GH-145273: warn when we can't find the standard library

[FFY00]

• Issue: Issue warning when the we can't find a valid standard library during path initialization #145273

[FFY00]

The main issue with this is that we don't have a way to disable these warnings, which is resolved by GH-145277.

[FFY00]

I fixed the tests, but unless we skip the platstdlib_dir check when running from source, the warning will show up on make, during the pybuilddir.txt generation.

I would prefer to drop the build_prefix check, and instead add -X pathconfig_warnings=0 to the -m sysconfig --generate-posix-vars call in the makefile.

Suggested change

	elif (not platstdlib_dir and not build_prefix) or not isdir(platstdlib_dir):
	elif not platstdlib_dir or not isdir(platstdlib_dir):

[FFY00]

I was thinking the symlink tests were failing on Windows because the getpath implementation of isdir/isfile didn't follow the link, but looking at the GetFileAttributesW documentation, it says it does follow symlinks.

This means the tests were failing because the symlink target did not exist, but in test_platform,PlatformTest.test_architecture_via_symlink for eg. I cannot see how that's possible? It uses test.support.PythonSymlink, which checks for it.

cpython/Lib/test/support/__init__.py

Lines 1700 to 1703 in 06b0920

	if os.path.lexists(self.real) and not os.path.exists(self.real):
	# App symlink appears to not exist, but we want the
	# real executable here anyway
	self.real = _winapi.GetModuleFileName(0)

[FFY00]

Ah, test.support.PythonSymlink is not setting PYTHONHOME correctly. It sets it based on the executable, and it knows this is incorrect on source builds, so it sets PYTHONPATH as a workaround.

cpython/Lib/test/support/__init__.py

Lines 1719 to 1721 in 06b0920

	self._env["PYTHONHOME"] = os.path.dirname(self.real)
	if sysconfig.is_python_build():
	self._env["PYTHONPATH"] = STDLIB_DIR

I am not familiar with the build on Windows, but looking at getpath I see that on source builds it set the prefix to VPATH.

[FFY00]

It's weird that it needs to set PYTHONHOME though. Did getpath perhaps not resolve the executable symlink in the past?

Anyway, the CI is still finishing, but test_platform has passed, so I guess it is fixed. That's enough debugging through CI for the day 😅