Really strange fit

Dear experts,

I'm experiencing a really strange behaviour while trying to perform something fairly simple.

I have a histogram, which needs to be fitted with a gaussian. I initialize the parameters of the gaussian and then fit, using a script.

The fit fails if the "B" option is specified. It also fails at the first invokation if no options were specified, and succeeds on the second attempt.

Now my problem is that I need to fit using the FitSlices* methods, and here the fit fails miserably in bins which are very happily fitted if ran manually. In this methods I don't have much control of the internals, namely I can't force it to perform the fit twice in each bin. Moreover, I really need to restrict some of the fit parameters, and that causes really bogus results.

Thanks for any help!

        Eduard

P.S. The root file with the histogram can be found at http://www.desy.de/~dich/misc/strangehisto.root and the script at
http://www.desy.de/~dich/misc/strangefit.C ROOT v.5.26
Calling as root -l strangefit.C
The resulting pads:

1(top): fit with "B" option
2(middle): single fit with no options
3(bottom): double fit with no options