Python 3.8, 3.9, and 3.10 link libcrypt.so.1 unexpectedy in releases 20230726 and later

[rmartin16]

With closure of #173 via a1e4e44, the releases were only expected to link libcrypt.so.1 from the _crypt so.

For Python 3.11 and 3.12, this is the case; however, Python 3.8, 3.9, and 3.10 still seem to link libcrypt.so.1 elsewhere.

For instance, if we consider cpython-3.10.13+20231002-x86_64-unknown-linux-gnu-install_only.tar.gz:

❯ ldd ~/Downloads/cpython-3.10.13+20231002-x86_64-unknown-linux-gnu-install_only/python/lib/libpython3.so 
	linux-vdso.so.1 (0x00007ffceffa8000)
	/home/user/Downloads/cpython-3.10.13+20231002-x86_64-unknown-linux-gnu-install_only/python/lib/../lib/libpython3.10.so.1.0 (0x00007f1dcb600000)
	libpthread.so.0 => /lib64/libpthread.so.0 (0x00007f1dccb2b000)
	libc.so.6 => /lib64/libc.so.6 (0x00007f1dcb41e000)
	libdl.so.2 => /lib64/libdl.so.2 (0x00007f1dccb26000)
	libm.so.6 => /lib64/libm.so.6 (0x00007f1dcb33d000)
	librt.so.1 => /lib64/librt.so.1 (0x00007f1dccb1f000)
	libcrypt.so.1 => not found
	libutil.so.1 => /lib64/libutil.so.1 (0x00007f1dccb1a000)
	/lib64/ld-linux-x86-64.so.2 (0x00007f1dccb4b000)

And if we consider cpython-3.11.6+20231002-x86_64-unknown-linux-gnu-install_only.tar.gz:

❯ ldd ~/Downloads/cpython-3.11.6+20231002-x86_64-unknown-linux-gnu-install_only/python/lib/libpython3.so 
	linux-vdso.so.1 (0x00007ffe85306000)
	/home/user/Downloads/cpython-3.11.6+20231002-x86_64-unknown-linux-gnu-install_only/python/lib/../lib/libpython3.11.so.1.0 (0x00007fd922000000)
	libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fd9237fb000)
	libc.so.6 => /lib64/libc.so.6 (0x00007fd921e1e000)
	libdl.so.2 => /lib64/libdl.so.2 (0x00007fd9237f6000)
	libm.so.6 => /lib64/libm.so.6 (0x00007fd921d3d000)
	librt.so.1 => /lib64/librt.so.1 (0x00007fd9237ef000)
	libutil.so.1 => /lib64/libutil.so.1 (0x00007fd9237ea000)
	/lib64/ld-linux-x86-64.so.2 (0x00007fd92381b000)

The same is also true of the Python binary. These tests were completed on Fedora 39; although, my current blocker is for recent Flatpak runtimes. I don't think this is a regression since a1e4e44; I had the same experience for 20230726 and 20230826.

Thank you for this work; it's greatly appreciated.

[indygreg]

Thanks for the report. Definitely unintended behavior.

This is on my backlog and I'll look at it the next time I sit down to hack on python-build-standalone. Probably not until late December at the earliest.

[indygreg]

I have a fix applied locally. Should push soon.

Turns out this was buggy behavior in CPython <3.11 configure script. See python/cpython#28881.