revert Timestamp and Timedelta constructors typing allowing NaTType return

[Dr-Irv]

Re discussion at #46171 (comment)

Has Timestamp and Timedelta constructors return union that includes NaTType

[Dr-Irv]

@mroeschke Not sure what to do here. pyright is reporting issues in other places in the code. See https://github.com/pandas-dev/pandas/runs/7863880652?check_suite_focus=true

I think that the level of pyright checking increased since the original PR was done. Or other changes made to the code were fine without having the constructors return NaTType so when we declare it that way, we need to fix some other places.

On the one hand, pyright is picking up some potential errors. On the other hand, making this simple change means there is more work to do to fix up other parts of the code that were not touched in the original PR.

I will see if I can workaround those issues.

[twoertwein]

I think the issue is that for mypy __new__ cannot return an object other than from the corresponding class. python/mypy#7188 suggests that mypy supports other type annotations for __new__ IF they are subclasses (so basically it still seems to not support them for Timestamp).

If they cannot be fixed now, it is probably best to add those files to https://github.com/pandas-dev/pandas/blob/main/pyright_reportGeneralTypeIssues.json one day when this ignore list is small enough we can add in-line ignore comments.

[Dr-Irv]

I think the issue is that for mypy __new__ cannot return an object other than from the corresponding class. python/mypy#7188 suggests that mypy supports other type annotations for __new__ IF they are subclasses (so basically it still seems to not support them for Timestamp).

Yes, that is it.

If they cannot be fixed now, it is probably best to add those files to https://github.com/pandas-dev/pandas/blob/main/pyright_reportGeneralTypeIssues.json one day when this ignore list is small enough we can add in-line ignore comments.

Thanks for this suggestion. Will do this in next commit.

[twoertwein]

Great that you were able to reduce the number of new files ignored by pyright's reportGeneralTypeIssues to only two!

Looks good to me, assuming the typing tests pass.

[twoertwein]

Probably the first catch of stubtest :) tz doesn't seem to exist at runtime: https://github.com/pandas-dev/pandas/runs/7865645731?check_suite_focus=true

[Dr-Irv]

Probably the first catch of stubtest :) tz doesn't seem to exist at runtime: https://github.com/pandas-dev/pandas/runs/7865645731?check_suite_focus=true

Thanks for finding that. I'll create tz and tzinfo methods for NaTType that return None

[Dr-Irv]

@mroeschke so this is good to go. I don't know what to do about the failure with the CircleCI Pipeline

[mroeschke]

Thanks for the quick resolution @Dr-Irv! Hoping the CircleCI issue is spurious, but the important part is the typing builds passed.