))
(class TPaveLabel*)0x1063cba8

This works for all objects deriving from TObject. However, a question remains. An object has a name if it derives from TNamed, not from TObject. For example, an arrow (TArrow) doesn’t have a name. In that case, the “name” is the name of the class. To know the name of an object, just click with the right button on it. The name appears at the top of the context menu. In case of multiple unnamed objects, a call to GetPrimitive("className") returns the instance of the class that was first created. To retrieve a later instance you can use GetListOfPrimitives(), which returns a list of all the objects on the pad. From the list you can select the object you need. Hiding an Object

Hiding an object in a pad can be made by removing it from the list of objects owned by that pad. This list is accessible by the GetListOfPrimitives() method of TPad. This method returns a pointer to a TList. Suppose we get the pointer to the object, we want to hide, call it obj (see paragraph above). We get the pointer to the list:

root[] li = gPad->GetListOfPrimitives()

Then remove the object from this list:

root[] li->Remove(obj)

The object will disappear from the pad as soon as the pad is updated (try to resize it for example). If one wants to make the object reappear:

root[] obj->Draw()

Caution, this will not work with composed objects, for example many histograms drawn on the same plot (with the option “same”). There are other ways! Try to use the method described here for simple objects.

9.3.2 The Coordinate Systems of a Pad

There are coordinate systems in a TPad: user coordinates, normalized coordinates (NDC), and pixel coordinates.

Pad coordinate systems

Pad coordinate systems The User Coordinate System

The most common is the user coordinate system. Most methods of TPad use the user coordinates, and all graphic primitives have their parameters defined in terms of user coordinates. By default, when an empty pad is drawn, the user coordinates are set to a range from 0 to 1 starting at the lower left corner. At this point they are equivalent of the NDC coordinates (see below). If you draw a high level graphical object, such as a histogram or a function, the user coordinates are set to the coordinates of the histogram. Therefore, when you set a point it will be in the histogram coordinates.

For a newly created blank pad, one may use TPad::Range to set the user coordinate system. This function is defined as:

void Range(float x1,float y1,float x2,float y2)

The arguments x1, x2 defines the new range in the x direction, and the y1, y2 define the new range in the y-direction.

root[] TCanvas MyCanvas ("MyCanvas")
root[] gPad->Range(-100,-100,100,100)

This will set the active pad to have both coordinates to go from -100 to 100, with the center of the pad at (0,0). You can visually check the coordinates by viewing the status bar in the canvas. To display the status bar select Event Status entry in the View canvas menu.

The status bar

The status bar The Normalized Coordinate System (NDC)

Normalized coordinates are independent of the window size and of the user system. The coordinates range from 0 to 1 and (0, 0) corresponds to the bottom-left corner of the pad. Several internal ROOT functions use the NDC system (3D primitives, PostScript, log scale mapping to linear scale). You may want to use this system if the user coordinates are not known ahead of time. The Pixel Coordinate System

The least common is the pixel coordinate system, used by functions such as DistanceToPrimitive() and ExecuteEvent(). Its primary use is for cursor position, which is always given in pixel coordinates. If (px,py) is the cursor position, px=0 and py=0 corresponds to the top-left corner of the pad, which is the standard convention in windowing systems. Using NDC for a particular Object

Most of the time, you will be using the user coordinate system. But sometimes, you will want to use NDC. For example, if you want to draw text always at the same place over a histogram, no matter what the histogram coordinates are. There are two ways to do this. You can set the NDC for one object or may convert NDC to user coordinates. Most graphical objects offer an option to be drawn in NDC. For instance, a line (TLine) may be drawn in NDC by using DrawLineNDC(). A latex formula or a text may use TText::SetNDC() to be drawn in NDC coordinates.

9.3.3 Converting between Coordinate Systems

There are a few utility functions in TPad to convert from one system of coordinates to another. In the following table, a point is defined by (px,py) in pixel coordinates, (ux,uy) in user coordinates, (ndcx,ndcy) in normalized coordinates, (apx, apy) are in absolute pixel coordinates.


TPad’s Methods


NDC to Pixel





Pixel to User






Double_t ux,uy

User to Pixel






Int_t px,py

User to absolute pixel






Int_t apx,apy

Absolute pixel to user






Double_t ux,uy

Note: all the pixel conversion functions along the Y axis consider that py=0 is at the top of the pad except PixeltoY() which assume that the position py=0 is at the bottom of the pad. To make PixeltoY() converting the same way as the other conversion functions, it should be used the following way (p is a pointer to a TPad):

p->PixeltoY(py - p->GetWh());

9.3.4 Dividing a Pad into Sub-pads

Dividing a pad into sub pads in order for instance to draw a few histograms, may be done in two ways. The first is to build pad objects and to draw them into a parent pad, which may be a canvas. The second is to automatically divide a pad into horizontal and vertical sub pads. Creating a Single Sub-pad

The simplest way to divide a pad is to build sub-pads in it. However, this forces the user to explicitly indicate the size and position of those sub-pads. Suppose we want to build a sub-pad in the active pad (pointed by gPad). First, we build it, using a TPad constructor:

root[] spad1 = new TPad("spad1","The first subpad",.1,.1,.5,.5)

One gives the coordinates of the lower left point (0.1, 0.1) and of the upper right one (0.5, 0.5). These coordinates are in NDC. This means that they are independent of the user coordinates system, in particular if you have already drawn for example a histogram in the mother pad. The only thing left is to draw the pad:

root[] spad1->Draw()

If you want more sub-pads, you have to repeat this procedure as many times as necessary. Dividing a Canvas into Sub-Pads

The manual way of dividing a pad into sub-pads is sometimes very tedious. There is a way to automatically generate horizontal and vertical sub-pads inside a given pad.

root[] pad1->Divide(3,2)
Dividing a pad into 6 sub-pads

Dividing a pad into 6 sub-pads

Dividing a pad into 6 sub-pads

Dividing a pad into 6 sub-pads

If pad1 is a pad then, it will divide the pad into 3 columns of 2 sub-pads. The generated sub-pads get names pad1_i where the index i=1 to nxm (in our case pad1_1, pad1_2pad1_6). The names pad1_1etc… correspond to new variables in CINT, so you may use them as soon as the executed method was pad->Divide(). However, in a compiled program, one has to access these objects. Remember that a pad contains other objects and that these objects may themselves be pads. So we can use the GetPrimitive() method:

TPad* pad1_1 = (TPad*)(pad1->GetPrimitive("pad1_1"))

One question remains. In case one does an automatic divide, how one can set the default margins between pads? This is done by adding two parameters to Divide(), which are the margins in x and y:

root[] pad1->Divide(3,2,0.1,0.1)

The margins are here set to 10% of the parent pad width.

9.3.5 Updating the Pad

For performance reasons, a pad is not updated with every change. For example, changing the coordinates of the pad does not automatically redraw it. Instead, the pad has a “bit-modified” that triggers a redraw. This bit is automatically set by:

In compiled code or in a long macro, you may want to access an object created during the paint process. To do so, you can force the painting with a TCanvas::Update(). For example, a TGraph creates a histogram (TH1) to paint itself. In this case the internal histogram obtained with TGraph::GetHistogram() is created only after the pad is painted. The pad is painted automatically after the script is finished executing or if you force the painting with TPad::Modified() followed by a TCanvas::Update(). Note that it is not necessary to call TPad::Modified() after a call to Draw(). The “bit-modified” is set automatically by Draw(). A note about the “bit-modified” in sub pads: when you want to update a sub pad in your canvas, you need to call pad->Modified() rather than canvas->Modified(), and follow it with a canvas->Update(). If you use canvas->Modified(), followed by a call to canvas->Update(), the sub pad has not been declared modified and it will not be updated. Also note that a call to pad->Update() where pad is a sub pad of canvas, calls canvas->Update() and recursively updates all the pads on the canvas.

9.3.6 Making a Pad Transparent

As we will see in the paragraph “Fill Attributes”, a fill style (type of hatching) may be set for a pad.

root[] pad1->SetFillStyle(istyle)

This is done with the SetFillStyle method where istyle is a style number, defined in “Fill Attributes”. A special set of styles allows handling of various levels of transparency. These are styles number 4000 to 4100, 4000 being fully transparent and 4100 fully opaque. So, suppose you have an existing canvas with several pads. You create a new pad (transparent) covering for example the entire canvas. Then you draw your primitives in this pad. The same can be achieved with the graphics editor. For example:

root[] .x tutorials/hist/h1draw.C
root[] TPad *newpad=new TPad("newpad","Transparent pad",0,0,1,1);
root[] newpad->SetFillStyle(4000);
root[] newpad->Draw();
root[] newpad->cd();
root[] // create some primitives, etc

9.3.7 Setting the Log Scale

Setting the scale to logarithmic or linear is an attribute of the pad, not the axis or the histogram. The scale is an attribute of the pad because you may want to draw the same histogram in linear scale in one pad and in log scale in another pad. Frequently, we see several histograms on top of each other in the same pad. It would be very inconvenient to set the scale attribute for each histogram in a pad.

Furthermore, if the logic was set in the histogram class (or each object) the scale setting in each Paint method of all objects should be tested.

If you have a pad with a histogram, a right-click on the pad, outside of the histograms frame will convince you. The SetLogx(), SetLogy() and SetLogz() methods are there. As you see, TPad defines log scale for the two directions x and y plus z if you want to draw a 3D representation of some function or histogram.

The way to set log scale in the x direction for the active pad is:

root[] gPad->SetLogx(1)

To reset log in the z direction:

root[] gPad->SetLogz(0)

If you have a divided pad, you need to set the scale on each of the sub-pads. Setting it on the containing pad does not automatically propagate to the sub-pads. Here is an example of how to set the log scale for the x-axis on a canvas with four sub-pads:

root[] TCanvas MyCanvas("MyCanvas","My Canvas")
root[] MyCanvas->Divide(2,2)
root[] MyCanvas->cd(1)
root[] gPad->SetLogx()
root[] MyCanvas->cd(2)
root[] gPad->SetLogx()
root[] MyCanvas->cd(3)
root[] gPad->SetLogx()

9.3.8 WaitPrimitive method

When the TPad::WaitPrimitive() method is called with no arguments, it will wait until a double click or any key pressed is executed in the canvas. A call to gSystem->Sleep(10) has been added in the loop to avoid consuming at all the CPU. This new option is convenient when executing a macro. By adding statements like:


You can monitor the progress of a running macro, stop it at convenient places with the possibility to interact with the canvas and resume the execution with a double click or a key press.

9.3.9 Locking the Pad

You can make the TPad non-editable. Then no new objects can be added, and the existing objects and the pad can not be changed with the mouse or programmatically. By default the TPad is editable.


9.4 Graphical Objects

In this paragraph, we describe the various simple 2D graphical objects defined in ROOT. Usually, one defines these objects with their constructor and draws them with their Draw() method. Therefore, the examples will be very brief. Most graphical objects have line and fill attributes (color, width) that will be described in “Graphical objects attributes”. If the user wants more information, the class names are given and he may refer to the online developer documentation. This is especially true for functions and methods that set and get internal values of the objects described here. By default 2D graphical objects are created in User Coordinates with (0, 0) in the lower left corner.

9.4.1 Lines, Arrows and Polylines

The simplest graphical object is a line. It is implemented in the TLine class. The line constructor is:

TLine(Double_t x1,Double_t y1,Double_t x2,Double_t y2)

The arguments x1, y1, x2, y2 are the coordinates of the first and second point. It can be used:

root[] l = new TLine(0.2,0.2,0.8,0.3)
root[] l->Draw()

The arrow constructor is:

TArrow(Double_t x1, Double_t y1,
       Double_t x2, Double_t y2,
       Float_t arrowsize, Option_t *option)

It defines an arrow between points x1,y1 and x2,y2. The arrow size is in percentage of the pad height. The option parameter has the following meanings:







Once an arrow is drawn on the screen, one can:

Different arrow formats

Different arrow formats

If FillColor is 0, an open triangle is drawn; else a full triangle is filled with the set fill color. If ar is an arrow object, fill color is set with:


Where icolor is the color defined in “Color and Color Palettes”.

The default-opening angle between the two sides of the arrow is 60 degrees. It can be changed with the method ar->SetAngle(angle), where angle is expressed in degrees.

A poly-line is a set of joint segments. It is defined by a set of N points in a 2D space. Its constructor is:

TPolyLine(Int_t n,Double_t* x,Double_t* y,Option_t* option)

Where n is the number of points, and x and y are arrays of n elements with the coordinates of the points. TPolyLine can be used by it self, but is also a base class for other objects, such as curly arcs.

9.4.2 Circles and Ellipses

An ellipse can be truncated and rotated. It is defined by its center (x1,y1) and two radii r1 and r2. A minimum and maximum angle may be specified (phimin,phimax). The ellipse may be rotated with an angle theta. All these angles are in degrees. The attributes of the outline line are set via TAttLine, of the fill area - via TAttFill class. They are described in “Graphical Objects Attributes”.

Different types of ellipses

Different types of ellipses

When an ellipse sector is drawn only, the lines between the center and the end points of the sector are drawn by default. By specifying the drawn option “only”, these lines can be avoided. Alternatively, the method SetNoEdges() can be called. To remove completely the ellipse outline, specify zero (0) as a line style.

The TEllipse constructor is:

TEllipse(Double_t x1, Double_t y1, Double_t r1, Double_t r2,
         Double_t phimin, Double_t phimax, Double_t theta)

An ellipse may be created with:

root[] e = new TEllipse(0.2,0.2,0.8,0.3)
root[] e->Draw()

9.4.3 Rectangles

The class TBox defines a rectangle. It is a base class for many different higher-level graphical primitives. Its bottom left coordinates x1, y1 and its top right coordinates x2, y2, defines a box. The constructor is:

TBox(Double_t x1,Double_t y1,Double_t x2,Double_t y2)

It may be used as in:

root[] b = new TBox(0.2,0.2,0.8,0.3)
root[] b->SetFillColor(5)
root[] b->Draw()
A rectangle with a border

A rectangle with a border

A TWbox is a rectangle (TBox) with a border size and a border mode. The attributes of the outline line and of the fill area are described in “Graphical Objects Attributes”

9.4.4 Markers

A marker is a point with a fancy shape! The possible markers are shown in the next figure.



The marker constructor is:

TMarker(Double_t x,Double_t y,Int_t marker)

The parameters x and y are the marker coordinates and marker is the marker type, shown in the previous figure. Suppose the pointer ma is a valid marker. The marker size is set via ma->SetMarkerSize(size), where size is the desired size. Note, that the marker types 1, 6 and 7 (the dots) cannot be scaled. They are always drawn with the same number of pixels. SetMarkerSize does not apply on them. To have a “scalable dot” a circle shape should be used instead, for example, the marker type 20. The default marker type is 1, if SetMarkerStyle is not specified. It is the most common one to draw scatter plots.

Different marker sizes

Different marker sizes

Different marker sizes

Different marker sizes

The user interface for changing the marker color, style and size looks like shown in this picture. It takes place in the editor frame anytime the selected object inherits the class TAttMarker.

Non-symmetric symbols should be used carefully in plotting. The next two graphs show how the misleading a careless use of symbols can be. The two plots represent the same data sets but because of a bad symbol choice, the two on the top appear further apart from the next example.

The use of non-symmetric markers

The use of non-symmetric markers

A TPolyMaker is defined by an array on N points in a 2D space. At each point x[i], y[i] a marker is drawn. The list of marker types is shown in the previous paragraph. The marker attributes are managed by the class TAttMarker and are described in “Graphical Objects Attributes”. The TPolyMarker constructor is:

TPolyMarker(Int_t n,Double_t *x,Double_t *y,Option_t *option)

Where x and y are arrays of coordinates for the n points that form the poly-marker.

9.4.5 Curly and Wavy Lines for Feynman Diagrams

This is a peculiarity of particle physics, but we do need sometimes to draw Feynman diagrams. Our friends working in banking can skip this part. A set of classes implements curly or wavy poly-lines typically used to draw Feynman diagrams. Amplitudes and wavelengths may be specified in the constructors, via commands or interactively from context menus. These classes are TCurlyLine and TCurlyArc. These classes make use of TPolyLine by inheritance; ExecuteEvent methods are highly inspired from the methods used in TPolyLine and TArc.

The picture generated by the tutorial macro feynman.C

The picture generated by the tutorial macro feynman.C

The TCurlyLine constructor is:

TCurlyLine(Double_t x1, Double_t y1, Double_t x2, Double_t y2,
           Double_t wavelength, Double_t amplitude)

The coordinates (x1, y1) define the starting point, (x2, y2) - the end-point. The wavelength and the amplitude are given in percent of the pad height.

The TCurlyArc constructor is:

TCurlyArc(Double_t x1, Double_t y1, Double_t rad,
          Double_t phimin, Double_t phimax,
          Double_t wavelength, Double_t amplitude)

The curly arc center is (x1, y1) and the radius is rad. The wavelength and the amplitude are given in percent of the line length. The parameters phimin and phimax are the starting and ending angle of the arc (given in degrees). Refer to $ROOTSYS/tutorials/graphics/feynman.C for the script that built the figure above.

9.4.6 Text and Latex Mathematical Expressions

Text displayed in a pad may be embedded into boxes, called paves (TPaveLabel), or titles of graphs or many other objects but it can live a life of its own. All text displayed in ROOT graphics is an object of class TText. For a physicist, it will be most of the time a TLatex expression (which derives from TText). TLatex has been conceived to draw mathematical formulas or equations. Its syntax is very similar to the Latex in mathematical mode. Subscripts and Superscripts

Subscripts and superscripts are made with the _ and ^ commands. These commands can be combined to make complex subscript and superscript expressions. You may choose how to display subscripts and superscripts using the 2 functions SetIndiceSize(Double_t) and SetLimitIndiceSize(Int_t). Examples of what can be obtained using subscripts and superscripts:

The expression


The expression


The expression













\(x_{1}^{y}\) Fractions

Fractions denoted by the / symbol are made in the obvious way. The #frac command is used for large fractions in displayed formula; it has two arguments: the numerator and the denominator. For example, the equation x = y + z 2 y 2 + 1 is obtained by following expression x=#frac{y+z/2}{y^{2}+1}. Roots

The #sqrt command produces the square ROOT of its argument; it has an optional first argument for other roots.

Example: #sqrt{10} #sqrt[3]{10} Delimiters

You can produce three kinds of proportional delimiters.

#[]{....} or “à la” Latex

#left[.....#right]big square brackets

#{}{....} or #left{.....#right}big curly brackets

#||{....} or #left|.....#right|big absolute value symbol

#(){....} or #left(.....#right)big parenthesis Changing Style in Math Mode

You can change the font and the text color at any moment using:

#font[font-number]{...} and #color[color-number]{...} Line Splitting

A TLatex string may be split in two with the following command: #splitline{top}{bottom}. TAxis and TGaxis objects can take advantage of this feature. For example, the date and time could be shown in the time axis over two lines with: #splitline{21 April 2003}{14:23:56}

9.4.7 Greek Letters

The command to produce a lowercase Greek letter is obtained by adding # to the name of the letter. For an uppercase Greek letter, just capitalize the first letter of the command name.

#alpha     #beta    #chi      #delta     #varepsilon  #phi
#gamma     #eta     #iota     #varphi    #kappa       #lambda
#mu        #nu      #omicron  #pi        #theta       #rho
#sigma     #tau     #upsilon  #varomega  #omega       #xi
#psi       #zeta    #Alpha    #Beta      #Chi         #Delta
#Epsilon   #Phi     #Gamma    #Eta       #Iota        #Kappa
#vartheta  #Lambda  #Mu       #Nu        #Omicron     #Pi
#Theta     #Rho     #Sigma    #Tau       #Upsilon     #Omega
#varsigma  #Xi      #Psi      #epsilon   #varUpsilon  #Zeta

9.4.8 Mathematical Symbols

TLatex can make mathematical and other symbols. A few of them, such as + and >, are produced by typing the corresponding keyboard character. Others are obtained with the commands as shown in the table above. Accents, Arrows and Bars

Symbols in a formula are sometimes placed one above another. TLatex provides special commands for that.

#hat{a} =hat

#check =inverted hat

#acute =acute

#grave =accent grave

#dot =derivative

#ddot =double derivative

#tilde =tilde

#slash =special sign. Draw a slash on top of the text between brackets for example

#slash{E}_{T}generates “Missing ET”

a _ is obtained with #bar{a}

a -> is obtained with #vec{a} Example 1

The script $ROOTSYS/tutorials/graphics/latex.C:

  TCanvas c1("c1","Latex",600,700);
  TLatex l;

  l.DrawLatex(0.1,0.8,"1) C(x) = d #sqrt{#frac{2}{#lambdaD}}
  l.DrawLatex(0.1,0.6,"2) C(x) = d #sqrt{#frac{2}{#lambdaD}}
  l.DrawLatex(0.1,0.4,"3) R = |A|^{2} =
  l.DrawLatex(0.1,0.2,"4) F(t) = #sum_{i=
The picture generated by the tutorial macro latex.C

The picture generated by the tutorial macro latex.C Example 2

The script $ROOTSYS/tutorials/graphics/latex2.C:

  TCanvas c1("c1","Latex",600,700);
  TLatex l;
  #rightarrowI#bar{I}, q#bar{q}");
  l.DrawLatex(0.5,0.3,"L_{em}=eJ^{#mu}_{em}A_{#mu} ,
The picture generated by the tutorial macro latex2.C

The picture generated by the tutorial macro latex2.C Example 3

The script $ROOTSYS/tutorials/graphics/latex3.C:

  TCanvas c1("c1");
  TPaveText pt(.1,.5,.9,.9);
  #frac{d#sigma}{dcos#theta} (e^{+}e^{-}
  #rightarrow f#bar{f} ) = ");
  pt.AddText("#left| #frac{1}{1 - #Delta#alpha} #right|^{2}
  pt.AddText("+ 4 Re #left{ #frac{2}{1 - #Delta#alpha} #chi(s)
  (1 + cos^{2}#theta) + 2 #hat{g}_{a}^{e}
  #hat{g}_{a}^{f} cos#theta) } #right}");
  pt.SetLabel("Born equation");
The picture generated by the tutorial macro latex3.C

The picture generated by the tutorial macro latex3.C

9.4.9 Text in a Pad

Text displayed in a pad may be embedded into boxes, called paves, or may be drawn alone. In any case, it is recommended to use a Latex expression, which is covered in the previous paragraph. Using TLatex is valid whether the text is embedded or not. In fact, you will use Latex expressions without knowing it since it is the standard for all the embedded text. A pave is just a box with a border size and a shadow option. The options common to all types of paves and used when building those objects are the following:

option = "T" top frame

option = "B" bottom frame

option = "R" right frame

option = "L" left frame

option = "NDC" x1,y1,x2,y2 are given in NDC

option = "ARC" corners are rounded

We will see the practical use of these options in the description of the more functional objects like TPaveLabels. There are several categories of paves containing text: TPaveLabel, TPaveText and TPavesText. TPaveLabels are panels containing one line of text. They are used for labeling.

TPaveLabel(Double_t x1, Double_t y1, Double_t x2, Double_t y2,
           const char *label, Option_t *option)

Where (x1, y1) are the coordinates of the bottom left corner, (x2,y2) - coordinates of the upper right corner. “label” is the text to be displayed and “option” is the drawing option, described above. By default, the border size is 5 and the option is “br”. If one wants to set the border size to some other value, one may use the method SetBorderSize(). For example, suppose we have a histogram, which limits are (-100,100) in the x direction and (0, 1000) in the y direction. The following lines will draw a label in the center of the histogram, with no border. If one wants the label position to be independent of the histogram coordinates, or user coordinates, one can use the option “NDC”. See “The Coordinate Systems of a Pad”.

root[] pl = new TPaveLabel(-50,0,50,200,"Some text")
root[] pl->SetBorderSize(0)
root[] pl->Draw()
PaveLabels drawn with different options

PaveLabels drawn with different options

A TPaveLabel can contain only one line of text. A TPaveText may contain several lines. This is the only difference. This picture illustrates and explains some of the points of TPaveText. Once a TPaveText is drawn, a line can be added or removed by brining up the context menu with the mouse.

PaveText examples

PaveText examples

A TPavesText is a stack of text panels (see TPaveText). One can set the number of stacked panels at building time. It has the following constructor: By default, the number of stacked panels is 5, option=br”.

TPavesText(Double_t x1, Double_t y1, Double_t x2, Double_t y2,
Int_t npaves, Option_t* option)
A PaveText example

A PaveText example

9.4.10 The TeX Processor TMathText

TMathText’s purpose is to write mathematical equations, exactly as TeX would do it. The syntax is the same as the TeX’s one.

The script $ROOTSYS/tutorials/graphics/tmathtex.C:

gives the following output:

A TMathText example

A TMathText example

TMathText uses plain TeX syntax and uses “\” as control instead of “#”. If a piece of text containing “\” is given to TLatex then TMathText is automatically invoked. Therefore, as histograms’ titles, axis titles, labels etc … are drawn using TLatex, the TMathText syntax can be used for them also.

9.5 Axis

The axis objects are automatically built by various high level objects such as histograms or graphs. Once build, one may access them and change their characteristics. It is also possible, for some particular purposes to build axis on their own. This may be useful for example in the case one wants to draw two axis for the same plot, one on the left and one on the right.

For historical reasons, there are two classes representing axis. TAxis * axis is the axis object, which will be returned when calling the TH1::GetAxis() method.

TAxis *axis = histo->GetXaxis()

Of course, you may do the same for Y and Z-axis. The graphical representation of an axis is done with the TGaxis class. The histogram classes and TGraph generate instances of this class. This is internal and the user should not have to see it.

9.5.1 Axis Title

The axis title is set, as with all named objects, by

axis->SetTitle("Whatever title you want");

When the axis is embedded into a histogram or a graph, one has to first extract the axis object:

h->GetXaxis()->SetTitle("Whatever title you want")

9.5.2 Axis Options and Characteristics

The axis options are most simply set with the styles. The available style options controlling specific axis options are the following:

TAxis *axis = histo->GetXaxis();
axis->SetAxisColor(Color_t color = 1);
axis->SetLabelColor(Color_t color = 1);
axis->SetLabelFont(Style_t font = 62);
axis->SetLabelOffset(Float_t offset = 0.005);
axis->SetLabelSize(Float_t size = 0.04);
axis->SetNdivisions(Int_t n = 510, Bool_t optim = kTRUE);
axis->SetNoExponent(Bool_t noExponent = kTRUE);
axis->SetTickLength(Float_t length = 0.03);
axis->SetTitleOffset(Float_t offset = 1);
axis->SetTitleSize(Float_t size = 0.02);

The getters corresponding to the described setters are also available. The general options, not specific to axis, as for instance SetTitleTextColor() are valid and do have an effect on axis characteristics.

9.5.3 Setting the Number of Divisions

Use TAxis::SetNdivisions(ndiv,optim) to set the number of divisions for an axis. The ndiv and optim are as follows:

For example:

ndiv = 0: no tick marks.

ndiv = 2: 2 divisions, one tick mark in the middle of the axis.

ndiv = 510: 10 primary divisions, 5 secondary divisions

ndiv = -10: exactly 10 primary divisions

9.5.4 Zooming the Axis

You can use TAxis::SetRange or TAxis::SetRangeUser to zoom the axis.

