[3.9] bpo-40854: Allow overriding sys.platlibdir via PYTHONPLATLIBDIR env-var (GH-20605)

[vstinner]

(cherry picked from commit 8f023a2)

https://bugs.python.org/issue40854

[vstinner]

bpo-1294959 added sys.platlibdir and configure --with-platlibdir to use /usr/lib64 on Fedora and SuSE to Python 3.9. The problem is that https://bugs.python.org/issue40854#msg370655 use case seems to be a regression compared to Python 3.9.

I now consider to backport the commit 8f023a2 to 3.9 since it looks like a fix part of bpo-1294959 rather than a bug fix.

I created this PR to backport PyConfig.platlibdir and PYTHONPLATLIBDIR env var to Python 3.9.

@ambv: Are you ok to include this change in Python 3.9 final?

cc @hroncok @mcepl @stratakis

[vstinner]

If this PR is merged, the documentation must be updated in the master branch to replace "3.9" with "3.10" (as I did in this backport).

[ambv]

I'm fine with this landing in 3.9.0 but unless it lands within the next hour or so it will miss Beta 2.

[mcepl]

LGTM, but, boy, this is getting really complicated! Originally, it was just a simple few lines hack …

+1

[vstinner]

LGTM, but, boy, this is getting really complicated! Originally, it was just a simple few lines hack …

A hack is acceptable for a downstream patch. But upstream, we usually write things differently and it requires more work. Also, when a feature is made upstream, new corner cases are discovered :-)

See also https://bugs.python.org/issue40684 "make install doesn't respect configure --with-platlibdir=lib64".

[vstinner]

I'm fine with this landing in 3.9.0 but unless it lands within the next hour or so it will miss Beta 2.

Ok, I merged my PR. Thanks @ambv ;-)

[vstinner]

If this PR is merged, the documentation must be updated in the master branch to replace "3.9" with "3.10" (as I did in this backport).

I created PR #20735.