> Given that there is a specific permission that can be assigned to the
> buildslave user, it seems appropriate that the tests that attempt to
> create symlinks should fail if the user does not have that
> permission.

No, the test should *not* fail. A failing test means "the test has
completed, and the outcome was not expected, due to a bug somewhere
(the operating system, Python, or the test case proper)". This is not
the case if the link cannot be created - the resulting behavior is
exactly the correct, and expected one (under the circumstances).

> Martin, do you still prefer something that wraps tests that call
> symlink and only executes those tests if the current user context has
> that permission?

Most definitely. That's what the notion of a "skipped" test is for
(as opposed to a "failed" one).