mypyc for the stdlib

[RonnyPfannschmidt]

with the classical approach to speedups in the stdlib being porting to c,
i was wondering how well it might work to use tools like mypyc to turn the python into c

possible candidates for that are at least logging, pathlib, warnings, pickle, single_dispatch

naive enthusiasm projects potentially big wins, but compatibility may be tricky

[da-woods]

Pickle has a C version already so presumably there isn't a huge benefit there.

Experience from Cython suggests maybe a 20% speed-up if you just compile Python code directly. I image mypyc would be a similar order of magnitude.

I think the idea of compiling bits of the standard library with Cython has been discussed and rejected before (mainly because it'd add a third party dependency). I don't know if mypyc is a little more "affiliated" and so would be more acceptable. Especially if the code didn't need to be adapted so could run without mypyc

[RonnyPfannschmidt]

One idea here is that mypyc would allow the stdlib to be more python + still get the gains

The other part is, that if mypyc eventually gains a hpy backend, there would be a implicit support for stdlib speedup of other python implementations

[hauntsaninja]

Compatibility is going to be a big issue if using mypyc.

Fwiw, using mypyc on black is about a 2x speed up (without too much optimisation). mypyc on mypy is a 3.5-4x speed up.