Proposed patch speeds up parsing positional arguments in PyArg_ParseTupleAndKeywords(). $ ./python -m perf timeit "1 .to_bytes(1, 'little', signed=False)" Unpatched: Median +- std dev: 2.01 us +- 0.09 us Patched: Median +- std dev: 1.23 us +- 0.03 us Currently names of all passed positional argument are looked up in the kwargs dictionary for checking if the argument is passed as positional and keyword argument. With the patch this is checked only if there are unhandled keyword arguments after parsing keyword arguments that don't passed as positional arguments. The same optimization also is applied to _PyArg_ParseTupleAndKeywordsFast() and like, but the effect is much smaller. The patch also includes tiny refactoring.
Oh, I missed this issue. Since I made minor cleanup recently, I took the liberty of rebasing your patch. I also pushed your "keywords => kwargs" change, just to make the patch simpler to review.
> Unpatched: Median +- std dev: 2.01 us +- 0.09 us > Patched: Median +- std dev: 1.23 us +- 0.03 us Oh wow, impressive speedup! As usual, good job Serhiy ;-) PyArg_ParseTupleAndKeywords-faster-positional-args-parse-2.patch: LGTM! Can you please push it?
Thank you for your reviews Inada and Victor. Thank you for rebasing the patch Victor. I were going first to try an alternative patch, maybe less efficient, but allowing to share more code. But since the code is changed so fast, it is better to commit the ready patch.
