Tracking issue for better testing for dist artifacts

[jieyouxu]

This is a tracking issue for improving the testing for dist artifacts (including but not limited to, dist compilers and dist components like rustc-src).

As of the time of writing, much of dist components are not really tested or exercised in a basic way (modulo tests through opt-dist, but not all dist artifacts go through opt-dist).

Example regressions

• rustc.exe -vV didn't exit successfully (exit code: 0xc000007b) on i686-pc-windows-gnu #136795

[jieyouxu]

I know that x86_64-unknown-linux-gnu and aarch64-unknown-linux-gnu dist artifacts do receive some degree of testing due to post-PGO opt-dist tests.

It's the apple and windows dist artifacts that I don't know if are tested at all.

[ChrisDenton]

Yes I think not testing CI dist artefacts is the default rather than the exception.

[Noratrieb]

If we call the issue like that, no one will bother to ever fix it of course. But this does actually affect more than just i686-pc-windows-gnu so someone might actually care.

[jieyouxu]

Zulip discussion: https://rust-lang.zulipchat.com/#narrow/channel/242791-t-infra/topic/Testing.20dist.20artifacts.3F

[Noratrieb]

A bootstrap PR broke tools completely after dist, #137469. Smoke tests would have prevented that as well.

[jieyouxu]

See also #138306. That is a case where the rustc-src dist component broke because we tried to use workspace lint table of the rust-lang/rust root manifest, which is not shipped as part of this dist component, so compiler sources no longer build on their own. I didn't catch this nor did the PR author because we would've needed to inspect the rustc-src dist component...