Notebook calls both __repr__ and _repr_html_

[EntilZha]

Observed Behavior:

1. Define _repr_html_ on class
2. In Jupyter/IPython notebook create an instance of the class
3. Showing the variable by it being the last line in a cell calls both __repr__ and _repr_html_

Expected Behavior:

Jupyter/IPython notebook should call only _repr_html_ if it is defined.

Rationale:

1. The return of repr is never shown to the user so there is no need to call it if _repr_html_ is defined already.
2. For object's whose repr function is expensive to call (eg large collections), _repr_html_ can be implemented efficiently by only showing a subset of the collection. This is allowed since although repr in general should adhere to its output being runnable by eval, _repr_html_ does not have to adhere to this since it produces HTML.

Minimal Example

class Custom:
    def __repr__(self):
        raise Exception("oops called repr")

    def _repr_html_(self):
        return "<h1>HTML repr</h1>"

In jupyter cell:

a = Custom()
a

This results in: https://gist.github.com/EntilZha/6b81fc9ced2540a5b7acdebc37c66346

Note, the HTML output is still printed despite the exception.

System Information

• Mac OS X 10.11.5
• Safari 9
• IPython 5.0.0
• Jupyter 1.0.0
• Python 3 via brew

Thanks!

[takluyver]

Sorry, this is by design. Although the notebook frontend prefers HTML, we store any output formats that are available, because if you convert the notebook document to other formats, a different output format may be used. We've discussed before whether we might filter e.g. image types for plots, but I don't think we'd ever skip the plain text repr.

The eval-able repr is a guideline rather than a rule, and there are common cases where large objects do not provide a full repr:

In [2]: numpy.zeros((1000, 1000))
Out[2]: 
array([[ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       ..., 
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.],
       [ 0.,  0.,  0., ...,  0.,  0.,  0.]])

[EntilZha]

Makes sense. I am guessing providing an option flag to tell ipython not to call repr also gets messy so the best solution is to truncate repr in the first place.

[takluyver]

If you want to provide a different repr in IPython, I think defining the _repr_pretty_(), which is intended for customising pretty-printing, should override __repr__. I haven't checked that __repr__ is not called at all in that case, though.

[EntilZha]

Still calls __repr__ even if _repr_pretty_ is defined. I think in any case it would be better for me to make __repr__ truncate since the main use case for it is inspecting data and I have various functions for persisting said data anyway. Thanks for the pointer to _repr_pretty_, might implement that as well possibly

[Carreau]

Closing as this is expected behavior.