Revert "Add support for attrs.fields (#15021)"

[jhance]

This reverts commit 391ed85.

The use of a Union item type for attr.Attribute would likely be more useful to callsites of attr.fields - otherwise this currently results in inferring builtins.object in many cases where we treat the fields as a variable-length tuple, rather than a fixed-length one. Applications of the fixed-length tuple seem limited since you would need to access it by index.

@Tinche as the original author of the commit.

[JukkaL]

One option would be to infer a variable-length tuple with union item type. The issue here is that mypy doesn't use unions for joins.

[Tinche]

@JukkaL I'm not in favor of this, see my comment over at #15021 (comment).

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

[ilevkivskyi]

Limiting the tuple special-casing mentioned in #15396 (comment) to cases where fallback instance is builtins.tuple, or maybe better to when custom fallback doesn't override tuple.__iter__. But I think there may be a much simpler solution: have you tried making Attribute covariant (unless there is a good reason to keep it invariant)? IIRC mypy is now smart enough to infer Attribute[object] as iteration item for covariant generics.

[jhance]

I don't think Attribute can be covariant because the validator object is clearly not. I looked into this yesterday.

[hauntsaninja]

Opened an alternate PR to fix this #15688

[Tinche]

@hauntsaninja thanks <3