GH-126789: fix some sysconfig data on late site initializations

[FFY00]

• Issue: sysconfig data inconsistencies caused by the disabling the site initialization on virtual environments #126789

[FFY00]

@hugovk could you maybe have a look at this PR when you have time, as I am currently the only active maintainer?

[FFY00]

There's an address sanitizer failure, but seems unrelated. Might be worth investigating still.

https://github.com/python/cpython/actions/runs/11827914917/job/32957010967?pr=126812

[miss-islington-app]

Thanks @FFY00 for the PR 🌮🎉.. I'm working now to backport this PR to: 3.12, 3.13.
🐍🍒⛏🤖

[freakboy3742]

@FFY00 This PR (and the 3.13 backport) appears to have broken the iOS and Android buildbots. venv tests are also impractical on those platforms due to the lack of subprocesses. I'm working on a fix now.

Update: #126941 is the PR; it also needs a backport to 3.13.

[FFY00]

@freakboy3742 sorry about that! I probably should have triggered the buildbot tests before merging.

I hadn't run into issues caused by the unavailability of subprocesses, so I forgot about that. This was an oversight on my part 🫤

[FFY00]

We could write the tests to not use subprocesses, but they depend on global state, which makes the test implementation tricky, so I just opted for using virtual environments, which unfortunately required the use of subprocesses. I don't think there is quite enough platform dependence in the tested code to warrant rewriting to be compatible with platforms that don't support subprocesses, but I am happy to hear other opinions on this.

[freakboy3742]

@FFY00 I'm not sure it's worth a major rewrite. Virtual environments aren't really useful in the iOS/Android/Emscripten context, because those platforms are already sandboxed and isolated by the app/webpage bundle that they're delivered in. Its similar to the question of whether you need venv inside a Docker container - except that in this case, there's no ability to call subprocess, which makes using a venv near impractical, rather than just an arguable design choice.

[terryjreedy]

Backport PRs #126918 and #126919 are merged.