Looks like the mypy-primer difference is due to python/mypy#17282 (comment)

I think it's debatable what should be done here then, but I think at least ndarray.nonzero and MaskedArray.nonzero should be aligned

Given the (noisy?) "Variadic tuple unpacking requires a star target" error, I'm inclined to suggest that just tuple[_1d_array[intp], ...] should be fine in both cases, it's also simpler to read / maintain - @jorenham any thoughts on that?

Given the (noisy?) "Variadic tuple unpacking requires a star target" error, I'm inclined to suggest that just tuple[_1d_array[intp], ...] should be fine in both cases, it's also simpler to read / maintain

It's pretty annoying that mypy doesn't support this. Given the primer diffs, and the fact that there are several other related mypy issues (python/mypy#19110, python/mypy#19109, python/mypy#19106, python/mypy#19105), I agree that we should fall back to using tuple[{}, ...] here. For future reference it could help to reference python/mypy#17282 (and perhaps this PR) there.

But either way, the shape-typing is already a good improvement, so this is still a win :)

Diff from mypy_primer, showing the effect of this PR on type check results on a corpus of open source code:

pandas (https://github.com/pandas-dev/pandas)
+ pandas/core/internals/managers.py:1527: error: "BlockPlacement" has no attribute "increment_above"  [attr-defined]

Thanks @MarcoGorelli