GeezerVison

DocTestCase.runTest() runs the examples with verbose=False, so there is no way to ask unittest for the details which doctest reports on its own. It now takes the verbosity from the test result, and reports every example if the test runner is asked for more than the test names.

$ ./python -m test -vv test_difflib
...
get_opcodes (difflib.SequenceMatcher)
Doctest: difflib.SequenceMatcher.get_opcodes ...
Trying:
    a = "qabxcd"
Expecting nothing
ok
...
get_opcodes (difflib.SequenceMatcher)
Doctest: difflib.SequenceMatcher.get_opcodes ... ok

The examples are written to the stream of the test runner, not to the standard output, so that they are not lost when the result buffers the output of the test.

Three changes make this reachable:

• unittest.TestResult has now a verbosity attribute, which it accepted but ignored. TextTestRunner.run() sets it, next to failfast, buffer and tb_locals, because a result class is free to filter what its constructor gets: test.libregrtest does.
• The unittest -v option is now counted, so that -vv means 3. It is parsed into the TestProgram, whose verbosity is already 1, so -v still means 2.
• The verbosity of regrtest is one less, so it is translated where the test runner is created.

A counted option is relative, so -v passed twice now means 3 rather than 2, and -q -v means 1 rather than 2.

This is the only goal of the issue which is left: the quiet levels already behave as it asks, and GH-134890 made every example a subtest.

• Issue: Better integration of doctest with unittest and libregrtest #108885