Can we cache pyc files

[paddymul]

Summary

uv has a --compile-bytecode option which precompiles the bytecode for packages across an entire venv. This results in slightly faster python startup time, but much slower package install time. compiling bytecode is deterministic depending on python version and python source file content. This should be eminently cacheable.

Example

This will be useful for local development where I install the same packages across many different venvs. I currently don't use --compile-bytecode because it's slow, but I would use it if the results were cached.

This also helps in github CI (I think). The astral-sh/setup-uv@v7 job caches the uv cache, not installed venvs. so you could save the --compile-bytecode cost on subsequent runs.

This wouldn't help docker builds, where you can save --compile-bytecode to its own layer. You get the same venv every time.

[konstin]

Does the just-in-time compilation for python source files cause noticeable problems for you? It's intentional that Python only bytecode compiles files that are imported, when they are imported. Bytecode compilation only matters for short running processes, such as CLIs, that aren't started repeatedly or run from a read only device (such as a frozen container).

[paddymul]

So I think in my local dev workflow, pyc is responsible for about 1.5 seconds of lag on a new env off the top of my head. I will run benchmarks for you. This pales in comparison to OS X gatekeeper slowdown of 10-17 seconds #17371 , but its still there.

It also shows up in CI, not a huge factor, but it does compound.

It just bugs me that the slowdown is there at all, maybe I should just let it go, but I think there is improvement to be had here.