fix message built for engine dying during task

[minrk]

The test_z_<foo>_crash tests actually test this, but they have long-since been skipped because people don't like the crash dialogs some OSes give.

[minrk]

Test results for commit 88d0730 merged into master
Platform: darwin

• python2.6: OK (libraries not available: cython matplotlib oct2py pygments pymongo qt rpy2 tornado wx wx.aui)
• python2.7: OK (libraries not available: oct2py rpy2 wx wx.aui)
• python3.2: OK (libraries not available: cython matplotlib oct2py pygments pymongo qt rpy2 wx wx.aui)

Not available for testing:

[fperez]

I wonder if we shouldn't instead tag those tests with a 'crash' label, omit them by default, but occasionally run them ourselves. We wouldn't expect regular users to run them all the time, but core devs could do so when desired.

[minrk]

I marked the crash tests with @attr('crash'), and then tweaked iptest, such that if no -a args are specified, then it is equivalent to iptest -a '!crash'. To run the crash tests, you can do iptest -a crash IPython.parallel. To run them all, I can't see a better way than iptest -a crash -a '!crash', given how nose handles tag-based selection (it only seems to do selection, not exclusion).

I think we can make use of tags to improve @skip_doctest, so that 'skipped' doctests do not show up as passing tests, as they do now, but it's not quite obvious to me precisely how.

[bfroehle]

Another option would be to use an environment flag, e.g. IPTEST_... to control whether or not these tests are skipped. This would avoid the issue of needing to remember the -a '!crash' flag.