@giltene I just committed a version that decodes properly the jHiccup log you posted. I had to add support for 16-bit counters to make it pass. The python version now supports 16, 32 and 64 bit counters.
The main features missing: auto-resize, double histograms, recorder.
Since I do not need them I'll let them for later as I need to update the wrk2 code now (for python side, I am only using the histogram decoding (64-bit) and aggregation.
I did some quick benchmark and decoding + add of a typical encoded histogram (2 digits precision or about 2KB per encoding) takes about 1msec with the python code on my macbook, which is not exactly great. Hopefully it should run much faster on the target environment (cloud servers). Most of the time is spent in the loop that adds counters one by one, which is pretty expensive in pure python. I suppose the Java version would be somewhat faster thanks to JIT.