Enable TypeAlias check by default

[JelleZijlstra]

And delete the machinery for disabling errors.

This made this check run on all tests, which helped me find a bug
(it triggered on all). I can split the test changes into a separate
PR if preferred.

I'll submit a separate PR to typeshed to disable this check for now.

Fixes #75. Fixes #86.

[JelleZijlstra]

The typeshed-primer failures are all Y026. We'll have to merge python/typeshed#6958 first for CI to become green.

[AlexWaygood]

Looks good, but I think we should maybe advertise this change more loudly, either in the README or the changelog (and maybe point to instructions elsewhere about how to disable certain codes in your flake8 config file?). If third-party users of this package are anything like typeshed, this check is going to cause a large number of hits.

[hauntsaninja]

Yeah, I have a tendency to think of typeshed as the main consumer of flake8-pyi, but that's not really the case.

It might be nice to collect the codemods we've applied to typeshed and link them in our CHANGELOG. It could also be nice to update stubgen output to be compliant with all these new lints.

[AlexWaygood]

Yeah, I have a tendency to think of typeshed as the main consumer of flake8-pyi, but that's not really the case.

It might be nice to collect the codemods we've applied to typeshed and link them in our CHANGELOG. It could also be nice to update stubgen output to be compliant with all these new lints.

Linking to the PRs in the CHANGELOG sounds like a great idea.

We could maybe also have somewhere to collect some of the scripts we've used for codemodding — either a new repo, or a subdirectory in this repo. WDYT?

[hauntsaninja]

A scripts directory here seems like a decent place, at the very least we're all here now!

@hoefling shared an excellent libcst script in python/typeshed#5564 (comment) that we could add on to.

[Akuli]

I don't like the idea of collecting everything into a scripts directory:

• We don't want to write tests for the scripts, or maintain them in general.
• We needed to write scripts for typeshed, because there's so many lines of stubs in typeshed. Most other stubs projects don't have anywhere near as much code, and should be fine with manually editing the lines that this plugin points out.
• The scripts assume the code is blacked, have a few weird corner cases, and aren't exactly clean code. (This applies mostly to my scripts, though.)

I think it would be better to instead improve error messages to say what you should do instead. Some of our error messages already work that way:

Y019 Use "_typeshed.Self" instead of "_S", e.g. "def bad_instance_method(self: Self, arg: bytes) -> Self: ..."

This makes writing a script easy enough to just do yourself if you need one. We could even include one in this repo, and if it's simple enough, it shouldn't be too much work to test and maintain.

Improving stubgen output would also be nice :)

[AlexWaygood]

I wasn't envisioning writing tests for the scripts. We could have a README file in the subdirectory that has a big "USE WITH CAUTION — UNTESTED" notice in it.

I also didn't mean to suggest that we needed to include every codemodding script we'd ever written in the subdirectory. It could be a selection.

[JelleZijlstra]

Looks good, but I think we should maybe advertise this change more loudly

I added a section listing codes users may not want to enable. Your Y027 can also get a mention there.

I'd be OK with adding cleanup scripts to this repo, but let's do that in a separate PR.

[AlexWaygood]

I'd be OK with adding cleanup scripts to this repo, but let's do that in a separate PR.

agreed.

[AlexWaygood]

LGTM