Hide empty cells using CSS

[martinRenou]

Should fix #387

[maartenbreddels]

Totally agree with this. I think we stripping cells without output in the code was more like a temporary solution, this is what we had to do from the beginning.

[vidartf]

Are there not some cases where a cell can initially have no outputs, but some later code (asynchronously) adds an output to the cell? This would not normally be an issue with nbconvert, but if we are doing progressive rendering, this might be an issue. Is it possible to use the :empty CSS selector here?

[martinRenou]

Good point, I agree using :empty or :blank would be better, because it would handle this case.

[vidartf]

:blank is still in the future: https://caniuse.com/#feat=mdn-css_selectors_blank

[martinRenou]

I've got an alternate solution, I'm going to open another PR

[martinRenou]

Alternate solution: #405

[martinRenou]

@vidartf

Are there not some cases where a cell can initially have no outputs, but some later code (asynchronously) adds an output to the cell?

Do you already have a specific use case in mind? I'm trying to figure out a scenario where this could happen.

[maartenbreddels]

display supports a display_id, so that you can change outputs of previous cells. e.g.:
https://nbviewer.jupyter.org/github/jupyter/nbconvert/blob/master/nbconvert/preprocessors/tests/files/Clear%20Output.ipynb

We don't support that actually with voila progressive rendering. I wonder if we should explicitly not support that, or if we should fix this somehow.

Note that if you output an empty Output widget, it will always work (since the cell contains output in the form of a widget)

[martinRenou]

display supports a display_id, so that you can change outputs of previous cells

I also thought about this, but for having a display_id you need to display something first no? With a first display call

[vidartf]

I also thought about this, but for having a display_id you need to display something first no? With a first display call

Yes, my question basically boils down to: If you do a a display call without any data, does that still make a difference for cell.outputs?

[martinRenou]

Yes, my question basically boils down to: If you do a a display call without any data, does that still make a difference for cell.outputs?

I actually tried and it seems you can not use display with no data. I thought of making an output placeholder with display(None) and it actually displays the string "None", so there is an output.

[vidartf]

This seems to work: display()

The interesting thing is that if I try this:

> handle = display(display_id=True)

> handle.update('world')

It doesn't actually work in neither classic notebook nor lab. No errors are raised, but the update isn't actually displayed. So I don't think it is a very common pattern (but strictly speaking, I would have expected that it should work).

[martinRenou]

Interesting. I agree it should work.

[vidartf]

(note, that I'm happy with the code in this PR as-is, since neither classic nor lab supports it, just making a note for the future about it)

[martinRenou]

I do agree.

I think I prefer the alternate solution though #405. It's basically the same behavior, only that instead of hiding empty cells we simply don't send them.

[martinRenou]

@maartenbreddels would you like to have another look at this one?

[SylvainCorlay]

I am fine with this solution.

[SylvainCorlay]

@maartenbreddels I had a small comment on this one that martin is fixing now. I think this should get in before the release.

[maartenbreddels]

I think so too, failure is unrelated. I'll wait a bit more before making a release, in case we need to merge more.

[martinRenou]

Done