Upgrade nbconvert

[martinRenou]

Revive #1003 and merge with #1014

closes #1003
closes #1014
closes #959

[minrk]

@martinRenou this mostly works now, but the lab template's font-awesome 5 is conflicting with the font-awesome 4 used on the nbviewer page, erasing the icons:

Not sure if there's a simple fix for that.

[martinRenou]

Thanks a lot for pushing on this @minrk !

but the lab template's font-awesome 5 is conflicting with the font-awesome 4 used on the nbviewer page, erasing the icons:

In Voila we do the following, maybe that will fix the nbviewer case:

<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@fortawesome/fontawesome-free@^5/css/all.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@fortawesome/fontawesome-free@^5/css/v4-shims.min.css" type="text/css" />

See https://github.com/voila-dashboards/voila/blob/2b6694191324109c884dc48cc93073977fdacca6/share/jupyter/voila/templates/lab/index.html.j2#L20

I am not sure where to add this though. I wonder if this should go directly in nbconvert.

[minrk]

We should update to fa5, but this backward-incompatibility is definitely an issue. The shims do work in layout.html (after our own css), but result in a different size than before:

before:

after:

[Carreau]

The navbar seem to be missing visually as well.

[Carreau]

One another big problem is that some notebooks completely fail.

For example, IRuby from the main page:

      File "/Users/bussonniermatthias/miniforge3/envs/nbv/share/jupyter/nbconvert/templates/lab/base.html.j2", line 162, in block 'data_svg'
        {{ output.data['image/svg+xml'] | clean_html }}
      File "src/lxml/html/clean.py", line 562, in lxml.html.clean.Cleaner.clean_html
      File "/Users/bussonniermatthias/miniforge3/envs/nbv/lib/python3.10/site-packages/lxml/html/__init__.py", line 873, in fromstring
        doc = document_fromstring(html, parser=parser, base_url=base_url, **kw)
      File "/Users/bussonniermatthias/miniforge3/envs/nbv/lib/python3.10/site-packages/lxml/html/__init__.py", line 759, in document_fromstring
        value = etree.fromstring(html, parser, **kw)
      File "src/lxml/etree.pyx", line 3254, in lxml.etree.fromstring
      File "src/lxml/parser.pxi", line 1908, in lxml.etree._parseMemoryDocument
    ValueError: Unicode strings with encoding declaration are not supported. Please use bytes input or XML fragments without declaration.

[Carreau]

I think listening on 0.0.0.0 by default is a security problem. Well, here on macos, it present me from doing so without root access so It's ok-ish, but i'd like to fix it.

[Carreau]

Slides are also quite broken.

For example you can compare with https://nbviewer.org/format/slides/github/astrojuanlu/talk-dataframes/blob/main/slides.ipynb

[Carreau]

Most of the reveal issue seem to be coming down to font-size being way to big.

Something like the following in the right place make it bearable.

div.reveal {
    font-size: 10px;

}

Now we do seem to have 2 sets of reveal controls on the page if you look carefully:

[Carreau]

So there are indeed 2 reveal. One done by nbconvert, and one by nbviewer.

Really this is crazy that we are embedding a full nbconvert with body|safe, that even include DOCTYPE ????. And of course the Reveal from nbconvert has no hooks, so we can't listen to the slide change event to hide the the nbviewer header.

The new nbconverts are really backward in term of reusing the templates...

[Carreau]

That is the part we can't do without changes to nbviewer.

[Carreau]

I think that's about as far as I we can reasonably take it in a single PR. It works, but there is still alot of work to do.

[Carreau]

I think we should leapfrog nbconvert 6.x and go directly to 7.x, I'm going to create a 1.x branch that is the current main, and suggest main become 2.x.

[minrk]

If that makes it easier, by all means!

[Carreau]

If that makes it easier, by all means!

It's more on the communication side, there will be likely so much change between 1.x and main, that it does make sens to bump major version number. This also let us re-do a 1.x at one point if ever necessary.

I've push a 1.x branch, and I'm going to merge this and keep iterating, once this is merged, main should not be considered stable as a number of notebook likely won't render yet, and reveal is still partially broken .

[martinRenou]

Thanks a lot for pushing on this!

[mgeier]

Would it be possible to get another nbconvert upgrade on https://nbviewer.org/?

I think the latest release should fix an issue where SVG images are not displayed correctly: jupyter/nbconvert#1863.
See also #1037.