opencv-python stubs

[Avasam]

ready for review, but do not merge until #5768 . numpy is too fundamental for this stub, and merging as-is will cause lots of issues when trying to use it with pyright strict typing
(I could make an ndarray protocol with all of its properties and methods, but I'd rather wait)

• I first generated the stubs using stubgen
  • Including manually re-running stubgen for dynamically generated modules that it wasn't picking up recursively (stubgen misses generated modules mypy#13904)
• also generated the stubs with pyright since it catches more return types (but can't do C module)
• satisfied pyright strict typing for all files but cv2/cv2.pyi.
• then added everything from https://github.com/microsoft/python-type-stubs/tree/main/cv2
• used the existing docstrings to complete signatures and overloads
• Included different pending fixes/adjustments:
  • Add types to cv.resize() function microsoft/python-type-stubs#227
  • Completed cv2.Error and cv2.error types microsoft/python-type-stubs#219
  • cv2 minor fixes microsoft/python-type-stubs#232
• Fixed misnamed parameters dts to dst
• Changed list params to Sequence and tuple params to aliases that include Sequence
• Removed docstrings (see Should stubs contain docstrings? #4881)
• Fixed overloads missing self parameter
• Validated against https://github.com/Avasam/Auto-Split

As noted in METADATA.toml, the stub is stuck on opencv-python v4.5 until opencv/opencv-python#676 is fixed (import cv2.cv2 causes an error).
cv2/cv2.pyi is incomplete, massive and you might just wanna skim over it. But everything else should get a proper look and is meant to be complete.

For later / Not in this PR:

• Fix unspecified encoding causing UnicodeDecodeError with utf-8 chars on Windows in check_new_syntax.py (Specify file encoding #8882) and Flake8-pyi (Specify encoding in open PyCQA/flake8-pyi#296)
• Complete # incomplete signatures using docstrings from the help() method.
• Complete return types enough to remove all temporary aliases.

[AlexWaygood]

Thanks! I can see a lot of work's gone into this.

Just a few questions from skimming through:

[AlexWaygood]

This is a bit weird. __all__ is empty, but there's nonetheless a bunch of stuff that should be considered re-exported from the module? 🧐

[Avasam]

__all__ is also empty in the source code and at runtime.
If __all__ is empty, should it be omitted?

Btw, cv2 adds a bunch of code and modules dynamically, not sure if that's why they have an empty __all__ there..

[AlexWaygood]

Adding the "DO-NOT-MERGE" label, as I just noticed the note at the top of the PR description :)

[Avasam]

Some update: work has started again on opencv stub generation in opencv/opencv#20370 . Hopefully this here PR never gets to be merged, or a slightly different version of it for older openCV support (as I've already mentioned, some people, myself included, are stuck on 4.5 for different reasons).

Keeping this here as a draft as it is a very useful comparison point (manual stubs vs automated stubs) to eventually get the wip official generated stubs as accurate as possible.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

streamlit (https://github.com/streamlit/streamlit)
+ lib/tests/streamlit/elements/image_test.py:21: note: ... from here:
- lib/tests/streamlit/elements/image_test.py:21:1: error: Cannot find implementation or library stub for module named "cv2"  [import]

[Avasam]

Given the recent issues with SQLAlchemy2, I no longer think it would be wise to backport cv2 types, even if they're available right now from a github install.

And thanks to opencv-python being updated with type generation before opencv 4.8, it means users will already have access to types (even if early versions) in opencv4.7 by installing from github.