fix PYO3_CONFIG_FILE env var not requesting rebuilds

[davidhewitt]

Candidate fix for #4757

I will verify this downstream and also add tests as soon as I'm satisfied this is the likely solution.

[davidhewitt]

Both pydantic/jiter#180 and pydantic/pydantic-core@7b01901 which test out #4758 are working as they should with this fix in place. So I am satisfied that this is the fix and we should get this merged asap.

[davidhewitt]

I pushed a crude test which should make it harder to make this mistake again. It would be nice to follow up with a more comprehensive way to test all cases of pyo3-build-config reading these env vars.

For now, is anyone available to review this? Otherwise I will merge in a few hours and start preparing the patch release.

[alexdotgov2]

(Hi, this is Alex's work github)

The fix looks prima facie reasonable to me, but I sadly don't have the time to deeply review.

I'm not a huge fan of the cfg() for tracking the env vars, but I'm not sure I have a better suggestion.

[davidhewitt]

Thanks, I will proceed to merge and get prepping a release.

I agree the cfg stuff is pretty nasty. I think what might be helpful is to refactor this stuff to have a ctx: &BuildScriptContext external argument which can then record all environment variable accesses (and then emit them on build scripts). But that would break a lot of pyo3-build-config APIs so would be a cleanup deferred until 0.24.

[alexdotgov2]

We should also use clippy to prohibit direct env var access and force people to go through the right internal functions -- rust-openssl/rust-openssl@e5bd08f is an example of this pattern.

[davidhewitt]

Agreed, while it wouldn't have caught this particular error, in combination with a better testing strategy it might prevent future mistakes. Opened #4761 to track the follow-ups.