If test_heapq is run before test_import on the current py3k head,
test_import will fail as per the attached traceback.

python -m test.regrtest -W test_heapq test_import > test_import.log

At a glance I can't see any obvious reason why test_heapq should have
any effect on test_import. Raising this bug while I try to narrow down.

An extra assert inside support.make_legacy_pyc confirms that
the .pyc being renamed into does in fact already exist.

Running test_import on its own or via regrtest when not preceded
by test_heapq runs with error.