TAxis::SetRange(Int_t binfirst,Int_t binlast)

The SetRange method parameters are bin numbers. They are not axis. For example if a histogram plots the values from 0 to 500 and has 100 bins, SetRange(0,10) will cover the values 0 to 50. The parameters for SetRangeUser are user coordinates. If the start or end is in the middle of a bin the resulting range is approximation. It finds the low edge bin for the start and the high edge bin for the high.

TAxis::SetRangeUser(Axis_t ufirst,Axis_t ulast)

Both methods, SetRange and SetRangeUser, are in the context menu of any axis and can be used interactively. In addition, you can zoom an axis interactively: click on the axis on the start, drag the cursor to the end, and release the mouse button.

9.5.5 Drawing Axis Independently of Graphs or Histograms

An axis may be drawn independently of a histogram or a graph. This may be useful to draw for example a supplementary axis for a graph. In this case, one has to use the TGaxis class, the graphical representation of an axis. One may use the standard constructor for this kind of objects:

TGaxis(Double_t xmin, Double_t ymin, Double_t xmax, Double_t ymax,
       Double_t wmin, Double_t wmax, Int_t ndiv = 510,
       Option_t* chopt,Double_t gridlength = 0)

The arguments xmin, ymin are the coordinates of the axis’ start in the user coordinates system, and xmax, ymax are the end coordinates. The arguments wmin and wmax are the minimum (at the start) and maximum (at the end) values to be represented on the axis; ndiv is the number of divisions. The options, given by the “chopt” string are the following:

Instead of the wmin,wmax arguments of the normal constructor, i.e. the limits of the axis, the name of a TF1 function can be specified. This function will be used to map the user coordinates to the axis values and ticks.

The constructor is the following:

TGaxis(Double_t xmin, Double_t ymin, Double_t xmax, Double_t ymax,
       const char* funcname, Int_t ndiv=510,
       Option_t* chopt, Double_t gridlength=0)

In such a way, it is possible to obtain exponential evolution of the tick marks position, or even decreasing. In fact, anything you like.

9.5.6 Orientation of Tick Marks on Axis

Tick marks are normally drawn on the positive side of the axis, however, if xmin = xmax, then negative.

9.5.7 Labels Position

Labels are normally drawn on side opposite to tick marks. However, chopt = '=': on Equal side. The function TAxis::CenterLabels() sets the bit kCenterLabels and it is visible from TAxis context menu. It centers the bin labels and it makes sense only when the number of bins is equal to the number of tick marks. The class responsible for drawing the axis TGaxis inherits this property. Orientation

Labels are normally drawn parallel to the axis. However, if xmin = xmax, then they are drawn orthogonal, and if ymin=ymax they are drawn parallel. Labels for Exponents

By default, an exponent of the form 10^N is used when the label values are either all very small or very large. One can disable the exponent by calling:


Note that this option is implicitly selected if the number of digits to draw a label is less than the fgMaxDigits global member. If the property SetNoExponent was set in TAxis (via TAxis::SetNoExponent), the TGaxis will inherit this property. TGaxis is the class responsible for drawing the axis. The method SetNoExponent is also available from the axis context menu.

Y-axis with and without exponent labels

Y-axis with and without exponent labels Number of Digits in Labels

TGaxis::fgMaxDigits is the maximum number of digits permitted for the axis labels above which the notation with 10^N is used. It must be greater than 0. By default fgMaxDigits is 5 and to change it use the TGaxis::SetMaxDigits method. For example to set fgMaxDigits to accept 6 digits and accept numbers like 900000 on an axis call:

TGaxis::SetMaxDigits(6) Tick Mark Positions

Labels are centered on tick marks. However, if xmin = xmax, then they are right adjusted. Label Formatting

Blank characters are stripped, and then the label is correctly aligned. The dot, if last character of the string, is also stripped. In the following, we have some parameters, like tick marks length and characters height (in percentage of the length of the axis, in user coordinates). The default values are as follows: Stripping Decimals

Use the TStyle::SetStripDecimals to strip decimals when drawing axis labels. By default, the option is set to true, and TGaxis::PaintAxis removes trailing zeros after the dot in the axis labels, e.g. {0, 0.5, 1, 1.5, 2, 2.5, etc.}

TStyle::SetStripDecimals (Bool_t strip=kTRUE)

If this function is called with strip=kFALSE, TGaxis::PaintAxis() will draw labels with the same number of digits after the dot, e.g. {0.0, 0.5, 1.0, 1.5, 2.0, 2.5, etc.} Optional Grid

chopt = 'W': cross-Wire Axis Binning Optimization

By default, the axis binning is optimized.

9.5.8 Axis with Time Units

Histograms’ axis can be defined as “time axis”. To do that it is enough to activate the SetTimeDisplay attribute on a given axis. If h is a histogram, it is done the following way:

h->GetXaxis()->SetTimeDisplay(1);    // X axis is a time axis

Two parameters can be adjusted in order to define time axis: the time format and the time offset. Time Format

It defines the format of the labels along the time axis. It can be changed using the TAxis method SetTimeFormat. The time format is the one used by the C function strftime(). It is a string containing the following formatting characters:

For the date:

%a abbreviated weekday name

%b abbreviated month name

%d day of the month (01-31)

%m month (01-12)

%y year without century

%Y year with century

For the time:

%H hour (24-hour clock)

%I hour (12-hour clock)

%p local equivalent of AM or PM

%M minute (00-59)

%S seconds (00-61)

%% %

The other characters are output as is. For example to have a format like dd/mm/yyyy one should do:


If the time format is not defined, a default one will be computed automatically. Time Offset

This is a time in seconds in the UNIX standard UTC format (the universal time, not the local one), defining the starting date of a histogram axis. This date should be greater than 01/01/95 and is given in seconds. There are three ways to define the time offset:

1- By setting the global default time offset:

TDatime da(2003,02,28,12,00,00);

If no time offset is defined for a particular axis, the default time offset will be used. In the example above, notice the usage of TDatime to translate an explicit date into the time in seconds required by SetTimeFormat.

2- By setting a time offset to a particular axis:

TDatime dh(2001,09,23,15,00,00);

3- Together with the time format using SetTimeFormat. The time offset can be specified using the control character %F after the normal time format. %F is followed by the date in the format: yyyy-mm-dd hh:mm:ss.

h->GetXaxis()->SetTimeFormat("%d/%m/%y%F2000-02-28 13:00:01");

Notice that this date format is the same used by the TDatime function AsSQLString. If needed, this function can be used to translate a time in seconds into a character string which can be appended after %F. If the time format is not specified (before %F) the automatic one will be used. The following example illustrates the various possibilities.

  TDatime da(2003,02,28,12,00,00);
  ct = new TCanvas("ct","Time on axis",0,0,600,600);
  ht1 = new TH1F("ht1","ht1",30000,0.,200000.);
  ht2 = new TH1F("ht2","ht2",30000,0.,200000.);
  ht3 = new TH1F("ht3","ht3",30000,0.,200000.);
  for (Int_t i=1;i<30000;i++) {
    Float_t noise = gRandom->Gaus(0,120);
  TDatime dh(2001,09,23,15,00,00);

The output is shown in the figure below. If a time axis has no specified time offset, the global time offset will be stored in the axis data structure. The histogram limits are in seconds. If wmin and wmax are the histogram limits, the time axis will spread around the time offset value from TimeOffset+wmin to TimeOffset+wmax. Until now all examples had a lowest value equal to 0. The following example demonstrates how to define the histogram limits relatively to the time offset value.

Time axis examples

Time axis examples

  // Define the time offset as 2003, January 1st
  TDatime T0(2003,01,01,00,00,00);
  int X0 = T0.Convert();

  // Define the lowest histogram limit as 2002,September 23rd
   TDatime T1(2002,09,23,00,00,00);
   int X1 = T1.Convert()-X0;

  // Define the highest histogram limit as 2003, March 7th
  TDatime T2(2003,03,07,00,00,00);
  int X2 = T2.Convert(1)-X0;

  TH1F * h1 = new TH1F("h1","test",100,X1,X2);

  TRandom r;
  for (Int_t i=0;i<30000;i++) {
    Double_t noise = r.Gaus(0.5*(X1+X2),0.1*(X2-X1));


The output is shown in the next figure. Usually time axes are created automatically via histograms, but one may also want to draw a time axis outside a “histogram context”. Therefore, it is useful to understand how TGaxis works for such axis. The time offset can be defined using one of the three methods described before. The time axis will spread around the time offset value. Actually, it will go from TimeOffset+wmin to TimeOffset+wmax where wmin and wmax are the minimum and maximum values (in seconds) of the axis. Let us take again an example. Having defined “2003, February 28 at 12h”, we would like to see the axis a day before and a day after.

A histogram with time axis X

A histogram with time axis X

A TGaxis can be created the following way (a day has 86400 seconds):

TGaxis *axis = new TGaxis(x1,y1,x2,y2,-100000,150000,2405,"t");

the “t” option (in lower case) means it is a “time axis”. The axis goes form 100000 seconds before TimeOffset and 150000 seconds after. So the complete macro is:

  c1 = new TCanvas("c1","Examples of TGaxis",10,10,700,500);
  TGaxis *axis = new TGaxis(-8,-0.6,8,-0.6,-100000,150000,2405,"t");

  TDatime da(2003,02,28,12,00,00);

The time format is specified with:


The macro gives the following output:

Thanks to the TLatex directive #splitline it is possible to write the time labels on two lines. In the previous example changing the SetTimeFormat line by:


will produce the following axis:

9.5.9 Axis Examples

To illustrate what was said, we provide two scripts. The first one creates the picture shown in the next figure.

The first axis example

The first axis example

The first script is:

  c1 = new TCanvas("c1","Examples of Gaxis",10,10,700,500);

  TGaxis *axis1 = new TGaxis(-4.5,-0.2,5.5,-0.2,-6,8,510,"");
  TGaxis *axis2 = new TGaxis(4.5,0.2,5.5,0.2,0.001,10000,510,"G");

  TGaxis *axis3 = new TGaxis(-9,-0.8,-9,0.8,-8,8,50510,"");
  TGaxis *axis4 = new TGaxis(-7,-0.8,7,0.8,1,10000,50510,"G");

  TGaxis *axis5 = new TGaxis(-4.5,-6,5.5,-6,1.2,1.32,80506,"-+");

  TGaxis *axis6 = new TGaxis(-4.5,0.6,5.5,0.6,100,900,50510,"-");
  TGaxis *axis7 = new TGaxis(8,-0.8,8,0.8,0,9000,50510,"+L");

  // one can make axis top->bottom. However because of a problem,
  // the two x values should not be equal
  TGaxis *axis8 = new TGaxis(6.5,0.8,6.499,-0.8,0,90,50510,"-");
The second axis example

The second axis example

The second example shows the use of the second form of the constructor, with axis ticks position determined by a function TF1:

void gaxis3a()

  TH2F *h2 = new TH2F("h","Axes",2,0,10,2,-2,2);
  TF1 *f1=new TF1("f1","-x",-10,10);
  TGaxis *A1 = new TGaxis(0,2,10,2,"f1",510,"-");
  A1->SetTitle("axis with decreasing values");

  TF1 *f2=new TF1("f2","exp(x)",0,2);
  TGaxis *A2 = new TGaxis(1,1,9,1,"f2");
  A2->SetTitle("exponential axis");

  TF1 *f3=new TF1("f3","log10(x)",0,800);
  TGaxis *A3 = new TGaxis(2,-2,2,0,"f3",505);
  A3->SetTitle("logarithmic axis");
An axis example with time display

An axis example with time display

// strip chart example
void seism() {

  TStopwatch sw; sw.Start();
  //set time offset
  TDatime dtime;
  TCanvas *c1 = new TCanvas("c1","Time on axis",10,10,1000,500);

  Float_t bintime = 1;
  // one bin = 1 second. change it to set the time scale
  TH1F *ht = new TH1F("ht","The ROOT seism",10,0,10*bintime);
  Float_t signal = 1000;

  for (Int_t i=1;i<2300;i++) {
    // Build a signal : noisy damped sine
    Float_t noise  = gRandom->Gaus(0,120);
    if (i > 700)
      noise += signal*sin((i-700.)*6.28/30)*exp((700.-i)/300.);
    //canvas can be edited during the loop
   printf("Real Time = %8.3fs,Cpu Time = %8.3fsn",sw.RealTime(),

9.6 Graphical Objects Attributes

9.6.1 Text Attributes

When a class contains text or derives from a text class, it needs to be able to set text attributes like font type, size, and color. To do so, the class inherits from the TAttText class (a secondary inheritance), which defines text attributes. TLatex and TText inherit from TAttText. Setting Text Alignment

Text alignment may be set by a method call. What is said here applies to all objects deriving from TAttText, and there are many. We will take an example that may be transposed to other types. Suppose “la” is a TLatex object. The alignment is set with:

root[] la->SetTextAlign(align)

The parameter align is a short describing the alignment:

align = 10*HorizontalAlign + VerticalAlign

For horizontal alignment, the following convention applies:

For vertical alignment, the following convention applies:

For example, align: 11 = left adjusted and bottom adjusted; 32 = right adjusted and vertically centered. Setting Text Angle

Use TAttText::SetTextAngle to set the text angle. The angle is the degrees of the horizontal.

root[] la->SetTextAngle(angle) Setting Text Color

Use TAttText::SetTextColor to set the text color. The color is the color index. The colors are described in “Color and Color Palettes”.

root[] la->SetTextColor(color) Setting Text Font

Use TAttText::SetTextFont to set the font. The parameter font is the font code, combining the font and precision: font = 10 * fontID + precision

root[] la->SetTextFont(font)

The table below lists the available fonts. The font IDs must be between 1 and 14. The precision can be:

When precision 0 is used, only the original non-scaled system fonts are used. The fonts have a minimum (4) and maximum (37) size in pixels. These fonts are fast and are of good quality. Their size varies with large steps and they cannot be rotated. Precision 1 and 2 fonts have a different behavior depending if True Type Fonts (TTF) are used or not. If TTF are used, you always get very good quality scalable and rotate-able fonts. However, TTF are slow. Precision 1 and 2 fonts have a different behavior for PostScript in case of TLatex objects:

For example: font = 62 is the font with ID 6 and precision 2.

Font’s examples

Font’s examples

The available fonts are:

Font ID


True Type name

Is italic




“Times New Roman”





“Times New Roman”





“Times New Roman”




helvetica-medium-r-norma l





helvetica-medium-o-norma l
















“Courier New”





“Courier New”





“Courier New”





“Courier New”










“Times New Roman”







This script makes the image of the different fonts:

  textc = new TCanvas("textc","Example of text",1);
  for (int i=1;i<15;i++) {
    cid = new char[8];
    sprintf(cid,"ID %d :",i);
    cid[7] = 0;
    lid = new TLatex(0.1,1-(double)i/15,cid);
    l = new TLatex(.2,1-(double)i/15,
                   "The quick brown fox is not here anymore")
} How to use True Type Fonts

You can activate the True Type Fonts by adding the following line in your .rootrc file.

Unix.*.Root.UseTTFonts:     true

You can check that you indeed use the TTF in your Root session. When the TTF is active, you get the following message at the start of a session: “Free Type Engine v1.x used to render TrueType fonts.” You can also check with the command:

gEnv->Print() Setting Text Size

Use TAttText::SetTextSize to set the text size.

root[] la->SetTextSize(size)

The size is the text size expressed in percentage of the current pad size.

The text size in pixels will be:

The user interface for changing the text color, size, font and allignment looks like shown in this picture. It takes place in the editor frame anytime the selected object inherits the class TAttText.

9.6.2 Line Attributes

All classes manipulating lines have to deal with line attributes: color, style and width. This is done by using secondary inheritance of the class TAttLine. The line color may be set by a method call. What is said here applies to all objects deriving from TAttLine, and there are many (histograms, plots). We will take an example that may be transposed to other types. Suppose “li” is a TLine object. The line color is set with:

root[] li->SetLineColor(color)

The argument color is a color number. The colors are described in “Color and Color Palettes”

The line style may be set by a method call. What is said here applies to all objects deriving from TAttLine, and there are many (histograms, plots). We will take an example that may be transposed to other types. Suppose “li” is a TLine object. The line style is set with:

root[] li->SetLineStyle(style)

The argument style is one of: 1=solid, 2=dash, 3=dot, 4=dash-dot.

The line width may be set by a method call. What is said here applies to all objects deriving from TAttLine, and there are many (histograms, plots). We will take an example that may be transposed to other types. Suppose “li” is a TLine object. The line width is set with:

root[] li->SetLineWidth(width)

The width is the width expressed in pixel units.

The user interface for changing the line color, line width and style looks like shown in this picture. It takes place in the editor frame anytime the selected object inherits the class TAttLine.

9.6.3 Fill Attributes

Almost all graphics classes have a fill area somewhere. These classes have to deal with fill attributes. This is done by using secondary inheritance of the class TAttFill. Fill color may be set by a method call. What is said here applies to all objects deriving from TAttFill, and there are many (histograms, plots). We will take an example that may be transposed to other types. Suppose “h” is a TH1F (1 dim histogram) object. The histogram fill color is set with:

root[] h->SetFillColor(color)

The color is a color number. The colors are described in “Color and color palettes”

Fill style may be set by a method call. What is said here applies to all objects deriving from TAttFill, and there are many (histograms, plots). We will take an example that may be transposed to other types. Suppose “h” is a TH1F (1 dim histogram) object. The histogram fill style is set with:

root[] h->SetFillStyle(style)

The convention for style is: 0:hollow, 1001:solid, 2001:hatch style, 3000+pattern number:patterns, 4000 to 4100:transparency, 4000:fully transparent, 4100: fully opaque.

Fill styles >3100 and <3999 are hatches. They are defined according to the FillStyle=3ijk value as follows:

The various patterns

The various patterns

9.6.4 Color and Color Palettes

At initialization time, a table of basic colors is generated when the first Canvas constructor is called. This table is a linked list, which can be accessed from the gROOT object (see TROOT::GetListOfColors()). Each color has an index and when a basic color is defined, two “companion” colors are defined:

The dark and bright colors are used to give 3-D effects when drawing various boxes (see TWbox, TPave, TPaveText, TPaveLabel, etc). If you have a black and white copy of the manual, here are the basic colors and their indices.

The basic ROOT colors

The basic ROOT colors

The list of currently supported basic colors (here dark and bright colors are not shown) are shown. The color numbers specified in the basic palette, and the picture above, can be viewed by selecting the menu entry Colors in the View canvas menu. The user may define other colors. To do this, one has to build a new TColor:

TColor(Int_t color,Float_t r,Float_t g,Float_t b,const char* name)

One has to give the color number and the three Red, Green, Blue values, each being defined from 0 (min) to 1(max). An optional name may be given. When built, this color is automatically added to the existing list of colors. If the color number already exists, one has to extract it from the list and redefine the RGB values. This may be done for example with:

root[] color=(TColor*)(gROOT->GetListOfColors()->At(index_color))
root[] color->SetRGB(r,g,b)

Where r, g and b go from 0 to 1 and index_color is the color number you wish to change.

The user interface for changing the fill color and style looks like shown in this picture. It takes place in the editor frame anytime the selected object inherits the class TAttFill. Color Palette (for Histograms)

Defining one color at a time may be tedious. The histogram classes (see Draw Options) use the color palette. For example, TH1::Draw("col") draws a 2-D histogram with cells represented by a box filled with a color CI function of the cell content. If the cell content is N, the color CI used will be the color number in colors[N]. If the maximum cell content is >ncolors, all cell contents are scaled to ncolors. The current color palette does not have a class or global object of its own. It is defined in the current style as an array of color numbers. The current palette can be changed with:

TStyle::SetPalette(Int_t ncolors,Int_t*color_indexes).

By default, or if ncolors <= 0, a default palette (see above) of 50 colors is defined. The colors defined in this palette are good for coloring pads, labels, and other graphic objects. If ncolors > 0 and colors = 0, the default palette is used with a maximum of ncolors. If ncolors == 1 && colors == 0, then a pretty palette with a spectrum Violet->Red is created. It is recommended to use this pretty palette when drawing lego(s), surfaces or contours. For example, to set the current palette to the “pretty” one, do:

root[] gStyle->SetPalette(1)

A more complete example is shown below. It illustrates the definition of a custom palette. You can adapt it to suit your needs. In case you use it for contour coloring, with the current color/contour algorithm, always define two more colors than the number of contours.

void palette() {
  // Example of creating new colors (purples)
  const Int_t colNum = 10;    // and defining of a new palette
  Int_t palette[colNum];
  for (Int_t i=0; i<colNum; i++) {
    // get the color and if it does not exist create it
    if (! gROOT->GetColor(230+i) ){
      TColor *color =
         new TColor(230+i,1-(i/((colNum)*1.0)),0.3,0.5,"");
    } else {
      TColor *color = gROOT->GetColor(230+i);
    palette[i] = 230+i;
  TF2 *f2 = new TF2("f2","exp(-(x^2)-(y^2))",-3,3,-3,3);
  // two contours less than the number of colors in palette

9.7 The Graphics Editor

A new graphics editor took place in ROOT v4.0. The editor can be activated by selecting the Editor menu entry in the canvas View menu or one of the context menu entries for setting line, fill, marker or text attributes. The following object editors are available for the current ROOT version.

9.7.1 TAxisEditor

This user interface gives the possibility for changing the following axis attributes:

9.7.2 TPadEditor

9.8 Copy and Paste

You can make a copy of a canvas using TCanvas::DrawClonePad. This method is unique to TCanvas. It clones the entire canvas to the active pad. There is a more general method TObject::DrawClone, which all objects descendent of TObject, specifically all graphic objects inherit. Below are two examples, one to show the use of DrawClonePad and the other to show the use of DrawClone.

9.8.1 Using the GUI

In this example we will copy an entire canvas to a new one with DrawClonePad. Run the script draw2dopt.C.

root[] .x tutorials/hist/draw2dopt.C

This creates a canvas with 2D histograms. To make a copy of the canvas follow the steps:

This copies the entire canvas and all its sub-pads to a new canvas. The copied canvas is a deep clone, and all the objects on it are copies and independent of the original objects. For instance, change the fill on one of the original histograms, and the cloned histogram retains its attributes. DrawClonePad will copy the canvas to the active pad; the target does not have to be a canvas. It can also be a pad on a canvas.

Different draw options

Different draw options

If you want to copy and paste a graphic object from one canvas or pad to another canvas or pad, you can do so with DrawClone method inherited from TObject. All graphics objects inherit the TObject::DrawClone method. In this example, we create a new canvas with one histogram from each of the canvases from the script draw2dopt.C.

Repeat these steps for one histogram on each of the canvases created by the script, until you have one pad from each type. If you wanted to put the same annotation on each of the sub pads in the new canvas, you could use DrawClone to do so. Here we added the date to each pad. The steps to this are:

The option in the DrawClone method argument is the Draw option for a histogram or graph. A call to TH1::DrawClone can clone the histogram with a different draw option.

9.8.2 Programmatically

To copy and paste the four pads from the command line or in a script you would execute the following statements:

root[] .x tutorials/hist/draw2dopt.C
root[] TCanvas c1("c1","Copy Paste",200,200,800,600);
root[] surfaces->cd(1); // get the first pad
root[] TPad *p1 = gPad;
root[] lego->cd(2);// get the next pad
root[] TPad *p2 = gPad;
root[] cont->cd(3);// get the next pad
root[] TPad *p3 = gPad;
root[] c2h->cd(4);// get the next pad
root[] TPad *p4 = gPad;
root[] // to draw the four clones
root[] c1->cd();
root[] p1->DrawClone();
root[] p2->DrawClone();
root[] p3->DrawClone();
root[] p4->DrawClone();

Note that the pad is copied to the new canvas in the same location as in the old canvas. For example if you were to copy the third pad of surf to the top left corner of the target canvas you would have to reset the coordinates of the cloned pad.

9.9 Legends

Legends for a graph are obtained with a TLegend object. This object points to markers, lines, boxes, histograms, graphs and represent their marker, line, fill attributes. Any object that has a marker or line or fill attribute may have an associated legend. A TLegend is a panel with several entries (class TLegendEntry) and is created by the constructor

TLegend(Double_t x1, Double_t y1, Double_t x2, Double_t y2,
        const char *header, Option_t *option)

The legend is defined with default coordinates, border size and option. The legend coordinates (NDC) in the current pad are x1, y1, x2, y2. The default text attributes for the legend are:

The title is a regular entry and supports TLatex. The default is no title (header = 0). The options are the same as for TPave; by default, they are “brand”. Once the legend box is created, one has to add the text with the AddEntry() method:

TLegendEntry* TLegend::AddEntry(TObject *obj,
                                const char *label,
                                Option_t *option)

The parameters are:

One may also use the other form of the method AddEntry:

TLegendEntry* TLegend::AddEntry(const char *name,
                                const char *label,
                                Option_t *option)

Here name is the name of the object in the pad. Other parameters are as in the previous case. Next example shows how to create a legend:

leg = new TLegend(0.4,0.6,0.89,0.89);
leg->AddEntry(fun1,"One Theory","l");
leg->AddEntry(fun3,"Another Theory","f");
leg->AddEntry(gr,"The Data","p");
// oops we forgot the blue line... add it after
              "#sqrt{2#pi} P_{T} (#gamma) latex  formula","f");
// and add a header (or "title") for the legend
leg->SetHeader("The Legend Title");

Here fun1, fun2, fun3 and gr are pre-existing functions and graphs. You can edit the TLegend by right clicking on it.

A legend example

A legend example

9.10 The PostScript Interface

To generate a PostScript (or encapsulated PostScript) file for a single image in a canvas, you can:

c1->Print("") // or

Next example prints the picture in the pad pointed by pad1.


The TPad::Print method has a second parameter called option. Its value can be:

You do not need to specify this second parameter; you can indicate by the filename extension what format you want to save a canvas in (i.e., canvas.gif, canvas.C, etc).

The size of the PostScript picture, by default, is computed to keep the aspect ratio of the picture on the screen, where the size along x is always 20 cm. You can set the size of the PostScript picture before generating the picture with a command such as:

TPostScript myps("",111)

The first parameter in the TPostScript constructor is the name of the file; the second one is the format option:

You can set the default paper size with:


You can resume writing again in this file with myps.Open(). Note that you may have several Post Script files opened simultaneously. Use TPostScript::Text(x,y,"string") to add text to a postscript file. This method writes the string in quotes into a PostScript file at position x, y in world coordinates.

9.10.1 Special Characters

The following characters have a special action on the PostScript file:

These special characters are printed as such on the screen. To generate one of these characters on the PostScript file, you must escape it with the escape character “@”. The use of these special characters is illustrated in several scripts referenced by the TPostScript constructor.

9.10.2 Writing Several Canvases to the Same PostScript File

The following sequence writes the canvas to “” and closes the postscript file:

TCanvas c1("c1");

If the Postscript file name finishes with “(”, the file remains opened (it is not closed). If the Postscript file name finishes with “)” and the file has been opened with “(”, the file is closed.

