Amaury, that was a good suggestion, but I did with Eric's help track down a mechanism to test for the presence of the symlink creation privilege. I prefer to have a proper check rather than to attempt to create one and test for the failure message because it doesn't have the potential to be confounded with other causes. I want to avoid issues such as where is the test symlink created and how is it cleaned up. If there's still desire to go that route after looking at this implementation, I'm not opposed to switching gears and using that technique.

This patch (20) uses ctypes (in the test suite; is that okay?) to determine if the symlink creation privilege is present, falling back to the assumption that it is, and integrates that with the test.support.has_symlink function (which is new in this issue thread).

Eric, can you test this new patch in your limited user environment and report what you find?