config: ini: should allow for `default=None` - currently returns `[]` then

[blueyed]

Will create a PR, but can use this then for a changelog.

[RonnyPfannschmidt]

You can also use the pr ID for the change log

[blueyed]

..which is not known upfront typically.

[RonnyPfannschmidt]

I fail to see how that is a problem with git

[blueyed]

I usually force-push later, yes, if that's what you mean - but here I was certain to have a PR without failing tests, and did not want to trigger CI (I cannot cancel Azure).
Anyway... to be fair you could easily guess it usually, but:

I'd be more interested on your opinion with regard to API changes, since default=None (the current default) gets you an empty list. I could also use default=unset explicitly for my use case - that would prevent API changes.

[RonnyPfannschmidt]

as far as i can tell its intentional and was added about a decade ago that way
so changing it would be a major breaking change

i haven't investigated the deeper details of this one, and i don't think i can in near future as it will require a proper history lesson spanning about 12-14 years of pytest config evolution

[blueyed]

I think it is just an oversight: for getoption it was done in 0365e5c.

[RonnyPfannschmidt]

hmm, im inclined to agree, im notsure how to best assess impact,

[blueyed]

Will consider it with #5968 again (code ref: https://github.com/pytest-dev/pytest/pull/6118/files#diff-b80bb977f96695d5bfe21ddd2680c4efR993).
AFAICS it should handle different type better anyway, and this could keep the old behavior then maybe, checking if "default" is passed as an arg or not etc.
I am using it already (6439ebc), so might stumble across odd things in this regard.

[blueyed]

Found something that changed in this regard: with https://github.com/pytest-dev/pytest/pull/4521/files#diff-b80bb977f96695d5bfe21ddd2680c4efR357 I was using default=notset myself (likely since None did not work), which then resulted in an empty list with #6118.
Since others might have used the same workaround I think we should check if it was provided in args/kwargs actually to see if it should be considered to have been set/given, or not.