   TCanvas c1("c1");
   c1.Print(""); // write canvas and keep the ps file open
   c1.Print("");  // canvas is added to ""
   c1.Print(""); // canvas is added to ""
                       // and ps file is closed

The TCanvas::Print("") mechanism is very useful, but it can be a little inconvenient to have the action of opening/closing a file being atomic with printing a page. Particularly if pages are being generated in some loop, one needs to detect the special cases of first and last page. The “[” and “]” can be used instead of “(” and “)” as shown in the next example.

c1.Print("[");      // no actual print; just open
for (i=0; i<10; ++i) {
   // fill canvas for context i
   c1.Print("");   // actually print canvas to
}  // end loop
c1.Print("]");     // no actual print; just close

The following script illustrates how to open a postscript file and draw several pictures. The generation of a new postscript page is automatic when TCanvas::Clear is called by object->Draw().

   TFile f("hsimple.root");
   TCanvas c1("c1","canvas",800,600);

   //select PostScript  output type
   Int_t type = 111;         //portrait  ps
   // Int_t type = 112;      //landscape ps
   // Int_t type = 113;      //eps

   //create a PostScript  file and set the paper size
   TPostScript ps("",type);
   ps.Range(16,24);          //set x,y of printed page

   //draw 3 histograms from file hsimple.root on separate pages
   c1.Update();              //force drawing in a script

The next example does the same:

   TFile f("hsimple.root");
   TCanvas c1("c1","canvas",800,600);

   //set x,y of printed page

   //draw 3 histograms from file hsimple.root on separate pages
   c1->Print("", "Portrait");

This following example shows two pages. The canvas is divided. TPostScript::NewPage must be called before starting a new picture. object->Draw does not clear the canvas in this case because we clear only the pads and not the main canvas. Note that c1->Update must be called at the end of the first picture.

   TFile *f1 = new TFile("hsimple.root");
   TCanvas *c1 = new TCanvas("c1");
   TPostScript *ps = new TPostScript("",112);

   // picture 1

   // picture 2

   // invoke PostScript  viewer

The next one does the same:

   TFile *f1 = new TFile("hsimple.root");
   TCanvas *c1 = new TCanvas("c1");

   // picture 1
   c1->Print("", "Landscape");

   // picture 2
   gSystem->Exec("gs");  // invoke PostScript  viewer

9.10.3 The Color Models

TPostScript (and TPDF) support two color models: RGB and CMYK. CMY and CMYK models are subtractive color models unlike RGB which is an additive. They are mainly used for printing purposes. CMY means Cyan Magenta Yellow to convert RGB to CMY it is enough to do: C=1-R, M=1-G and Y=1-B. CMYK has one more component K (black). The conversion from RGB to CMYK is:

 Double_t Black   = TMath::Min(TMath::Min(1-Red,1-Green),1-Blue);
 Double_t Cyan    = (1-Red-Black)/(1-Black);
 Double_t Magenta = (1-Green-Black)/(1-Black);
 Double_t Yellow  = (1-Blue-Black)/(1-Black);

CMYK add the black component which allows to have a better quality for black printing. TPostScript (and TPDF) support the CMYK model. To change the color model use:


9.11 The PDF Interface

Like PostScript, PDF is a vector graphics output format allowing a very high graphics output quality. The functionnalities provided by this class are very similar to those provided by TPostScript`.

Compare to PostScript output, the PDF files are usually smaller because some parts of them can be compressed.

PDF also allows to define table of contents. This facility can be used in ROOT. The following example shows how to proceed:

   TCanvas* canvas = new TCanvas("canvas");
   TH1F* histo = new TH1F("histo","test 1",10,0.,10.);
   canvas->Print("plots.pdf(","Title:One bin filled");
   canvas->Print("plots.pdf","Title:Two bins filled");
   canvas->Print("plots.pdf","Title:Three bins filled");
   canvas->Print("plots.pdf","Title:Four bins filled");
   canvas->Print("plots.pdf)","Title:The fourth bin content is 2");

Each character string following the keyword “Title:” makes a new entry in the table of contents.

9.12 Create or Modify a Style

All objects that can be drawn in a pad inherit from one or more attribute classes like TAttLine, TAttFill, TAttText, TAttMarker. When objects are created, their default attributes are taken from the current style. The current style is an object of the class TStyle and can be referenced via the global variable gStyle (in TStyle.h). See the class TStyle for a complete list of the attributes that can be set in one style.

ROOT provides several styles called:

The “Default” style is created by:

TStyle *default = new TStyle("Default","Default Style");

The “Plain” style can be used if you want to get a “conventional” PostScript output or if you are working on a monochrome display. The following example shows how to create it.

TStyle *plain  = new TStyle("Plain",
                            "Plain Style(no colors/fill areas)");

You can set the current style by:


You can get a pointer to an existing style by:

TStyle *style = gROOT->GetStyle(style_name);

You can create additional styles by:

TStyle *st1 = new TStyle("st1","my style");
st1->cd();  // this becomes now the current style gStyle

In your rootlogon.C file, you can redefine the default parameters via statements like:


Note that when an object is created, its attributes are taken from the current style. For example, you may have created a histogram in a previous session and saved it in a file. Meanwhile, if you have changed the style, the histogram will be drawn with the old attributes. You can force the current style attributes to be set when you read an object from a file by calling ForceStyle before reading the objects from the file.


When you call gROOT->ForceStyle() and read an object from a ROOT file, the object’s method UseCurrentStyle is called. The attributes saved with the object are replaced by the current style attributes. You call also call myObject->UseCurrentStyle() directly. For example if you have a canvas or pad with your histogram or any other object, you can force these objects to get the attributes of the current style by:


The description of the style functions should be clear from the name of the TStyle setters or getters. Some functions have an extended description, in particular:

9.13 3D Viewers

ROOT provides several viewers capable of displaying 3D content:

The X3D and GL viewers are created as external windows, associated with a pad, and displaying the same content as it. Only these external viewers are detailed here - for Pad (TPad, TView classes) you should refer to “Graphical Containers: Canvas and Pad” and the class definitions.

All viewers use a common architecture to publish 3D objects to the viewer - described in “Common 3D Viewer Architecture” below. In most cases, you will not need to use this, working instead with a package, such as the “The Geometry Package”, which provides comprehensive, high level functionality to create and place objects into complex 3D scenes, and uses the viewer architecture internally to show the result in your chosen viewer.

9.13.1 Invoking a 3D viewer

A 3D viewer can be created in a script by passing the appropriate option to Draw()when attaching the drawn object(s) to a pad. For a fuller explanation of pads, attaching objects with Draw() etc. refer to “Graphical Containers: Canvas and Pad”.

root[] myShapes->Draw("ogl");

Valid option strings are:

If no option is passed to Draw() then the “pad” is used by default. If you already have content in a pad, which you would like to display in one of the external viewers you can select from the canvas View menu / View With, and pick the viewer type.

Invoking external 3D viewers from canvas menus

Invoking external 3D viewers from canvas menus

Note: A current limitation means that when an external viewer is created the pad is no longer redrawn. When the external viewer is closed, clicking in the pad will refresh.

9.13.2 The GL Viewer

The GL Viewer uses (or compliant libraries such as ) to generate high quality, high-performance 3D renderings, with sophisticated lighting, materials and rendering styles for 3D scenes. Many users will be able to take advantage of hardware acceleration of the underlying OpenGL commands by their computer’s video card, resulting is considerable performance gains - up to interactive manipulation of 1000’s of complex shapes in real-time.

The GL Viewer is supported on all official ROOT platforms (assuming you have suitable libraries), and is the main 3D viewer, which development effort is concentrated upon. As OpenGL® is a trademark we refer to our viewer built on this technology as the ‘GL Viewer’. The code for it can be found under $ROOTSYS/gl.

The GL 3D Viewer

The GL 3D Viewer

You can manipulate the viewer via the GUI or via the base TGLViewer object behind the interface. These are detailed below - see also $ROOTSYS/tutorials/gl/glViewerExercise.C. Projections Modes (Cameras)

The GL Viewer supports two basic types of camera, which affect how the 3D world is projected onto the 2D render area:

You can select the active camera from the viewer’s Camera menu on the top menu bar. There are three perspective camera choices:

In each case the perspective camera is constrained to keep the chosen floor plane, defined by a pair of world axes, appearing level at all times - i.e. there is no banking of the ‘horizon’ that you experience when a plane rolls. There are also three orthographic camera choices:

Orthographic projections are generally constrained to look down one of the global axes of the world, with the other two axes lying horizontal/vertical on the viewer window. Therefore, XOY has the X-axis horizontal, the Y-axis vertical. You can always confirm the orientation and constraints of the camera in the world by enabling axis drawing in the “Guides” tab - see sections “Guides” and “Clipping” below. For orthographic camera a ruler-depicting current scene units is also available.

You can also pick the current camera by obtaining a handle to the GL Viewer object behind the interface:

TGLViewer * v = (TGLViewer *)gPad->GetViewer3D();

calling the method TGLViewer::SetCurrentCamera with one of the TGLViewer::ECameraType types:


See also $ROOTSYS/tutorials/gl/glViewerExercise.C. Adjusting Cameras

The interactions with the camera are summarized above. In each case the interaction is listed, along with description and user actions required to achieve it. For all cameras you can reset the original default view, framing the entire scene, by double clicking any mouse button.

GL Viewer camera interactions

GL Viewer camera interactions

For the Zoom interaction you can use the following modifiers combinations to adjust the sensitivity:

The modifiers must be applied after the zoom action has started (right mouse button is down).

Note for orthographic cameras:

Note for perspective cameras:

Configure the camera by calling the methods SetPerspectiveCamera() or SetOrthographicCamera() of TGLViewer:

TGLViewer * v = (TGLViewer *)gPad->GetViewer3D();
v->SetPerspectiveCamera (camera,fov,dolly,center,hRotate,vRotate);

Note - you can configure any of the six cameras in the viewer at any time, but you will not see the result until the camera is made current. Draw Styles

The GL Viewer supports three different rendering modes, which are applied to all the objects in your scene, but not Clip Shapes and Guides (See “Clipping” and “Manipulators”). These are shown below, along with the key used to activate the style.

GL Viewer draw styles

GL Viewer draw styles

Filled Polygons Wireframe Outline Enable with ‘r’ key Enable with ‘w’ key Enable with ‘t’ key Solid polygons, with hidden surface Object edges in color, with Combination of Filled Polygons removal, color surface materials, no surface filling/hiding. and Outline styles. Solid opacity, specular reflection etc. shapes with edges. Black background. Black background. White background.

Call method TGLViewer::SetStyle with one of TGLRnrCtx::EDrawStyleflags kFill, kOutline, kWireFrame:

v->SetStyle(TGLRnrCtx::kFill); Lighting / Style

The GL viewer creates five diffuse lights (left, right, top, bottom, and front) arranged around the 3D scene. These lights are carried with the camera - that is they are always in same position relative to your eye - the left light always shines from the left.

Light controls are located: Viewer Controls Pane ‘Style’.

Each light has a checkbox to enable/disable it. Set lights on/off with TGLLightSet::SetLight e.g.

v->GetLightSet()->SetLight(TGLLightSet::kLightBottom, kFALSE); Clipping

The GL viewer supports interactive clipping, enabling you to remove sections of your 3D scene and the shapes, revealing internal details.

GL Viewer interactive box clipping

GL Viewer interactive box clipping

The controls for clipping can be found under: Viewer Controls Pane ‘Clipping’ tab.

Two clipping ‘shapes’ are currently supported:

Pick the type from the radio buttons - only one (or none) may be active at one time.

The clip object can be adjusted by:

To show and/or directly manipulate the object check the ‘Show / Edit in Viewer’ checkbox. The clip object is drawn in semi-transparent light brown. The current manipulator is attached to it, allowing you direct control over its position, scale and rotation. See “Manipulators” section below for details on using viewer manipulators.

The clip plane is described by the standard plane equation: ax+by+cz+d=0, where the factors a, b, c, d are entered into the edit boxes, and applied using the ‘Apply’ button.

The clip box is described by its center position, entered in the ‘Center X’, ‘Center Y’ and ‘Center Z’ edit boxes, and its lengths (extents) entered in the ‘Length X’, ‘Length Y’ and ‘Length Z’ edit boxes.

This clipping is achieved using OpenGL clip plane support; as such, there are certain limitations:

Set the current clip object with TGLClipSet::SetClipType


Configure the clip object with TGLClipSet::SetClipState

Double_t planeEq[4] = {0.5,1.0,-1.0, 2.0};
v->GetClipSet()->SetClipState(TGLClipSet::kClipPlane, planeEq);

As with cameras, any clip can be configured at any time, but you must set the clip current to see the effect. Manipulators

Manipulators are GUI ‘widgets’ or controls attached to a 3D object in the viewer, allowing a direct manipulation of the object’s geometry. There are three manipulators for the three basic geometries transformations. In each case, the manipulator consists of three components, one for each local axis of the object, shown in standard colors: red (X), green (Y) and blue (Z).

GL Viewer object manipulators

GL Viewer object manipulators

Activate the manipulator by moving the mouse over one of these components (which turns yellow to indicate active state). Click with left mouse and drag this active component to perform the manipulation. Toggle between the manipulator types using the ‘x’, ‘c’, ‘v’ keys while the mouse cursoris above the manipulator. Note: Manipulators cannot be controlled via the API at present. Guides

Guides are visual aids drawn into the viewer world. Controls for these are under the “Guides” tab:

Viewer Controls Pane Guides Tab

Axes show the world (global) frame coordinatedirections: X (red), Y (green) and Z (blue). The negative portion of the axis line is shown in dark color, the positive in bright. The axis name and minimum / maximum values are labeled in the same color. There are three options for axes drawing - selected by radio buttons:

For edge axes, the zero value for each axis is marked on the axis line with a colored sphere. For origin axes, a single white sphere is shown at the origin.

Edge axes are depth clipped - i.e. are obscured by 3D objects in front of them. Origin axes (which generally pass through the middle of the 3D scene) are not depth clipped - so always visible.

A single orange sphere of fixed view port (window) size can be shown at any arbitrary position. Enable / disable the drawing with ‘Show’ checkbox. Enter X/Y/Z position in the edit boxes to set position. Initial position is at the center of the scene.

Set the guides using TGLViewer::SetGuideState e.g. to enable edge axes, and enable a reference marker at world position 50, 60, 100:

Double_t refPos[3] = {50.0,60.0,100.0};
v->SetGuideState(TGLUtil::kAxesEdge, kTRUE, refPos); Selecting Scene Shapes

You can select a single shape from your scene by pressing ‘Shift’ key, pointing and left clicking anywhere on the shape in the viewer. Selection is currently shown by drawing the shape-bounding box (not depth clipped) in white (polygon or wire frame render styles) or red (outline render style). Manipulators supported by the shape are drawn in red, green and blue while the non-supported ones are drawn in grey. To deselect a shape, either select another, or shift/click anywhere on the background (empty space) in the viewer. You cannot select Manipulators or Guides (Axes / Reference Marker). Editing Shapes

When a shape is selected, the viewer’s control pane shows the user interface that allows you to review and adjust the color and geometry properties of the shape.

Note: At present modifications to the shapes are local to the viewer - they are not propagated back to external objects/client that published to the viewer. The changes are preserved only until the viewer is closed. In some cases, this will never be feasible as there is not a one-to-one correspondence between a shape in the viewer and a single external object in which the modification could be stored. Colors / Style

Viewer Controls Pane ‘Style’ tab.

A full description of OpenGL materials, colors and lighting is beyond the scope of this document. You should refer to the OpenGL programming manual (Red Book) for a full discussion. In most cases adjustment of the Diffuse color material + Opacity/Shine properties is sufficient to achieve desired results.

A shape has four-color materials (components):

For each of these you can select the component via the radio buttons. Each component can have the red, green and blue values for the component adjusted via the sliders. You can apply this adjustment to the shape itself, or to all shapes sharing a common ‘family’. Shapes of the same family have external objects with the same TObject name string. You can also adjust the ‘Opacity’ and ‘Shine’ for the shapes materials via the sliders. Geometry

Viewer Controls Pane ‘Geometry’ tab.

Review and modify the shapes X/Y/Z center and scaling factors via the edit boxes. Selection and editing of shapes is not available via the API at present. Outputting Viewer Contents

The current viewer rendering can be output to an external EPS or PDF, using the options under the ‘File’ menu on the top menu bar. The file is named ‘viewer.eps’ or ‘viewer.pdf’ and written to the current ROOT directory.

9.13.3 The X3D Viewer

The X3D viewer is a fairly simple and limited viewer, capable of showing basic lines and polygons. It lacks the quality, performance and more advanced features of the GL Viewer, and additionally is not supported on Windows. It is not actively developed and you are encouraged to use the GL Viewer out of preference. The below table presents the main interactions - these are repeated in the Help dialog of the viewer.

Action KeyActionKey

Wireframe Mode wRotate about xx a

Hidden Line Mode eRotate about yy b

Hidden Surface Mode rRotate about zz c

Move object down uAuto-rotate about x1 2 3

Move object up iAuto-rotate about y4 5 6

Move object left lAuto-rotate about z7 8 9

Move object right hToggle controls styleo

Move object forward jToggle stereo displays

Move object backward kToggle blue stereo viewd

Adjust focus (stereo mode) [ ] { }Toggle double bufferf

Rotate object Left mouse button down + move.

9.13.4 Common 3D Viewer Architecture

The 3D Viewer Architecture provides a common mechanism for viewer clients to publish 3D objects to it. It enables:

The architecture consists of:

A typical interaction between viewer and client using these, taken from TGeoPainter is:

TVirtualViewer3D * viewer = gPad->GetViewer3D();
// Does viewer prefer local frame positions?
Bool_t localFrame = viewer->PreferLocalFrame();
// Perform first fetch of buffer from the shape and try adding it to the viewer
const TBuffer3D &buffer = shape.GetBuffer3D(TBuffer3D::kCore |
TBuffer3D::kBoundingBox |
Int_t reqSections = viewer->AddObject(buffer, &addDaughters);

// If the viewer requires additional sections fetch from the shape
// (if possible) and add again
if (reqSections != TBuffer3D::kNone)
shape.GetBuffer3D(reqSections, localFrame);

Together these allow clients to publish objects to any one of the 3D viewers free of viewer specific drawing code. They allow our simple x3d viewer, and considerably more sophisticated OpenGL one to both work with both geometry libraries (g3d and geom) efficiently.

In addition to external viewers, created in separate windows, this architecture is also used by internal TPad drawing when it requires 3D projections. Publishing to a viewer consists of the following steps:

1- Create / obtain viewer handle.

2- Begin scene on viewer.

3- Fill mandatory parts of TBuffer3D describing object.

4- Add to viewer.

5- Fill optional parts of TBuffer3D as requested by viewer.

[ …. repeat 3/4/5 as required for other/child objects]

6- End scene on viewer.

You should attach the top-level node of your external geometry (or the manager) to a TPad object using TObject::Draw(), and perform the publishing to the viewer in your object’s TObject::Paint() overloaded method. See “Scene Rebuilds”, and example scripts, for more details. Creating / Obtaining Viewer Handle

External viewers are bound to a TPad object (this may be removed as a requirement in the future). You can create or obtain the current viewer handle via the method:

TVirtualViewer3D * v = gPad->GetViewer3D("type");

Here the “type” string defines the viewer type - currently one of:

If no type is passed (null string), and there is no current viewer, then the type is defaulted to “pad”. If no type is passed and there is a current viewer, then this is returned - hence once a viewer is created it can be obtained elsewhere by:

TVirtualViewer3D * v = gPad->GetViewer3D(); Opening / Closing Scenes

Objects must be added to viewer between BeginScene() and EndScene() calls e.g.

// Add objects
viewer ->EndScene();

These calls enable the viewer to suspend redraws, and perform internal caching/setup. If the object you attach to the pad derives from TAtt3D, then the pad will take responsibility for calling BeginScene() and EndScene() for you. You can always test if the scene is already open for object addition with:

Overview of 3D viewer architecture

Overview of 3D viewer architecture

Note: the x3d viewer does not support rebuilding of scenes - objects added after the first Open/Close Scene pair will be ignored. Describing Objects - Filling TBuffer3D

The viewers behind the TVirtualViewer3D interface differ greatly in their capabilities e.g.

Similarly some viewer clients are only capable of providing positions in master frame, cannot provide bounding boxes etc. Additionally we do not want to incur the cost of expensive tessellation operations if the viewer does not require them. To cope with these variations the TBuffer3D objects are filled by negotiation with the viewer.

TBuffer3D class hierarchy

TBuffer3D class hierarchy

TBuffer3D classes are conceptually divided into enumerated sections: kCore, kBoundingBox, kRaw - see the class diagram and the file TBuffer3D.h for more details. The TBuffer3D methods SectionsValid(), SetSectionsValid(), ClearSectionsValid() are used to test, set, clear these section validity flags e.g.

if (buffer.SectionsValid(TBuffer3D:: kShapeSpecific)) {

The sections found in the base TBuffer3D (kCore/kBoundingBox/kRawSizes/kRaw) are sufficient to describe any tessellated shape in a generic fashion. An additional kShapeSpecific section is added in TBuffer3D derived classes, allowing a more abstract shape description (“a sphere of inner radius x, outer radius y”). This enables a viewer, which knows how to draw (tessellate) the shape itself to do so, while providing a generic fallback suitable for all viewers. The rules for client negotiation with the viewer are:

If the viewer requires more sections to be completed (kRaw/kRawSizes) TBuffer3D::AddObject() will return flags indicating which ones, otherwise it returns kNone. If requested, you must fill the buffer, mark these sections valid, and call TBuffer3D::AddObject again, to complete adding the object. For example, in out TGeo geometry package, in TGeoPainter::PaintShape, we perform the negotiation with viewer:

TVirtualViewer3D * viewer = gPad->GetViewer3D();
if (shape.IsA() != TGeoCompositeShape::Class()) {
   // Does viewer prefer local frame positions?
   Bool_t localFrame = viewer->PreferLocalFrame();
   // Perform first fetch of buffer from the shape and adding
   // it to the viewer
   const TBuffer3D &buffer = shape.GetBuffer3D(TBuffer3D::kCore |
   TBuffer3D::kBoundingBox |
   TBuffer3D::kShapeSpecific, localFrame);
   Int_t reqSections = viewer->AddObject(buffer, &addDaughters);
   // If the viewer requires additional sections fetch from the
   // shape (if possible) and add again
   if (reqSections != TBuffer3D::kNone) {
      shape.GetBuffer3D(reqSections, localFrame);
      viewer->AddObject(buffer, &addDaughters);

The buffer is supplied/filled by the appropriate TShape::GetBuffer3D() and TShape::FillBuffer3D overloads e.g. for a sphere in TGeoSphere.

const TBuffer3D &TGeoSphere::GetBuffer3D(Int_t reqSections,
Bool_t localFrame) const {
   // Fills a static 3D buffer and returns a reference.
   static TBuffer3DSphere buffer;
   // Filling of kBoundingBox is defered to TGeoBBox, and
   // kCore on up to TGeoShape
   TGeoBBox::FillBuffer3D(buffer, reqSections, localFrame);
   // Complete kShapeSpecific section for sphere
   if (reqSections & TBuffer3D::kShapeSpecific) {
      buffer.fRadiusInner  = fRmin;
      buffer.fRadiusOuter  = fRmax;
   // Complete kRawSizes section
   if (reqSections & TBuffer3D::kRawSizes) {
   // Complete kRaw tesselation section
   if ((reqSections & TBuffer3D::kRaw) &&
        buffer.SectionsValid(TBuffer3D::kRawSizes)) {
      // Transform points to master frame if viewer requires it
      // The fLocalFrame flag and translation matrix will have
      // already been set in TGeoShape::FillBuffer3D() as required
      if (!buffer.fLocalFrame)
         TransformPoints(buffer.fPnts, buffer.NbPnts());
   return buffer;

Note: Shape Specific TBuffer3D Derived Classes

Currently we provide the following shape specific classes, which the GL Viewer can take advantage of (see TBuffer3D.h and TBuffer3DTypes.h)

See the above example from TGeoSphere::GetBuffer3D and also equivalent functions in TGeoTube, TGeoTubeSeg and TGeoCtub. Anyone is free to add new TBuffer3D classes, but it should be clear that one or more viewers will require updating to be able to take advantage of them. Hence we only provide classes which existing viewers can benefit from. The number of native shapes in GL Viewer will be expanded in the future. Master / Local Reference Frames

The Core section of TBuffer3D contains two members relating to reference frames:

If fLocalFrame is false, fLocalMaster should contain an identity matrix. This is set by default, and can be reset using the TBuffer3D::SetLocalMasterIdentity() method. Bounding Boxes

You are not obliged to complete the kBoundingBox section, as any viewer requiring one internally (GL Viewer) will build it if you do not provide. However to do this the viewer will force you to provide the (expensive) raw tessellation, and the resulting box will be axis aligned with the overall scene, which is non-ideal for rotated shapes. As we need to support orientated (rotated) bounding boxes, TBuffer3D requires the 6 vertices of the box. We also provide a convenience function, TBuffer::SetAABoundingBox(), for simpler case of setting an axis aligned bounding box. The bounding box should be filled in same frame (local / master) as the rest of the TBuffer3D, and inaccordance with fLocalFrame flag.

A typical example from TGeoBBox::FillBuffer3D:

   if (reqSections & TBuffer3D::kBoundingBox) {
      Double_t halfLengths[3] = { fDX, fDY, fDZ };
      buffer.SetAABoundingBox(fOrigin, halfLengths);
      if (!buffer.fLocalFrame) {
         TransformPoints(buffer.fBBVertex[0], 8);
   } Logical and Physical Objects

Some viewers can support two types of object placement:

The second case is very typical in geometry packages, e.g. ROOT’s TGeo package, GEANT4 etc, where we have very large number repeated placements of relatively few unique “shapes”.

Some viewers (GL Viewer only at present) are able to take advantage of this by identifying unique logical shapes from the fID logical ID member of TBuffer3D. If repeated addition of the same fID is found, the shape is cached already - and the costly tessellation does not need to be sent again. The viewer can also perform internal GL specific caching (display lists) with considerable performance gains in these cases. For this to work correctly the logical object in must be described in TBuffer3D in the local reference frame, complete with the local/master translation. In some cases you will not have a real object you can reasonably set TBuffer3D::fID to, or the object is recycled or temporary. To suppress internal caching in the GL Viewer in these cases, set TBuffer3D::fID to 0 (null).

The viewer indicates it can support local frame objects through the TVirtualViewer3D interface method: PreferLocalFrame(). If this returns kTRUE you can make repeated calls to AddObject(), with TBuffer3D containing the same fID, and different fLocalMaster placements.

For viewers supporting logical/physical objects, the TBuffer3D content refers to the properties of the logical object, with the exception of:

attributes, which can be varied for each physical object.

