Support for static hints on std `replace()` functions

[Beinsezii]

Python's dataclasses module offers a replace() which allows for a theoretically safe interface to update frozen structs

import copy, dataclasses

@dataclasses.dataclass(frozen=True)
class Struc:
    a: int = 3
    b: str = "string"

data = Struc()
data2 = dataclasses.replace(data, a="word")  # Runs, but should lint regardless
data3 = copy.replace(data, c=False)  # Does not run, should lint

So while replace has a sig of **changes: Any, since we know in practice that it will TypeError if the wrong data is provided, to me it makes sense to have static checks for the args of replace against the field of the object provided, similar to standard assignment. The same could likely be applied to 3.13's copy.replace.

The reason I think this function in particular is suited to pyright unlike setattr or similar is that replace isn't intended for runtime reflection but rather a means for working with immutable classes.

[erictraut]

The function dataclasses.replace is defined in the typeshed stub dataclasses.pyi:

def replace(obj: _DataclassT, /, **changes: Any) -> _DataclassT: ...

The function copy.replace is defined in the typeshed stub copy.pyi:

def replace(obj: _SR, /, **changes: Any) -> _SR: ...

In both cases, pyright is correctly applying typing rules for these calls.

I think you're proposing for pyright to ignore the typeshed type definitions and override them with custom logic. That's technically possible, but it's something I try to avoid as a rule. I've broken this rule a few times in the past and I've come to regret it because it results in behaviors that are hard to explain. It also becomes a maintenance burden if and when the typeshed stubs change.

For these reasons, I'm going to reject this enhancement request. I'm willing to revisit this decision if there's sufficient signal from other pyright users in the form of upvotes. I'm going to close this issue for now because there are currently no plans to implement it.

[Beinsezii]

MyPy also just ignores it, and will report dataclass.replace being used with wrong keys/types.

[chazmo03]

Here's the MR for addressing this in mypy: python/mypy#14849

[trim21]

I think this will be resolve by python/typeshed#14819