This is illustrated by this notebook:
https://gist.github.com/anonymous/e6cb1acd8bfbfda35bf1
The code in the notebook is adapted from an example in the matplotlib docs. I've modified the example so that the ax_update function simply raises an error instead of doing anything useful. This is supposed to represent a buggy callback function that raises an exception.
Using the nbagg backend, if I run the notebook and then zoom in somewhere on the right-hand axis, I do not get a traceback printed in the notebook despite the fact that ax_update is being called and raising an exception.
Using the macosx backend doing the exact same thing does print out a traceback after manipulating the right-hand axes in the GUI window.
Not getting information about errors raised by callback functions makes it very difficult to debug callbacks using the nbagg backend. It would be much better if errors were raised to the level of the user.
This is illustrated by this notebook:
https://gist.github.com/anonymous/e6cb1acd8bfbfda35bf1
The code in the notebook is adapted from an example in the matplotlib docs. I've modified the example so that the
ax_updatefunction simply raises an error instead of doing anything useful. This is supposed to represent a buggy callback function that raises an exception.Using the nbagg backend, if I run the notebook and then zoom in somewhere on the right-hand axis, I do not get a traceback printed in the notebook despite the fact that
ax_updateis being called and raising an exception.Using the macosx backend doing the exact same thing does print out a traceback after manipulating the right-hand axes in the GUI window.
Not getting information about errors raised by callback functions makes it very difficult to debug callbacks using the nbagg backend. It would be much better if errors were raised to the level of the user.