Re: working around bugs and limitations?

Thank you. The documentation for TTreeFormula is incomplete and quite complicated, but following TTreePlayer::Scan I was able to get it to work. Most of the complexity is due to the need to handle an arbitrary tree structure, rather than my much simpler TNtuple.

TTreeFormula is vastly more robust against invalid expressions than is CINT, and I have been unable to crash it in several dozen attempts. Now I don't need to call CINT at all, completely avoiding the problem. Thanks!

Tom Roberts

Philippe Canal wrote:
> Hi Tom,
>
> The class you are looking for is TTreeFormula.
> You can see a simple-ish and compact use case
> in the implementation of TTreePlayer::Scan.
> (TTreeViewer also indirectly uses TTreeFormula,
> via calls to TTree::Draw, which is implemented
> in the class TSelectorDraw).
>
> Cheers,
> Philippe.
>
> -----Original Message-----
> From: owner-roottalk_at_pcroot.cern.ch [mailto:owner-roottalk_at_pcroot.cern.ch]
> On Behalf Of Tom Roberts
> Sent: Thursday, February 15, 2007 3:06 PM
> To: 'ROOT Talk'
> Subject: [ROOT] working around bugs and limitations?
>
> I am trying to present my users with a stable and user-friendly
> application based on Root. This is proving to be unexpectedly difficult.
>
> I have found it is not possible to compile a function containing
> expressions entered by the user, as CINT simply does not play nice with
> others, and if an expression is invalid it can cause the entire Root
> instance to cease working (e.g. error message "cannot create file" in
> code without any file creation, or lots of "busy" functions).
>
> [Is it possible to add a function with the functionality of
> ProcessLineSync(), but is guaranteed to return to the
> caller, even if the string contains a .L command and the
> file has compilation errors? Such errors must NOT cause
> side effects like making the program cease to work. I need
> it callable from a macro, so cannot use longjmp().]
>
>
> As a workaround, I would like to use TFormula. Unfortunately the
> documentation is both sparse and obscure. Here's what I want to do:
> Given a TNtuple, I want to read its events one at a time, and for each
> event evaluate several user-entered expressions that use the names of
> its branches (plus arithmetic operators and the usual math functions).
> Those expression values are used to apply cuts and generate a histogram.
> [Yes, TTreeViewer can do something similar, but its code is obscure and
> its user interface is usable only by experts.]
>
> So some questions about TFormula:
> There are fNpar parameters, initially zero with a max of 1000.
> They can be given names, can the expression refer to them by name?
> Can I use the first ~30 parameters, give them the names of the
> branches, and set their values when I read each event?
> How can I set fNpar to do that?
> TFormula changes fNpar internally, in obscure ways -- what is that?
> Some of those changes are linked to magic values of the expression
> number (100,200,...) -- what is that?
> It looks like additional tokens can be added -- how might I add
> a token with the name of a branch that evaluates to the value
> of a specific float variable?
> I would play with this, but I can find no obvious way to set fNpar so
> that I can start....
>
> The code comments refer to a fictional class TNtupleFormula -- this
> looks like it might be precisely what I want. What happened to it? Why
> isn't there a class that simply evaluates formulas within a TNtuple (the
> code I see is obscure and complicated)?
>
>
> Tom Roberts
>
>
Received on Sat Feb 17 2007 - 03:38:16 CET