I came across a problem of wanting to get asymmetric error bars on an efficiency histogram that is built with BayesDivide(). My input histograms have non-integer bin contents though, and that is not handled properly: the contents are rounded to integers. Usually this goes unnoticed, unless the bin contents are really small, as in my case.
I adapted TGraphAsymmErrors.cxx and TGraphAsymmErrors.h to work with floats. I tested standalone, not within root, but the changes are trivial and should be easy to merge. I attach the diff’s.
Could this be included?