"python -m unittest" does nothing useful (runs 0 tests). It would be good to have it default to discovery, which removes the useless behaviour *and* makes the default invocation for test discovery shorter.

The existing discover subcommand would enter the deprecation process, right?

@Eric

No. "python -m unittest discover ..." will still be needed where you want to pass arguments to discovery ("python -m unittest ..." still has a different meaning where you pass arguments). All this will do is give the previously useless "python -m unittest" the same meaning as "python -m unittest discover".

Submitting a complete patch file for this:

1. Modified Lib/unittest/main.py to implement the change
2. Modified Lib/unittest/test/test_program.py to make the tests pass without warnings from regrtest.py
3. Updated the documentation in Doc/library/unittest.rst
4. Updated Misc/NEWS

Sorry for the late reply. I don't think the patch as provided is sufficient as it would interfere with people causing unittest.main() directly from a test module. Checking that TestProgram.loader is None is the way to detect we have been launched from the command line via "python -m unittest ..".

Committed revision 86649.

Michael,

I feared this might be the case. Without knowing the module deeply, it's not easy to be aware of all the possible paths, preconditions, and consequences (and tests obviously didn't catch that). A suggestion - maybe documenting these parts would allow you to get more meaningful help with the module. Or at least add the tests necessary to verify them.

Eli - I quite agree. TestProgram is a *particularly* obscure part of unittest. A much better solution (well - both would be ideal) would be to refactor the code so that it isn't so obscure.

TestProgram is an artefact of unittest's history and should be a series of functions rather than a class.