Handle pytest_warning_captured in workers

[nicoddemus]

Currently the warnings are no longer being transferred from workers to master.

[nicoddemus]

Oh oh. What to do about the item parameter of the pytest_warning_captured hook?

[RonnyPfannschmidt]

@nicoddemus its a unfixable error as is

[nicoddemus]

Yep, in hindsight that should have been nodeid, not item. 😞 See my PR.

[RonnyPfannschmidt]

@nicoddemus since we don't have a process for experimenting with api and pluggy doesn't have a tool for renames or changes of hook parameters, a new hook name is needed

[nicoddemus]

Perhaps, but as it stands at least the warnings are displayed, so I think the current PR is an improvement over not showing anything.

[RonnyPfannschmidt]

None instead of a item is a breach of api

[nicoddemus]

Not according to the docs:

item (pytest.Item|None) – The item being executed if when is "runtest", otherwise None.

So None is definitely a possible value for item, and hook implementations should deal with it accordingly.

[nicoddemus]

Just to be clear: I'm not saying this is perfect, but we have to take in account a few points:

1. This hook was just introduced, so I think only pytest itself uses it at the moment.
2. The long term solution is to either introduce a new hook as you suggest or a new nodeid parameter to the existing hook, possibly deprecating item (we will need to introduce such support to pluggy).

All of the above are very long term (at least until the next minor release cycle), so the approach proposed here is not a breach of API (according to the documentation) and at least will show warnings from workers. The price? Well, item will always be None when using xdist, which pytest already handles gracefully and as said I suspect no other plugins make use of this hook anyway.

Because of those reasons I still think the current PR is an improvement over the current state of things.

[RonnyPfannschmidt]

Indeed my mistake, if none is valid its good to go in