@mikeb01 nice job! I hope you will add support for packaging (rpm/debian). Note that I now have my wrk2 fork working pretty well with your previous commit (no more local copy of the hdr_histogram.c code).
Would be great to have a small set of "standard" perf test to see how the various implementations compare, for example define a standard fill algorithm that reflects typical histograms and see how much time it is needed to encode and decode the histogram. I am doing performance measurement in the python version - more so on the decoding side and python loops on arrays are slow. The python decode and add function is too slow for my requirements and probably 1 to 2 orders of magnitude slower than the C version. I'm working on a way to speed this up without using any compiler and so far looking good with the use of numpy+ctypes (I want to keep it pure python so I'd like to avoid using Cython, Numba...).