Chapter: Input Output

WARNING: This documentation is not maintained anymore. Some part might be obsolete or wrong, some part might be missing but still some valuable information can be found there. Instead please refer to the ROOT Reference Guide and the ROOT Manual. If you think some information should be imported in the ROOT Reference Guide or in the ROOT Manual, please post your request to the ROOT Forum or via a Github Issue.

13 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.

13.1 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.

13.1.1 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.

Byte Value Name Description
1 -> 4 root Root file identifier
5 -> 8 fVersion File format version
9 -> 12 fBEGIN Pointer to first data record
13 -> 16 [13->20] fEND Pointer to first free word at the EOF
17 -> 20 [21->28] fSeekFree Pointer to FREE data record
21 -> 24 [29->32] fNbytesFree Number of bytes in FREE data record
25 -> 28 [33->36] nfree Number of free data records
29 -> 32 [37->40] fNbytesName Number of bytes in TNamed at creation time
33 -> 33 [41->41] fUnits Number of bytes for file pointers
34 -> 37 [42->45] fCompress Zip compression level
38 -> 41 [46->53] fSeekInfo Pointer to TStreamerInfo record
42 -> 45 [54->57] fNBytesInfo Number of bytes in TStreamerInfo record
46 -> 63 [58->75] fCompress 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.

13.1.2 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

13.1.3 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):

Byte Value Name Description
1 -> 4 Nbytes Length of compressed object (in bytes)
5 -> 6 Version TKey version identifier
7 -> 10 ObjLen Length of uncompressed object
11 -> 14 Datime Date and time when object was written to file
15 -> 16 KeyLen Length of the key structure (in bytes)
17 -> 18 Cycle Cycle of key
19 -> 22 [19->26] SeekKey Pointer to record itself (consistency check)
23 -> 26 [27->34] SeekPdir Pointer to directory header
27 -> 27 [35->35] lname 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 | Data bytes associated to the object

You see a reference to TKey. It is explained in detail in the next section.

13.1.4 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.

13.1.5 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.

13.1.6 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

13.2 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

13.2.1 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.

13.2.2 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. Cling 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()

13.2.3 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. ROOT 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

13.2.4 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.

13.2.5 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.

13.2.6 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

13.2.7 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);

13.2.8 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");

13.2.9 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, Cling 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()

13.2.10 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:

13.3 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 Cling 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.

13.3.1 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, rootcling 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 dictionary by the rootcling utility. This is the rootcling statement in the $ROOTSYS/test/Makefile:

@rootcling -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.

13.3.2 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

13.3.3 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;                 //->

13.3.4 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]

13.3.5 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 integer. The actual size of disk (before compression) is determined by the parameter next to the data member declaration. For example:

Double32_t m_data;     //[min,max<,nbits>]

If the comment is absent or does not contain min, max, nbits, the member is saved as a float.

If min and max are present, they are saved with 32 bits precision. min and max can be explicit values or expressions of values known to Cling (e.g. pi).

If nbits is present, the member is saved as int with nbits bits precision. For more details see the io tutorials double32.C.

Compression and precision of Double32_t
Compression and precision of Double32_t

13.3.6 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

13.3.7 Streamers with Special Additions

Most of the time you can let rootcling 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. Note that an alternative for those cases is to consider write a schema evolution rule.

First, you need to tell rootcling not to build a Streamerfor you. The input to the rootcling 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 rootcling not to generate a Streamer. In the example, you can see the Event class is the only one for which rootcling is instructed not to generate a Streamer.

#ifdef __ROOTCLING__

#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 rootcling 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);

13.3.8 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.

13.3.9 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.

13.3.10 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.

13.4 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++.

13.4.1 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.

13.4.2 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).

13.4.3 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.

13.4.4 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 accessible from TProcessID::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 a fUniqueID, a pointer to this unique object may be found using fObjects->At(fUniqueID). In the same way, when a TRef::GetObject is called, GetObject uses its own fUniqueID to find the pointer to the referenced object. See TProcessID::GetObjectWithID and PutObjectWithID. 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();

13.4.5 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 Cling 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 Cling script GetWebHistogram.C.

13.4.6 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.

13.5 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.

13.5.1 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

13.5.2 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.

13.5.3 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

13.5.4 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.

13.5.5 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.

