PoC: Use logging for warnings

[erlend-aasland]

Another experiment: tear out the custom warning functionality, and instead use logging.warning.

Pro: simplified error handling code
Con: no line numbers for warnings¹

Footnotes

1. well, there were never usable line numbers for warnings anyway ↩

[erlend-aasland]

Could be a nice yak-shave for #35

[erlend-aasland]

Another nice side effect is that post this, we could easily sprinkle Argument Clinic with debug and info log messages; it could significantly help improve the Argument Clinic debug experience.

[erlend-aasland]

What do you think, should we upstream this, @AlexWaygood? After this, it should be trivial to get rid of the global clinic object in the remaining fail().

[vstinner]

After this, it should be trivial to get rid of the global clinic object in the remaining fail().

Can you explain how this change make it easier to remove the global clinic object in fail()?

[erlend-aasland]

Can you explain how this change make it easier to remove the global clinic object in fail()?

warn() and fail() currently share implementation in warn_or_fail; tearing them out one at the time makes for better targeted PRs that are easier to review, and have smaller diffs.

[vstinner]

But how do you avoid access clinic global variable thanks to the logging module? You still need to retrieve filename and line number location somehow, no?

[erlend-aasland]

But how do you avoid access clinic global variable thanks to the logging module? You still need to retrieve filename and line number location somehow, no?

We never had line numbers for warn(); they are called from places in clinic.py where we don't have the line number at hand. IMO, it is better to acknowledge that fact, and just produce the warning. We have the filename, as you see from the PR :)

[vstinner]

FYI I wrote python#114752 to remove the usage global variable clinic.