Nbconvert: Respect isolated metadata flag when converting to HTML

[takluyver]

The live notebook looks for an "isolated": true flag in the metadata to put SVG/HTML in an iframe. However, conversion to static HTML appears to ignore this flag.

I suspect that this is the cause of IRkernel/IRkernel#198.

[flying-sheep]

maybe this can work with static documents by doing <iframe src="data:...">

[takluyver]

It looks like there's a srcdoc attribute, so you can do <iframe srcdoc="<h1>Boo</h1>"/>. Ugly, but better than a base-64 data URL. Not supported in IE, though.

[flying-sheep]

neither does it support src, so let’s use this for IE.

srcdoc sounds good. we only need to escape " and put it into srcdoc then?

[flying-sheep]

maybe we should special-case SVG and use <img src="data:...">?

[takluyver]

I think the srcdoc approach we described before is nicer than base64-ing text data.

[flying-sheep]

i meant using imgs generally for SVG. but the most important thing is to get it in :wink:

[mpacer]

Is this still open?

Also, given the concerns elsewhere (https://github.com/jupyter/nbconvert/issues/328) that seem pro data URIs why is there anti data URI sentiment here? Is that not a valid solution?

[flying-sheep]

the other thread doesn’t mention srcdoc, so that should work.

HTML/XML attribute values (strings) can span multiple lines, so we just need to think about quoting. would it be enough to

1. replace (in the HTML code) & with & (so no entities will be expanded in the string)
2. replace " with " (so the string isn’t ended prematurely)

this would result in this doc

<div title="baz&quot;" id='a'>
  &hellip;
</div>

being emedded like so:

<iframe srcdoc="<span title=&quot;baz&amp;quot;&quot; id='a'>
    &amp;hellip;
</span>"></iframe>

which makes it still more readable than base64 encoded data.

to make it even more readable, we could just replace the & in all entities that expand to ". there’s ", the decimal and hexadecimal entities, and what else?

[takluyver]

@michaelpacer data URIs are an OK if somewhat ugly way of embedding binary data like images into a text-based format like HTML. Encoding HTML, which is already text, into unreadable base64 data seems rather inelegant. That's why I prefer the srcdoc approach.

[mpacer]

I can revisit this if desired after the 5.2 release.

[flying-sheep]

this is definitely still wanted, thank you!

[MSeal]

Removing milestone altogether as no one is pushing this change into PR form.