Pytest 6.2 warns about the deprecated --strict flag

[Zalathar]

As of pytest 6.2, the --strict flag is deprecated and issues a warning. Because we include this flag in our pytest.ini (and various other places), all of our pytest invocations produce this annoying warning, making it harder to understand the test results.

(The --strict-markers flag was introduced in pytest 4.5 as an alias for the existing --strict flag, in the hope that --strict could be eventually repurposed into a catch-all strict mode flag.)

In #2862 I tried to fix this by replacing --strict with --strict-markers in our repo. This almost worked, except that check-pytest43 now fails because pytest 4.3 it doesn't understand --strict-markers and errors out.

Before going too far down the rabbit hole of trying to fix that, I thought I'd open an issue for discussion of what we should actually do about this whole situation.

[Zalathar]

I understand that we want to try to support older pytest versions, but is there any particular reason to stick with 4.3?

Back in January of 2020, pytest appears to have stopped active maintenance of pytest 4.6 in favour of best-effort community maintenance.

[Zalathar]

For context, the last time we bumped our pytest requirement was #2126 in late 2019.

[Zalathar]

I think I've figured out how to make pytest 4.3 not see --strict-markers in pytest.ini, so I might end up reopening #2862 after all.

[Zac-HD]

I'm perfectly happy to bump our minimum Pytest version to 4.6 and move on - according to the general rule "we support whatever upstream still supports, plus older things if they're still in widespread use and doing so is not inconvenient to us volunteers". Since 4.3 is now inconvenient as well as EOL upstream, it's time to drop support.

On the other hand, if you can make old versions ignore new flags we might as well add --strict-config too!

[Zalathar]

At this point the workaround is easier than dropping support (since I've already written the workaround), but I wouldn't object to dropping support anyway.

[Zac-HD]

I'm voting to drop support for <4.6; it's about time and it saves us the slightly-awkward workaround.