Treat "env changed" as a warning, not a failure

[vstinner]

No description provided.

[vstinner]

Example on AMD64 FreeBSD15 NoGIL PR: https://buildbot.python.org/#/builders/2369/builds/2

1 test failed:
    test.test_free_threading.test_capi
513 tests OK.
(..)
Result: FAILURE
*** Error code 2

[vstinner]

A few years ago, I was tracking buildbot failures more closely and I wanted be notified when anything bad happened. I added "env changed" to libregrtest when a test passed but the test modified a common resource, leaked a subprocess/thread, etc. I treated "env changed" so I would get an email notification for example.

Nowadays, we have a way larger set of buildbot workers. It became really hard to keep track of all of them.

I propose to treat "env changed" as a warning (orange), instead of a failure (red).

@encukou @zware: Are you ok with this change?

[encukou]

I don't look at warnings very often.
Does anyone else? If not, the "env changed" wouldn't get noticed/fixed. I don't think we want that.

[vstinner]

I do check warnings (orage builds) time to time.

Does anyone else? If not, the "env changed" wouldn't get noticed/fixed. I don't think we want that.

Last months (years), I didn't notice any major issue reported by "env changed". It's usually files, processes or threads which are not cleanly properly by the test itself, not by the tested code.

"env changed" occurs randomly and so a buildbot blinks between red and orange.