Yes, tests are only run after a change is committed and pushed into
Mercurial; this is done by BuildBot https://www.python.org/dev/buildbot/ .

So it's a good idea to run tests before submitting a patch or committing a change.  No matter how trivial a change seems, it should always be tested first.  Every programmer has a few stories of "this can't possibly fail" changes that fail, sometimes spectacularly.

(One of mine: I rewrote some C string-handling code for a product
that supported 4 or 5 different Unixes and processor architectures,
tried it on one of them, and concluded it was fine.  It segfaulted on
exactly one architecture.  Unfortunately this was discovered by a VP
who was demoing to a customer at the time.  I got a talking-to about
that one.)

Running the tests finds a simple problem: there's no longer an 'import
zipfile' statement.  I'll add the import inside the _make_zipfile()
function.  This is against PEP 8, strictly speaking, but it means
importing shutil doesn't always import zipfile; it'll only import the
module if it's actually needed.  (I'll probably do the same for the
import of tarfile.)

Derek, thanks for your patch!