Re: really crazy crash

Note that the recipe indicated by Lorenzo works everywhere, including Windows.

Rene Brun

Tom Roberts wrote:
> Lorenzo Moneta wrote:
>> If you have a so long macro I recommend you first to compile it with
>> AClic,
>> do :
>> .L test5.C+
>> In this case the #includes must be there
>
> I distribute my macro to many users, most of whom are physicists, not
> computer experts, and many of whom use Windows. I cannot depend on a
> C++ compiler being installed on the user's system. So I don't want to
> do this unless absolutely required.
>
> Axel has indicated he found this particular problem and has fixed it,
> so I'll not change anything I'm doing, for now (though I will put in
> all the #include-s). I still need to address the actual problem in my
> big macro (related to TTreeFormula in some way I have not yet isolated
> into a short test).
>
>
> Tom Roberts
>
>
>> Lorenzo
>>
>> On Apr 22, 2008, at 11:19 PM, Tom Roberts wrote:
>>
>>> I am trying to debug a >2000 line Root macro by isolating the
>>> problem into a small macro (it works on 5.12-16 but crashes on
>>> 5.18). I keep getting frustrated by seemingly trivial crashes that
>>> do not seem to be the real problem. Here is another one that crashes
>>> on 5.16 and 5.18 (Linux, Fedora Core 8).
>>>
>>>
>>> I thought CINT permitted one to omit the #includes in a macro file.
>>> This seems to not be the case, except for the remarkable fact that
>>> it works, but crashes on ".q".
>>>
>>> Specifically: if I do:
>>> .L test5.C
>>> .q
>>> root crashes.
>>>
>>> If I do:
>>> .X test5.C
>>> .q
>>> it prints "Ntuple 'VirtualDetector/Det1' opened" (i.e. the macro
>>> worked correctly) and then crashes on the .q.
>>>
>>> Indeed, all that is needed is a declaration "TNtuple *ntuple" in the
>>> class to make this crash on .q.
>>>
>>> If I uncomment the #include, all is well. Note that no #include of
>>> TFile.h seems to be needed, and attempts to display this problem
>>> with TFile all failed.
>>>
>>> What is the difference between TFile and TNtuple?
>>> Do I really need to #include the class header for every Root class I
>>> use in a macro?
>>> How does the macro execute correctly but crash on .q?
>>>
>>> (my big macro does crash on .q, but it normally calls
>>> gApplication->Terminate(0) which does not crash, so this is not a
>>> problem to my users.)
>>>
>>>
>>> Tom Roberts
>>> // test5.C
>>>
>>> //#include <TNtuple.h>
>>>
>>> class PlotNtuple {
>>> TFile *file;
>>> TNtuple *ntuple;
>>> public:
>>> PlotNtuple(const char *filename, const char *ntuplename);
>>> void scan();
>>> };
>>>
>>> PlotNtuple::PlotNtuple(const char *filename, const char *ntuplename)
>>> {
>>> file = 0;
>>> ntuple = 0;
>>>
>>> file = new TFile(filename,"READ");
>>> if(file->IsZombie()) {
>>> printf("Cannot read '%s'\n",filename);
>>> file = 0;
>>> return;
>>> }
>>> ntuple = dynamic_cast<TNtuple*>(file->Get(ntuplename));
>>> if(!ntuple) {
>>> printf("Cannot find NTuple '%s'\n",ntuplename);
>>> return;
>>> }
>>> printf("Ntuple '%s' opened\n",ntuplename);
>>> }
>>>
>>> void PlotNtuple::scan()
>>> {
>>> }
>>>
>>>
>>> void test5()
>>> {
>>> PlotNtuple *pn=new
>>> PlotNtuple("g4beamline.root","VirtualDetector/Det1");
>>> pn->scan();
>>> }
>>> <g4beamline.root>
>>
>
Received on Wed Apr 23 2008 - 16:05:50 CEST