Pytest: 5.3 changes plurality of keys in testdir.run().parseoutcomes()

[altendky]

While testdir.run().parseoutcomes() used to give results such as {'warnings': 1} (plural) it now gives {'warning': 1} (singular). This broke the tests in pytest-twisted... and while I setup nightly 'cron' builds I apparently didn't bother to notice when they broke awhile ago. :[ I do not know what the proper answer is here especially now that we have multiple versions out with each approach. Since pytest-twisted is still supporting py2, I don't get to just update to latest pytest functionality across the board. Though yes, py2 support will at some point be dropped when the pain is sufficient.

I'll just normalize the parseoutcomes() dict keys myself but wanted to document the change and see what opinions there are on it.

When pytest-twisted broke:
last good pytest==5.2.4: https://travis-ci.org/pytest-dev/pytest-twisted/jobs/614201584
first bad pytest==5.3.0: https://travis-ci.org/pytest-dev/pytest-twisted/jobs/614718766

Introduced in #5989 and https://github.com/pytest-dev/pytest/commit/d863c30c743138ed29cdf786cb18c41c24983fe1.

There seems to be little testing being done on parseoutcomes().
https://github.com/pytest-dev/pytest/blob/e98176cf50c5b2d3df965ec409d7f62fefcd9ebd/testing/test_pytester.py#L687-L706
https://github.com/pytest-dev/pytest/blob/e98176cf50c5b2d3df965ec409d7f62fefcd9ebd/testing/test_capture.py#L1161-L1191

[Zac-HD]

5990 makes sense for user-facing reporting, but IMO the parseoutcomes keys should not vary with their values, and I would therefore be happy to accept a pull request that ensures we always use plurals.

[blueyed]

Sounds related: #6382

[altendky]

Looks like #6382 covers error[s] so yeah, definitely related and a start. Perhaps we need an explicit singular/plural table for the outcomes which can be used for both pleasant human readable output, and for undoing that human readable output..

[blueyed]

/cc @MarcoGorelli

[MarcoGorelli]

Really sorry for the breakage I introduced with my issue / PR! Gonna be pretty busy for the next two weeks, but if this isn't addressed by then I'll take a look

[koscinskic]

If we could, myself and a few others can look into ensuring a system that functions solely on plural-only keys while retaining proper singular / plural output.

[MarcoGorelli]

@koscinskic sure, go ahead :)