Improve mypy coverage

[schnellerhase]

Mypy check was only run on source code before, with the introduction of auto-generated stubs #970 this is not sufficient anymore.

Note: editable install based to avoid missing the autogenerated stubs FEniCS/dolfinx#4096 (comment).

[jhale]

Not sure about this; running locally mypy -p basix inside the root folder (so that we see the pyproject.toml mypy settings, but not any source folders) with a non-editable install gives:

/Users/jack.hale/fenicsx-main/lib/python3.12/site-packages/basix/_basixcpp.pyi:522: error: Overloaded function signature 2 will never be matched: signature 1's parameter type(s) are the same or broader  [overload-cannot-match]
/Users/jack.hale/fenicsx-main/lib/python3.12/site-packages/basix/finite_element.py:556: error: Incompatible return value type (got "list[list[Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str]]]", expected "list[list[ndarray[tuple[Any, ...], dtype[Any]]]]")  [return-value]
/Users/jack.hale/fenicsx-main/lib/python3.12/site-packages/basix/finite_element.py:565: error: Incompatible return value type (got "list[list[Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str]]]", expected "list[list[ndarray[tuple[Any, ...], dtype[Any]]]]")  [return-value]
Found 3 errors in 2 files (checked 13 source files)

[jhale]

Have one error left which annoyingly is related to the generated stubs:

@overload
def tabulate_polynomial_set(celltype: CellType, polytype: PolysetType, d: int, n: int, x: Annotated[ArrayLike, dict(dtype='float32', shape=(None, None), order='C', writable=False)]) -> Annotated[ArrayLike, dict(dtype='float32')]: ...

@overload
def tabulate_polynomial_set(celltype: CellType, polytype: PolysetType, d: int, n: int, x: Annotated[ArrayLike, dict(dtype='float64', shape=(None, None), order='C', writable=False)]) -> Annotated[ArrayLike, dict(dtype='float64')]: ...

Gives:

/Users/jack.hale/fenicsx-main/lib/python3.12/site-packages/basix/_basixcpp.pyi:522: error: Overloaded function signature 2 will never be matched: signature 1's parameter type(s) are the same or broader  [overload-cannot-match]
Found 1 error in 1 file (checked 13 source files)

I noticed this in pattern_stub.txt:

basix._basixcpp.tabulate_polynomial_set:
    def tabulate_polynomial_set(celltype: CellType, polytype: PolysetType, d: int, n: int, x: Annotated[ArrayLike, dict(dtype='float64', writable=False, shape=(None, None), order='C')]) -> Annotated[ArrayLike, dict(dtype='float64', )]: ...

Maybe it needs applying differently?

[jhale]

OK, it is now triggering but my fixes, which work locally, are not good for the CI. I guess there are subtle differences.

https://github.com/FEniCS/basix/actions/runs/22392500731/job/64817819539#step:8:14

[jhale]

There is no need for INSTALL_TIME and the .pyi file can be installed as a standard CMake target.

[jhale]

This nearly works. The last renaming issue I cannot get the pattern_stub.txt to match tabulate_polynomial_set at runtime, so that I can replace it.

https://nanobind.readthedocs.io/en/latest/typing.html#pattern-files

[schnellerhase]

What's the downside of using INSTALL_TIME in both cases and not just for windows? This gives now very different install behaviours depending on platform.

[jhale]

I read the manual and it says that install_time should only be used when the standard
mode doesn't work. I think the standard mode makes a lot of sense in the context of CMake - build artifacts are always built and then installed. Install time is a hack.

[jhale]

Win32 is always 'odd' - its philosophy on how software should be shipped and linked is fundamentally different to Unix. It may be possible to get the build mode to work on Win32 as well but I don't fancy prodding the CI for hours. Maybe I get a windows box soon.