implementation of pytest lastfailed usecases that don't prematurely remove tests from the failure set

[RonnyPfannschmidt]

with the current ad hoc implementation its not possible to track all edge cases while remembering the last failure set over into the next iteration

for implementing lastfailed that does not subtract passing tests prematurely from the last failure set
it is necessary to spec out a exact and well defined behavior for the detail interactions between
lastfailed, xdist and sets of tests and failures

there are a few boundaries that create huge potential for edge-cases
the ones i'm aware of while creating the ticket are

• current collection vs full collection
• current selection
• collector failure vs test failure
• known set of tests vs actual set o f tests (currently untracked and preventing adding newly found tests to the set of tests to the set of tests to run)

[nicoddemus]

Could you provide a few examples @RonnyPfannschmidt? From the description I can't quite figure out what the edge cases are.

[RonnyPfannschmidt]

I'm planning to, I cut it short at the description since writing the actual spec covering the areas I mentioned takes a while and is error prone as well

[nicoddemus]

OK, thanks. 😄