semanal: Check that async for/with is inside an async function

[shoracek]

Description

Added a check that async for/with is in an async function.
The code and error message are similar to those checking await.

Fixes #10594

Test Plan

I used an expanded example from the issue as a unit test. The example was extended to also cover for loop, with expression and usage outside of a function.

Also fixed another test case using async with expression outside of an async function.

[jhance]

We should probably have an error code. Also, the format string for the message should be in the message registry.

[JelleZijlstra]

This can probably just go into the misc code. It's a syntax error at runtime, so I don't see much reason for users to want to skip this error.

[jhance]

Is there a reason the ast module itself doesn't raise for syntax?

[JelleZijlstra]

Some SyntaxErrors are raised after parsing in the Python compiler, e.g. invalid nonlocal declarations, some things with async. So mypy has to replicate such checks in semanal.

[shoracek]

I have kept the error code as misc, made the messages use terminology that the Python's syntax error uses, and added the message to the registry. Although, now that I am looking closer at the possible errors, there's a syntax code, so it probably could have been better to use that.

[jhance]

Yeah, syntax seems better if it already exists.

[shoracek]

Used syntax as the error code and also set it to be a blocking error, since Python will not run code with this error.

[JelleZijlstra]

An async genexp is actually legal in a non-async function: def f(): return (x async for x in y). Could you add a test for that?

[JelleZijlstra]

Not sure we should use a blocking error. There's a risk that this check has a bug or that a future Python version changes the rules, and in that case it would be useful for users to be able to skip this error. I don't think we should be using blocking errors for anything we can recover from.

[shoracek]

As suggested, I've switched it back to non-blocking error, removed the error for generator expression, and added the remaining expressions into the test.

[jhance]

It looks good. Can we change True in ... to any(genr.is_async) for all of the cases where we have something like that? Afterwards I think we can merge this. IMO this is less odd and easy to read.

[shoracek]

Done.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