weird setupstate behaviour for yield tests about pytest HOT 12 CLOSED

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

To be more precise and make sure i understand: collecting yield-based tests needs setupstate and when further items are collected, the state may be teared down and thus when the yield-tests actually run, the setupstate needs to be recreated a second time. Right?

Now i wonder: why can the setup state not be created a second time?
And if that cannot happen, could the machinery be rewritten to use pytest_generate_tests?

feel free to also point me to the repo and exact code you are talking about.

from pytest.

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

ronny, what's the status of this issue and can you provide an exact example of what shoudl be fixed here?

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

im not sure what/how to fix, the problem is that yield tests need a setupstate in collection,
in the case of moint the generate a local wiki instance and prepare data
since the setupstate is removed when tests in a different hierachy run, the yielded closures break since the setup state changed

from pytest.

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

can you provide a minimal failing code example?

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

finally had a good idea for a minimal case

from pytest.

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

I assume, yield/setupstate is not an issue anymore, given that moinmoin now uses a more modern version.

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

its still an isse, moinmoin has a hack in place to work around

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

basically it will stay an issue till we provide a different way to have partial failures without stopping the complete test

from pytest.

Original comment by holger krekel (BitBucket: hpk42, GitHub: hpk42):

I am not sure if we should pursue any improvements to yield-based tests. In any case, i don't know what the actual idea here is for improving yield-based setup. Doing larger scale refactorings just for better yield support does not seem worthwhile to me. It might be acceptable if better comes out as a side effect and the refactoring is useful for other reasons.

If nothing happens on this issue for a couple of months i'd like to close it.

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

i hit yet another proejct where this prevents pytest usage

the suggested solution is to drop collecting yield items, and make reporting prettier later

from pytest.

Original comment by Ronny Pfannschmidt (BitBucket: RonnyPfannschmidt, GitHub: RonnyPfannschmidt):

addresses #16 - replace Generator with pytest_pyfunc_interpret_result

from pytest.

finally addressed in #1714

from pytest.