[red-knot] Add support for calling type[…]

[sharkdp]

Summary

This fixes the non-diagnostics part of #15948, but is mostly an experiment to see if the new ecosystem checks work as intended (see failing mypy primer run here). I saw a few false positives when expressions of type type[<dynamic base>] were called, so this seemed rather easy to fix.

Test Plan

New Markdown tests.

Negative diff on the ecosystem checks:

zipp (https://github.com/jaraco/zipp)
- error: lint:call-non-callable
-    --> /tmp/mypy_primer/projects/zipp/zipp/__init__.py:393:16
-     |
- 392 |     def _next(self, at):
- 393 |         return self.__class__(self.root, at)
-     |                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Object of type `type[Unknown]` is not callable
- 394 |
- 395 |     def is_dir(self):
-     |
- 
- Found 9 diagnostics
+ Found 8 diagnostics

arrow (https://github.com/arrow-py/arrow)
+     |
+     |
+ warning: lint:unused-ignore-comment
+    --> /tmp/mypy_primer/projects/arrow/arrow/arrow.py:576:66
+ 574 |                 values.append(1)
+ 575 |
+ 576 |             floor = self.__class__(*values, tzinfo=self.tzinfo)  # type: ignore[misc]
+     |                                                                  -------------------- Unused blanket `type: ignore` directive
+ 577 |
+ 578 |             if frame_absolute == "week":
- error: lint:call-non-callable
-     --> /tmp/mypy_primer/projects/arrow/arrow/arrow.py:1080:16
-      |
- 1078 |           dt = self._datetime.astimezone(tz)
- 1079 |
- 1080 |           return self.__class__(
-      |  ________________^
- 1081 | |             dt.year,
- 1082 | |             dt.month,
- 1083 | |             dt.day,
- 1084 | |             dt.hour,
- 1085 | |             dt.minute,
- 1086 | |             dt.second,
- 1087 | |             dt.microsecond,
- 1088 | |             dt.tzinfo,
- 1089 | |             fold=getattr(dt, "fold", 0),
- 1090 | |         )
-      | |_________^ Object of type `type[Unknown]` is not callable
- 1091 |
- 1092 |       # string output and formatting
-      |

black (https://github.com/psf/black)
- 
-     |
-     |
- error: lint:call-non-callable
-    --> /tmp/mypy_primer/projects/black/src/blib2to3/pgen2/grammar.py:135:15
- 133 |         Copy the grammar.
- 134 |         """
- 135 |         new = self.__class__()
-     |               ^^^^^^^^^^^^^^^^ Object of type `type[@Todo]` is not callable
- 136 |         for dict_attr in (
- 137 |             "symbol2number",
- Found 328 diagnostics
+ Found 327 diagnostics

[AlexWaygood]

The diff in the mypy_primer run looks great! It's unfortunate that it shows up as a "failed check" on the PR, though -- is that deliberate?

[sharkdp]

It's unfortunate that it shows up as a "failed check" on the PR, though -- is that deliberate?

Signalling a non-empty diff as a 'failure' (that you need to look at) seems fine to me? Once we add commenting on PRs, we could potentially turn that into a check that always succeeds.

[AlexWaygood]

It's unfortunate that it shows up as a "failed check" on the PR, though -- is that deliberate?

Signalling a non-empty diff as a 'failure' (that you need to look at) seems fine to me? Once we add commenting on PRs, we could potentially turn that into a check that always succeeds.

hmm, I suspect this will be quite confusing for new contributors. Though having said that, contributors have also been confused over at typeshed by the mypy_primer comments. (A "sea of red" in the comment can be a good thing, since it shows that diagnostics are going away! But red usually connotes "bad" 😄)

[sharkdp]

These will be fixed by #16512

[AlexWaygood]

I think overall I'd much prefer a "comment on the PR" mypy_primer check rather than have the check fail. But the changes this PR makes look great!

[sharkdp]

I think overall I'd much prefer a "comment on the PR" mypy_primer check rather than have the check fail

Ok, I'll look into it.