Python 3.7, Windows and symlinks, oh my!
I'll investigate & fix next week.

@ned-deily this was quite a rabbit hole, and I'm as far as I can go. Unfortunately, not far enough. See Discourse for details:https://discuss.python.org/t/23903/11

I'll run buildbots to check if this is fine for non-Windows platforms, but I don't think CPython 3.7 can ship it :(

🤖 New build scheduled with the buildbot fleet by@encukou for commit773e47e 🤖

If you want to schedule another build, you need to add the🔨 test-with-buildbots label again.

[...] but I don't think CPython 3.7 can ship it :(

Thank you, Petr, for all your work on this across all the active branches! Given your analysis of the current 3.7 patch's shortcoming and give that 3.7 is scheduled to reach end-of-life in just a few weeks, I agree with your conclusion that it would be too risky and disruptive to officially accept this change for 3.7. As you note, third-party distributors who plan to provide support for 3.7 beyond its official end-of-life date are free to use and adapt the patch for their customers and they will have the opportunity to provide updates to it as necessary. Considering that the shortcomings that this change address have existed and been documented for many years and considering that all subsequent versions of Python (i.e. 3.8 on) will have this fix available, I do not think there is a need to consider either rushing a fix into 3.7 before its end-of-life or extending 3.7's life.

All buildbot failures look unrelated -- cryptography/network, USAN/ASAN, unsupported platforms/missing tools (MSVC toolset version, Tk, Wasm, Windows ARM), timeouts, a readline fail.

They shouldn't prevent building Python 3.7 from source.