Strange segfault

ROOT
1

Here’s a basic piece of code:

{
  std::vector<TH1F*> e;

  for(int i=0;i<24;i++) { 
    e.push_back(new TH1F(Form("e%d",i),"",100,0.9,10)); 
    std::cout << e.at(i)->Integral() << std::endl; 
  }
  
}

When I run it, it gives me a seg fault. I am running root version 5.32/00 . Any clues as to what’s going on?

2

Update: checked under 5.34/01, same issue. What am I not getting?

3

[code]{
gSystem->Exec(“rm -f AutoDictvectorTH1F*”);
gInterpreter->GenerateDictionary(“vector<TH1F*>”, “TH1F.h;vector”);

std::vector<TH1F*> e;

for(int i = 0; i < 24; i++) {
TH1F *h = new TH1F(TString::Format(“e%d”,i), “”, 100, 0.9, 10);
e.push_back(h); // CINT limitation: no “new TH1F” allowed here
std::cout << e.at(i)->Integral() << std::endl;
}
}[/code]

4

Hi Willie,

so it looks like CINT cannot handle simultaneous construction of the TH1Fs and
vector::push_back. I tried another thing which was to create the vector with
the right size and just place the pointers at the correct index which also
worked.

Is there a pattern one should generally avoid here or is the answer just
"compile your code" since this should work?

Cheers,

b.

5 

It looks like CINT  can't handle simultaneous new's and
push backs, as honk already mentioned.  And a software engineer colleague of mine is
telling me that [b]in general[/b] embedding a construction in a push back is known to be dangerous.

Anyway, here's the modification that worked.

[code]
{
  std::vector<TH1F*> e;

  for(int i=0;i<24;i++) { 
    TH1F* tmp =new TH1F(Form("e%d",i),"",100,0.9,10); 
    e.push_back(tmp);

    std::cout << e.at(i)->Integral() << std::endl; 

  }
}
[/code]

It looks like CINT can’t handle simultaneous new’s and
push backs, as honk already mentioned. And a software engineer colleague of mine is
telling me that in general embedding a construction in a push back is known to be dangerous.

Anyway, here’s the modification that worked.

{
  std::vector<TH1F*> e;

  for(int i=0;i<24;i++) { 
    TH1F* tmp =new TH1F(Form("e%d",i),"",100,0.9,10); 
    e.push_back(tmp);

    std::cout << e.at(i)->Integral() << std::endl; 

  }
}
6

I cannot see how that could that be dangerous (even in non-general cases). All
arguments to vector::push_back will be evaluated before it is called, so even
if new threw an exception the vector would still be in a consistent state. The
story might be different if one needs to work with a broken std library
implementation or a broken compiler, but that should hardly be the case for
std::vector (even less so in general).

My take would have been that variables should in general be declared in
the smallest possible scope – and allocating the histogram in the
vector::push_back does that.

7

So you think this is just a CINT deficiency?