failures are not easily visible in pqm make check error output

[Martin Pool]

pqm failures messages, which I think are the result of 'make check', start out with a lot of output about skipped tests. This is really not helpful: I want to know about the actual failures. There are subunit files if more information is needed.

example:
bzr selftest: /home/pqm/pqm/bin/pqm
  /home/pqm/source/bzr.dev/bzrlib
  bzr-2.1.3dev python-2.6.5 Linux-2.6.32-27-server-x86_64-with-Ubuntu-10.04-lucid

blackbox.test_dpush.TestDpushStrictWithChanges.test_push_with_revision(uncommitted) SKIP 0ms
Text attachment: reason
------------
dpush does not handle --revision
------------

blackbox.test_dpush.TestDpushStrictWithChanges.test_push_with_revision(pending-merges) SKIP 0ms
Text attachment: reason
------------
dpush does not handle --revision
------------

blackbox.test_dpush.TestDpushStrictWithChanges.test_push_with_revision(out-of-sync-trees) SKIP 0ms
Text attachment: reason
------------
dpush does not handle --revision
------------
....

[Martin Pool (mbp)]

The simplest way to fix this would be to just change 'make check' back out of using subunit output.

Pros:
 * the failure email will be more concise
 * the subunit stream seems reasonably often not to be able to be parsed by subunit so it's not reliably useful
 * other people running make check will not be surprised by reams of output
 * this will avoid some complexity about checking the output that's sometimes caused pqm to fail open

Cons:
 * perhaps people do like subunit? (I like the idea in general but I don't find it very useful with pqm).
 * changing it might break something in pqm

[Martin Pool (mbp)]

lp:~mbp/bzr/760387-no-subunit turns off subunit. On the whole I think I'd like to try without it.

[Robert Collins (lifeless)]

FWIW I find subunit errors very useful and reliable. Of course, I'm
not hacking on bzr now, but we get them back from ec2 with Launchpad
pretty well.

[Martin Pool (mbp)]

Robert and spm have cowboyed a change that will make pqm not report skipped tests; if that works we can commit it to pqm.