Refactor threading layer priority tests to not use stdout/stderr

[stuartarchibald]

As title, this prevents output from e.g diagnostics being able to
influence the outcome of the tests.

Fixes #7458

[esc]

Although the PR didn't touch this line, I would recommend to change:

env = os.environ.copy()

into

env = {}

Copying the entire environment is a bit dangerous. For tests, it is better to start with a clean slate to avoid whatever variables exist in the environment to leak into the test. May not become an issue in practice, but I believe it's prudent to make sure it really doesn't.

[esc]

It looks great. A few minor things and it'll be good to go!

[stuartarchibald]

Although the PR didn't touch this line, I would recommend to change:

env = os.environ.copy()

into

env = {}

Copying the entire environment is a bit dangerous. For tests, it is better to start with a clean slate to avoid whatever variables exist in the environment to leak into the test. May not become an issue in practice, but I believe it's prudent to make sure it really doesn't.

Agree, am happy to make this change whilst editing these tests. It was, however, the env copy that actually triggered and demonstrated the problem with these tests in the first place!

[esc]

Thank you for the changes, looks good now!

[stuartarchibald]

Turns out that doing #7749 (comment) leads to:

Fatal Python error: _Py_HashRandomization_Init: failed to get random numbers to initialize Python
Python runtime state: preinitialized

on windows, probably because the sysroot is cleared. 1d7f47b puts back the copy of the env as it's likely the easiest way to obtain a suitable environment that doesn't require tracking required Python env vars.

[esc]

OK, let's go with the original.