Traceback in cell execution error

[mupdt]

We're executing Notebooks during CI to make sure the content is always up-to-date.

However, if the execution fails, we get the following error message fro nbconvert:

[NbConvertApp] Converting notebook /u/matej/Test.ipynb to notebook
[NbConvertApp] Executing notebook with kernel: python2
[NbConvertApp] ERROR | Error while converting '/u/matej/Test.ipynb'
Traceback (most recent call last):
...UNINFORMATIVE INTERNAL STACK TRACE...
File "/u/matej/git/open-source/nbconvert/nbconvert/preprocessors/execute.py", line 113, in preprocess_cell
raise CellExecutionError(msg)
CellExecutionError
[1] 110973 exit 1 jupyter nbconvert --to notebook --stdout --execute ~/Test.ipynb

The message only says CellExecutionError, which makes it hard to pinpoint the cell and code where things went wrong.

This patch adds a traceback to the error message. It now looks like this:

[NbConvertApp] Converting notebook /u/matej/Test.ipynb to notebook
[NbConvertApp] Executing notebook with kernel: python2
[NbConvertApp] ERROR | Error while converting '/u/matej/Test.ipynb'
Traceback (most recent call last):
...UNINFORMATIVE INTERNAL STACK TRACE...
File "/u/matej/git/open-source/nbconvert/nbconvert/preprocessors/execute.py", line 113, in preprocess_cell
raise CellExecutionError(msg)
CellExecutionError: An error occurred while executing the following cell:
------------------
import dakjdnsak
------------------
ImportError: No module named dakjdnsak
[1] 111347 exit 1 jupyter nbconvert --to notebook --stdout --execute ~/Test.ipynb

[mpacer]

This looks like useful functionality and a straightforward change.

I'd prefer to see a test to make sure that we never break this functionality once it's included.

I could imagine testing against the content of a traceback might be an issue… but it seems like that would be the appropriate approach for testing this feature.

[takluyver]

I think it should be sufficient to test that the error string is shown. I.e. have a cell:

raise Exception("Can you see me?")

Then run that in nbconvert and assert that the exception shows "Can you see me?".

[Carreau]

Is that necessary ? Wouldn't message be passed to Exception if there is no init ?

[mupdt]

Oh, yes, that's right. Will remove.

[takluyver]

@mupdt982 it looks like we're waiting on a couple of changes here (test, remove unnecessary __init__). Just wanted to check you haven't forgotten about it. :-)

[mupdt]

@takluyver: I've got a fix now, will push soon.

[mupdt]

Please take another look.

• removed the explicit super constructor call,
• added tests (I reused existing notebooks for the tests; I also added a test that verifies utf-8 is handled correctly).

[takluyver]

Thanks, that looks good to me.