fix[next]: Reduce type ignores in client code

[DropD]

Description

Partner-PR in icon4py: C2SM/icon4py#1096 (make sure this passes tests first)

Radically reduce the amount of required type ignores in client code which uses the from gt4py import next as gtx pattern.

• fixes next: type hints for dsl decorators need to be improved. #2496
• fixes next: dimension types are not valid types for static type checking purposes #2503

Experimenting with this in icon4py reveals more ways in which gt4py blocks client code type checking, which have not been addressed in this PR so far:

• next: builtin function type hints block client type checking #2507
• subtly different versions of next: builtin function type hints block client type checking #2507 where mypy reports a mismatch in dtypes, which have not been looked into further so far

Requirements

• All fixes and/or new features come with corresponding tests.
• Important design decisions have been documented in the appropriate ADR inside the docs/development/ADRs/ folder.

[egparedes]

Thanks for the PR. I just have a comment and a question

[egparedes]

Question: why is this needed? What error is mypy generating here?

[DropD]

one of these for each member:

src/gt4py/next/__init__.py:42:1: error: Module "gt4py.next.ffront.fbuiltins" does not explicitly export attribute "uint32"  [attr-defined]

I theorize it is because to determine which attributes are actually in fbuiltins.__all__, mypy would have to execute code. We are generating and concatenating lists, no idea why we do all of that dynamically at import time...

[egparedes]

My guess is that this ignore won't be needed after the suggestion above about defining fbuiltins.__all__ statically

[DropD]

Still no, the problem being that we don't sin = make_builtin("sin") etc, but instead we do globals()[key] = BuiltInFunction(val) in a loop.

One way to fix this would be to do the analog of sin = math.sin for each of these in an if TYPE_CHECKING block. The other would be to do it more manually in the first place.

[DropD]

I am going with manually unrolling the loops for now (the asserts will catch if there is a forgotten assignment)

[DropD]

cscs-ci run default

[egparedes]

Looks correct to me, thanks! I have some comments and questions, many of them optional or just for my own understanding.

[egparedes]

I would remove this from this PR. I'll try to create a tiny PR later to fix properly the versioningit warnings following this strategy.

[DropD]

This is only here so I can run CI in the meantime.

[DropD]

cscs-ci run default