As a minimum requirement all clients must be capable of filling the raw tessellation of the object buffer, in the master reference frame. Conversely viewers must always be capable of displaying the object described by this buffer. If either does not meet this requirement the object may not be displayed. Scene Rebuilds

TBuffer3D::AddObject is not an explicit command to the viewer - it may for various reasons decide to ignore it:

In all these cases TBuffer3D::AddObject() returns kNone, as it does for successful addition, indicating it does not require further information about this object. Hence you should not try to make any assumptions about what the viewer did with the object. The viewer may decide to force the client to rebuild (republish) the scene, obtaining a different collection of objects, if the internal viewer state changes .e.g. significant camera move. It does this presently by forcing a repaint on the attached TPad object - hence you should attach you master geometry object to the pad (via TObject::Draw()), and perform the publishing to the viewer in response to TObject::Paint(). Physical IDs

TVirtualViewer3D provides for two methods of object addition:

virtual Int_t AddObject(const TBuffer3D &buffer,
                        Bool_t * addChildren = 0)
virtual Int_t AddObject(UInt_t physicalID,
                        const TBuffer3D & buffer,
                        Bool_t *addChildren = 0)

If you use the first (simple) case a viewer using logical/physical pairs will generate sequential IDs for each physical object internally. Scene rebuilds will require destruction and recreation of all physical objects. For the second you can specify an identifier from the client side, which must be unique and stable - i.e. the IDs of a published object is consistent, regardless of changes in termination of contained child geometry branches. In this case the viewer can safely cache the physical objects across scene rebuilds, discarding those no longer of interest. Child Objects

In many geometries there is a rigid containment hierarchy, and so if the viewer is not interested in a certain object due to limits/size then it will also not be interest in any of the contained branch of siblings. Both TBuffer3D::AddObject() methods have an addChildren return parameter. The viewer will complete this (if passed) indicating if children of the object just sent are worth sending. Recycling TBuffer3D

Once add TBuffer3D::AddObject() has been called, the contents are copied to the viewer’s internal data structures. You are free to destroy this TBuffer3D, or recycle it for the next object if suitable. Examples

For an example of a simple geometry, working in master reference frame examine the code under $ROOTSYS/g3d. For a more complex example, which works in both master and local frames, and uses logical/physical division of shape geometry and placement, examine the code under $ROOTSYS/geom - in particular TGeoShape hierarchy, and the painter object TGeoPainter (under geopainter) where the negotiation with the viewer is performed.

10 Folders and Tasks

10.1 Folders

A TFolder is a collection of objects visible and expandable in the ROOT object browser. Folders have a name and a title and are identified in the folder hierarchy by an “UNIX-like” naming convention. The base of all folders is //root. It is visible at the top of the left panel in the browser. The browser shows several folders under //root.

New folders can be added and removed to/from a folder.

10.2 Why Use Folders?

One reason to use folders is to reduce class dependencies and improve modularity. Each set of data has a producer class and one or many consumer classes. When using folders, the producer class places a pointer to the data into a folder, and the consumer class retrieves a reference to the folder.

The consumer can access the objects in a folder by specifying the path name of the folder.

Here is an example of a folder’s path name:


One does not have to specify the full path name. If the partial path name is unique, it will find it; otherwise it will return the first occurrence of the path.

The first diagram shows a system without folders. The objects have pointers to each other to access each other’s data. Pointers are an efficient way to share data between classes. However, a direct pointer creates a direct coupling between classes. This design can become a very tangled web of dependencies in a system with a large number of classes.

In the second diagram, a reference to the data is in the folder and the consumers refer to the folder rather than each other to access the data. The naming and search service provided by the ROOT folders hierarchy provides an alternative. It loosely couples the classes and greatly enhances I/O operations. In this way, folders separate the data from the algorithms and greatly improve the modularity of an application by minimizing the class dependencies.

In addition, the folder hierarchy creates a picture of the data organization. This is useful when discussing data design issues or when learning the data organization. The example below illustrates this point.

10.3 How to Use Folders

Using folders means to build a hierarchy of folders, posting the reference to the data in the folder by the producer, and creating a reference to the folder by the user.

10.3.1 Creating a Folder Hierarchy

To create a folder hierarchy you add the top folder of your hierarchy to //root. Then you add a folder to an existing folder with the TFolder::AddFolder method. This method takes two parameters: the name and title of the folder to be added. It returns a pointer of the newly created folder.

The code below creates the folder hierarchy shown in the browser. In this macro, the folder is also added to the list of browsable. This way, it is visible in the browser on the top level.

   // Add the top folder of my hierary to //root
   TFolder *aliroot=gROOT->GetRootFolder()->AddFolder("aliroot",
                                   "aliroot top level folders");
   // Add the hierarchy to the list of browsables

   // Create and add the constants folder
   TFolder *constants=aliroot->AddFolder("Constants",
                                         "Detector constants");

   // Create and add the pdg folder to pdg
   TFolder *pdg = constants->AddFolder("DatabasePDG","PDG database");

   // Create and add the run folder
   TFolder *run = aliroot->AddFolder("Run","Run dependent folders");

   // Create and add the configuration folder to run
   TFolder *configuration = run->AddFolder("Configuration",
                                           "Run configuration");

   // Create and add the run_mc folder
   TFolder *run_mc = aliroot->AddFolder("RunMC",
                     "MonteCarlo run dependent folders");

   // Create and add the configuration_mc folder to run_mc
   TFolder *configuration_mc = run_mc->AddFolder("Configuration",
                                    "MonteCarlo run configuration");

10.3.2 Posting Data to a Folder (Producer)

A TFolder can contain other folders as shown above or any TObject descendents. In general, users will not post a single object to a folder; they will store a collection or multiple collections in a folder. For example, to add an array to a folder:

TObjArray *array;

10.3.3 Reading Data from a Folder (Consumer)

One can search for a folder or an object in a folder using the TROOT::FindObjectAny method. It analyzes the string passed as its argument and searches in the hierarchy until it finds an object or folder matching the name. With FindObjectAny, you can give the full path name, or the name of the folder. If only the name of the folder is given, it will return the first instance of that name. A string-based search is time consuming. If the retrieved object is used frequently or inside a loop, you should save a pointer to the object as a class data member. Use the naming service only in the initialization of the consumer class. When a folder is deleted, any reference to it in the parent or other folder is deleted also.

   // or ...

By default, a folder does not own the object it contains. You can overwrite that with TFolder::SetOwner. Once the folder is the owner of its contents, the contents are deleted when the folder is deleted. Some ROOT objects are automatically added to the folder hierarchy. For example, the following folders exist on start up:

//root/ROOT Files with the list of open Root files

//root/Classes with the list of active classes

//root/Geometries with active geometries

//root/Canvases with the list of active canvases

//root/Styles with the list of graphics styles

//root/Colors with the list of active colors

For example, if a file myFile.root is added to the list of files, one can retrieve a pointer to the corresponding TFile object with a statement like:

   TFile *myFile = (TFile*)gROOT->FindObjectAny(
   TFile *myFile = (TFile*)gROOT->FindObjectAny("myFile.root");

10.4 Tasks

Tasks can be organized into a hierarchy and displayed in the browser. The TTask class is the base class from which the tasks are derived. To give task functionality, you need to subclass the TTask class and override the Exec method. An example of TTask subclassesis $ROOTSYS/tutorials/MyTasks.cxx. The script that creates a task hierarchy and adds it to the browser is $ROOTSYS/tutorials/tasks.C. Here is a part of MyTasks.cxx that shows how to subclass from TTask.

// A set of classes deriving from TTask see macro tasks.C. The Exec
// function of each class prints one line when it is called.
#include "TTask.h"
class MyRun : public TTask {
   MyRun() { ; }
   MyRun(const char *name,const char *title);
   virtual ~MyRun() { ; }
   void Exec(Option_t *option="");
   ClassDef(MyRun,1)         // Run Reconstruction task

class MyEvent : public TTask {
   MyEvent() { ; }
   MyEvent(const char *name,const char *title);
   virtual ~MyEvent() { ; }
   void Exec(Option_t *option="");
   ClassDef(MyEvent,1)   // Event Reconstruction task

Later in MyTasks.cxx, we can see examples of the constructor and overridden Exec() method:

MyRun::MyRun(const char *name,const char *title):TTask(name,title)
void MyRun::Exec(Option_t *option)
   printf("MyRun executingn");

Each TTask derived class may contain other TTasks that can be executed recursively. In this way, a complex program can be dynamically built and executed by invoking the services of the top level task or one of its subtasks. The constructor of TTask has two arguments: the name and the title. This script creates the task defined above, and creates a hierarchy of tasks.

// Show the tasks in a browser. To execute a Task, select
// "ExecuteTask" in the context menu see also other functions in the
// TTask context menu, such as:
//           -setting a breakpoint in one or more tasks
//           -enabling/disabling one task, etc
void tasks() {
   gROOT->ProcessLine(".L MyTasks.cxx+");

   TTask *run = new MyRun("run","Process one run");
   TTask *event = new MyEvent("event","Process one event");
   TTask *geomInit = new MyGeomInit("geomInit",
                         "Geometry Initialisation");
   TTask *matInit    = new MyMaterialInit("matInit",
   TTask *tracker = new MyTracker("tracker","Tracker manager");
   TTask *tpc     = new MyRecTPC("tpc","TPC Reconstruction");
   TTask *its     = new MyRecITS("its","ITS Reconstruction");
   TTask *muon    = new MyRecMUON("muon","MUON Reconstruction");
   TTask *phos    = new MyRecPHOS("phos","PHOS Reconstruction");
   TTask *rich    = new MyRecRICH("rich","RICH Reconstruction");
   TTask *trd     = new MyRecTRD("trd","TRD Reconstruction");
   TTask *global  = new MyRecGlobal("global","Global Reconstruction");

   // Create a hierarchy by adding sub tasks

   // Add the top level task

   // Add the task to the browser
   new TBrowser;
Tasks in the ROOT browser

Tasks in the ROOT browser

Note that the first line loads the class definitions in MyTasks.cxx with ACLiC. ACLiC builds a shared library and adds the classes to the CINT dictionary. See “Adding a Class with ACLiC”.

To execute a TTask, you call the ExecuteTask method. ExecuteTask will recursively call:

If the top level task is added to the list of ROOT browseable objects, the tree of tasks can be seen in the ROOT browser. To add it to the browser, get the list of browseable objects first and add it to the collection.


The first parameter of the Add method is a pointer to a TTask, the second parameter is the string to show in the browser. If the string is left out, the name of the task is used.

After executing, the script above the browser will look like in this figure.

10.5 Execute and Debug Tasks

The browser can be used to start a task, set break points at the beginning of a task or when the task has completed. At a breakpoint, data structures generated by the execution up this point may be inspected asynchronously and then the execution can be resumed by selecting the “Continue” function of a task.

A task may be active or inactive (controlled by TTask::SetActive). When a task is inactive, its sub tasks are not executed. A task tree may be made persistent, saving the status of all the tasks.

11 Input/Output

This chapter covers the saving and reading of objects to and from ROOT files. It begins with an explanation of the physical layout of a ROOT file. It includes a discussion on compression, and file recovery. Then we explain the logical file, the class TFile and its methods. We show how to navigate in a file, how to save objects and read them back. We also include a discussion on Streamers. Streamers are the methods responsible to capture an objects current state to save it to disk or send it over the network. At the end of the chapter is a discussion on the two specialized ROOT files: TNetFile and TWebFile.

11.1 The Physical Layout of ROOT Files

A ROOT file is like a UNIX file directory. It can contain directories and objects organized in unlimited number of levels. It also is stored in machine independent format (ASCII, IEEE floating point, Big Endian byte ordering). To look at the physical layout of a ROOT file, we first create one. This example creates a ROOT file and 15 histograms, fills each histogram with 1000 entries from a Gaussian distribution, and writes them to the file.

   char name[10], title[20];
   TObjArray Hlist(0);      // create an array of Histograms
   TH1F* h;                 // create a pointer to a histogram
   // make and fill 15 histograms and add them to the object array
   for (Int_t i = 0; i < 15; i++) {
      sprintf(title,"histo nr:%d",i);
      h = new TH1F(name,title,100,-4,4);
   // open a file and write the array to the file
   TFile f("demo.root","recreate");

The example begins with a call to the TFile constructor. This class is describing the ROOT file (that has the extension “.root”). In the next section, we will cover TFile in details. The last line of the example closes the file. To view its contents we need to open it again, and to create a TBrowser object by:

root[] TFile f("demo.root")
root[] TBrowser browser;
The browser with 15 created histograms

The browser with 15 created histograms

You can check if the file is correctly opened by:

   TFile f("demo.root");
   if (f.IsZombie()) {
      cout << "Error opening file" << endl;
   } else {

Once we have the TFile object, we can call the TFile::Map() method to view the physical layout. The output prints the date/time, the start record address, the number of bytes in the record, the class name of the record and the compression factor.

root[] f.Map()
20051208/124502  At:100    N=114       TFile
20051208/124502  At:214    N=413       TH1F           CX =  2.35
20051208/124502  At:627    N=410       TH1F           CX =  2.36
20051208/124502  At:1037   N=396       TH1F           CX =  2.45
20051208/124502  At:1433   N=400       TH1F           CX =  2.42
20051208/124502  At:1833   N=402       TH1F           CX =  2.41
20051208/124502  At:2235   N=416       TH1F           CX =  2.33
20051208/124502  At:2651   N=406       TH1F           CX =  2.39
20051208/124502  At:3057   N=403       TH1F           CX =  2.40
20051208/124502  At:3460   N=411       TH1F           CX =  2.36
20051208/124502  At:3871   N=400       TH1F           CX =  2.42
20051208/124502  At:4271   N=409       TH1F           CX =  2.38
20051208/124502  At:4680   N=409       TH1F           CX =  2.38
20051208/124502  At:5089   N=420       TH1F           CX =  2.32
20051208/124502  At:5509   N=406       TH1F           CX =  2.40
20051208/124502  At:5915   N=405       TH1F           CX =  2.40
20051208/124503  At:6320   N=3052      StreamerInfo   CX =  3.16
20051208/124503  At:9372   N=732       KeysList
20051208/124503  At:10104  N=53        FreeSegments
20051208/124503  At:10157  N=1         END

Here we see the fifteen histograms (TH1F’s) with the first one starting at byte 148. We also see an entry TFile. You may notice that the first entry starts at byte 100. The first 100 bytes are taken by the file header.

11.1.1 The File Header

This table shows the file header information. When fVersion is greater than 1000000, the file is a large file (> 2 GB) and the offsets will be 8 bytes long. The location in brackets are the location in the case of a large file.


Value Name


1 -> 4


Root file identifier

5 -> 8


File format version

9 -> 12


Pointer to first data record

13 -> 16 [13->20]


Pointer to first free word at the EOF

17 -> 20 [21->28]


Pointer to FREE data record

21 -> 24 [29->32]


Number of bytes in FREE data record

25 -> 28 [33->36]


Number of free data records

29 -> 32 [37->40]


Number of bytes in TNamed at creation time

33 -> 33 [41->41]


Number of bytes for file pointers

34 -> 37 [42->45]


Zip compression level

34 -> 37 [46->53]


Pointer to TStreamerInfo record

34 -> 37 [54->57]


Number of bytes in TStreamerInfo record

34 -> 37 [58->75]


Universal Unique ID

The first four bytes of the file header contain the string “root” which identifies a file as a ROOT file. Because of this identifier, ROOT is not dependent on the “.root” extension. It is still a good idea to use the extension, just for us to recognize them easier. The nfree and value is the number of free records. This variable along with FNBytesFree keeps track of the free space in terms of records and bytes. This count also includes the deleted records, which are available again.

11.1.2 The Top Directory Description

The 84 bytes after the file header contain the top directory description, including the name, the date and time it was created, and the date and time of the last modification.

20010404/092347  At:64        N=84        TFile

11.1.3 The Histogram Records

What follows are the 15 histograms, in records of variable length.

20010404/092347  At:148       N=380       TH1F           CX =  2.49
20010404/092347  At:528       N=377       TH1F           CX =  2.51

The first 4 bytes of each record is an integer holding the number of bytes in this record. A negative number flags the record as deleted, and makes the space available for recycling in the next writing. The rest of bytes in the header contain all the information to identify uniquely a data block on the file. It is followed by the object data.

The next table explains the values in each individual record. If the key is located past the 32 bit file limit (> 2 GB) then some fields will be 8 bytes instead of 4 bytes (values between the brackets):


Value Name


1 -> 4


Length of compressed object (in bytes)

5 -> 6


TKey version identifier

7 -> 10


Length of uncompressed object

11 -> 14


Date and time when object was written to file

15 -> 16


Length of the key structure (in bytes)

17 -> 18


Cycle of key

19 -> 22 [19->26]


Pointer to record itself (consistency check)

23 -> 26 [27->34]


Pointer to directory header

27 -> 27 [35->35]


Number of bytes in the class name

28 -> … [36->..

.] | ClassName

| Object Class Name


| lname

| Number of bytes in the object name


| Name

| lName bytes with the name of the object


| lTitle

| Number of bytes in the object title


| Title

| Title of the object



| Data bytes associated to the object

You see a reference to TKey. It is explained in detail in the next section.

11.1.4 The Class Description List (StreamerInfo List)

The histogram records are followed by the StreamerInfo list of class descriptions. The list contains the description of each class that has been written to file.

20010404/092347  At:5854   N=2390   StreamerInfo   CX =  3.41

The class description is recursive, because to fully describe a class, its ancestors and object data members have to be described also. In demo.root, the class description list contains the description for:

This description is implemented by the TStreamerInfo class, and is often referred to as simply StreamerInfo. You can print a file’s StreamerInfolist with the TFile::ShowStreamerInfo method. Below is an example of the output. Only the first line of each class description is shown. The demo.root example contains only TH1F objects. Here we see the recursive nature of the class description; it contains the StreamerInfoof all the classes needed to describe TH1F.

root[] f.ShowStreamerInfo()
StreamerInfo for class: TH1F, version=1
  BASE     TH1         offset=0 type= 0 1-Dim histogram base class
  BASE     TArrayF     offset=0 type= 0 Array of floats

StreamerInfo for class: TH1, version=3
  BASE     TNamed      offset=0 type=67 The basis for named object(name,title)
  BASE     TAttLine    offset=0 type=0  Line attributes
  BASE     TAttFill    offset=0 type=0  Fill area attributes
  BASE     TAttMarker  offset=0 type=0  Marker attributes
  Int_t    fNcells     offset=0 type=3  number bins(1D),cells(2D)+U/Overflows
  TAxis    fXaxis      offset=0 type=61 X axis descriptor
  TAxis    fYaxis      offset=0 type=61 Y axis descriptor
  TAxis    fZaxis      offset=0 type=61 Z axis descriptor
  Short_t  fBarOffset  offset=0 type=2  (1000*offset) for barcharts or legos
  Short_t  fBarWidth   offset=0 type=2  (1000*width) for bar charts or legos
  Stat_t   fEntries    offset=0 type=8  Number of entries//continued...
  Stat_t   fTsumw      offset=0 type=8  Total Sum of weights
  Stat_t   fTsumw2     offset=0 type=8  Total Sum of squares of weights
  Stat_t   fTsumwx     offset=0 type=8  Total Sum of weight*X
  Stat_t   fTsumwx2    offset=0 type=8  Total Sum of weight*X*X
  Double_t fMaximum    offset=0 type=8  Maximum value for plotting
  Double_t fMinimum    offset=0 type=8  Minimum value for plotting
  Double_t fNormFactor offset=0 type=8  Normalization factor
  TArrayD  fContour    offset=0 type=62 Array to display contour levels
  TArrayD  fSumw2      offset=0 type=62 Array of sum of squares of weights
  TString  fOption     offset=0 type=65 histogram options
  TList*   fFunctions  offset=0 type=63 ->Pointer to list of functions(fits,user)

StreamerInfo for class: TNamed, version=1
StreamerInfo for class: TAttLine, version=1
StreamerInfo for class: TAttFill, version=1
StreamerInfo for class: TAttMarker, version=1
StreamerInfo for class: TArrayF, version=1
StreamerInfo for class: TArray, version=1
StreamerInfo for class: TAxis, version=6
StreamerInfo for class: TAttAxis, version=4

ROOT allows a class to have multiple versions, and each version has its own description in form of a StreamerInfo. Above you see the class name and version number. The StreamerInfolist has only one description for each class/version combination it encountered. The file can have multiple versions of the same class, for example objects of old and new versions of a class can be in the same file. The StreamerInfois described in detail in the section on Streamers.

11.1.5 The List of Keys and the List of Free Blocks

The last three entries on the output of TFile::Map() are the list of keys, the list of free segments, and the address where the data ends.. When a file is closed, it writes a linked list of keys at the end of the file. This is what we see in the third to the last entry. In our example, the list of keys is stored in 732 bytes beginning at byte# 8244.

20010404/092347    At:8244      N=732       KeysList
20010404/092347    At:8976      N=53        FreeSegments
20010404/092347    At:9029      N=1         END

The second to last entry is a list of free segments. In our case, this starts 8976 and is not very long, only 53 bytes, since we have not deleted any objects. The last entry is the address of the last byte in the file.

11.1.6 File Recovery

A file may become corrupted or it may be impossible to write it to disk and close it properly. For example if the file is too large and exceeds the disk quota, or the job crashes or a batch job reaches its time limit before the file can be closed. In these cases, it is imperative to recover and retain as much information as possible. ROOT provides an intelligent and elegant file recovery mechanism using the redundant directory information in the record header.

If a file that has been not properly closed is opened again, it is scanned and rebuilt according to the information in the record header. The recovery algorithm reads the file and creates the saved objects in memory according to the header information. It then rebuilds the directory and file structure. If the file is opened in write mode, the recovery makes the correction on disk when the file is closed; however if the file is opened in read mode, the correction can not be written to disk. You can also explicitly invoke the recovery procedure by calling the TFile::Recover() method. You can recover the directory structure, but you cannot save what you recovered to the file on disk. In the following example, we interrupted and aborted the previous ROOT session, causing the file not to be closed. When we start a new session and attempt to open the file, it gives us an explanation and status on the recovery attempt.

root[] TFile f("demo.root")
Warning in <TFile::TFile>: file demo.root probably not closed, trying to recover successfully recovered 15 keys

11.2 The Logical ROOT File: TFile and TKey

We saw that the TFile::Map() method reads the file sequentially and prints information about each record while scanning the file. It is not feasible to support only sequential access and hence ROOT provides random or direct access, i.e. reading a specified object at a time. To do so, TFile keeps a list of TKeys, which is essentially an index to the objects in the file. The TKey class describes the record headers of objects in the file. For example, we can get the list of keys and print them. To find a specific object on the file we can use the TFile::Get() method.

root[] TFile f("demo.root")
root[] f.GetListOfKeys()->Print()
TKey Name = h0, Title = histo nr:0, Cycle = 1
TKey Name = h1, Title = histo nr:1, Cycle = 1
TKey Name = h2, Title = histo nr:2, Cycle = 1
TKey Name = h3, Title = histo nr:3, Cycle = 1
TKey Name = h4, Title = histo nr:4, Cycle = 1
TKey Name = h5, Title = histo nr:5, Cycle = 1
TKey Name = h6, Title = histo nr:6, Cycle = 1
TKey Name = h7, Title = histo nr:7, Cycle = 1
TKey Name = h8, Title = histo nr:8, Cycle = 1
TKey Name = h9, Title = histo nr:9, Cycle = 1
TKey Name = h10, Title = histo nr:10, Cycle = 1
TKey Name = h11, Title = histo nr:11, Cycle = 1
TKey Name = h12, Title = histo nr:12, Cycle = 1
TKey Name = h13, Title = histo nr:13, Cycle = 1
TKey Name = h14, Title = histo nr:14, Cycle = 1
root[] TH1F *h9 = (TH1F*)f.Get("h9");

The TFile::Get() finds the TKey object with name “h9”. Using the TKey info it will import in memory the object in the file at the file address #3352 (see the output from the TFile::Map above). This is done by the Streamer method that is covered in detail in a later section. Since the keys are available in a TList of TKeys we can iterate over the list of keys:

   TFile f("demo.root");
   TIter next(f.GetListOfKeys());
   TKey *key;
   while ((key=(TKey*)next())) {
      printf("key: %s points to an object of class: %s at %dn",

The output of this script is:

root[] .x iterate.C
key: h0 points to an object of class: TH1F at 150
key: h1 points to an object of class: TH1F at 503
key: h2 points to an object of class: TH1F at 854
key: h3 points to an object of class: TH1F at 1194
key: h4 points to an object of class: TH1F at 1539
key: h5 points to an object of class: TH1F at 1882
key: h6 points to an object of class: TH1F at 2240
key: h7 points to an object of class: TH1F at 2582
key: h8 points to an object of class: TH1F at 2937
key: h9 points to an object of class: TH1F at 3293
key: h10 points to an object of class: TH1F at 3639
key: h11 points to an object of class: TH1F at 3986
key: h12 points to an object of class: TH1F at 4339
key: h13 points to an object of class: TH1F at 4694
key: h14 points to an object of class: TH1F at 5038

In addition to the list of keys, TFile also keeps two other lists: TFile::fFree is a TList of free blocks used to recycle freed up space in the file. ROOT tries to find the best free block. If a free block matches the size of the new object to be stored, the object is written in the free block and this free block is deleted from the list. If not, the first free block bigger than the object is used. TFile::fListHead contains a sorted list (TSortedList) of objects in memory. The diagram below illustrates the logical view of the TFile and TKey.

ROOT File/Directory/Key description

ROOT File/Directory/Key description

11.2.1 Viewing the Logical File Contents

TFile is a descendent of TDirectory, which means it behaves like a TDirectory. We can list the contents, print the name, and create subdirectories. In a ROOT session, you are always in a directory and the directory you are in is called the current directory and is stored in the global variable gDirectory. Let us look at a more detailed example of a ROOT file and its role as the current directory. First, we create a ROOT file by executing a sample script.

root[] .x $ROOTSYS/tutorials/hsimple.C

Now you should have hsimple.root in your directory. The file was closed by the script so we have to open it again to work with it. We open the file with the intent to update it, and list its contents.

root[] TFile f ("hsimple.root","UPDATE")
TFile** hsimple.root
TFile* hsimple.root
KEY: TH1F hpx;1 This is the px distribution
KEY: TH2F hpxpy;1 py vs px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple ntuple;1 Demo ntuple

It shows the two lines starting with TFile followed by four lines starting with the word “KEY”. The four keys tell us that there are four objects on disk in this file. The syntax of the listing is:

KEY: <class> <variable>;<cycle number> <title>

For example, the first line in the list means there is an object in the file on disk, called hpx. It is of the class TH1F (one-dimensional histogram of floating numbers). The object’s title is “This is the px distribution”. If the line starts with OBJ, the object is in memory. The <class> is the name of the ROOT class (T-something). The <variable> is the name of the object. The cycle number along with the variable name uniquely identifies the object. The <title> is the string given in the constructor of the object as title.

The structure of TFile

The structure of TFile

The figure shows a TFile with five objects in the top directory (kObjA;1, kObjA;2, kObjB;1, kObjC;1 and kObjD;1). ObjA is on file twice with two different cycle numbers. It also shows four objects in memory (mObjE, mObjeF, mObjM, mObjL). It also shows several subdirectories.

