Enable RMM Debug Logging via Python

[harrism]

Description

Enables RMM debug logging in Python and exposes controls for the level of logging. Fixes #1336.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[wence-]

Some minor possible cleanup (mostly documentation), looks good to me overall

[wence-]

Thanks @harrism ! Looks really good. A bunch of very minor polish/nitpick comments, but nothing to hold things up (other than the flake8 failure in linting on line 111 of logger.pyx).

[wence-]

other than the flake8 failure in linting on line 111 of logger.pyx

I committed the fix for that one.

[bdice]

I started a review but I'd like to wait until some of @wence-'s comments can be resolved before diving into this at a closer level.

For the moment, I'm wondering whether should_log should be a private function, or maybe replaced by something like a private member (or method) _compiled_log_level that can be used for direct determination of the compiled setting.

[vyasr]

I just commented on the threads where I was tagged but can provide a more comprehensive review once more of the open questions have been addressed.

[harrism]

My latest changes expose the C++ enum directly, renamed to rmm.logging_level and with its names capitalized to be more Pythonic. The functions that take a level now only accept instances of rmm.logging_level, not ints or strings. This simplifies the logic and testing a lot. It also warns if you try to set a logging level that will not be logged due to compilation level.

Regarding should_log: we could make it private, but I thought it might come in handy for anyone trying to detect or understand why any custom log messages are not being logged. It directly mirrors a C++ function in spdlog.

I could expose logging functions to enable adding custom messages to the log from Python code as well. Separate PR.

[wence-]

Minor fixes for link syntax

[wence-]

From my perspective this looks great as is. I (hopefully) fixed the doc-build failures 🤞 which were due to bad links. Not sure if @bdice will still have comments

[harrism]

Thanks for fixing my sphinx links @wence-, and for all your suggestions.

[vyasr]

Following @wence-'s lead, I'm trying out the semantic review comments! Overall this PR is looking really good now, just a few things that I'd like to see resolved before we can merge.

FYI I would guess that the issue you were seeing where any integral value was accepted for the enum is closely related to cython/cython#5610.

[harrism]

/merge