[red-knot] A FunctionType can be a subtype of Callable (but never the other way around)

[MatthewMckee4]

Summary

Partially fixes #16953

Test Plan

Update is_subtype_of.md

Im unsure how thoroughly this needs to be tested, im not sure if it does for many more function types?

[github-actions]

mypy_primer results

Changes were detected when running on open source projects

pybind11 (https://github.com/pybind/pybind11)
- error[lint:invalid-parameter-default] /tmp/mypy_primer/projects/pybind11/pybind11/setup_helpers.py:415:9: Default value of type `Literal[no_recompile]` is not assignable to annotated parameter type `(str, str, /) -> bool`
- Found 276 diagnostics
+ Found 275 diagnostics

rich (https://github.com/Textualize/rich)
- error[lint:invalid-parameter-default] /tmp/mypy_primer/projects/rich/rich/console.py:1881:9: Default value of type `Literal[currentframe]` is not assignable to annotated parameter type `() -> FrameType | None`
- Found 586 diagnostics
+ Found 585 diagnostics

black (https://github.com/psf/black)
- error[lint:invalid-return-type] /tmp/mypy_primer/projects/black/src/black/trans.py:2509:12: Object of type `Literal[is_valid_index]` is not assignable to return type `(int, /) -> bool`
- error[lint:invalid-argument-type] /tmp/mypy_primer/projects/black/src/black/lines.py:684:56: Object of type `Literal[is_import]` cannot be assigned to parameter `first_leaf_matches` of bound method `is_fmt_pass_converted`; expected type `((Leaf, /) -> bool) | None`
- Found 235 diagnostics
+ Found 233 diagnostics

[MatthewMckee4]

Ive just added the test that is failing, I'm not sure why it is failing though, any help is appreciated.

[MatthewMckee4]

Maybe an interesting point, and a question i have, should:

is_subtype_of(CallableTypeFromFunction[f], CallableTypeFromFunction[g]))
=>
is_subtype_of(TypeOf[f], TypeOf[g])

I guess this just depends on how we define FunctionLiteral arms in is_subtype_of. I would think that this equality should hold, but interested to see if there are any points against this

[carljm]

Barring the one test that should be changed, this looks good to me, and reasonably well tested!

[carljm]

For good measure, we could assert this is not true the other way around.

[MatthewMckee4]

Which one?

[carljm]

I meant both, but after reviewing the rest of the PR I think this is already adequately tested.

[carljm]

Maybe an interesting point, and a question i have, should:

is_subtype_of(CallableTypeFromFunction[f], CallableTypeFromFunction[g]))
=>
is_subtype_of(TypeOf[f], TypeOf[g])

No, this should not hold, for the same reason mentioned in my review above. TypeOf[some_function] is a singleton function literal type, which has no subtypes besides itself and Never. It is not a general callable type whose inhabitants are all functions with a substitutable signature (that's what we get from CallableTypeFromFunction[some_function]).

[MatthewMckee4]

Maybe an interesting point, and a question i have, should:

is_subtype_of(CallableTypeFromFunction[f], CallableTypeFromFunction[g]))
=>
is_subtype_of(TypeOf[f], TypeOf[g])

No, this should not hold, for the same reason mentioned in my review above. TypeOf[some_function] is a singleton function literal type, which has no subtypes besides itself and Never. It is not a general callable type whose inhabitants are all functions with a substitutable signature (that's what we get from CallableTypeFromFunction[some_function]).

Okay sounds good, thank you for the response and review. I think i thought TypeOf[some_function] was more similar to CallableTypeFromFunction[some_function] than it is.