11.2.2 The Current Directory

When you create a TFile object, it becomes the current directory. Therefore, the last file to be opened is always the current directory. To check your current directory you can type:

root[] gDirectory->pwd()

This means that the current directory is the ROOT session (Rint). When you create a file, and repeat the command the file becomes the current directory.

root[] TFile f1("AFile1.root");
root[] gDirectory->pwd()

If you create two files, the last becomes the current directory.

root[] TFile f2("AFile2.root");
root[] gDirectory->pwd()

To switch back to the first file, or to switch to any file in general, you can use the TDirectory::cd method. The next command changes the current directory back to the first file.

root[] gDirectory->pwd()

Note that even if you open the file in “READ” mode, it still becomes the current directory. CINT also offers a shortcut for gDirectory->pwd() and gDirectory->ls(), you can type:

root[] .pwd
root[] .ls
TFile**        AFile1.root
TFile*         AFile1.root

To return to the home directory where we were before:

root[] gROOT->cd()
(unsigned char)1
root[] gROOT->pwd()

11.2.3 Objects in Memory and Objects on Disk

The TFile::ls() method has an option to list the objects on disk (“-d”) or the objects in memory (“-m”). If no option is given it lists both, first the objects in memory, then the objects on disk. For example:

root[] TFile *f = new TFile("hsimple.root");
root[] gDirectory->ls("-m")
TFile**         hsimple.root
TFile*         hsimple.root

Remember that gDirectory is the current directory and at this time is equivalent to “f”. This correctly states that no objects are in memory.

The next command lists the objects on disk in the current directory.

root[] gDirectory->ls("-d")
TFile**         hsimple.root
TFile*         hsimple.root
KEY: TH1F     hpx;1    This is the px distribution
KEY: TH2F     hpxpy;1  py vs px
KEY: TProfile hprof;1  Profile of pz versus px
KEY: TNtuple  ntuple;1 Demo ntuple

To bring an object from disk into memory, we have to use it or “Get” it explicitly. When we use the object, ROOT gets it for us. Any reference to hprof will read it from the file. For example drawing hprof will read it from the file and create an object in memory. Here we draw the profile histogram, and then we list the contents.

root[] hprof->Draw()
<TCanvas::MakeDefCanvas>: created default TCanvas with name c1
root[] f->ls()
TFile** hsimple.root
TFile* hsimple.root
OBJ: TProfile hprof Profile of pz versus px : 0
KEY: TH1F hpx;1 This is the px distribution
KEY: TH2F hpxpy;1 py vs px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple ntuple;1 Demo ntuple

We now see a new line that starts with OBJ. This means that an object of class TProfile, called hprof has been added in memory to this directory. This new hprof in memory is independent from the hprof on disk. If we make changes to the hprof in memory, they are not propagated to the hprof on disk. A new version of hprof will be saved once we call Write.

You may wonder why hprof is added to the objects in the current directory. hprof is of the class TProfile that inherits from TH1D, which inherits from TH1. TH1 is the basic histogram. All histograms and trees are created in the current directory (also see “Histograms and the Current Directory”). The reference to “all histograms” includes objects of any class descending directly or indirectly from TH1. Hence, our TProfile hprof is created in the current directory f.There was another side effect when we called the TH1::Draw method. CINT printed this statement:

<TCanvas::MakeDefCanvas>: created default TCanvas with name c1

It tells us that a TCanvas was created and it named it c1. This is where ROOT is being nice, and it creates a canvas for drawing the histogram if no canvas was named in the draw command, and if no active canvas exists. The newly created canvas, however, is NOT listed in the contents of the current directory. Why is that? The canvas is not added to the current directory, because by default ONLY histograms and trees are added to the object list of the current directory. Actually, TEventList objects are also added to the current directory, but at this time, we don’t have to worry about those. If the canvas is not in the current directory then where is it? Because it is a canvas, it was added to the list of canvases.

This list can be obtained by the command gROOT->GetListOfCanvases()->ls(). The ls() will print the contents of the list. In our list, we have one canvas called c1. It has a TFrame, a TProfile, and a TPaveStats.

root[] gROOT->GetListOfCanvases()->ls()
Canvas Name=c1 Title=c1
Option=TCanvas fXlowNDC=0 fYlowNDC=0 fWNDC=1 fHNDC=1
Name= c1 Title= c1
Option=TFrame  X1= -4.000000 Y1=0.000000 X2=4.000000 Y2=19.384882
OBJ: TProfile hprof   Profile of pz versus px : 0
TPaveText  X1=-4.900000 Y1=20.475282 X2=-0.950000 Y2=21.686837 title
TPaveStats X1=2.800000  Y1=17.446395 X2=4.800000  Y2=21.323371 stats

Lets proceed with our example and draw one more histogram, and we see one more OBJ entry.

root[] hpx->Draw()
root[] f->ls()
TFile**         hsimple.root
TFile*         hsimple.root
OBJ: TProfile hprof    Profile of pz versus px : 0
OBJ: TH1F     hpx      This is the px distribution : 0
KEY: TH1F     hpx;1    This is the px distribution
KEY: TH2F     hpxpy;1  py vs px
KEY: TProfile hprof;1  Profile of pz versus px
KEY: TNtuple  ntuple;1 Demo ntuple

TFile::ls() loops over the list of objects in memory and the list of objects on disk. In both cases, it calls the ls() method of each object. The implementation of the ls method is specific to the class of the object, all of these objects are descendants of TObject and inherit the TObject::ls() implementation. The histogram classes are descendants of TNamed that in turn is a descent of TObject. In this case, TNamed::ls() is executed, and it prints the name of the class, and the name and title of the object. Each directory keeps a list of its objects in the memory. You can get this list by TDirectory::GetList(). To see the lists in memory contents you can do:

OBJ: TProfile   hprof   Profile of pz versus px : 0
OBJ: TH1F       hpx     This is the px distribution : 0

Since the file f is the current directory (gDirectory), this will yield the same result:

root[] gDirectory->GetList()->ls()
OBJ: TProfile   hprof   Profile of pz versus px : 0
OBJ: TH1F       hpx     This is the px distribution : 0

11.2.4 Saving Histograms to Disk

At this time, the objects in memory (OBJ) are identical to the objects on disk (KEY). Let’s change that by adding a fill to the hpx we have in memory.

root[] hpx->Fill(0)

Now the hpx in memory is different from the histogram (hpx) on disk. Only one version of the object can be in memory, however, on disk we can store multiple versions of the object. The TFile::Write method will write the list of objects in the current directory to disk. It will add a new version of hpx and hprof.

root[] f->Write()
root[] f->ls()
TFile**         hsimple.root
TFile*         hsimple.root
OBJ: TProfile hprof  Profile of pz versus px : 0
OBJ: TH1F     hpx    This is the px distribution : 0
KEY: TH1F     hpx;2  This is the px distribution
KEY: TH1F     hpx;1  This is the px distribution
KEY: TH2F     hpxpy;1 py vs px
KEY: TProfile hprof;2 Profile of pz versus px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple  ntuple;1        Demo ntuple
The file before and after the call to Write

The file before and after the call to Write

The TFile::Write method wrote the entire list of objects in the current directory to the file. You see that it added two new keys: hpx;2 and hprof;2 to the file. Unlike memory, a file is capable of storing multiple objects with the same name. Their cycle number, the number after the semicolon, differentiates objects on disk with the same name. If you wanted to save only hpx to the file, but not the entire list of objects, you could use the TH1::Writemethod of hpx:

root[] hpx->Write()

A call to obj->Write without any parameters will call obj->GetName() to find the name of the object and use it to create a key with the same name. You can specify a new name by giving it as a parameter to the Write method.

root[] hpx->Write("newName")

If you want to re-write the same object, with the same key, use the overwrite option.

root[] hpx->Write("",TObject::kOverwrite)

If you give a new name and use the kOverwrite, the object on disk with the matching name is overwritten if such an object exists. If not, a new object with the new name will be created.

root[] hpx->Write("newName",TObject::kOverwrite)

The Write method did not affect the objects in memory at all. However, if the file is closed, the directory is emptied and the objects on the list are deleted.

root[] f->Close()
root[] f->ls()
TFile**     hsimple.root
TFile*      hsimple.root

In the code snipped above, you can see that the directory is now empty. If you followed along so far, you can see that c1 which was displaying hpx is now blank. Furthermore, hpx no longer exists.

root[] hpx->Draw()
Error: No symbol hpx in current scope

This is important to remember, do not close the file until you are done with the objects or any attempt to reference the objects will fail.

11.2.5 Histograms and the Current Directory

When a histogram is created, it is added by default to the list of objects in the current directory. You can get the list of histograms in a directory and retrieve a pointer to a specific histogram.

   TH1F *h = (TH1F*)gDirectory->Get("myHist"); // or
   TH1F *h = (TH1F*)gDirectory->GetList()->FindObject("myHist");

The method TDirectory::GetList() returns a TList of objects in the directory. You can change the directory of a histogram with the SetDirectory method.


If the parameter is 0, the histogram is no longer associated with a directory.


Once a histogram is removed from the directory, it will no longer be deleted when the directory is closed. It is now your responsibility to delete this histogram object once you are finished with it. To change the default that automatically adds the histogram to the current directory, you can call the static function:


In this case, you will need to do all the bookkeeping for all the created histograms.

11.2.6 Saving Objects to Disk

In addition to histograms and trees, you can save any object in a ROOT file. For example to save a canvas to the ROOT file you can use either TObject::Write() or TDirectory::WriteTObject(). The example:

root[] c1->Write()

This is equivalent to:

root[] f->WriteTObject(c1)

For objects that do not inherit from TObject use:

root[] f->WriteObject(ptr,"nameofobject")

Another example:

root[] TFile *f = new TFile("hsimple.root","UPDATE")
root[] hpx->Draw()
<TCanvas::MakeDefCanvas>: created default TCanvas with name c1
root[] c1->Write()
root[] f->ls()
TFile**        hsimple.root
TFile*         hsimple.root
OBJ: TH1F      hpx      This is the px distribution : 0
KEY: TH1F     hpx;2   This is the px distribution
KEY: TH1F     hpx;1   This is the px distribution
KEY: TH2F     hpxpy;1 py vs px
KEY: TProfile hprof;2 Profile of pz versus px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple  ntuple;1   Demo ntuple
KEY: TCanvas  c1;1    c1

11.2.7 Saving Collections to Disk

All collection classes inherit from TCollection and hence inherit the TCollection::Write() method. When you call TCollection::Write() each object in the container is written individually into its own key in the file. To write all objects into one key you can specify the name of the key and use the optionTObject::kSingleKey. For example:

root[] TList * list = new TList;
root[] TNamed * n1, * n2;
root[] n1 = new TNamed("name1","title1");
root[] n2 = new TNamed("name2","title2");
root[] list->Add(n1);
root[] list->Add(n2);
root[] gFile->WriteObject(list,"list",TObject::kSingleKey);

11.2.8 A TFile Object Going Out of Scope

There is another important point to remember about TFile::Close and TFile::Write. When a variable is declared on the stack in a function such as in the code below, it will be deleted when it goes out of scope.

void foo() {
   TFile f("AFile.root","RECREATE");

As soon as the function foohas finished executing, the variable f is deleted. When a TFile object is deleted an implicit call to TFile::Close is made. This will save only the file descriptor to disk. It contains the file header, the StreamerInfolist, the key list, the free segment list, and the end address. See “The Physical Layout of ROOT Files”. The TFile::Close does not make a call to Write(), which means that the objects in memory will not be saved in the file. You need to explicitly call TFile::Write() to save the object in memory to file before the exit of the function.

void foo() {
   TFile f("AFile.root","RECREATE");
   ... stuff ...

To prevent an object in a function from being deleted when it goes out of scope, you can create it on the heap instead of on the stack. This will create a TFile object f, that is available on a global scope, and it will still be available when exiting the function.

void foo() {
   TFile *f = new TFile("AFile.root","RECREATE");

11.2.9 Retrieving Objects from Disk

If you have a ROOT session running, please quit and start fresh.

We saw that multiple versions of an object with the same name could be in a ROOT file. In our example, we saved a modified histogram hpx to the file, which resulted in two hpx's uniquely identified by the cycle number: hpx;1 and hpx;2. The question is how we can retrieve the right version of hpx. When opening the file and using hpx, CINT retrieves the one with the highest cycle number. To read the hpx;1 into memory, rather than the hpx:2 we would get by default, we have to explicitly get it and assign it to a variable.

root[] TFile *f1 = new TFile("hsimple.root")
root[] TH1F *hpx1; f1->GetObject("hpx;1",hpx)
root[] hpx1->Draw()

11.2.10 Subdirectories and Navigation

The TDirectory class lets you organize its contents into subdirectories, and TFile being a descendent of TDirectory inherits this ability. Here is an example of a ROOT file with multiple subdirectories as seen in the ROOT browser. To add a subdirectory to a file use TDirectory::mkdir. The example below opens the file for writing and creates a subdirectory called “Wed011003”. Listing the contents of the file shows the new directory in the file and the TDirectory object in memory.

root[] TFile *f = new TFile("AFile.root","RECREATE")
root[] f->mkdir("Wed011003")
(class TDirectory*)0x1072b5c8
root[] f->ls()
TFile**         AFile.root
TFile*          AFile.root
TDirectory*           Wed011003       Wed011003
KEY: TDirectory       Wed011003;1     Wed011003

We can change the current directory by navigating into the subdirectory, and after changing directory; we can see that gDirectory is now “Wed011003”.

root[] f->cd("Wed011003")
root[] gDirectory->pwd()

In addition to gDirectory we have gFile, another global that points to the current file. In our example, gDirectory points to the subdirectory, and gFile points to the file (i.e. the files’ top directory).

root[] gFile->pwd()

Use cd() without any arguments to return to the file’s top directory.

root[] f->cd()

Change to the subdirectory again, and create a histogram. It is added to the current directory, which is the subdirectory “Wed011003”.

root[] f->cd("Wed011003")
root[] TH1F *histo = new TH1F("histo","histo",10,0,10)
root[] gDirectory->ls()
TDirectory* Wed011003   Wed011003
OBJ: TH1F      histo   histo : 0

If you are in a subdirectory and you want to have a pointer to the file containing the subdirectory, you can do:

root[] gDirectory->GetFile()

If you are in the top directory gDirectory is the same as gFile. We write the file to save the histogram on disk, to show you how to retrieve it later.

root[] f->Write()
root[] gDirectory->ls()
TDirectory*             Wed011003       Wed011003
OBJ: TH1F      histo   histo : 0
KEY: TH1F      histo;1 histo

When retrieving an object from a subdirectory, you can navigate to the subdirectory first or give it the path name relative to the file. The read object is created in memory in the current directory. In this first example, we get histo from the top directory and the object will be in the top directory.

root[] TH1 *h; f->GetObject("Wed011003/histo;1",h)

If file is written, a copy of histo will be in the top directory. This is an effective way to copy an object from one directory to another. In contrast, in the code box below, histo will be in memory in the subdirectory because we changed the current directory.

root[] f->cd("Wed011003")
root[] TH1 *h; gDirectory->GetObject("histo;1",h)

Note that there is no warning if the retrieving was not successful. You need to explicitly check the value of h, and if it is null, the object could not be found. For example, if you did not give the path name the histogram cannot be found and the pointer to h is null:

root[] TH1 *h; gDirectory->GetObject("Wed011003/histo;1",h)
root[] h
(class TH1*)0x10767de0
root[] TH1 *h; gDirectory->GetObject("histo;1",h)
root[] h
(class TH1*)0x0

To remove a subdirectory you need to use TDirectory::Delete. There is no TDirectory::rmdir. The Delete method takes a string containing the variable name and cycle number as a parameter.

void Delete(const char *namecycle)

The namecycle string has the format name;cycle. The next are some rules to remember:

For example to delete a directory from a file, you must specify the directory cycle:

root[] f->Delete("Wed011003;1")

Some other examples of namecycle format are:

11.3 Streamers

To follow the discussion on Streamers, you need to know what a simple data type is. A variable is of a simple data type if it cannot be decomposed into other types. Examples of simple data types are longs, shorts, floats, and chars. In contrast, a variable is of a composite data type if it can be decomposed. For example, classes, structures, and arrays are composite types. Simple types are also called primitive types, basic types, and CINT sometimes calls them fundamental types.

When we say, “writing an object to a file”, we actually mean writing the current values of the data members. The most common way to do this is to decompose (also called the serialization of) the object into its data members and write them to disk. The decomposition is the job of the Streamer. Every class with ambitions to be stored in a file has a Streamerthat decomposes it and “streams” its members into a buffer.

The methods of the class are not written to the file, it contains only the persistent data members. To decompose the parent classes, the Streamercalls the Streamerof the parent classes. It moves up the inheritance tree until it reaches an ancestor without a parent. To serialize the object data members it calls their Streamer. They in turn move up their own inheritance tree and so forth. The simple data members are written to the buffer directly. Eventually the buffer contains all simple data members of all the classes that make up this particular object. Data members that are references (as MyClass &fObj;) are never saved, it is always the responsibility of the object’s constructor to set them properly.

11.3.1 Automatically Generated Streamers

A Streamerusually calls other Streamers: the Streamerof its parents and data members. This architecture depends on all classes having Streamers, because eventually they will be called. To ensure that a class has a Streamer, rootcint automatically creates one in the ClassDef macro that is defined in $ROOTSYS/include/Rtypes.h. ClassDef defines several methods for any class, and one of them is the Streamer. The automatically generated Streameris complete and can be used as long as no customization is needed.

The Event class is defined in $ROOTSYS/test/Event.h. Looking at the class definition, we find that it inherits from TObject. It is a simple example of a class with diverse data members.

class Event : public TObject {
   TDirectory    *fTransient;            //! current directory
   Float_t     fPt;                   //! transient value
char           fType[20];
Int_t          fNtrack;
Int_t          fNseg;
Int_t          fNvertex;
UInt_t         fFlag;
Float_t        fTemperature;
EventHeader    fEvtHdr;           //|| don't split
TClonesArray  *fTracks;           //->
TH1F          *fH;                //->
Int_t          fMeasures[10];
Float_t        fMatrix[4][4];
Float_t       *fClosestDistance;  //[fNvertex]

The Event class is added to the CINT dictionary by the rootcint utility. This is the rootcint statement in the $ROOTSYS/test/Makefile:

@rootcint -f EventDict.cxx -c Event.h EventLinkDef.h

The EventDict.cxx file contains the automatically generated Streamerfor Event:

void Event::Streamer(TBuffer &R__b){
   // Stream an object of class Event.
   if (R__b.IsReading()) {
      Event::Class()->ReadBuffer(R__b, this);
   } else {
      Event::Class()->WriteBuffer(R__b, this);

When writing an Event object, TClass::WriteBuffer is called. WriteBuffer writes the current version number of the Event class, and its contents into the buffer R__b. The Streamercalls TClass::ReadBuffer when reading an Event object. The ReadBuffer method reads the information from buffer R__b into the Event object.

11.3.2 Transient Data Members (//!)

To prevent a data member from being written to the file, insert a “!” as the first character after the comment marks. It tells ROOT not to save that data member in a root file when saving the class. For example, in this version of Event, the fPt and fTransient data members are not persistent.

class Event : public TObject {
   TDirectory    *fTransient; //! current directory
   Float_t fPt;               //! transient value

11.3.3 The Pointer to Objects (//->)

The string “->” in the comment field of the members *fH and *fTracks instruct the automatic Streamer to assume these will point to valid objects and the Streamerof the objects can be called rather than the more expensive R__b << fH. It is important to note that no check is done on the validity of the pointer value. In particular if the pointer points, directly or indirectly, back to the current object, this will result in an infinite recursion and the abrupt end of the process.

TClonesArray  *fTracks;            //->
TH1F          *fH;                 //->

11.3.4 Variable Length Array

When the Streamercomes across a pointer to a simple type, it assumes it is an array. Somehow, it has to know how many elements are in the array to reserve enough space in the buffer and write out the appropriate number of elements. This is done in the class definition. For example:

class Event : public TObject {
   char           fType[20];
   Int_t          fNtrack;
   Int_t          fNseg;
   Int_t          fNvertex;
   Float_t       *fClosestDistance;   //[fNvertex]

The array fClosestDistance is defined as a pointer of floating point numbers. A comment mark (//), and the number in square brackets tell the Streamerthe length of the array for this object. In general the syntax is:

<simple type> *<name>//[<length>]

The length cannot be an expression. If a variable is used, it needs to be an integer data member of the class. It must be defined ahead of its use, or in a base class.

The same notation also applies to variable length array of object and variable length array of pointer to objects.

MyObject *obj; //[fNojbs]
MyObject **objs; //[fDatas]

11.3.5 Double32_t

Math operations very often require double precision, but on saving single usually precision is sufficient. For this purpose we support the typedef Double32_t which is stored in memory as a double and on disk as a float or interger. The actual size of disk (before compression) is determined by the parameter next to the data member declartion. For example:

Double32_t m_data;     //[min,max<,nbits>]

If the comment is absent or does not contain min, max, nbit, the member is saved as a float.

If min and max are present, they are saved as a 32 bits precision. min and max can be explicit values or be expressions of values known to CINT (e.g. “pi").

If nbits is present, the member is saved as int with ‘nbit’. For more details see the io tutorials double32.C.

Compression and precision of Double32_t

Compression and precision of Double32_t

11.3.6 Prevent Splitting (//|| )

If you want to prevent a data member from being split when writing it to a tree, append the characters || right after the comment string. This only makes sense for object data members. For example:

EventHeader    fEvtHdr;       //|| do not split the header

11.3.7 Streamers with Special Additions

Most of the time you can let rootcint generate a Streamer for you. However if you want to write your own Streameryou can do so. For some classes, it may be necessary to execute some code before or after the read or write block in the automatic Streamer. For example after the execution of the read block, one can initialize some non persistent members. There are two reasons why you would need to write your own Streamer: 1) if you have a non-persistent data member that you want to initialize to a value depending on the read data members; 2) if you want or need to handle the schema evolution on your own. In addition, the automatic Streamerdoes not support C-structures. It is best to convert the structure to a class definition.

First, you need to tell rootcint not to build a Streamerfor you. The input to the rootcint command (in the makefile) is a list of classes in a LinkDef.h file. For example, the list of classes for Event is listed in $ROOTSYS/test/EventLinkDef.h. The “-” at the end of the class name tells rootcint not to generate a Streamer. In the example, you can see the Event class is the only one for which rootcint is instructed not to generate a Streamer.

#ifdef __CINT__

#pragma link off all globals;
#pragma link off all classes;
#pragma link off all functions;
#pragma link C++ class EventHeader+;
#pragma link C++ class Event-;
#pragma link C++ class HistogramManager+;
#pragma link C++ class Track+;

#pragma link C++ class EventHeader+;

The “+” sign tells rootcint to use the new Streamersystem introduced in ROOT 3.0. The following is an example of a customized Streamerfor Event. The Streamer takes a TBuffer as a parameter, and first checks to see if this is a case of reading or writing the buffer.

void Event::Streamer(TBuffer &R__b) {
   if (R__b.IsReading()) {
      Event::Class()->ReadBuffer(R__b, this);
      fTransient = gDirectory;       //save current directory
      fPt= TMath::Sqrt(fPx*fPx + fPy*fPy + fPz*fPz);
   } else {
      Event::Class()->WriteBuffer(R__b, this);

11.3.8 Writing Objects

The Streamer decomposes the objects into data members and writes them to a buffer. It does not write the buffer to a file, it simply populates a buffer with bytes representing the object. This allows us to write the buffer to a file or do anything else we could do with the buffer. For example, we can write it to a socket to send it over the network. This is beyond the scope of this chapter, but it is worthwhile to emphasize the need and advantage of separating the creation of the buffer from its use. Let us look how a buffer is written to a file. The dictionary for a class needs to be loaded before any object of that type can be saved.

The TObject::Write method does the following:

In other words, the TObject::Write calls the Streamer method of the class to build the buffer. The buffer is in the key and the key is written to disk. Once written to disk the memory consumed by the buffer part is released. The key part of the TKey is kept.

A diagram of a streamed TH1F in the buffer

A diagram of a streamed TH1F in the buffer

The key consumes about 60 bytes, whereas the buffer, since it contains the object data, can be very large.

11.3.9 Ignore Object Streamers

Your class can ignore the TObject Streamerwith the MyClass->Class::IgnoreObjectStreamer() method. When the class kIgnoreTObjectStreamerbit is set (by calling the IgnoreTObjectStreamermethod), the automatically generated Streamerwill not call TObject::Streamer, and the TObject part of the class is not streamed to the file. This is useful in case you do not use the TObject fBits and fUniqueIDdata members. You gain space on the file, and you do not loose functionality if you do not use the fBits and fUniqueID.See “The Role of TObject” on the use of fBits and fUniqueID.

11.3.10 Streaming a TClonesArray

When writing a TClonesArray it bypasses by default the Streamerof the member class and uses a more efficient internal mechanism to write the members to the file. You can override the default and specify that the member class Streameris used by setting the TClonesArray::BypassStreamer bit to false:

   TClonesArray *fTracks;
   fTracks->BypassStreamer(kFALSE);    // use the member Streamer

When the kBypassStreamer bit is set, the automatically generated Streamercan call directly the method TClass::WriteBuffer. Bypassing the Streamer improves the performance when writing/reading the objects in the TClonesArray. However, the drawback is when a TClonesArray is written with split=0 bypassing the Streamer, the StreamerInfoof the class in the array being optimized, one cannot later use the TClonesArray with split > 0. For example, there is a problem with the following scenario: a class Foo has a TClonesArray of Bar objects the Foo object is written with split=0 to Tree T1. In this case the StreamerInfo for the class Bar is created in optimized mode in such a way that data members of the same type are written as an array improving the I/O performance. In a new program, T1 is read and a new Tree T2 is created with the object Foo in split > 1.

When the T2branch is created, the StreamerInfo for the class Bar is created with no optimization (mandatory for the split mode). The optimized Bar StreamerInfo is going to be used to read the TClonesArray in T1. The result will be Bar objects with data member values not in the right sequence. The solution to this problem is to call BypassStreamer(kFALSE) for the TClonesArray. In this case, the normal Bar::Streamer function will be called. The Bar::Streamer function works OK independently if the Bar StreamerInfohad been generated in optimized mode or not.

11.4 Pointers and References in Persistency

An object pointer as a data member presents a challenge to the streaming software. If the object pointed to is saved every time, it could create circular dependencies and consume a large amount of disk space. The network of references must be preserved on disk and recreated upon reading the file.

If you use independent I/O operations for pointers and their referenced objects you can use the TRef class. Later in this section is an example that compares disk space, memory usage, and I/O times of C++ pointers and TRefs. In general, a TRef is faster than C++ but the advantage of a C++ pointer is that it is already C++.

11.4.1 Streaming C++ Pointers

When ROOT encounters a pointer data member it calls the Streamer of the object and labels it with a unique object identifier. The object identifier is unique for one I/O operation. If there is another pointer to the object in the same I/O operation, the first object is referenced i.e. it is not saved again. When reading the file, the object is rebuilt and the references recalculated.

Streaming object pointers

Streaming object pointers

In this way, the network of pointers and their objects is rebuilt and ready to use the same way it was used before it was persistent. If the pointer hold the address of an object which in embedded in another object (as opposed to being pointed to by a pointer), the object will be duplicate at read time. To avoid this, make the pointer a transient data member.

