subunit-filter --no-success doesn't filter out all successes

I don't know the internal details, but I'm definitely getting some weird results from this command. The output here is a test run from pqm where there was one failure, and I'm trying to figure out what that was:

$ zcat pqm-stdout.gz | subunit-filter --no-skip --no-success --no-passthrough | subunit2pyunit
bzrlib.tests.blackbox.test_info.TestInfo.test_info_locking ... ok
bzrlib.tests.blackbox.test_revert.TestRevert.test_revert_dirname ... ok
...

There are 40 lines that end with 'ok'. (And otherwise 77 lines of output).

This is with subunit trunk:
  132 Jonathan Lange 2010-08-05 [merge]
      Move the bulk of subunit-ls into an importable Python module.

Which seems to be post the 0.0.6 release.

Also note that I checked with 'subunit-stats' to make sure:
$ zcat pqm-stdout.gz | subunit-filter --no-success --no-skip --no-passthrough | subunit-stats
Total tests: 41
Passed tests: 40
Failed tests: 1
Skipped tests: 0
Seen tags:

So --no-success leaves me with 40 *passed* tests. (These may be KnownFailure?)

Note that I *want* failures and errors, but *only* those. If I just use subunit-filter and do some manual filtering (for all the 'time' comments, etc that are just noise), I do find these lines:
test: bzrlib.tests.blackbox.test_info.TestInfo.test_info_locking
xfail: bzrlib.tests.blackbox.test_info.TestInfo.test_info_locking [ multipart

I was thinking to change the bzr test suite to actually filter out more junk, because right now I get a 7MB file with 99.9% of that garbage talking about skipped tests. (And my mail host is configured to reject emails >10MB, so if I have a suite with actual failures, it can easily exceed the cap.)

However, even without that, there still seems to be something wrong if I'm getting 'ok' from subunit2pyunit, but I'm passing --no-success and --no-skip to the rest of the code.

I'm uploading the pqm output, but I'll note that I've gotten this behavior with *all* pqm output that I've analyzed.

It may be that we just need "--no-xfail" ?