JamesTheAwesomeDude
commented
5 months ago Also, this will require adding some logic to choose the best implementation.
I think import-time is the way to go for this, though there seem to be a few ways to approach it:
- Benchmark all the implementations when the module is imported, like [can't remember which package, but it's a popularish one] does
- Depend on a 3rd-party Python library such as
py-cpuinfo - Roll our own CPU-detection code
Stealvendor the code from an existing 3rd-party Python library
To avoid the pain trilemma of "reinvent a wheel, poorly; or torture users with Hellfire by pinning a dependency version; or risk breakage by using a non-pinned dependency", I think that vendoring might be a good choice.
Fix whatever cursed compile errors are blocking that.