11.4.2 Motivation for the TRef Class

If the object is split into several files or into several branches of one or more TTrees, standard C++ pointers cannot be used because each I/O operation will write the referenced objects, and multiple copies will exist. In addition, if the pointer is read before the referenced object, it is null and may cause a run time system error. To address these limitations, ROOT offers the TRef class.

TRef allows referencing an object in a different branch and/or in a different file. TRef also supports the complex situation where a TFile is updated multiple times on the same machine or a different machine. When a TRef is read before its referenced object, it is null. As soon as the referenced object is read, the TRef points to it. In addition, one can specify an action to be taken by TRef in the case it is read before its reference object (see“Action on Demand” below).

11.4.3 Using TRef

A TRef is a lightweight object pointing to any TObject. This object can be used instead of normal C++ pointers in case:

Below is a line from the example in $ROOTSYS/test/Event.cxx.

   TRef   fLastTrack;             //pointer to last track
   Track *track = (Track*)fTracks->ConstructedAt(fNtrack++);
   // Save reference to last Track in the collection of Tracks
   fLastTrack = track;

The track and its reference fLastTrack can be written with two separate I/O calls in the same or in different files, in the same or in different branches of a TTree. If the TRef is read and the referenced object has not yet been read, TRef will return a null pointer. As soon as the referenced object will be read, TRef will point to it.

11.4.4 How Does It Work?

A TRef is itself a TObject with an additional transient pointer fPID. When a TRef is used to point to a TObject*R, for example in a class with

   TRef  P;

one can do:

   P = R;      //to set the pointer

When the statement P = Ris executed, the following happens:

After having set P, one can immediately return the value of R using P.GetObject(). This function returns the fObjects[fUniqueID] from the fPID object.

When the TRef is written, the process id number pidf of fPID is written in addition to the TObject part of the TRef (fBits,fUniqueID). When the TRef is read, its pointer fPID is set to the value stored in the TObjArray of TFile::fProcessIDs(fProcessIDs[pidf]).

When a referenced object is written, TObject::Streamer writes the pidf in addition to the standard fBits and fUniqueID. When TObject::Streamer reads a reference object, the pidf is read. At this point, the referenced object is entered into the table of objects of the TProcessID corresponding to pidf.

WARNING: If MyClass is the class of the referenced object, The TObject part of MyClass must be streamed. One should not call MyClass::Class()->IgnoreTObjectStreamer(). TProccessID and TUUID

A TProcessID uniquely identifies a ROOT job. The TProcessID title consists of a TUUID object, which provides a globally unique identifier. The TUUID class implements the UUID (Universally Unique Identifier), also known as GUID (Globally Unique Identifier). A UUID is 128 bits long, and if generated according to this algorithm, is either guaranteed to be different from all other UUID generated until 3400 A.D. or extremely likely to be different.

The TROOT constructor automatically creates a TProcessID. When a TFile contains referenced objects, the TProcessID object is written to the file. If a file has been written in multiple sessions (same machine or not), a TProcessID is written for each session. The TProcessID objects are used by TRef to uniquely identify the referenced TObject.

When a referenced object is read from a file (its bit kIsReferenced is set), this object is entered into the objects table of the corresponding TProcessID. Each TFile has a list of TProcessIDs (see TFile::fProcessIDs) also accessiblefromTProcessID::fgPIDs(for all files). When this object is deleted, it is removed from the table via the cleanup mechanism invoked by the **TObject** destructor. Each **TProcessID** has a table (TObjArray *fObjects) that keeps track of all referenced objects. If a referenced object has afUniqueID, a pointer to this unique object may be found usingfObjects->At(fUniqueID). In the same way, when a **TRef::GetObject** is called,GetObjectuses its ownfUniqueIDto find the pointer to the referenced object. SeeTProcessID::GetObjectWithIDandPutObjectWithID`. Object Number

When an object is referenced, a unique identifier is computed and stored in both the fUniqueID of the referenced and referencing object. This uniqueID is computed by incrementing by one the static global in TProcessID::fgNumber. The fUniqueID is the serial object number in the current session. One can retrieve the current fgNumber value by calling the static function TProcessID::GetObjectCount at any time or can set this number by TProcessID::SetObjectCount. To avoid a growing table of fObjects in TProcessID, in case, for example, one processes many events in a loop, it might be necessary to reset the object number at the end of processing of one event. See an example in $ROOTSYS/test/Event.cxx (look at function Build). The value of ObjectNumbermay be saved at the beginning of one event and reset to this original value at the end of the event. These actions may be nested.

   saveNumber = TProcessID::GetObjectCount();

11.4.5 Action on Demand

The normal behavior of a TRef has been described above. In addition, TRef supports “Actions on Demand”. It may happen that the referenced object is not yet in the memory, on a separate file or not yet computed. In this case, TRef is able to execute automatically an action: How to Select This Option?

In the definition of the TRef data member in the original class, do:

    TRef fRef;   //EXEC:execName points to something

When the special keyword "EXEC:" is found in the comment field of the member, the next string is assumed to be the name of a TExec object. When a file is connected, the dictionary of the classes on the file is read in memory (see TFile::ReadStreamerInfo). When the TStreamerElement object is read, a TExec object is automatically created with the name specified after the keyword "EXEC:" in case a TExec with a same name does not already exist.

The action to be executed via this TExec can be specified with:

One can compute a pointer to an existing TExec with a name with:

   TExec *myExec = gROOT->GetExec(execName);

The parameter actionCommand is a string containing a CINT instruction. Examples:

   myExec->SetAction(".x script.C");

When a TRef is de-referenced via TRef::GetObject, its TExec** is automatically executed. The TExec function/script can do one or more of the following:**


As soon as an object is returned to GetObject, the fUniqueID of the TRef is set to the fUniqueID of the referenced object. At the next call to GetObject, the pointer stored in fPid:fObjects[fUniqueID] will be returned directly. An example of action on demand is in $ROOTSYS/test/Event.h:

   TRef    fWebHistogram;       //EXEC:GetWebHistogram

When calling fWebHistogram.GetObject(), the function GetObject will automatically invoke the script GetWebHistogram.C via the interpreter. An example of a GetWebHistogram.C script is shown below:

void GetWebHistogram() {
   TFile *f=TFile::Open("");
   TH1 *h6 = (TH1*)gDirectory->Get("h6");
   delete f;

In the above example, a call to fWebHistogram.GetObject() executes the script with the function GetWebHistogram. This script connects a file with histograms: pippa.root on the ROOT Web site and returns the object h6 to TRef::GetObject.

 TRef    fWebHistogram;           //EXEC:GetWebHistogram()

Note that if the definition of the TRef fWebHistogram had been changed the compiled or interpreted function GetWebHistogram() would have been called instead of the CINT script GetWebHistogram.C.

11.4.6 Array of TRef

When storing multiple TRefs, it is more efficient to use a TRefArray. The efficiency is due to having a single pointer fPID for all TRefs in the array. It has a dynamic compact table of fUniqueIDs. We recommend that you use a TRefArray rather then a collection of TRefs.


The 3 arrays mytracks,pions and muons may be written separately.

11.5 Schema Evolution

Schema evolution is a problem faced by long-lived data. When a schema changes, existing persistent data can become inaccessible unless the system provides a mechanism to access data created with previous versions of the schema. In the lifetime of collaboration, the class definitions (i.e. the schema) are likely to change frequently. Not only can the class itself change, but any of its parent classes or data member classes can change also. This makes the support for schema evolution necessary.

ROOT fully supports schema evolution. The next figure below illustrates some of the scenarios.

The ROOT schema evolution

The ROOT schema evolution

The top half represents different versions of the shared library with the class definitions. These are the in-memory class versions. The bottom half represents data files that contain different versions of the classes.

In case of a mismatch between the in-memory version and the persistent version of a class, ROOT maps the persistent one to the one in memory. This allows you to change the class definition at will, for example:

The schema evolution for objects written on disk and in memory

The schema evolution for objects written on disk and in memory

ROOT supports schema evolution by keeping a class description of each version of the class that was ever written to disk, with the class. When it writes an object to file, it also writes the description of the current class version along with it. This description is implemented in the StreamerInfoclass.

11.5.1 The TStreamerInfo Class

Each class has a list of StreamerInfo objects, one for each version of the class if that version was written to disk at least once. When reading an object from a file, the system uses the StreamerInfo list to decode an object into the current version. The StreamerInfo is made up of TStreamerElements . Each describes one persistent data member of the class. By default, all data members of a class are persistent. To exclude a data member (i.e. make it not persistent), add a “!” after the comment marks. For example the pointer *fPainter of a TH1 is not persistent:

   TVirtualHistPainter* fPainter //!pointer to histogram painter

11.5.2 The TStreamerElement Class

A TStreamerElement describes a data member of a simple type, object, array, pointer, or container. The offset in the TStreamerElement is the starting address of the data for that data member.

BASE   TNamed        offset=  0 type=67 The basis for a named object
BASE   TAttLine      offset= 28 type= 0 Line attributes

In this example, the TNamed data starts at byte 0, and TAttLine starts at byte 28. The offset is machine and compiler dependent and is computed when the StreamerInfo is analyzed. The types are defined in the file TStreamerInfo.h and listed here:

enum EReadWrite {
kBase=0,   kChar=1,kShort=2,kInt=3,kLong=4,
kFloat=5,         kCounter=6,kCharStar=7, kDouble=8,kUChar=11,
kUShort=12,         kUInt=13,kULong=14,kBits=15,kOffsetL=20,
kOffsetP=40,  kObject=61,kAny=62,kObjectp=63,kObjectP=64,
kTString=65,  kTObject=66,kTNamed=67,kSkip=100,kSkipL=120,
kSkipP=140,    kConv=200,       kConvL=220,kConvP=240,kStreamer=500,
kStreamLoop=501,    kMissing=99999

The method TClass::GetStreamerInfo analyzes the StreamerInfo the same way it would be analyzed by referring to the class. While analyzing the StreamerInfo, it computes the offsets. The type field is the type of the TStreamerElement. It is specific to the StreamerInfo definition.

11.5.3 Example: TH1 StreamerInfo

In the StreamerInfo of the TH1 class we see the four base classes: TNamed, TAttLine, TAttFill, andTAttMarker. These are followed by a list of the data members. Each data member is implemented by a TStreamerElement object.

root[] TH1::Class()->GetStreamerInfo()->ls()
StreamerInfo for class: TH1, version=3
BASE    TNamed  offset=  0 type=67 The basis for a named object
BASE    TAttLine      offset= 28 type= 0 Line attributes
BASE    TAttFill      offset= 40 type= 0 Fill area attributes
BASE    TAttMarker    offset= 48 type= 0 Marker attributes
Int_t   fNcells       offset= 60 type= 3 number of bins(1D
TAxis   fXaxis        offset= 64 type=61 X axis descriptor
TAxis   fYaxis        offset=192 type=61 Y axis descriptor
TAxis   fZaxis        offset=320 type=61 Z axis descriptor
Short_t fBarOffset    offset=448 type= 2(1000*offset)for bar charts or legos
Short_t fBarWidth     offset=450 type= 2 (1000*width)for bar charts or legos
Stat_t  fEntries      offset=452 type= 8 Number of entries
Stat_t  fTsumw        offset=460 type= 8 Total Sum of weights
Stat_t  fTsumw2       offset=468 type= 8 Total Sum of squares of weights
Stat_t  fTsumwx       offset=476 type= 8 Total Sum of weight*X
Stat_t  fTsumwx2      offset=484 type= 8 Total Sum of weight*X*X
Double_t fMaximum     offset=492 type= 8 Maximum value for plotting
Double_t fMinimum     offset=500 type= 8 Minimum value for plotting
Double_t fNormFactor  offset=508 type= 8 Normalization factor
TArrayD  fContour     offset=516 type=62 Array to display contour levels
TArrayD  fSumw2       offset=528 type=62 Array of sum of squares of weights
TString  fOption      offset=540 type=65 histogram options
TList*   fFunctions   offset=548 type=63 ->Pointer to list of functions
i= 0, TNamed       type= 67, offset=  0, len=1, method=0
i= 1, TAttLine     type=  0, offset= 28, len=1, method=142484480
i= 2, TAttFill     type=  0, offset= 40, len=1, method=142496992
i= 3, TAttMarker   type=  0, offset= 48, len=1, method=142509704
i= 4, fNcells      type=  3, offset= 60, len=1, method=0
i= 5, fXaxis       type= 61, offset= 64, len=1, method=1081287424
i= 6, fYaxis       type= 61, offset=192, len=1, method=1081287548
i= 7, fZaxis       type= 61, offset=320, len=1, method=1081287676
i= 8, fBarOffset   type= 22, offset=448, len=2, method=0
i= 9, fEntries     type= 28, offset=452, len=8, method=0
i=10, fContour     type= 62, offset=516, len=1, method=1081287804
i=11, fSumw2       type= 62, offset=528, len=1, method=1081287924
i=12, fOption      type= 65, offset=540, len=1, method=1081288044
i=13, fFunctions   type= 63, offset=548, len=1, method=1081288164

11.5.4 Optimized StreamerInfo

The entries starting with “i = 0” is the optimized format of the StreamerInfo. Consecutive data members of the same simple type and size are collapsed and read at once into an array for performance optimization.

i= 0, TNamed       type= 67, offset=  0, len=1, method=0
i= 1, TAttLine     type=  0, offset= 28, len=1, method=142484480
i= 2, TAttFill     type=  0, offset= 40, len=1, method=142496992
i= 3, TAttMarker   type=  0, offset= 48, len=1, method=142509704

For example, the five data members beginning with fEntiesand the three data members beginning with fMaximum, are put into an array called fEntries (i = 9) with the length 8.

i= 9, fEntries     type= 28, offset=452, len=8, method=0

Only simple type data members are combined, object data members are not combined. For example the three axis data members remain separate. The “method” is a handle to the method that reads the object.

11.5.5 Automatic Schema Evolution

When a class is defined in ROOT, it must include the ClassDef macro as the last line in the header file inside the class definition. The syntax is:


The version number identifies this particular version of the class. When a class has version 0 it is not stored in a root file but its base class(es) is(are). The reason can be that this class has no data members worth saving or all real info is in the base classes. The version number is written to the file in the Streamer by the call TBuffer::WriteVersion. You, as the designer of the class, do not need to do any manual modification in the Streamer. ROOT schema evolution mechanism is automatic and handled by the StreamerInfo.

11.5.6 Manual Schema Evolution

If you have written your own Streamer as described in the section “Streamers with Special Additions”, you will have to manually add code for each version and manage the evolution of your class. When you add or remove data members, you must modify the Streamer by hand. ROOT assumes that you have increased the class version number in the ClassDef statement and introduced the relevant test in the read part of the Streamer. For example, if a new version of the Event class above includes a new member: Int_t fNew the ClassDef statement should be changed to ClassDef(Event,2) and the following lines should be added to the read part of the Streamer:

   if (R__v > 1) R__b >> fNew;
   else fNew = 0;        // set to some default value

If, in the same new version 2 you remove the member fH, you must add the following code to read the histogram object into some temporary object and delete it:

if (R__v) < 2 {
   TH1F *dummy = 0;
   R__b >> dummy;
   delete dummy;

Our experience with manual schema evolution shows that it is easy to make and mismatches between Streamer writers and readers are frequent and increase as the number of classes increase. We recommend you use rootcint generated Streamers whenever you can, and profit from the automatic schema evolution.

11.5.7 Building Class Definitions with the StreamerInfo

A ROOT file’s StreamerInfolist contains the description of all versions of all classes in the file. When a file is opened the StreamerInfois read into memory and it provides enough information to make the file browsable. The TStreamerInfoenables us to recreate a header file for the class in case the compiled class is not available. This is done with the TFile::MakeProject method. It creates a directory with the header files for the named classes and a makefile to compile a shared library with the class definitions.

11.5.8 Example: MakeProject

To explain the details, we use the example of the ATLFast project that is a fast simulation for the ATLAS experiment. The complete source for ATLFast can be down loaded at Once we compile and run ATLFast we get a ROOT file called atlfast.root, containing the ATLFast objects. When we open the file, we get a warning that the file contains classes that are not in the CINT dictionary. This is correct since we did not load the class definitions.

root[] TFile f("atlfast.root")
Warning in <TClass::TClass>: no dictionary for class TMCParticle is available
Warning in <TClass::TClass>: no dictionary for class ATLFMuon available

We can see the StreamerInfofor the classes:

root[] f.ShowStreamerInfo()
StreamerInfo for class: ATLFMuon, version=1
BASE  TObject      offset=  0 type=66 Basic ROOT object
BASE  TAtt3D       offset=  0 type= 0 3D attributes
Int_t m_KFcode     offset=  0 type= 3 Muon KF-code
Int_t m_MCParticle offset=  0 type= 3 Muon position in MCParticles list
Int_t m_KFmother   offset=  0 type= 3 Muon mother KF-code
Int_t m_UseFlag    offset=  0 type= 3 Muon energy usage flag
Int_t m_Isolated   offset=  0 type= 3 Muon isolation (1 for isolated)
Float_t m_Eta      offset=  0 type= 5 Eta coordinate
Float_t m_Phi      offset=  0 type= 5 Phi coordinate
Float_t m_PT       offset=  0 type= 5 Transverse energy
Int_t   m_Trigger  offset=  0 type= 3 Result of trigger...

However, when we try to use a specific class we get a warning because the class is not in the CINT dictionary. We can create a class using gROOT->GetClass() which makes a fake class from the StreamerInfo.

// Build a 'fake' class
root[] gROOT->GetClass("ATLFMuon")
(const class TClass*)0x87e5c08
// The fake class has a StreamerInfo
root[] gROOT->GetClass("ATLFMuon")->GetStreamerInfo()->ls()
StreamerInfo for class: ATLFMuon, version=1
  BASE    TObject       offset=  0 type=66 Basic ROOT object
  BASE    TAtt3D        offset=  0 type= 0 3D attributes
  Int_t   m_KFcode      offset= 16 type= 3 Muon KF-code
  Int_t   m_MCParticle  offset= 20 type= 3 Muon position in MCParticles list
  Int_t   m_KFmother    offset= 24 type= 3 Muon mother KF-code
  Int_t   m_UseFlag     offset= 28 type= 3 Muon energy usage flag
  Int_t   m_Isolated    offset= 32 type= 3 Muon isolation
  Float_t m_Eta         offset= 36 type= 5 Eta coordinate
  Float_t m_Phi         offset= 40 type= 5 Phi coordinate
  Float_t m_PT          offset= 44 type= 5 Transverse energy
  Int_t   m_Trigger     offset= 48 type= 3 Result of trigger
  i= 0, TObject         type= 66, offset=  0, len=1, method=0
  i= 1, TAtt3D          type=  0, offset=  0, len=1, method=142684688
  i= 2, m_KFcode        type= 23, offset= 16, len=5, method=0
  i= 3, m_Eta           type= 25, offset= 36, len=3, method=0
  i= 4, m_Trigger       type=  3, offset= 48, len=1, method=0

MakeProject has three parameters:

MakeProject(const char *dirname,const char *classes,Option_t *option)

The first is the directory name in which to place the generated header files. The second parameter is the name of the classes to include in the project. By default, all classes are included. It recognizes the wild card character *, for example, “ATLF*” includes all classes beginning with ATLF. The third parameter is an option with the following values:

This example makes a directory called MyProject that will contain all class definitions from the atlfast.root file. The necessary makefile to build a shared library are also created, and since the ‘++’ is appended, the shared library is also loaded.

root[] f.MakeProject("MyProject","*", "recreate++")
MakeProject has generated 0 classes in MyProject
MyProject/MAKE file has been generated
Shared lib MyProject/ has been generated
Shared lib MyProject/ has been dynamically linked

The contents of MyProject:

root[]     .! ls MyProject
ATLFCluster.h      ATLFJet.h           ATLFMiscMaker.h     ATLFTrack.h
TMCParticle.h      ATLFClusterMaker.h  ATLFJetMaker.h      ATLFMuon.h
ATLFElectron.h     ATLFMCMaker.h       ATLFMuonMaker.h     ATLFElectronMaker.h
ATLFMaker.h        ATLFPhoton.h        ATLFHistBrowser.h   ATLFMisc.h
ATLFPhotonMaker.h  ATLFTrackMaker.h    ATLFTrigger.h       ATLFTriggerMaker.h
LinkDef.h          MAKE              MyProjectProjectDict.h
MyProjectProjectDict.cxx               MyProjectProjectDict.o

Now you can load the shared library in any consecutive root session to use the atlfast classes.

root[]ATLFMuon muon

This is an example of a generated header file:

//   This class has been generated by TFile::MakeProject
//     (Thu Apr  5 10:18:37 2001 by ROOT version 3.00/06)
//      from the TStreamerInfo in file atlfast.root
#ifndef ATLFMuon_h
#define ATLFMuon_h
#include "TObject.h"
#include "TAtt3D.h"
class ATLFMuon : public TObject , public TAtt3D {
   Int_t     m_KFcode;           //Muon KF-code
   Int_t     m_MCParticle;       //Muon position in MCParticles list
   Int_t     m_KFmother;         //Muon mother KF-code
   Int_t     m_UseFlag;          //Muon energy usage flag
   Int_t     m_Isolated;         //Muon isolation (1 for isolated)
   Float_t   m_Eta;              //Eta coordinate
   Float_t   m_Phi;              //Phi coordinate
   Float_t   m_PT;               //Transverse energy
   Int_t     m_Trigger;          //Result of trigger
   ATLFMuon() {;}
   virtual ~ATLFMuon() {;}
   ClassDef(ATLFMuon,1) //

11.6 Migrating to ROOT 3

We will distinguish the following cases:

Case A: You have your own Streamer method in your class implementation file. This also means that you have specified MyClass in the LinkDef.h file.

 void MyClass::Streamer(TBuffer &R__b) {
   // Stream an object of class MyClass.
   if (R__b.IsReading()) {
      UInt_t R__s, R__c;
      Version_t R__v = R__b.ReadVersion(&R__s, &R__c);
      if (R__v > 1) {
         MyClass::Class()->ReadBuffer(R__b, this, R__v, R__s, R__c);
      // process old versions before automatic schema evolution
      R__b >> xxxx;
      R__b >> .. etc
      R__b.CheckByteCount(R__s, R__c, MyClass::IsA()); // end of old versions
   } else

Case B: You use the automatic Streamer in the dictionary file.

Case C: You use the automatic Streamer in the dictionary file and you already use the option “+” in the LinkDef file. If the old automatic Streamer does not contain any statement using the function WriteArray, you have nothing to do, except running rootcint again to regenerate the new form of the Streamer function, otherwise proceed like for case B.

11.7 Compression and Performance

ROOT uses a compression algorithm based on the well-known gzip algorithm. It supports nine levels of compression. The default for ROOT is one. The compression level can be set with the method TFile::SetCompressionLevel. The experience with this algorithm shows that a compression level of 1.3 for raw data files and around two on most DST files is the optimum. The choice of one for the default is a compromise between the time it takes to read and write the object vs. the disk space savings.

To specify no compression, set the level to zero.

We recommend using compression when the time spent in I/O is small compared to the total processing time. If the I/O operation is increased by a factor of 5 it is still a small percentage of the total time and it may compress the data by a factor of 10. On the other hand if the time spend on I/O is large, compression may have a large impact on the program’s performance.

The compression factor, i.e. the savings of disk space, varies with the type of data. A buffer with a same value array is compressed so that the value is only written once. For example, a track has the mass of a pion that it is always the same, and the charge of the pion that is either positive or negative. For 1000 pions, the mass will be written only once, and the charge only twice (positive and negative). When the data is sparse, i.e. when there are many zeros, the compression factor is also high.

Compression level


Write Time (sec)

Read Time (sec.)

















The time to uncompress an object is small compared to the compression time and is independent of the selected compression level. Note that the compression level may be changed at any time, but the new compression level will only apply to newly written objects. Consequently, a ROOT file may contain objects with different compression levels. This table shows four runs of the demo script that creates 15 histograms with different compression parameters. To make the numbers more significant, the macro was modified to create 1000 histograms. We have included two more examples to show the impact of compression on Trees in the next chapter.

11.8 Remotely Access to ROOT Files via a rootd

Reading and writing ROOT files over the net can be done by creating a TNetFile object instead of a TFile object. Since the TNetFile class inherits from the TFile class, it has exactly the same interface and behavior. The only difference is that it reads and writes to a remote rootd daemon.

11.8.1 TNetFile URL

TNetFile file names are in standard URL format with protocol “root”. The following are valid TNetFile URL’s:


The only difference with the well-known http URL’s is that the root of the remote file tree is the remote user’s home directory. Therefore an absolute pathname requires a // after the host or port (as shown in the last example above). Further the expansion of the standard shell characters, like ~, $, .., etc. is handled as expected. The default port on which the remote rootd listens is 1094 and TNetFile (actually by TUrl that is used by TNetFile) assumes this default port. The port number has been allocated by the IANA and is reserved for ROOT.

11.8.2 Remote Authentication

Connecting to a rootd daemon requires a remote user id and password. TNetFile supports several ways for you to provide your login information:

11.8.3 A Simple Session

root[] TFile *f1 = TFile::Open("local/file.root","update")
root[] TFile *f2 = TFile::Open("root://","new")
Name (pcna49a:rdm):
root[] TFile *f3 = TFile::Open("")
KEY: TH1F hpx;1 This is the px distribution
KEY: TH2F hpxpy;1 py vs px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple ntuple;1 Demo ntuple
root[] hpx.Draw()

11.8.4 The rootd Daemon

The rootd daemon works with the TNetFile class. It allows remote access to ROOT database files in read or read/write mode. The rootd daemon can be found in the directory $ROOTSYS/bin. It can be started either via inetd or by hand from the command line (no need to be super user). Its performance is comparable with NFS but while NFS requires all kind of system permissions to setup, rootd can be started by any user. The simplest way to start rootd is by starting it from the command line while being logged in to the remote machine. Once started rootd goes immediately in the background (does not need &) and you can log out from the remote node. The only required argument is the range of ports (specified using -p port1-port2). rootd will listen on the first available port in this range. You can also specify -p 0-N to search relative to the service port specified in /etc/services. If a single port is specified (rootd -p 1094) then no search is made. Unless started by inetd (rootd -i), it prints information about the found port, something like: ROOTD_PORT=5151, ROOTD_PID=14433 before spawning the daemon. This way the user knows what was used (eval `rootd` will set these as variables in Bourne-like shells). Also, rootd shows an error message (as well as sending the syslog message) if there is any problem binding the port or forking the daemon.

Using TNetFile you can now read and write files on the remote machine.

In the example below, rootd runs on the remote node under user id minuser and searches for an available port into the range 1094-1098. It finds and listens to port 1094. When creating a TNetFile object you have to specify the same port number 1094 and use minuser (and corresponding password) as login id. When rootd is started in this way, you can only login with the user id under which rootd was started on the remote machine.

hpsalo[]     telnet
login:     minuser
<fsgi02>     rootd -p 1094-1098
<fsgi02>     exit
hpsalo[]     root
root[]     TFile *f = TFile::Open("root://","new")
Name (     minuser

However, you can make many connections since the original rootd will fork (spawn) a new rootd that will service the requests from the TNetFile. The original rootd keeps listening on the specified port for other connections. Each time a TNetFile makes a connection; it gets a new private rootd that will handle its requests. At the end of a ROOT, session when all TNetFiles are closed only the original rootd will stay alive ready to service future TNetFiles.

