For test discovery to work where a dependent module is optional, you end up needing to do something like what is done inhttp://hg.python.org/cpython/rev/15ddd683c321:-crypt = support.import_module('crypt')+def setUpModule():+    # this import will raise unittest.SkipTest if _crypt doesn't exist,+    # so it has to be done in setUpModule for test discovery to work+    global crypt+    crypt = support.import_module('crypt')That's kind of ugly. It would be better if unittest recognized SkipTest at import time during test discovery

Agreed and it should be easy to implement.

FTR there is already an alternative to setupmodule:try:    import moduleexcept ImportError:    module = None@unittest.skipUnless(module, 'requires module')class ModuleTests(unittest.TestCase):    passThis idiom is more lines than support.import_module, but works for non-stdlib tests too, and is useful when you don’t want the whole file to be skipped if the module is missing (like in distutils’ test_sdist where zlib can be missing).

FWIW the difference between support.import_module and the try/except/skip is that usually the former is used when *all* the tests require the imported module, whereas the latter is used when only some of the tests requires it.

Without the proposed enhancement, you could also combine Éric's approach with the original patch by doing something like:try:    support.import_module(module)except SkipTest:    module = Nonedef setUpModule():    if module is None:        raise SkipTest()

Should be: "module = support.import_module('module')"

Still, raising SkipTest from the toplevel is useful when some toplevel setup code otherwise depends on the missing module.

I agree that raising SkipTest (or subclasses thereof, such as ResourceDenied) at module level should be supported.  That would mean no changes would be needed in most of the should-be-skipped-but-fail-instead tests listed inissue 16748 to make test discovery play nicely, and in fact the changes to test_crypt could be mostly reverted.Personally, I don't find either of the suggestions given as alternates to what I did in test_crypt to be particularly prettier, not that what I did is pretty either.

Here's a patch that I believe nicely handles the raising of unittest.SkipTest at module level while doing test discovery.  It adds a _make_skipped_test function to unittest.loader, and an ``except case.SkipTest`` clause to TestLoader._find_tests.  For our own test package, this covers non-enabled resources as well.Here's some example output:"""P:\cpython>python -m unittest discover -vLib/test/ "test_curs*"test_curses (unittest.loader.ModuleSkipped) ... skipped "Use of the 'curses' resource not enabled"----------------------------------------------------------------------Ran 1 test in 0.001sOK (skipped=1)[102289 refs, 38970 blocks]"""

The patch looks good - can you add a test and documentation for this?

Sure can.  With a little luck, I'll have the patch ready later today; with less luck it'll be sometime later this week.

I think this patch should cover the test and Doc changes necessary.  Of course, let me know if it doesn't :)

Thanks for the review, Ezio.  I've made some changes and here's the new patch.

LGTM.

New changeset61ce6deb4577 by Ezio Melotti in branch 'default':#16935: unittest now counts the module as skipped if it raises SkipTest, instead of counting it as an error.  Patch by Zachary Ware.http://hg.python.org/cpython/rev/61ce6deb4577

Applied, thanks for the patch!SkipTest should probably be documented, but that's a separate issue :)

New changeset22b6b59c70e6 by Ezio Melotti in branch 'default':#16935: update test_crypt now that unittest discover understands SkipTest.http://hg.python.org/cpython/rev/22b6b59c70e6

New changesete4f9a2d2be42d5a2cdd624f8ed7cdf5028c5fbc3 by Victor Stinner in branch 'master':bpo-30813: Fix unittest when hunting refleaks (#2502)https://github.com/python/cpython/commit/e4f9a2d2be42d5a2cdd624f8ed7cdf5028c5fbc3

New changeset714afccf6e7644d21ce1a39e90bf83cb0c9a74f1 by Victor Stinner in branch '3.5':bpo-30813: Fix unittest when hunting refleaks (#2502) (#2506)https://github.com/python/cpython/commit/714afccf6e7644d21ce1a39e90bf83cb0c9a74f1

New changeset22d4e8fb99b16657eabfe7f9fee2d40a5ef882f6 by Victor Stinner in branch '3.6':bpo-30813: Fix unittest when hunting refleaks (#2502) (#2505)https://github.com/python/cpython/commit/22d4e8fb99b16657eabfe7f9fee2d40a5ef882f6