feat: trait typing

[maartenbreddels]

Took #788 and vidartf#1 did some fixes for py < 3.10 and squashed it to make rebasing to master easier.

Closes #788

[maartenbreddels]

Failures are expected due to not being on master (where this would be green).

Also, would require a bit more work to get all the other types supported.

As @rmorshea said in #788 (comment) there is no way but to copy-paste the __new__ overloads and modify them for each trait. I have to say that even with more advanced typing, I don't see us avoid this, since each trait is a bit different.

[blink1073]

I'd say partial support is a good start, we can follow up with the rest.

[maartenbreddels]

I'd like to add support to a few more, and see if we can do without the Self and simply add some more copy-pasting.

[maartenbreddels]

My guess is that the windows failures are solved by our good friend @vidartf who is always ahead of what we need for windows: davidfritzsche/pytest-mypy-testing#34
Is that correct @vidartf ? Does the error in CI look familiar to you?

[maartenbreddels]

cc @rmorshea

Btw, I've moved away from overloading __new__, but we are now overloading __init__, by annotating self.

class Float(TraitType[G, S]):
    """A float trait."""

    default_value = 0.0
    info_text = "a float"

    @t.overload
    def __init__(
        self: "Float[float, t.Union[int, float]]",
        default_value: t.Union[float, Sentinel] = ...,
        allow_none: Literal[False] = ...,
        **kwargs: t.Any,
    ):
        ...

    @t.overload
    def __init__(
        self: "Float[t.Optional[int], t.Union[int, float, None]]",
        default_value: t.Union[float, Sentinel, None] = ...,
        allow_none: Literal[True] = ...,
        **kwargs: t.Any,
    ):
        ...

    def __init__(self, default_value=Undefined, allow_none=False, **kwargs):
        self.min = kwargs.pop("min", -float("inf"))
        self.max = kwargs.pop("max", float("inf"))
        super().__init__(default_value=default_value, allow_none=allow_none, **kwargs)

This looks way less scary than what we suggested in #788

[maartenbreddels]

Typing is already paying off

[blink1073]

Okay, so basically we need to wait for a new release of mypy, and then if there is not a release of davidfritzsche/pytest-mypy-testing#34, skip the typing tests on Windows.

[blink1073]

mypy 1.0 has been released. It seems to me that pytest-mypy-testing isn't going to get a release. I would vote against depending on it, and instead have a set of files that we run mypy directly against.

[saulshanabrook]

Not to get too off-topic here, but @blink1073 I am also looking at using pytest-mypy-testing in a project but also am seeing that it is out of date.

I was thinking about trying instead to use MyPy's own unit testing framework, which is built on pytest, and was curious if you all had considered that as well.

• Pytest plugin: https://github.com/python/mypy/blob/5e2702fa753ed1625d5788dcb1b71ac4a50989fd/mypy/test/data.py
• Docs on test plugin: https://github.com/python/mypy/blob/master/test-data/unit/README.md

I wasn't sure if anyone was using this outside of MyPy though.

[maartenbreddels]

Interesting, I think it is tricky to get typing right without testing both successes and failures.

[saulshanabrook]

@maartenbreddels Yes I agree. I wasn't sure if you were implying that the mypy unit testing framework doesn't support testing failures as well, but here is a simple example of using it verify some things that pass and some that fail: https://github.com/python/mypy/blob/54635dec2379e2ac8b65b6ef07778015c69cfb6a/test-data/unit/check-basic.test#L93-L99

[saulshanabrook]

I just had a go at trying to use MyPy's builtin testing framework in my package, and it seems to work OK! Only requires one monkeypatch...

egraphs-good/egglog-python#6

You can see the test file here which verifies that one way of calling a method passes MyPy and the other raises an error: https://github.com/metadsl/egg-smol-python/blob/db300cba4413f2b91cd0b8c3c9ae9cecab089ad1/test-data/unit/check-basic.test

[blink1073]

I wonder if the mypy folks would support making the plugin public and not require the monkey-patch

[blink1073]

All green, let's gooo!