11.8.5 Starting rootd via inetd

If you expect to often connect via TNetFile to a remote machine, it is more efficient to install rootd as a service of the inetd super daemon. In this way, it is not necessary for each user to run a private rootd. However, this requires a one-time modification of two system files (and super user privileges to do so). Add to /etc/services the line: rootd 1094/tcp. To /etc/inetd.conf the line:

rootd stream tcp nowait root /usr/local/root/bin/rootd rootd -i

After these changes force inetd to reread its configuration file with: “kill -HUP <pid inetd>”. It is not necessary to specify a port number in the URL given to TNetFile when the setup done this way. TNetFile assumes the default port to be 1094 as specified above in the /etc/services file.

11.8.6 Command Line Arguments for rootd

rootd supports the following arguments:

0 = no debug (default) 1 = minimum

2 = medium3 = maximum

11.9 Reading ROOT Files via Apache Web Server

By adding one ROOT specific module to your Apache web server, you can distribute ROOT files to any ROOT user. There is no longer a need to send your files via FTP and risking (out of date) histograms or other objects. Your latest up-to-date results are always accessible to all your colleagues. To access ROOT files via a web server, create a TWebFile object instead of a TFile object with a standard URL as file name. For example:

root[] TWebFile f("")
KEY: TH1F hpx;1 This is the px distribution
KEY: TH2F hpxpy;1 py vs px
KEY: TProfile hprof;1 Profile of pz versus px
KEY: TNtuple ntuple;1 Demo ntuple
root[] hpx.Draw()

Since TWebFile inherits from TFile all TFile operations work as expected. However, due to the nature of a web server a TWebFile is a read-only file. A TWebFile is ideally suited to read relatively small objects (like histograms or other data analysis results). Although possible, you don’t want to analyze large TTree's via a TWebFile.

Here follows a step-by-step recipe for making your Apache 1.1 or 1.2 web server ROOT aware:

11.9.1 Using the General Open Function of TFile

To make life simple we provide a general function to open any type of file (except shared memory files of class TMapFile). This functionality is provided by the static TFile::Open() function:

TFile *TFile::Open(const Text_t *name,Option_t *option="",
const Text_t *title="",Int_t compress,Int_t netopt)

Depending on the name argument, the function returns a TFile, a TNetFile or a TWebFile object. In case a TNetFile URL specifies a local file, a TFile object will be returned (and of course no login information is needed). The arguments of the Open() function are the same as the ones for the TFile constructor.

Using ReOpen() method it is possible to reopen a file with a different access mode, like from READ to UPDATE or from NEW, CREATE, RECREATE, UPDATE to READ. Thus the mode argument can be either “READ” or “UPDATE”. The method returns:

11.10 XML Interface

A new module xml as implemented by Sergey Linev (GSI). It is an optional package that can be used to save a canvas into file.xml file format instead of file.root. XML files do not have any advantages compared to the normal ROOT files, except that the information in these files can be edited via a normal editor. The main motivation for this new format is to facilitate the communication with other non ROOT applications. Currently writing and reading XML files is limited to ROOT applications. It is our intention to develop a simple reader independent of the ROOT libraries that could be used as an example for real applications.

The XML format should be used only for small data volumes, typically histogram files, pictures, geometries, calibrations. The XML file is built in memory before being dumped to disk. Like for normal ROOT files, XML files use the same I/O mechanism exploiting the ROOT/CINT dictionary. Any class having a dictionary can be saved in XML format. This first implementation does not support subdirectories or trees.

The shared library may be loaded dynamically via gSystem->Load("libRXML"). This library is also automatically loaded by the plug-in manager as soon a XML file is created. To create an XTM file, simply specify a filename with an .xml extension when calling TFile::Open. TFile::Open will recognize that you are trying to open an XML file and return a TXMLFile object. When a XML file is open in write mode, one can use the normal TObject::Write to write an object in the file.

   // example of a session saving a histogram to a XML file
   TFile *f = TFile::Open("Example.xml","recreate");
   TH1F *h = new TH1F("h","test",1000,-2,2)
   delete f;
   // example of a session saving a histogram to a XML file
   TFile *f = TFile::Open("Example.xml");
   TH1F *h = (TH1F*)f->Get("h");

The canvas can be saved as a XML file format via File menu / Save or Save As menu entries. One can do also:


12 Trees

12.1 Why Should You Use a Tree?

In the “Input/Output” chapter, we saw how objects can be saved in ROOT files. In case you want to store large quantities of same-class objects, ROOT has designed the TTree and TNtuple classes specifically for that purpose. The TTree class is optimized to reduce disk space and enhance access speed. A TNtuple is a TTree that is limited to only hold floating-point numbers; a TTree on the other hand can hold all kind of data, such as objects or arrays in addition to all the simple types.

When using a TTree, we fill its branch buffers with leaf data and the buffers are written to disk when it is full. Branches, buffers, and leafs, are explained a little later in this chapter, but for now, it is important to realize that each object is not written individually, but rather collected and written a bunch at a time.

This is where the TTree takes advantage of compression and will produce a much smaller file than if the objects were written individually. Since the unit to be compressed is a buffer, and the TTree contains many same-class objects, the header of the objects can be compressed.

The TTree reduces the header of each object, but it still contains the class name. Using compression, the class name of each same-class object has a good chance of being compressed, since the compression algorithm recognizes the bit pattern representing the class name. Using a TTree and compression the header is reduced to about 4 bytes compared to the original 60 bytes. However, if compression is turned off, you will not see these large savings.

The TTree is also used to optimize the data access. A tree uses a hierarchy of branches, and each branch can be read independently from any other branch. Now, assume that Px and Py are data members of the event, and we would like to compute Px2 + Py2 for every event and histogram the result.

If we had saved the million events without a TTree we would have to:

We would have to do that a million times! This is very time consuming, and we really do not need to read the entire event, every time. All we need are two little data members (Px and Py). On the other hand, if we use a tree with one branch containing Px and another branch containing Py, we can read all values of Px and Py by only reading the Px and Py branches. This makes the use of the TTree very attractive.

12.2 A Simple TTree

This script builds a TTree from an ASCII file containing statistics about the staff at CERN. This script, staff.C and its input file staff.dat are in $ROOTSYS/tutorials/tree.

   // example of macro to read data from an ascii file and
   // create a root file with an histogram and a TTree

   // the structure to hold the variables for the branch

   struct staff_t {
      Int_t cat;
      Int_t division;
      Int_t flag;
      Int_t age;
      Int_t service;
      Int_t children;
      Int_t grade;
      Int_t step;
      Int_t nation;
      Int_t hrweek;
      Int_t cost;
   staff_t staff;
   // continued...
   // open the ASCII file
   FILE *fp = fopen("staff.dat","r");
   char line[81];
   // create a new ROOT file
   TFile *f = new TFile("staff.root","RECREATE");
   // create a TTree
   TTree *tree = new TTree("T","staff data from ascii file");
   // create one branch with all information from the stucture
   // fill the tree from the values in ASCII file
   while (fgets(&line,80,fp)) {
   // check what the tree looks like

The script declares a structure called staff_t, with several integers representing the relevant attribute of a staff member. It opens the ASCII file, creates a ROOT file and a TTree. Then it creates one branch with the TTree::Branch method. The first parameter of the Branch method is the branch name. The second parameter is the address from which the first leaf is to be read. In this example it is the address of the structure staff. Once the branch is defined, the script reads the data from the ASCII file into the staff_t structure and fills the tree. The ASCII file is closed, and the ROOT file is written to disk saving the tree. Remember, trees and histograms are created in the current directory, which is the file in our example. Hence an f->Write()saves the tree.

12.3 Show an Entry with TTree::Show

An easy way to access one entry of a tree is the use the TTree::Show method. For example to look at the 10th entry in the staff.root tree:

root[] TFile f("staff.root")
root[] T->Show(10)
======> EVENT:10
 Category        = 361
 Flag            = 15
 Age             = 51
 Service         = 29
 Children        = 0
 Grade           = 7
 Step            = 13
 Hrweek          = 40
 Cost            = 7599
 Division        = PS
 Nation          = FR

A helpful command to see the tree structure meaning the number of entries, the branches and the leaves, is TTree::Print.

root[] T->Print()
*Tree    :T         : staff data from ascii file                     *
*Entries :3354      : Total = 245417 bytes  File  Size =        59945*
*                     Tree compression factor =   2.90               *
*Br    0 :staff     :Category/I:Flag:Age:Service:Children:Grade:...  *
*         | Cost                                                     *
*Entries :3354 : Total Size  = 154237 bytes  File Size = 32316       *
*Baskets :   3 : Basket Size =  32000 bytes  Compression= 2.97       *

12.5 Scan a Variable the Tree with TTree::Scan

The TTree::Scan method shows all values of the list of leaves separated by a colon.

root[] T->Scan("Cost:Age:Children")
*    Row   *      Cost *       Age *  Children *
*        0 *     11975 *        58 *         0 *
*        1 *     10228 *        63 *         0 *
*        2 *     10730 *        56 *         2 *
*        3 *      9311 *        61 *         0 *
*        4 *      9966 *        52 *         2 *
*        5 *      7599 *        60 *         0 *
*        6 *      9868 *        53 *         1 *
*        7 *      8012 *        60 *         1 *

12.6 The Tree Viewer

Activating the tree viewer

Activating the tree viewer

The tree viewer is a quick and easy way to examine a tree. To start the tree viewer, open a file and object browser. Right click on a TTree and select StartViewer. You can also start the tree viewer from the command line. First load the viewer library.

root[] TFile f("staff.root")
root[] T->StartViewer()

If you want to start a tree viewer without a tree, you need to load the tree player library first:

root[] gSystem->Load("")
root[] new TTreeViewer()

The figure above shows how the tree viewer looks like for the example file staff.root. The left panel contains the list of trees and their branches; in this case there is only one tree. You can add more trees with the File-Open command to open the file containing the new tree, then use the context menu on the right panel, select SetTreeName and enter the name of the tree to add. On the right are the leaves or variables in the tree. You can double click on any leaf to a histogram it.

The toolbar in the upper part can be used for user commands, changing the drawing option and the histogram name. The lower part contains three picture buttons that draw a histogram, stop the current command, and refresh the tree.

The three check buttons toggle the following:

Hist- the histogram drawing mode;

Scan- enables redirecting of TTree::Scancommand in an ASCII file;

Rec - enables recording of the last issued command.

Below them there are two text widgets for specifying the input and output event lists. A Tree Viewer session is made by the list of user-defined expressions and cuts, applying to a specified tree. A session can be saved using File / SaveSource menu or the SaveSource method from the context menu of the right panel. This will create a macro having as default name treeviewer.C that can be ran at any time to reproduce the session.

Besides the list of user-defined expressions, a session may contain a list of RECORDS. A record can be produced in the following way: dragging leaves/expression on X/Y/Z; changing drawing options; clicking the RED button on the bottom when happy with the histogram

NOTE that just double clicking a leaf will not produce a record: the histogram must be produced when clicking the DRAW button on the bottom-left. The records will appear on the list of records in the bottom right of the tree viewer. Selecting a record will draw the corresponding histogram. Records can be played using the arrow buttons near to the record button. When saving the session, the list of records is being saved as well.

Records have a default name corresponding to the Z: Y: X selection, but this can be changed using SetRecordName() method from the right panel context menu. You can create a new expression by right clicking on any of theE() boxes. The expression can be dragged and dropped into any of the boxes (X, Y, Z, Cut, or Scan). To scan one or more variables, drop them into the Scan box, then double click on the box. You can also redirect the result of the scan to a file by checking the Scan box on top.

When the “Rec” box is checked, the Draw and Scan commands are recorded in the history file and echoed on the command line. The “Histogram” text box contains the name of the resulting histogram. By default it is htemp. You can type any name, if the histogram does not exist it will create one. The Option text box contains the list of Draw options. See “Draw Options”. You can select the options with the Options menu. The Command box lets you enter any command that you could also enter on the command line. The vertical slider on the far left side can be used to select the minimum and maximum of an event range. The actual start and end index are shown in on the bottom in the status window.

There is an extensive help utility accessible with the Help menu. The IList and OList are to specify an input list of entry indices and a name for the output list respectively. Both need to be of type TList and contain integers of entry indices. These lists are described below in the paragraph “Error! Reference source not found.”.

A couple of graphs

A couple of graphs

The first one is a plot of the age distribution, the second a scatter plot of the cost vs. age. The second one was generated by dragging the age leaf into the Y-box and the cost leaf into the X-box, and pressing the Draw button. By default, this will generate a scatter plot. Select a different option, for example "lego" to create a 2D histogram.

12.7 Creating and Saving Trees

This picture shows the TTree class:

The TTree class

The TTree class

To create a TTree we use its constructor. Then we design our data layout and add the branches. A tree can be created by giving a name and title:

   TTree t("MyTree","Example Tree");

12.7.1 Creating a Tree from a Folder Hierarchy

An alternative way to create a tree and organize it is to use folders (see “Folders and Tasks”). You can build a folder structure and create a tree with branches for each of the sub-folders:

   TTree folder_tree("MyFolderTree","/MyFolder");

The second argument "/MyFolder"is the top folder, and the “/” signals the TTree constructor that this is a folder not just the title. You fill the tree by placing the data into the folder structure and calling TTree::Fill.

12.7.2 Tree and TRef Objects


This call requests the construction of an optional branch supporting table of references (TRefTable). This branch (TBranchRef) will keep all the information needed to find the branches containing referenced objects at each Tree::Fill, the branch numbers containing the referenced objects are saved in the table of references. When the Tree header is saved (via TTree::Write for example), the branch is saved, keeping the information with the pointers to the branches having referenced objects. Enabling this optional table, allow TTree::Draw to automatically load the branches needed to dereference a TRef (or TRefArray) object.

12.7.3 Autosave

Autosave gives the option to save all branch buffers every n byte. We recommend using Autosave for large acquisitions. If the acquisition fails to complete, you can recover the file and all the contents since the last Autosave. To set the number of bytes between Autosave you can use the TTree::SetAutosave() method. You can also call TTree::Autosave in the acquisition loop every nentry.

12.7.4 Trees with Circular Buffers

When a TTree is memory resident, you set it up so that it retains retain only the last few entries. For example, this can be very useful for monitoring purpose.

   void TTree::SetCircular(Long64_t maxEntries);

where maxEntries is the maximum number of entries to be kept in the buffers. When the number of entries exceeds this value, the first entries in the Tree are deleted and the buffers used again. An example of a script using a circular buffer is shown below:

void circular() {
   gROOT->cd(); //make sure that the Tree is memory resident
   TTree *T = new TTree("T","test circular buffers");
   TRandom r;
   Float_t px,py,pz;
   Double_t random;
   UShort_t i;
   for (i = 0; i < 65000; i++) {
      pz = px*px + py*py;
      random = r.Rndm();

12.7.5 Size of TTree in the File

When writing a TTree to a file, if the file size reaches the value stored in the TTree::GetMaxTreeSize(), the current file is closed and a new file is created. If the original file is named “myfile.root”, subsequent files are named “myfile_1.root”, “myfile_2.root”, etc.

Currently, the automatic change of file is restricted to the case where the tree is in the top level directory. The file should not contain sub-directories. Before switching to a new file, the tree header is written to the current file, then the current file is closed. To process the multiple files created by ChangeFile(), one must use a TChain.

The new file name has a suffix “_N” where N is equal to fFileNumber+1. By default a Root session starts with fFileNumber=0. One can set fFileNumber to a different value via TTree::SetFileNumber(). In case a file named “_N” already exists, the function will try a file named “__N”, then “___N”, etc. The maximum tree size can be set via the static function TTree::SetMaxTreeSize(). The default value of fgMaxTreeSize is 1.9 GB. If the current file contains other objects (like TH1 and TTree), these objects are automatically moved to the new file.

12.7.6 User Info Attached to a TTree Object

The function TTree::GetUserInfo() allows adding any object defined by a user to the tree that is not depending on the entry number. For example:


12.7.7 Indexing a Tree

Use TTree::BuildIndex(), to build an index table using expressions depending on the value in the leaves.

tree->BuildIndex(majorname, minorname);

The index is built in the following way:

Once the index is computed, using the TTree::GetEntryWithIndex(majornumber, minornumber) one entry can be retrieved. Example:

   // to create an index using leaves Run and Event
   // to read entry corresponding to Run=1234 and Event=56789

Note that majorname and minorname may be expressions using original tree variables e.g.: “run-90000”, “event +3*xx”. In case an expression is specified, the equivalent expression must be computed when calling GetEntryWithIndex(). To build an index with only majorname, specify minorname="0" (default).

Note that once the index is built, it can be saved with the TTree object with:

   tree.Write();     //if the file has been open in "update" mode

The most convenient place to create the index is at the end of the filling process just before saving the tree header. If a previous index was computed, it is redefined by this new call.

Note that this function can also be applied to a TChain. The return value is the number of entries in the Index (< 0 indicates failure).

12.8 Branches

The organization of branches allows the designer to optimize the data for the anticipated use. The class for a branch is called TBranch. If two variables are independent, and the designer knows the variables will not be used together, they should be placed on separate branches. If, however, the variables are related, such as the coordinates of a point, it is most efficient to create one branch with both coordinates on it. A variable on a TBranch is called a leaf (yes - TLeaf). Another point to keep in mind when designing trees is that branches of the same TTree can be written to separate files. To add a TBranch to a TTree we call the method TTree::Branch(). Note that we DO NOT use the TBranch constructor.

The TTree::Branch method has several signatures. The branch type differs by what is stored in it. A branch can hold an entire object, a list of simple variables, contents of a folder, contents of a TList, or an array of objects. Let’s see some examples. To follow along you will need the shared library First, check if it is in $ROOTSYS/test. If it is, copy it to your own area. If it is not there, you have to build it by typing make in $ROOTSYS/test.

12.9 Adding a Branch to Hold a List of Variables

As in the very first example (staff.root) the data we want to save is a list of simple variables, such as integers or floats. In this case, we use the following TTree::Branch signature:


The first parameter is the branch name.

The second parameter is the address from which the first variable is to be read. In the code above, “event” is a structure with one float and three integers and one unsigned integer. You should not assume that the compiler aligns the elements of a structure without gaps. To avoid alignment problems, you need to use structures with same length members. If your structure does not qualify, you need to create one branch for each element of the structure.

The leaf name is NOT used to pick the variable out of the structure, but is only used as the name for the leaf. This means that the list of variables needs to be in a structure in the order described in the third parameter.

This third parameter is a string describing the leaf list. Each leaf has a name and a type separated by a “/” and it is separated from the next leaf by a “:”.


The example on the next line has two leafs: a floating-point number called temp and an integer named ntrack.


The type can be omitted and if no type is given, the same type as the previous variable is assumed. This leaf list has three integers called ntrack, nseg, and nvtex.


There is one more rule: when no type is given for the very first leaf, it becomes a float (F). This leaf list has three floats called temp, mass, and px.


The symbols used for the type are:

The type is used for a byte count to decide how much space to allocate. The variable written is simply the block of bytes starting at the starting address given in the second parameter. It may or may not match the leaf list depending on whether or not the programmer is being careful when choosing the leaf address, name, and type.

By default, a variable will be copied with the number of bytes specified in the type descriptor symbol. However, if the type consists of two characters, the number specifies the number of bytes to be used when copying the variable to the output buffer. The line below describes ntrack to be written as a 16-bit integer (rather than a 32-bit integer).


With this Branch method, you can also add a leaf that holds an entire array of variables. To add an array of floats use the f[n] notation when describing the leaf.

   Float_t  f[10];

You can also add an array of variable length:

   TFile *f = new TFile("peter.root","recreate");
   Int_t nPhot;
   Float_t E[500];
   TTree* nEmcPhotons = new TTree("nEmcPhotons","EMC Photons");

See “Example 2: A Tree with a C Structure” below ($ROOTSYS/tutorials/tree/tree2.C) and staff.C at the beginning of this chapter.

12.10 Adding a TBranch to Hold an Object

To write a branch to hold an event object, we need to load the definition of the Event class, which is in $ROOTSYS/test/ (if it doesn’t exist type make in $ROOTSYS/test). An object can be saved in a tree if a ROOT dictionary for its class has been generated and loaded.

root[] .L

First, we need to open a file and create a tree.

root[]     TFile *f = new TFile("AFile.root","RECREATE")
root[]     TTree *tree = new TTree("T","A Root Tree")

We need to create a pointer to an Event object that will be used as a reference in the TTree::Branch method. Then we create a branch with the TTree::Branch method.

root[]     Event *event = new Event()
root[]     tree->Branch("EventBranch","Event",&event,32000,99)

To add a branch to hold an object we use the signature above. The first parameter is the name of the branch. The second parameter is the name of the class of the object to be stored. The third parameter is the address of a pointer to the object to be stored.

Note that it is an address of a pointer to the object, not just a pointer to the object.

The fourth parameter is the buffer size and is by default 32000 bytes. It is the number of bytes of data for that branch to save to a buffer until it is saved to the file. The last parameter is the split-level, which is the topic of the next section. Static class members are not part of an object and thus not written with the object. You could store them separately by collecting these values in a special “status” object and write it to the file outside of the tree. If it makes sense to store them for each object, make them a regular data member.

12.10.1 Setting the Split-level

To split a branch means to create a sub-branch for each data member in the object. The split-level can be set to 0 to disable splitting or it can be set to a number between 1 and 99 indicating the depth of splitting.

If the split-level is set to zero, the whole object is written in its entirety to one branch. The TTree will look like the one on the right, with one branch and one leaf holding the entire event object.

A split and not split tree

A split and not split tree

When the split-level is 1, an object data member is assigned a branch. If the split-level is 2, the data member objects will be split also, and a split level of 3 its data members objects, will be split. As the split-level increases so does the splitting depth. The ROOT default for the split-level is 99. This means the object will be split to the maximum. Memory Considerations when Splitting a Branch

Splitting a branch can quickly generate many branches. Each branch has its own buffer in memory. In case of many branches (say more than 100), you should adjust the buffer size accordingly. A recommended buffer size is 32000 bytes if you have less than 50 branches. Around 16000 bytes if you have less than 100 branches and 4000 bytes if you have more than 500 branches. These numbers are recommended for computers with memory size ranging from 32MB to 256MB. If you have more memory, you should specify larger buffer sizes. However, in this case, do not forget that your file might be used on another machine with a smaller memory configuration. Performance Considerations when Splitting a Branch

A split branch is faster to read, but slightly slower to write. The reading is quicker because variables of the same type are stored consecutively and the type does not have to be read each time. It is slower to write because of the large number of buffers as described above. See "

Performance Benchmarks" for performance impact of split and non-split mode. Rules for Splitting

When splitting a branch, variables of different types are handled differently. Here are the rules that apply when splitting a branch.

   // STL vector of vectors of TAxis*
   vector<vector<TAxis *> >  fVectAxis;
   // STL map of string/vector
   map<string,vector<int> >  fMapString;
   // STL deque of pair
   deque<pair<float,float> > fDequePair;

12.10.2 Exempt a Data Member from Splitting

If you are creating a branch with an object and in general you want the data members to be split, but you want to exempt a data member from the split. You can specify this in the comment field of the data member:

class Event : public TObject {
      EventHeader    fEvtHdr;      //|| Don't split the header

12.10.3 Adding a Branch to Hold a TClonesArray

ROOT has two classes to manage arrays of objects. The TObjArray can manage objects of different classes, and the TClonesArray that specializes in managing objects of the same class (hence the name Clones Array). TClonesArray takes advantage of the constant size of each element when adding the elements to the array. Instead of allocating memory for each new object as it is added, it reuses the memory. Here is an example of the time a TClonesArray can save over a TObjArray. We have 100,000 events, and each has 10,000 tracks, which gives 1,000,000,000 tracks. If we use a TObjArray for the tracks, we implicitly make a call to new and a corresponding call to delete for each track. The time it takes to make a pair of new/delete calls is about 7 s (10-6). If we multiply the number of tracks by 7 s, (1,000,000,000 * 7 * 10-6) we calculate that the time allocating and freeing memory is about 2 hours. This is the chunk of time saved when a TClonesArray is used rather than a TObjArray. If you do not want to wait 2 hours for your tracks (or equivalent objects), be sure to use a TClonesArray for same-class objects arrays. Branches with TClonesArrays use the same method (TTree::Branch) as any other object described above. If splitting is specified the objects in the TClonesArray are split, not the TClonesArray itself.

12.10.4 Identical Branch Names

When a top-level object (say event), has two data members of the same class the sub branches end up with identical names. To distinguish the sub branch we must associate them with the master branch by including a “.” (a dot) at the end of the master branch name. This will force the name of the sub branch to be master.sub branch instead of simply sub branch. For example, a tree has two branches Trigger and MuonTrigger, each containing an object of the same class (Trigger). To identify uniquely the sub branches we add the dot:


If Trigger has three members, T1, T2, T3, the two instructions above will generate sub branches called: Trigger.T1, Trigger.T2, Trigger.T3, MuonTrigger.T1, MuonTrigger.T2, andMuonTrigger.T3.

12.11 Adding a Branch with a Folder

Use the syntax below to add a branch from a folder:


This method creates one branch for each element in the folder. The method returns the total number of branches created.

12.12 Adding a Branch with a Collection

This Branch method creates one branch for each element in the collection.

   tree->Branch(*aCollection, 8000, 99);
   // Int_t TTree::Branch(TCollection *list, Int_t bufsize,
   //                     Int_t splitlevel, const char *name)

The method returns the total number of branches created. Each entry in the collection becomes a top level branch if the corresponding class is not a collection. If it is a collection, the entry in the collection becomes in turn top level branches, etc. The split level is decreased by 1 every time a new collection is found. For example if list is a TObjArray*

In case a collection element is a TClonesArray, the special Tree constructor for TClonesArray is called. The collection itself cannot be a TClonesArray. If name is given, all branch names will be prefixed with name_.

IMPORTANT NOTE1: This function should not be called if splitlevel<1. IMPORTANT NOTE2: The branches created by this function will have names corresponding to the collection or object names. It is important to give names to collections to avoid misleading branch names or identical branch names. By default collections have a name equal to the corresponding class name, e.g. the default name of TList is “TList”.

12.13 Examples for Writing and Reading Trees

The following sections are examples of writing and reading trees increasing in complexity from a simple tree with a few variables to a tree containing folders and complex Event objects. Each example has a named script in the $ROOTSYS/tutorials/tree directory. They are called tree1.C to tree4.C. The examples are:

Each script contains the main function, with the same name as the file (i.e. tree1), the function to write - tree1w, and the function to read - tree1r. If the script is not run in batch mode, it displays the tree in the browser and tree viewer. To study the example scripts, you can either execute the main script, or load the script and execute a specific function. For example:

// execute the function that writes, reads, shows the tree
root[]     x tree1.C
// use ACLiC to build shared library, check syntax, execute
root[] x tree1.C++
// Load the script and select a function to execute
root[]     L tree1.C
root[]     tree1w()
root[]     tree1r()

12.14 Example 1: A Tree with Simple Variables

This example shows how to write, view, and read a tree with several simple (integers and floating-point) variables.

