typing.TYPE_CHECKER object to change types depending on which type checker is being run

[zunda-arrow]

One issue I have run into as a library dev is varying support for typing features or bugs for both of the type checkers. typing.TYPE_CHECKER would allow me to change type definition based on the type checker so that I can support all static analysis tools more effectively.

I propose that TYPE_CHECKER would be a dict with a "name" and "version" key. Type checkers could also optionally add other keys/values.

TYPE_CHECKER = {
    "name": "type-checker-name",
    # `VersionInfo` would be implemented by each type checker to match their versioning scheme.
    "version": VersionInfo(...),
}

Using it for types would look like this:

if TYPE_CHECKER.get("name") == "pyright" and TYPE_CHECKER.get("version") > (1, 1, 290):
    T = TypeVar("T", default=int)
else:
    T = TypeVar("T")

At runtime I think the TYPE_CHECKER object should be an empty dictionary.

[srittau]

There was a related discussion in #1129 about feature flags or a similar field.

[hmc-cs-mdrissi]

While I prefer feature flag approach so that new versions of type checker automatically enable new features, type checker name approach is still fine. I think version is too detailed though and would be maintenance pain to track.