text failure report is too gassy

[Martin Pool]

A typical testtools bzr failure is

======================================================================
ERROR: bzrlib.plugins.launchpad.test_lp_directory.TestDebuntuExpansions.test_ubuntu_natty_distroseries_expansion
----------------------------------------------------------------------
_StringException: Text attachment: log
------------

------------
Text attachment: traceback
------------
Traceback (most recent call last):
 File "/usr/lib/python2.4/site-packages/testtools/runtest.py", line 128, in _run_user
   return fn(*args)
 File "/usr/lib/python2.4/site-packages/testtools/testcase.py", line 368, in _run_test_method
   testMethod()
 File "/home/pqm/bzr-pqm-workdir/home/+trunk/bzrlib/plugins/launchpad/test_lp_directory.py", line 423, in test_ubuntu_natty_distroseries_expansion
   self.assertURL('ubuntu/natty/foo', 'ubuntu:natty/foo', series='natty')
 File "/home/pqm/bzr-pqm-workdir/home/+trunk/bzrlib/plugins/launchpad/test_lp_directory.py", line 383, in assertURL
   self.directory._resolve(shortcut, factory))
 File "/home/pqm/bzr-pqm-workdir/home/+trunk/bzrlib/plugins/launchpad/lp_directory.py", line 77, in _resolve
   if result.scheme in ('ubuntu', 'debianlp'):
AttributeError: 'tuple' object has no attribute 'scheme'
------------

----------------------------------------------------------------------

There is a lot of unnecessary padding here.
 * if the attachment's empty, don't show it (perhaps bzr should not attach it)
 * obviously it's a traceback, you don't need to tell me
 * the dashes and blanks are a bit excessive

[Martin Packman (gz)]

I agree with these comments, plus the top two levels of the traceback aren't needed.

Fixing this just in bzrlib is on my todo list along with bug 625597 but making the default testtools output more concise would also be good.

[Robert Collins (lifeless)]

An empty attachment is important for some projects; if bzr doesn't want it, it shouldn't attach it. (OTOH I think that that attachment has a single blank line : its not empty).

There can be many tracebacks - -0, -1, -2 etc (occurs when server threads blow up, or cleanups error). If you genuinely don't want to see the attachment label for the attachment called traceback, I guess we could do that but it seems counter productive to me.

As for the = and -, patches appreciated.

[Jonathan Lange (jml)]

The blank lines are gone in trunk, fwiw.

[Jonathan Lange (jml)]

As my patch communicates, I agree with Martin that the first attachment – the primary traceback – should be without the attachment label.

[Martin Pool (mbp)]

here's an example subunit output from bzr; thanks for tackling this