12.14.1 Writing the Tree

Below is the function that writes the tree (tree1w). First, the variables are defined (px, py, pz, random and ev). Then we add a branch for each of the variables to the tree, by calling the TTree::Branch method for each variable.

void tree1w(){

   // create a tree file tree1.root - create the file, the Tree and
   // a few branches
   TFile f("tree1.root","recreate");
   TTree t1("t1","a simple Tree with simple variables");
   Float_t px, py, pz;
   Double_t random;
   Int_t ev;

   // fill the tree
   for (Int_t i=0; i<10000; i++) {
      pz = px*px + py*py;
      random = gRandom->Rndm();
      ev = i;
   // save the Tree heade; the file will be automatically closed
   // when going out of the function scope
} Creating Branches with A single Variable

This is the signature of TTree::Branch to create a branch with a list of variables:

   TBranch* TTree::Branch(const char* name,void* address,
                          const char* leaflist,
                          Int_t bufsize = 32000)

The first parameter is the branch name. The second parameter is the address from which to read the value. The third parameter is the leaf list with the name and type of each leaf. In this example, each branch has only one leaf. In the box below, the branch is named px and has one floating point type leaf also called px.

   t1.Branch("px",&px,"px/F"); Filling the Tree

First we find some random values for the variables. We assign px and py a Gaussian with mean = 0 and sigma = 1 by calling gRandom->Rannor(px,py), and calculatepz. Then we call the TTree::Fill() method. The call t1.Fill() fills all branches in the tree because we have already organized the tree into branches and told each branch where to get the value from. After this script is executed we have a ROOT file called tree1.root with a tree called t1. There is a possibility to fill branches one by one using the method TBranch::Fill(). In this case you do not need to call TTree::Fill() method. The entries can be set by TTree::SetEntries(Double_t n). Calling this method makes sense only if the number of existing entries is null.

12.14.2 Viewing the Tree

The tree1.root file and its tree in the browser and a leaf histogram

The tree1.root file and its tree in the browser and a leaf histogram

In the right panel of the ROOT object browse are the branches: ev, px, py, pz, and random. Note that these are shown as leaves because they are “end” branches with only one leaf. To histogram a leaf, we can simply double click on it in the browser. This is how the tree t1 looks in the Tree Viewer. Here we can add a cut and add other operations for histogramming the leaves. See “The Tree Viewer”. For example, we can plot a two dimensional histogram.

The tree viewer

The tree viewer

12.14.3 Reading the Tree

The tree1r function shows how to read the tree and access each entry and each leaf. We first define the variables to hold the read values.

   Float_t px, py, pz;

Then we tell the tree to populate these variables when reading an entry. We do this with the method TTree::SetBranchAddress. The first parameter is the branch name, and the second is the address of the variable where the branch data is to be placed. In this example, the branch name is px. This name was given when the tree was written (see tree1w). The second parameter is the address of the variable px.

   t1->SetBranchAddress("px",&px); GetEntry

Once the branches have been given the address, a specific entry can be read into the variables with the method TTree::GetEntry(n). It reads all the branches for entry (n) and populates the given address accordingly. By default, GetEntry() reuses the space allocated by the previous object for each branch. You can force the previous object to be automatically deleted if you call mybranch.SetAutoDelete(kTRUE) (default is kFALSE).

Consider the example in $ROOTSYS/test/Event.h. The top-level branch in the tree T is declared with:

   Event *event = 0;
   // event must be null or point to a valid object;
   // it must be initialized

When reading the Tree, one can choose one of these 3 options:

Option 1:

   for (Int_t i = 0; i<nentries; i++) {
      //the object event has been filled at this point

This is the default and recommended way to create an object of the class Event.It will be pointed by event.

At the following entries, event will be overwritten by the new data. All internal members that are TObject* are automatically deleted. It is important that these members be in a valid state when GetEntry is called. Pointers must be correctly initialized. However these internal members will not be deleted if the characters “->” are specified as the first characters in the comment field of the data member declaration.

The pointer member is read via the pointer->Streamer(buf) if “->” is specified. In this case, it is assumed that the pointer is never null (see pointer TClonesArray *fTracks in the $ROOTSYS/test/Event example). If “->” is not specified, the pointer member is read via buf >> pointer. In this case the pointer may be null. Note that the option with “->” is faster to read or write and it also consumes less space in the file.

Option 2 - the option AutoDelete is set:

   TBranch *branch = T.GetBranch("event");
   for (Int_t i=0; i<nentries; i++) {
      T.GetEntry(i); // the object event has been filled at this point

At any iteration, the GetEntry deletes the object event and a new instance of Event is created and filled.

Option 3 - same as option 1, but you delete the event yourself:

   for (Int_t i=0; i<nentries; i++) {
      delete event;
      event = 0;      //EXTREMELY IMPORTANT
      // the objrect event has been filled at this point

It is strongly recommended to use the default option 1. It has the additional advantage that functions like TTree::Draw (internally calling TTree::GetEntry) will be functional even when the classes in the file are not available. Reading selected branches is quicker than reading an entire entry. If you are interested in only one branch, you can use the TBranch::GetEntry method and only that branch is read. Here is the script tree1r:

void tree1r(){
   // read the Tree generated by tree1w and fill two histograms
   // note that we use "new" to create the TFile and TTree objects,
   // to keep them alive after leaving this function.
   TFile *f = new TFile("tree1.root");
   TTree *t1 = (TTree*)f->Get("t1");
   Float_t px, py, pz;
   Double_t random;
   Int_t ev;
   // create two histograms
   TH1F *hpx   = new TH1F("hpx","px distribution",100,-3,3);
   TH2F *hpxpy = new TH2F("hpxpy","py vs px",30,-3,3,30,-3,3);
   //read all entries and fill the histograms
   Int_t nentries = (Int_t)t1->GetEntries();
   for (Int_t i=0; i<nentries; i++) {
   // We do not close the file. We want to keep the generated
   // histograms we open a browser and the TreeViewer
   if (gROOT->IsBatch()) return;
   new TBrowser ();

   //In the browser, click on "ROOT Files", then on "tree1.root"
   //You can click on the histogram icons in the right panel to draw
   //them in the TreeViewer, follow the instructions in the Help.

12.15 Example 2: A Tree with a C Structure

The executable script for this example is $ROOTSYS/tutorials/tree/tree2.C.In this example we show:

A C structure (struct) is used to build a ROOT tree. In general we discourage the use of C structures, we recommend using a class instead. However, we do support them for legacy applications written in C or FORTRAN. The example struct holds simple variables and arrays. It maps to a Geant3 common block /gctrak/.This is the definition of the common block/structure:

const Int_t MAXMEC = 30;

typedef struct {
   Float_t  vect[7];
   Float_t  getot;
   Float_t  gekin;
   Float_t  vout[7];
   Int_t    nmec;
   Int_t    lmec[MAXMEC];
   Int_t    namec[MAXMEC];
   Int_t    nstep;
   Int_t    pid;
   Float_t  destep;
   Float_t  destel;
   Float_t  safety;
   Float_t  sleng;
   Float_t  step;
   Float_t  snext;
   Float_t  sfield;
   Float_t  tofg;
   Float_t  gekrat;
   Float_t  upwght;
} Gctrak_t;

When using Geant3, the common block is filled by Geant3 routines at each step and only the TTree::Fill method needs to be called. In this example we emulate the Geant3 step routine with the helixStep function. We also emulate the filling of the particle values. The calls to the Branch methods are the same as if Geant3 were used.

void helixStep(Float_t step, Float_t *vect, Float_t *vout)
   // extrapolate track in constant field
   Float_t field = 20; // field in kilogauss
   enum Evect {kX,kY,kZ,kPX,kPY,kPZ,kPP};
   vout[kPP] = vect[kPP];

   Float_t h4    = field*2.99792e-4;
   Float_t rho   = -h4/vect[kPP];
   Float_t tet   = rho*step;
   Float_t tsint = tet*tet/6;
   Float_t sintt = 1 - tsint;
   Float_t sint  = tet*sintt;
   Float_t cos1t = tet/2;
   Float_t f1 = step*sintt;
   Float_t f2 = step*cos1t;
   Float_t f3 = step*tsint*vect[kPZ];
   Float_t f4 = -tet*cos1t;
   Float_t f5 = sint;
   Float_t f6 = tet*cos1t*vect[kPZ];

   vout[kX]  = vect[kX]  + (f1*vect[kPX] - f2*vect[kPY]);
   vout[kY]  = vect[kY]  + (f1*vect[kPY] + f2*vect[kPX]);
   vout[kZ]  = vect[kZ]  + (f1*vect[kPZ] + f3);
   vout[kPX] = vect[kPX] + (f4*vect[kPX] - f5*vect[kPY]);
   vout[kPY] = vect[kPY] + (f4*vect[kPY] + f5*vect[kPX]);
   vout[kPZ] = vect[kPZ] + (f4*vect[kPZ] + f6);

12.15.1 Writing the Tree

void tree2w() {
   // write tree2 example
   //create a Tree file tree2.root
   TFile f("tree2.root","recreate");

   //create the file, the Tree
   TTree t2("t2","a Tree with data from a fake Geant3");
   // declare a variable of the C structure type
   Gctrak_t gstep;

   // add the branches for a subset of gstep

   //Initialize particle parameters at first point
   Float_t px,py,pz,p,charge=0;
   Float_t vout[7];
   Float_t mass  = 0.137;
   Bool_t newParticle = kTRUE;
   gstep.step    = 0.1;
   gstep.destep  = 0;
   gstep.nmec    = 0;     = 0;

   //transport particles
   for (Int_t i=0; i<10000; i++) {
      //generate a new particle if necessary (Geant3 emulation)
      if (newParticle) {
      px = gRandom->Gaus(0,.02);
      py = gRandom->Gaus(0,.02);
      pz = gRandom->Gaus(0,.02);
      p  = TMath::Sqrt(px*px+py*py+pz*pz);
      charge = 1;
      if (gRandom->Rndm() < 0.5) charge = -1;    += 1;
         gstep.vect[0] = 0;
         gstep.vect[1] = 0;
         gstep.vect[2] = 0;
         gstep.vect[3] = px/p;
         gstep.vect[4] = py/p;
         gstep.vect[5] = pz/p;
         gstep.vect[6] = p*charge;
         gstep.getot   = TMath::Sqrt(p*p + mass*mass);
         gstep.gekin   = gstep.getot - mass;
         newParticle   = kFALSE;
      // fill the Tree with current step parameters

      //transport particle in magnetic field (Geant3 emulation)
      helixStep(gstep.step, gstep.vect, vout);
      //make one step
      //apply energy loss
      gstep.destep   = gstep.step*gRandom->Gaus(0.0002,0.00001);
      gstep.gekin -= gstep.destep;
      gstep.getot  = gstep.gekin + mass;
      gstep.vect[6]= charge*TMath::Sqrt(gstep.getot*gstep.getot
                      - mass*mass);
      gstep.vect[0] = vout[0];
      gstep.vect[1] = vout[1];
      gstep.vect[2] = vout[2];
      gstep.vect[3] = vout[3];
      gstep.vect[4] = vout[4];
      gstep.vect[5] = vout[5];
      gstep.nmec    = (Int_t)(5*gRandom->Rndm());
      for (Int_t l=0; l<gstep.nmec; l++) gstep.lmec[l] = l;
      if  (gstep.gekin < 0.001) newParticle = kTRUE;
      if  (TMath::Abs(gstep.vect[2]) > 30) newParticle = kTRUE;
   //save the Tree header. The file will be automatically
   // closed when going out of the function scope
} Adding a Branch with a Fixed Length Array

At first, we create a tree and create branches for a subset of variables in the C structureGctrak_t. Then we add several types of branches. The first branch reads seven floating-point values beginning at the address of 'gstep.vect'. You do not need to specify &gstep.vect, because in C and C++ the array variable holds the address of the first element.

   t2.Branch("gekin",&gstep.gekin,"gekin/F"); Adding a Branch with a Variable Length Array

The next two branches are dependent on each other. The first holds the length of the variable length array and the second holds the variable length array. The lmec branch reads nmec number of integers beginning at the address gstep.lmec.


The variable nmec is a random number and is reset for each entry.

   gstep.nmec = (Int_t)(5*gRandom->Rndm()); Filling the Tree

In this emulation of Geant3, we generate and transport particles in a magnetic field and store the particle parameters at each tracking step in a ROOT tree.

12.15.2 Analysis

In this analysis, we do not read the entire entry we only read one branch. First, we set the address for the branch to the file dstep, and then we use the TBranch::GetEntry method. Then we fill a histogram with the dstep branch entries, draw it and fit it with a Gaussian. In addition, we draw the particle’s path using the three values in the vector. Here we use the TTree::Draw method. It automatically creates a histogram and plots the 3 expressions (see Trees in Analysis).

void tree2r() {

   // read the Tree generated by tree2w and fill one histogram
   // we are only interested by the destep branch

   // note that we use "new" to create the TFile and TTree objects because we
   // want to keep these objects alive when we leave this function
   TFile *f = new TFile("tree2.root");
   TTree *t2 = (TTree*)f->Get("t2");
   static Float_t destep;
   TBranch *b_destep = t2->GetBranch("destep");

   //create one histogram
   TH1F *hdestep = new TH1F("hdestep","destep in Mev",100,1e-5,3e-5);
   //read only the destep branch for all entries
   Int_t nentries = (Int_t)t2->GetEntries();
   for (Int_t i=0;i<nentries;i++) {
      // fill the histogram with the destep entry

   // we do not close the file; we want to keep the generated histograms;
   // we fill a 3-d scatter plot with the particle step coordinates
   TCanvas *c1 = new TCanvas("c1","c1",600,800);


   gPad->SetFillColor(37);                       // continued...
   if (gROOT->IsBatch()) return;

   // invoke the x3d viewer

12.16 Example 3: Adding Friends to Trees

In this example, we will show how to extend a tree with a branch from another tree with the Friends feature.

12.16.1 Adding a Branch to an Existing Tree

You may want to add a branch to an existing tree. For example, if one variable in the tree was computed with a certain algorithm, you may want to try another algorithm and compare the results. One solution is to add a new branch, fill it, and save the tree. The code below adds a simple branch to an existing tree. Note that the kOverwrite option in the Write method overwrites the existing tree. If it is not specified, two copies of the tree headers are saved.

void tree3AddBranch() {
   TFile f("tree3.root","update");
   Float_t new_v;
   TTree *t3 = (TTree*)f->Get("t3");
   TBranch *newBranch = t3-> Branch("new_v",&new_v,"new_v/F");
   //read the number of entries in the t3
   Int_t nentries = (Int_t)t3->GetEntries();
   for (Int_t i = 0; i < nentries; i++){
      new_v= gRandom->Gaus(0,1);
   t3->Write("",TObject::kOverwrite); // save only the new version of
                                      // the tree

Adding a branch is often not possible because the tree is in a read-only file and you do not have permission to save the modified tree with the new branch. Even if you do have the permission, you risk loosing the original tree with an unsuccessful attempt to save the modification. Since trees are usually large, adding a branch could extend it over the 2GB limit. In this case, the attempt to write the tree fails, and the original data is may also be corrupted. In addition, adding a branch to a tree enlarges the tree and increases the amount of memory needed to read an entry, and therefore decreases the performance. For these reasons, ROOT offers the concept of friends for trees (and chains). We encourage you to use TTree::AddFriend rather than adding a branch manually.

12.16.2 TTree::AddFriend

A tree keeps a list of friends. In the context of a tree (or a chain), friendship means unrestricted access to the friends data. In this way it is much like adding another branch to the tree without taking the risk of damaging it. To add a friend to the list, you can use the TTree::AddFriendmethod. The TTree (tree) below has two friends (ft1 and ft2) and now has access to the variables a,b,c,i,j,k,l and m.

The AddFriend method has two parameters, the first is the tree name and the second is the name of the ROOT file where the friend tree is saved. AddFriend automatically opens the friend file. If no file name is given, the tree called ft1 is assumed to be in the same file as the original tree.


If the friend tree has the same name as the original tree, you can give it an alias in the context of the friendship:

   tree.AddFriend("tree1 = tree","friendfile1.root");

Once the tree has friends, we can use TTree::Draw as if the friend’s variables were in the original tree. To specify which tree to use in the Draw method, use the syntax:


If the variablename is enough to identify uniquely the variable, you can leave out the tree and/or branch name.

For example, these commands generate a 3-d scatter plot of variable “var” in the TTree tree versus variable v1 inTTree ft1versus variablev2in **TTree**ft2`.


The picture illustrates the access of the tree and its friends with a Draw command.

When AddFriend is called, the ROOT file is automatically opened and the friend tree (ft1) header is read into memory. The new friend (ft1) is added to the list of friends of tree. The number of entries in the friend must be equal or greater to the number of entries of the original tree. If the friend tree has fewer entries, a warning is given and the missing entries are not included in the histogram.

Use TTree::GetListOfFriends to retrieve the list of friends from a tree.

When the tree is written to file (TTree::Write), the friends list is saved with it. Moreover, when the tree is retrieved, the trees on the friends list are also retrieved and the friendship restored. When a tree is deleted, the elements of the friend list are also deleted. It is possible to declare a friend tree that has the same internal structure (same branches and leaves) as the original tree, and compare the same values by specifying the tree.


The example code is in $ROOTSYS/tutorials/tree/tree3.C. Here is the script:

void tree3w() {
   // Example of a Tree where branches are variable length arrays
   // A second Tree is created and filled in parallel.
   // Run this script with .x tree3.C
   // In the function treer, the first Tree is open.
   // The second Tree is declared friend of the first tree.
   // TTree::Draw is called with variables from both Trees.
   const Int_t kMaxTrack = 500;
   Int_t ntrack;
   Int_t stat[kMaxTrack];
   Int_t sign[kMaxTrack];
   Float_t px[kMaxTrack];
   Float_t py[kMaxTrack];
   Float_t pz[kMaxTrack];
   Float_t pt[kMaxTrack];
   Float_t zv[kMaxTrack];
   Float_t chi2[kMaxTrack];
   Double_t sumstat;

   // create the first root file with a tree
   TFile f("tree3.root","recreate");
   TTree *t3 = new TTree("t3","Reconst ntuple");

   // create the second root file with a different tree
   TFile fr("tree3f.root","recreate");
   TTree *t3f = new TTree("t3f","a friend Tree");

   // Fill the trees
   for (Int_t i=0;i<1000;i++) {
      Int_t nt = gRandom->Rndm()*(kMaxTrack-1);
      ntrack = nt;
      sumstat = 0;
      // set the values in each track
      for (Int_t n=0;n<nt;n++) {
         stat[n] = n%3;
         sign[n] = i%2;
         px[n]   = gRandom->Gaus(0,1);
         py[n]   = gRandom->Gaus(0,2);
         pz[n]   = gRandom->Gaus(10,5);
         zv[n]   = gRandom->Gaus(100,2);
         chi2[n] = gRandom->Gaus(0,.01);
         sumstat += chi2[n];
         pt[n]   = TMath::Sqrt(px[n]*px[n] + py[n]*py[n]);
   // Write the two files

// Function to read the two files and add the friend
void tree3r()         {
   TFile *f = new TFile("tree3.root");
   TTree *t3 = (TTree*)f->Get("t3");
   // Add the second tree to the first tree as a friend
   // Draw pz which is in the first tree and use pt
   // in the condition. pt is in the friend tree.

// This is executed when typing .x tree3.C
void tree3() {

12.17 Example 4: A Tree with an Event Class

This example is a simplified version of $ROOTSYS/test/MainEvent.cxx and where Event objects are saved in a tree. The full definition of Event is in $ROOTSYS/test/Event.h. To execute this macro, you will need the library $ROOTSYS/test/ If it does not exist you can build the test directory applications by following the instruction in the $ROOTSYS/test/README file.

In this example we will show

12.17.1 The Event Class

Event is a descendent of TObject. As such it inherits the data members of TObject and its methods such as Dump() and Inspect()andWrite(). In addition, because it inherits from TObject it can be a member of a collection. To summarize, the advantages of inheriting from a TObject are:

Below is the list of the Event data members. It contains a character array, several integers, a floating-point number, and an EventHeader object. The EventHeader class is described in the following paragraph. Event also has two pointers, one to a TClonesArray of tracks and one to a histogram. The string “->” in the comment field of the members *fTracks and *fH instructs the automatic Streamer to assume that the objects *fTracks and *fH are never null pointers and that fTracks->Streamer can be used instead of the more time consuming form R__b << fTracks.

class Event : public TObject {
      char                 fType[20];
      Int_t                fNtrack;
      Int_t                fNseg;
      Int_t                fNvertex;
      UInt_t               fFlag;
      Float_t              fTemperature;
      EventHeader          fEvtHdr;
      TClonesArray        *fTracks;            //->
      TH1F                *fH;                 //->
      Int_t                fMeasures[10];
      Float_t              fMatrix[4][4];
      Float_t             *fClosestDistance;   //[fNvertex]
      static TClonesArray *fgTracks;
      static TH1F         *fgHist;
      // ... list of methods
      ClassDef(Event,1)  //Event structure

12.17.2 The EventHeader Class

The EventHeader class (also defined in Event.h) does not inherit from TObject. Beginning with ROOT 3.0, an object can be placed on a branch even though it does not inherit from TObject. In previous releases branches were restricted to objects inheriting from the TObject. However, it has always been possible to write a class not inheriting from TObject to a tree by encapsulating it in a TObject descending class as is the case in EventHeader and Event.

class EventHeader {
      Int_t   fEvtNum;
      Int_t   fRun;
      Int_t   fDate;
      // ... list of methods
      ClassDef(EventHeader,1)      //Event Header

12.17.3 The Track Class

The Track class descends from TObject since tracks are in a TClonesArray (i.e. a ROOT collection class) and contains a selection of basic types and an array of vertices. Its TObject inheritance enables Track to be in a collection and in Event is a TClonesArray of Tracks.

class Track : public TObject {
      Float_t   fPx;         //X component of the momentum
      Float_t   fPy;         //Y component of the momentum
      Float_t   fPz;         //Z component of the momentum
      Float_t   fRandom;     //A random track quantity
      Float_t   fMass2;      //The mass square of this particle
      Float_t   fBx;         //X intercept at the vertex
      Float_t   fBy;         //Y intercept at the vertex
      Float_t   fMeanCharge; //Mean charge deposition of all hits
      Float_t   fXfirst;     //X coordinate of the first point
      Float_t   fXlast;      //X coordinate of the last point
      Float_t   fYfirst;     //Y coordinate of the first point
      Float_t   fYlast;      //Y coordinate of the last point
      Float_t   fZfirst;     //Z coordinate of the first point
      Float_t   fZlast;      //Z coordinate of the last point
      Float_t   fCharge;     //Charge of this track
      Float_t   fVertex[3];  //Track vertex position
      Int_t     fNpoint;     //Number of points for this track
      Short_t   fValid;      //Validity criterion

      // method definitions ...
      ClassDef(Track,1)          //A track segment

12.17.4 Writing the Tree

We create a simple tree with two branches both holding Event objects. One is split and the other is not. We also create a pointer to an Event object (event).

void tree4w() {
   // check to see if the event class is in the dictionary
   // if it is not load the definition in
   if (!TClassTable::GetDict("Event")) {
   // create a Tree file tree4.root
   TFile f("tree4.root","RECREATE");
   // create a ROOT Tree
   TTree t4("t4","A Tree with Events");
   // create a pointer to an Event object
   Event *event = new Event();
   // create two branches, split one
   t4.Branch("event_branch", "Event", &event,16000,2);
   t4.Branch("event_not_split", "Event", &event,16000,0);

   // a local variable for the event type
   char etype[20];

   // fill the tree
   for (Int_t ev = 0; ev <100; ev++) {
      Float_t sigmat, sigmas;
      Int_t ntrack   = Int_t(600 + 600 *sigmat/120.);
      Float_t random = gRandom->Rndm(1);
      event->SetHeader(ev, 200, 960312, random);
      for(UChar_t m = 0; m < 10; m++) {
         event->SetMeasure(m, Int_t(gRandom->Gaus(m,m+1)));
                                                  // continued...
      // fill the matrix
      for(UChar_t i0 = 0; i0 < 4; i0++) {
         for(UChar_t i1 = 0; i1 < 4; i1++) {
      // create and fill the Track objects
      for (Int_t t = 0; t < ntrack; t++) event->AddTrack(random);
      t4.Fill();      // Fill the tree
      event->Clear(); // Clear before reloading event
   f.Write();            // Write the file header
   t4.Print();           // Print the tree contents

12.17.5 Reading the Tree

First, we check if the shared library with the class definitions is loaded. If not we load it. Then we read two branches, one for the number of tracks and one for the entire event. We check the number of tracks first, and if it meets our condition, we read the entire event. We show the fist entry that meets the condition.

void tree4r() {
   // check if the event class is in the dictionary
   // if it is not load the definition in
   if (!TClassTable::GetDict("Event")) {
   // read the tree generated with tree4w

   // note that we use "new" to create the TFile and TTree objects, because we
   // want to keep these objects alive when we leave this function.
   TFile *f = new TFile("tree4.root");
   TTree *t4 = (TTree*)f->Get("t4");

   // create a pointer to an event object for reading the branch values.
   Event *event = new Event();
   // get two branches and set the branch address
   TBranch *bntrack = t4->GetBranch("fNtrack");
   TBranch *branch  = t4->GetBranch("event_split");

   Int_t nevent = t4->GetEntries();
   Int_t nselected = 0;
   Int_t nb = 0;
   for (Int_t i=0; i<nevent; i++) {
      //read branch "fNtrack"only

      // reject events with more than 587 tracks
      if (event->GetNtrack() > 587)continue;

      // read complete accepted event in memory
      nb += t4->GetEntry(i);

     // print the first accepted event
     if (nselected == 1) t4->Show();
     // clear tracks array

   if (gROOT->IsBatch()) return;
   new TBrowser();

Now, let’s see how the tree looks like in the tree viewer.

The tree viewer with tree4 example

The tree viewer with tree4 example

You can see the two branches in the tree in the left panel: the event branch is split and hence expands when clicked on. The other branch event not split is not expandable and we can not browse the data members.

The TClonesArray of tracks fTracks is also split because we set the split level to 2. The output on the command line is the result of tree4->Show(). It shows the first entry with more than 587 tracks:

======> EVENT:26
 event_split     =
 fUniqueID       = 0
 fBits           = 50331648
 fType[20]       = 116 121 112 101 49 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 fNtrack         = 585
 fNseg           = 5834
 fNvertex        = 17
 fFlag           = 0
 fTemperature    = 20.044315
 fEvtHdr.fEvtNum = 26
 fEvtHdr.fRun    = 200
 fEvtHdr.fDate   = 960312
 fTracks         = 585
 fTracks.fUniqueID = 0, 0, 0, 0, 0, 0, 0, 0, 0, 0

12.18 Example 5: Import an ASCII File into a TTree

The method TTree::ReadFile can be used to automatic define the structure of the TTree and read the data from a formatted ascii file.