Support Decorators Maintaining Signatures of Generic Functions

[WillAyd]

As a follow up to #27393 I think I made a mistake in just typing some of the return values to be Callable. The following script highlights that issue:

from pandas.util._decorators import Appender

def func1(num: int) -> int:
    return num

@Appender("")
def func2(num: int) -> int:
    return num

reveal_type(func1)
reveal_type(func2)

If you run that from the project root you'll get something like the following:

$ mypy foo.py
foo.py:10: note: Revealed type is 'def (num: builtins.int) -> builtins.int'
foo.py:11: note: Revealed type is 'def (*Any, **Any) -> Any'

We probably want to maintain the types of the signature for func2 so I think need to use TypeVar for generic support.

More info on the mypy side in:

python/mypy#3113
python/mypy#1551

So I think need something like:

T = TypeVar('T', bound=Callable[..., Any])
# then decorators can return
-> Callable[[T], T]

[topper-123]

Does reveal_type return the proper types if this change is made?

[topper-123]

Possibly related to #13875.

[WillAyd]

@simonjayhawkins you recently had a solution for this right? Interested in extending it too pandas.util._decorators?

[simonjayhawkins]

@WillAyd I'm reopening this as #28128 did not close this issue. solved for Appender and Substitution only

dd95741 will fix deprecate_kwarg and probably similar for rewrite_axis_style_signature

[WillAyd]

Sounds good - thanks for taking a look

[jbrockmendel]

Does being py37+ make this any more solvable?

[WillAyd]

This should already be resolved here:

pandas/pandas/_typing.py

Line 101 in 6f2c60e

F = TypeVar("F", bound=FuncType)