Stanislav Terliakov

I'm marking this draft for now, because Eric Traut seems to be against this idea, let's wait for some decision to settle.

Is the terrific Windows failure at test teardown really caused by my changes?..

There is one severe false positive, caused by `overload`. I have a test case for it now and will try to resolve this problem tomorrow, then go through the rest....

Hm, and this primer diff looks marginally better. * `starlette`: the only false positive bit here, as far as I understand. How critical is `partial+ParamSpec` support? It is fundamentally different...

I have cleaned up code here a little bit, this should be now ready for review. #17355 is the follow-up patch I promised for `functools.partial` based on this (marked draft...

Cc @JelleZijlstra in a hope for review here:) This PR is a prerequisite for #17355 which overlaps with @ilevkivskyi ongoing ParamSpec work. I'm also fine with abandoning this and doing...

Diff review: * Starlette: false positive on a weird case (partial + overload), resolved in #17355 * Optuna: true positive (`fn(**Any)` where `fn(*args: P.args, **kwargs: P.kwargs)`) * Code (homeassistant): true...

Wow, it's the same problem as in the first workflow run for this PR. I hardly know anything about Windows, but still doesn't seem to be caused by the changes...

Yes, the most recent implementation is less forgiving for `.annotate()`, but positional arguments are indeed missed. https://github.com/typeddjango/django-stubs/blob/ffcebb3f0a71c4a2c3fd5038d29b3a5bcbf73248/mypy_django_plugin/transformers/querysets.py#L246-L275 Since the alias resolution logic is fairly non-trivial, we have several options here:...

cc @cdce8p as original author - could you have a look, please?