Improve typing of PropagateHandler and InterceptHandler

[Delgan]

See #654 (comment) and #449 (comment).

[quantumpacket]

The typehint can actually be put on the first declaration as it's the first declaration that is being used to determine the type for the same named variable. This works for Mypy, Pyright and Pyre.

        try:
            level: int | str = logger.level(record.levelname).name
        except ValueError:
            level = record.levelno

[Delgan]

@quantumpacket Is it considered best practice? I felt it might be confusing for the reader, so I decided to explicitly declare the level types before the assignment.

[quantumpacket]

Not sure what would consider it best practice or not, it's allowed by all type checkers listed. I guess it's a matter of preference at this point. 🤷‍♂️

[Delgan]

Ok, I'll leave it as is then, but thanks for the heads-up.

[quantumpacket]

Was up all night reading about frames/stacks and I came up with another way to get the depth value, which works in my tests (need your confirmation), and is a bit cleaner with reduced cyclomatic complexity.

depth = 1

for frame in traceback.extract_stack():
    if frame.filename == logging.__file__:
        depth += 1

Python, a thousand and one ways to do the same thing. 😆

[Delgan]

Haha, nice find!
It passes all tests in tests/test_interception.py so I think you're good!

This implementation indeed looks pretty good, at the cost of a bit more memory usage. I personally prefer to deal directly with the frame object than the FrameSummary abstraction.

[quantumpacket]

@Delgan if you still want to use inspect.currentframe() then this should work (works on my tests):

        frame, depth = inspect.currentframe(), 1

        while frame:
            if inspect.getfile(frame) == logging.__file__:
                depth += 1

            frame = frame.f_back

[Delgan]

I don't really have any particular preference. I've just documented a solution that solves the problems you mentioned, so I didn't dwell on it too much since I'm planning to replace these classes anyway. But thanks for sharing.