[email protected]: do not symlink as pip3 and wheel3

[fxcoudert]

Fixes #62911

[SMillerDev]

I think this warrants a revbump

[fxcoudert]

Revision-bumping sadly will not remove the previously created symlinks, will it? (since they're created in postinstall)
But brew cleanup should do it, I think.

[SMillerDev]

I'm not sure about that, but at least it'll make sure everyone gets the fix. Which was what I was more worried about.

[iMichka]

Maybe if you revbump [email protected] to override these symlinks?

Just a note: last time we made a Python migration, we were bitten by a similar issue if I remember well. But it's hard to test upfront to avoid this mistake.

[SMillerDev]

Maybe we could make an audit/rubocop to make sure only one formula has these? @MikeMcQuaid probably has a better idea on how feasible that would be.

[MikeMcQuaid]

Maybe if you revbump [email protected] to override these symlinks?

Yes, this could work.

Maybe we could make an audit/rubocop to make sure only one formula has these? @MikeMcQuaid probably has a better idea on how feasible that would be.

Specifying "only one formula" probably isn't feasible. Having brew audit check "no python formula except PYTHON_ALLOWED_FORMULA has these symlinks" seems sensible.

Something like https://github.com/Homebrew/brew/blob/7f422af7897f4f58cc84079ccb6f1d25ac66a759/Library/Homebrew/dev-cmd/audit.rb#L985-L994 will allow checking this only if the formula is installed. I'd suggest this be done before this PR is merged.

[fxcoudert]

I'd suggest this be done before this PR is merged

The original reporter's use case (CI testing) would benefit from merging this as-is, and looking at systematic improvements later. Without revision bump, it will only affect new users and CI systems, which is still important.

[SMillerDev]

It would only affect those that build from source right? Or do you plan on overwriting the bottles?

[fxcoudert]

My plan was overwriting the bottles: isn't that what the merging bot does, when there is no revision?

[SMillerDev]

I don't think it does.

[fxcoudert]

@SMillerDev see for example this PR: #58881
It was auto-merged, does not have a revision bump, and lead to two commits: 4f17008 and 731f2b1

[BrewTestBot]

@fxcoudert has triggered a merge.

[MikeMcQuaid]

The original reporter's use case (CI testing) would benefit from merging this as-is, and looking at systematic improvements later. Without revision bump, it will only affect new users and CI systems, which is still important.

My issue is: here we are several days later and no-one has opened an issue or a PR to track this work. We've got to get better at adding regression tests like this or things get forgotten and will repeatedly reoccur in future. This work should block just getting a fix out there and is definitely more important than e.g. migrating more formulae.

[fxcoudert]

here we are several days later

Testing finished about 24 hours ago, and merging was yesterday.

no-one has opened an issue or a PR to track this work

I have not forgotten about it, even though I have not filed an issue, this tab is open in my browser because I know it's in my to-do list. All help is welcome, especially to write an audit, as I'm totally ignorant of the brew codebase.

Also, there already is an open issue, which was purposely not closed, at #62911

[fxcoudert]

And now there is a tracking issue for the brew side: Homebrew/brew#8949

[MikeMcQuaid]

@fxcoudert Thanks for opening the tracking issue 👍🏻