Add support for per-module enabling/disabling of error codes

[hauntsaninja]

Feature

#9172 implemented global error code disabling/enabling as discussed in #8975. It seems like it was always the intention to consider per-module configuration (#8975 (comment)), but we don't currently have an open issue for it.

Pitch

My personal use case is for tests. Typing tests is often just a chore, but it's nice to have whatever type safety you can get for free. I'd want to be able to do:

[mypy]
strict = True

[mypy-tests.*]
allow_untyped_defs = True
allow_untyped_calls = True
disable_error_codes = ["var-annotated"]

(For this use case, I think there's actually a fair argument to be made that we should suppress var-annotated aka "needs type annotation" errors when checking untyped defs. If this strikes you as reasonable, I can open another issue. Edit: see #8558 for this)

[gvanrossum]

Seems reasonable.

You have a typo, the option is actually called disable_error_code. This triggered surprise in me, because at first I thought you were talking about show_error_codes. Is it too late to rename that flag to ignore_error_code? I guess a problem with that is that the inverse would be dont-ignore-error-code, which is kind of awkward.

[mcous]

Second this request. For me, I'd really like to disable func-returns-value in my tests but keep it active in source

[JukkaL]

This sounds like a useful feature. Another potential use case would be migrating legacy code -- you might initially ignore some error codes in code that wasn't originally written with type checking in mind, but you'd want to enable all error codes in new files.

[chadrik]

Another potential use case would be migrating legacy code

Exactly this. I wrote a tool called mypy-runner for filtering errors using error codes (first using my own error codes, then the official ones once that feature arrived) as well regexes. I'd love to not have to use my tool at all! One of the big downsides of the runner-based approach is that there's a major performance penalty: dmypy rechecks files with errors (regardless of modified timestamps), and from mypy's perspective all those files still have errors, so reruns of dmypy on our very large codebase are slow.

[nipunn1313]

Just ran into a use case for this! I'm migrating our (large) codebase to mypy-protobuf 2.0 - and it requires some backward incompatible changes to enums in pyi stubs. Ideally, I'm able to do this in many smaller diffs.

However, if a migrated module tries to import from a non-migrated module, it hits a name-defined error.

Ideally I can disable that error specifically - and reenable it later when the non-migrated module gets migrated.

Will be using ignore_errors for now - but it seems like an excessive hammer

[levsa]

This would be useful to turn off override errors when using fixtures in tests, like:

class A:
     @pytest.fixture
     def b():
        pass

     @pytest.fixture
     def a(b):
          pass


class B(A):
     @pytest.fixture
     def c():
        pass

     @pytest.fixture
     def a(b, c):
          pass