DOC: limit error traceback to one line for expected exceptions (GH10715)

[jorisvandenbossche]

Closes #10715

This gives you an output like

In [4]: s = pd.Series([1,3,5,np.nan,6,8])

In [5]: s.iloc[11]
---------------------------------------------------------------------------
IndexError: single positional indexer is out-of-bounds

for input

.. ipython:: python
   :okexcept:

   s = pd.Series([1,3,5,np.nan,6,8])
   s.iloc[10]

I think we would always want such a truncated traceback in the docs in cases that exceptions are expected, so I don't think it is needed to make it an extra new option.
I only should check for cases where the actual message spans multiple lines (now it is hardcoded to give me the last line of the error traceback)

@jreback with this we don't have to use the code-block s in eg whatsnew for the expect behaviour of to_datetime

[jreback]

@jorisvandenbossche

yep, actually should prob go thru the code and fix up cases where we are doing this (so that it shows the actual trace-back rather then the copy/pasted code-block). but that is prob some work; maybe make a new issue.

[shoyer]

Looks good, although my inclination is to remove one more line of the traceback:

In [5]: s.iloc[11]
IndexError: single positional indexer is out-of-bounds

[jorisvandenbossche]

@shoyer yep, that's also certainly possible. I deliberately included the first (the line) and the last line (the message), in the idea that in this way it stands out a bit clearer as an error message. But if people prefer it without line, no problem for me and easy to change.

[jreback]

I prefer having just one line of the error message as @shoyer suggests.

[jorisvandenbossche]

Updated with one line (already had it done locally, but forgot to push ..)

[jreback]

@jorisvandenbossche doc builds are stopping since this commit
https://travis-ci.org/pydata/pandas/jobs/76637871#L1800

any thoughts?

[jorisvandenbossche]

Hmm, stupid I didn't look at the travis builds (it builds OK locally). Thanks for noting!

It halts in enhancingperf.rst, but I did not change anything there. The only reason I could think is because there is one okexcept in that file. But locally that one is parsed fine (will try later on a linux machine)

[jreback]

gr8 thanks!