Added fix for display hook call output format

[MSeal]

Hit an issue were under complex conditions, that are hard to express in a simple notebook, the displayhook was building output array elements in the old tuple format instead of the kwarg format. This causes (ignore py2 package path in the stack trace example):

TypeErrorTraceback (most recent call last)
<ipython-input-7-dcf49c89b1af> in <module>()
----> 1 output.outputs[0]

/usr/local/lib/python2.7/dist-packages/IPython/utils/capture.pyc in outputs(self)
    112                 display(o)
    113         """
--> 114         return [ RichOutput(**kargs) for kargs in self._outputs ]

Tracing down the fix was to adjust the callable which was not updated. Tracing the capture code throughout the project was a real headache as the global state is touched a lot along the way, so I made a spec just to prove that the output format conforms rather than express where that hook callable is made consequentially.

Py2 backport PR: #11102

[MSeal]

Failure seems unrelated to PR? FAIL: IPython.core.tests.test_completer.test_omit__names?

[Carreau]

Failure seems unrelated to PR? FAIL: IPython.core.tests.test_completer.test_omit__names?

That might be due to a recent release of Jedi. I'll investigate.

[MSeal]

@Carreau Any update on this?

[takluyver]

I think this makes sense. The corresponding change to displaypub was in #10837, which shipped in 6.3. Reopening to rerun tests.

[takluyver]

...which probably won't help, because they're failing on master as well. @Carreau , did you get anywhere with the test_omit__names failure?

[MSeal]

Looks like the tests are passing now against the master changes

[Carreau]

Yes we've commented the tests.

I'm flabbergasted our bot did not react to the PR merge. I guess we also need to backport that on 6.x. and do a 5.x, and 6.x release.

[lumberbot-app]

There seem to be a conflict, please backport manually

[Carreau]

Ah, no my bad, I merged the backport, and not on master that's why.