If python is installed into a symlink'ed directory, the variables
"sys.exec_prefix" and "sys.executable" can contain different paths.
Therefore, the respective test in build_ext.py fails (line 202)
and a wrong library search directory (-L.) is set.

The attached patch fixes this issue, by using os.path.samefile() instead
of the string comparison to see whether two files are identical
irrespective of their paths.

Thanks for the report and patch.

I just tested that When the executable itself is symlinked (not installed, just linked from an existing installation), sys.exec_prefix and sys.executable are different. This means that a unit test (to prevent regressions) will be easy to write with the helpers in test.test_support (2.7)/test.support (3.x)/distutils2.tests.support.

os.path.samefile is not supported on Windows before 3.2; I guess we’ll have to live with that bug and just document it :(

Do you want to add tests for this? I’ll do it if you don’t.

(Out of curiosity, why are there three signed-off-by lines? Is there a secret community of distutils bug fixers using git?)

symlinks are not supported at all on Windows before 3.2, so we’re good, except for cygwin.

Of course there is a secret distutils comunity using git! ;-) The three ppl mentioned there found and fixed the problem, that's why.

But back to the issue: Can we use a switch or wrapper function to change the test method for this path dependent on the OS? I'm asking because in windows there might be no symlink either and everything just works(tm).

> The three ppl mentioned there found and fixed the problem, that's why.
The real question behind the joke was: Are you part of the same community? Where does this fix come from?

> But back to the issue: Can we use a switch or wrapper function to
> change the test method for this path dependent on the OS?
Symlink support detection is not trivial, it depends on platform, version and user rights (the docs say Windows Vista has a notion of “right to create a symlink”). See Lib/test/symlink_support.py for some scary code.  For our bug, I have hope that a simple “hasattr(os.path, 'samefile'” will be enough to decide whether to use this function or string comparison.

Do you want to add tests for this bug?

Two people started working on this during the first Montreal sprint.