13.5.6 Manual Data Model Evolution Capabilities

The automatic data model schema evolution implemented in ROOT makes it possible to read back the serialized data object in the situation when the definition of the classes those objects represent changed slightly (some of the data members were removed or some new ones added). It is also possible to manually specify the rules for more sophisticated data transformations done while reading to load the serialized objects into data structures that changed quite significantly.

ROOT provides two interface enabling users to specify the conversion rules. The first way is to define a rule in the dictionary file and the second way is to insert it to the TClass object using the C++ API.

There are two types of conversion rules. The first of them, the normal rules, are the ones that should be used in the most of the cases. They provide a buffered input data and an address of the in-memory target object and allow user to specify the conversion function mapping the data being read to the output format. The second type of the rules, the raw rules, also provide the pointer to the target object but the input is a raw TBuffer object containing the input data member declared as an input to the rule. This type of a rule is provided mainly to handle the file format changes that couldn’t have been handled otherwise and in general should not be used unless there is no other option. The dictionaries

The most convenient place to specify the conversion rules is a dictionary. One can do that either in a LinkDef file or in the selection xml file being fed to genreflex. The syntax of the rules is the following:

#pragma read                                              \
    sourceClass="ClassA"                                  \
    source="double m_a; double m_b; double m_c"           \
    version="[4-5,7,9,12-]"                               \
    checksum="[12345,123456]"                             \
    targetClass="ClassB"                                  \
    target="m_x"                                          \
    embed="true"                                          \
    include="iostream,cstdlib"                            \
    code="{m_x = onfile.m_a * onfile.m_b * onfile.m_c; }" \

#pragma readraw           \
      sourceClass="TAxis" \
      source="fXbins"     \
      targetClass="TAxis" \
      target="fXbins"     \
      version="[-5]"      \
      include="TAxis.h"   \
Float_t * xbins=0; \
Int_t n = buffer.ReadArray( xbins ); \
fXbins.Set( xbins ); \
<ioread sourceClass="ClassA"
        source="double m_a; double m_b; double m_c"
   m_x = onfile.m_a * onfile.m_b * onfile.m_c;
]] >

<ioreadraw sourceClass="TAxis"
      Float_t *xbins = 0;
      Int_t n = buffer.ReadArray( xbins ) ;
      fXbins.Set( xbins );
]] >

The variables in the rules have the following meaning:

The user can assume that in the provided code snippet the following variables will be defined:

The user provided code snippets have to consist of valid C++ code. The system can do

some preprocessing before wrapping the code into function calls and declare some variables to facilitate the rule definitions. The user can expect the following variables being predeclared: The C++ API

The schema evolution C++ API consists of two classes: ROOT::TSchemaRuleSet and ROOT::TSchemaRule. Objects of the TSchemaRule class represent the rules and their fields have exactly the same meaning as the ones of rules specified in the dictionaries. TSchemaRuleSet objects manage the sets of rules and ensure their consistency. There can be no conflicting rules in the rule sets. The rule sets are owned by the TClass objects corresponding to the target classes defined in the rules and can be accessed using TClass::{Get|Adopt}SchemaRules.

13.5.7 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 rootcling generated Streamers whenever you can, and profit from the automatic schema evolution.

13.5.8 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.

13.5.9 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 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 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) //

13.6 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 Bytes Write Time (sec) Read Time (sec.)
0 1,004,998 4.77 0.07
1 438,366 6.67 0.05
5 429,871 7.03 0.06
9 426,899 8.47 0.05

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.

13.6.1 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 or one of its derivations, for example a TXNetFile or a TDavixFile object. In case a 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:

13.7 Remotely Access to ROOT Files

ROOT files can be accessed remotely in many ways, on the base of the protocol URL. Among the most popular are XRootD (protocols ‘root://’ and ‘xrd://’) and a Web server (protocl ‘http://’ or ‘https://’).

The rootd daemon is deprecated and has been removed in version 6.16/00.

Please refer to the XRootD documentation for starting and ensuring that such a daemon is running.

Reading and writing ROOT files over the net can be done by creating a TFile object using the static method TFile::Open() object. This will instantiate the appropriate derivation of TFile to handle the request. Inheriting from the TFile class, the returned instance will have exactly the same interface and behavior of TFile. The only difference is that it reads and writes from a remote service. In the example below the file is read via a web server through the TDavixFile plug-in.

13.7.1 A Simple Session

root[] TFile *f1 = TFile::Open("local/file.root","update")
root[] TFile *f2 = TFile::Open("root://","new")
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()

13.8 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/Cling 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: