#2899 Please take a look at this branch and lmk if this fix is acceptable. Not seeing hyperfast execution with tox -p auto (approx 10 processes on my laptop takes about 70-80s), seems a bit faster with setting processes to 4 or 6. All tests are passing though with this change (both initial and subsequent runs as previously only the latter were passing).

FWIW Opus 4.5 located the issue, I was just steering the ship. Please feel free to take over from here and make fixes.

Please do not use LLMs to generate PRs. This is a huge mess of code (and a bunch of docs that don't belong) that is impossible to review.

FWIW I thought this was opened as a Draft

seems like the issue is with

force_subprocess_stdout = patch.object(
            subprocess,
            "Popen",
            partial(subprocess.Popen, stdout=f),
        )

Hope this helps someone