> With patch #25, I get 6 test failures: test_glob, test_os,
> test_platform, test_posixpath, test_shutil, and test_tarfile.
> 
> This is interesting because the failures are different depending on how
> I compile Python. Running in WOW64 (32-bit Python on 64-bit Windows),
> the tests fail for not having the required privilege.

I'm unable to reproduce this. I'm using an automated script to checkout, patch, build, and test the code. You can probably use the same script as me; it's available as https://svn.jaraco.com/jaraco/python/incubator/test-python-symlink-patch.py . It should work on your system so long as you have 'patch' and 'svn' in the path and Visual Studio 9.0 in the default location. Would you be willing to try it, so at least we can eliminate the technique as a variable?

When I run the script as guest or admin, the 32-bit regression tests all seem to pass (I occasionally get a transient failure in test_os). The 64-bit ones, however, fail even without applying the patch. I haven't yet figured out why this is. There are plenty of failures in the build, so I was guessing the build is broken.

The fact that it's failing for not having the required privilege indicates to me that the lib/test/symlink_support.py isn't loading properly. Can you try 'from lib.test import symlink_support' and let me know what response you get?

> Running a 64-bit
> Python, the tests fail various asserts (see attached
> patch25results_python64.txt).

These asserts look very much like the ones I was seeing early on before I had addressed symlink artifacts in the tests. Why you would experience different behavior in the different architectures is still unclear.

In the meantime, I'm going to try to see if I can find a case where the import of symlink_support would fail unexpectedly.