Detect missing or incorrect overload defaults

[MarcoGorelli]

Feature

Is it in-scope for either mypy or stubtest to detect incorrect or missing defaults in overloads?

Incorrect default

from typing import Literal, overload, reveal_type

@overload
def foo(a: Literal[True] = ...) -> None: ...
@overload
def foo(a: Literal[False]) -> int: ...
def foo(a: bool = False) -> None | int:
    return 1 if not a else None

reveal_type(foo())

Here, the default for a is False, yet the overload has a: Literal[True] = ..., meaning that foo() will be revealed to be of type None instead of type int

Missing overload example

from typing import Literal, overload

@overload
def foo(a: Literal[True]) -> None: ...
@overload
def foo(a: Literal[False]) -> int: ...
def foo(a: bool = False) -> None | int:
    return 1 if not a else None

The default for a is False, so the second overload should probably include a: Literal[False] = ...?

Pitch

I tried putting together a little static analysis tool to detect simple cases of the above: https://github.com/MarcoGorelli/fix-overload-defaults

It only uses ast parsing, and is very simple, so it can't detect much, but it was already enough to find issues in a few libraries where I tried it:

• TYP: svd overload incorrectly noted Literal[False] to be the default for compute_uv numpy/numpy#29331
• TYP: correct default value of unicode in chararray.__new__ and of full in polyfit numpy/numpy#29340
• fix(python): Fix missing overload defaults in read_ods and tree_format pola-rs/polars#23442
• fix(python): Overload for eager default in Schema.to_frame was False instead of True pola-rs/polars#23413
• fix: type annotation default for named: Literal[False] was missing in DataFrame.iter_rows narwhals-dev/narwhals#2778

So, I figured that this might be broadly useful?

I'm much more bothered about detecting incorrect defaults than potentially missing ones. Might this be in-scope for mypy or mypy.stubtest?