Problem with float to int conversion in plotting poisson errors from TH1F

[timherr]

---

ROOT Version: 6.06

---

Hello,

I have a TH1F filled with integer numbers. With this two options I enforce Poisson errors in plotting:

        dataHist->SetBinErrorOption(TH1::kPoisson);
        dataHist->Sumw2(false);

Everything works fine except for one bin.
For one bin I set the bin content to 2.
If I call GetBinContent for that bin I get: 2
If I call GetBinErrorUp for that bin I get: 1.29953
If I call GetBinErrorLow for that bin I get: 1.82725
But it should be 2.63786 and 1.29181 respectively.
For a bin content of 1 it would be 2.29953 and 0.827246 respectively.

So, what causes the problem is the double to int conversion in the GetBinErrorLow/high function: https://root.cern.ch/doc/master/TH1_8cxx_source.html#l08413
In my case the 2 is internally a 1.9999999 and it gets rounded to 1.
If I fill the histogramm with 2.00001 everything is fine.

Is it a option to change the line 8413 and 8443 from

Int_t n = int(c);

to

Int_t n = int(c+0.5);
if (c<-0.5) n = int(c-0.5);

or would this cause other problems?

As temporary workaround I add 0.00001 to each bin before plotting. Is there a better solution to avoid this?

Thanks,
Tim

[couet]

Have you tried to use TH1D instead of TH1F ?

[timherr]

I just tried. It does not solve the problem. But I figured out, that I set the bin content not exactly to 2. It is already internally a 1.9999999. If I round the number before setting the bin content then everything works fine. This is what I will do now.
Thanks