BUG: _repr_latex is called in jupyter qtconsole

[chris-b1]

xref #11778

[jreback]

hmm, is this on master?

[jorisvandenbossche]

There are some issues at IPython/Jupyter for this: jupyter/qtconsole#65, jupyter/qtconsole#56 (IRkernel has the same problem for the display of their dataframes)

@chris-b1 Would you be able to test if this is solved in qtconsole master?

[jorisvandenbossche]

I think this is a reason to consider reverting #11778

[jreback]

cc @nbonnotte

[jorisvandenbossche]

Or, another option is making the use of the _repr_latex_ optional, and defaulting to False. So somebody who wants this feature for nbconvert, can switch it on, but knowing that it will bug in qtconsole.

[jreback]

so we could have an option display.latex.repr=False maybe?

[chris-b1]

Yeah, this also happens on qtconsole master. If there was a way to make this in_qtconsole check work again, that'd be a way to avoid an option.

But since they've restructured all the ipython/jupyter stuff, I don't know how (if it's possible) to do that.

[nbonnotte]

This goes beyond my current use of pandas or jupyter, so I'm not really qualified to say anything.

That being said, I like the idea of being able to nicely export my notebooks containing dataframes to LaTeX. And, if I understand correctly, this is more a bug of the jupyter qtconsole rather than pandas, right? So, if anything, my current understanding of the situation does not convince me that reverting #11778 is absolutely necessary. But maybe it's just because I use jupyter notebooks, and I feel closer to exporting them to LaTeX rather than using the qtconsole.

[jorisvandenbossche]

And, if I understand correctly, this is more a bug of the jupyter qtconsole rather than pandas, right? So, if anything, my current understanding of the situation does not convince me that reverting #11778 is absolutely necessary. But maybe it's just because I use jupyter notebooks

The qtconsole is used in eg Spyder, which I suppose quite some pandas users will use.
To be clear, ideally we find another solution or workaround so reverting #11778 is not needed. But if not, we cannot ship pandas like this in any case, as it would break the interactive usage for lots of users.

[jorisvandenbossche]

A possible idea is to do something similar as what sympy does, so that the user has to call a function to activate the latex repr support (sympy.init_printing()).
E.g. a pd.options.display.init_latex_repr() (although this being a function is not really consistent with the other options). Calling the function would then add the _repr_latex_ method to DataFrame on the fly.

[nbonnotte]

Would it still be possible to convert a notebook to LaTeX?

And a naive question: what prevents in_qtconsole from working? Is it because jupyter does not want us to know to which front-end it is connected?

[jorisvandenbossche]

Would it still be possible to convert a notebook to LaTeX?

Yes, but you will just have call this function in the beginning of your notebook that you want to convert to latex to activate the latex repr

And a naive question: what prevents in_qtconsole from working? Is it because jupyter does not want us to know to which front-end it is connected?

As far as I understand, yes, the kernel does not know in which frontend it is run (notebook or qtconsole or ... are all ipykernel.zmqshell.ZMQInteractiveShell ). And if you want to convert a notebook to nbconvert, the latex repr should already be stored inside of the notebook.

[jreback]

@jorisvandenbossche so what should we do about this, revert the actual _repr_latex_ method?

The other changes are fine.

[jorisvandenbossche]

I think the best would be something like the following:

Have a new pd.options.display.latex_repr option, which defaults to False, and when setting it to True, this triggers a callback function that adds the _latex_repr_ method dynamically to DataFrame.

[chris-b1]

It doesn't even need to be a callback, _repr_latex can just return None (if the option is False) and the qtconsole will fallback. I'll submit a PR here shortly.

[jorisvandenbossche]

Ah, yes, that seems also to work. Even simpler!