unittest: do not use TestCase.debug with --pdb

[blueyed]

Reverts #1890, which needs to
be fixed/addressed in another way
(#5996).

Fixes #5991.

[blueyed]

Decided to revert to sane behavior, until this is addressed properly (#5996 is a start).

[nicoddemus]

This is a regression I'm afraid, we need to find a new way to properly handle skips...

[blueyed]

Sure - but it was a regression / bug from the beginning.
Therefore I'd rather revert it already, and fix / re-try it later.

[blueyed]

Should I do this against master maybe instead? ;)

Seriously, I am aware that it reverts a previous change, but that one is just wrong, and caused multiple (and more serious) regressions to start with.

I've came across this several times, and it had to be adjusted / fixed multiple times.
The original author never reacted to my pings, and might not even be using pytest anymore.
We should not protect this kind of "feature" (which is really a hack after all).
The proper / better way might just be to add/support "--unittest-debug" - using "--pdb" for that was just a misunderstanding to begin with (actually people are using that by default, and it causes all teardowns to be skipped then (IIRC)).

[blueyed]

But since this conflicts anyway already, and me being tired of pointing it out (and people not caring), I'll just close it for now.
(I'm not using unittests myself btw)

[blueyed]

@nicoddemus
/cc @pytest-dev/core
To summarize: pytest uses .debug() in unittests, which is meant to be used only manually for debugging unittests.
Somebody came along and found that tests are being teared down already when using --pdb for post-portem debugging, which closed their GUI.
This was in 2016 already (#1890), and a constant source of issues, e.g. reported via pytest-django, but also here (the latest #5991).

After investigating, and giving it a try myself to fix it I've concluded that it is the best to revert this for now, especially since the issue mentions a specific workaround already, and the more obvious would be to just put a pdb.set_trace() where you need it (before the failure).

Anyway, what I really do not get is that "reverting this mis-feature" is being rejected as being a regression itself..!
Like I've mentioned above the original author never reacted to my concerns about this, and might not even be using pytest anymore - but still this bug is being protected from being reverted?!

I would like to raise awareness that we should strive for rather reverting a broken feature than carrying it along (there were several fixups already after #1890).

[nicoddemus]

You got a point there. I'm on vacation at the moment so I can't look at the code now, but I will pick it up again when I get back.

[nicoddemus]

I agree, see #5996 (comment).