Attached is an attempt at fixing the sporadic failures of test_mailbox on the AMD64 Windows buildbot. It fails due to access errors on some directories which leads me to believe the helper functions in test.support should fix the problem.
I think the support functions just ignore errors. Isn't this going to continue to leave garbage on the buildbot filesystem without fixing the underlying problem? I wonder if this is a variation on the usual Windows access errors, in which case perhaps that is the best we can do...
Actually, the support functions (as of 3.3) attempt to work around the access errors. They attempt to wait (to a point) for a successful operation before returning to the caller. See for details. It is usually the case that the previous operation has not yet finished as to the cause of the access error.
