HTML Reader: Regression for figure without figcaption

[davidar]

HTML figures without a corresponding figcaption should fall back to the alt text of the img (as was the behaviour prior to #3813) rather than having no caption at all.

$ pandoc -v
pandoc 2.0.5
$ echo '<figure><img src="foo" alt="bar"></figure>' | pandoc -f html -t markdown
![](foo)
$ echo '<img src="foo" alt="bar">' | pandoc -f html -t markdown
![bar](foo)

A related issue is that figcaption is ignored when it contains block tags like div or p:

$ echo '<figure><img src="foo" alt="bar"><figcaption><p>baz</p></figcaption></figure>' \
  | pandoc -f html -t markdown
![](foo)
$ echo '<figure><img src="foo" alt="bar"><figcaption><div>baz</div></figcaption></figure>' \
  | pandoc -f html -t markdown
![](foo)
$ echo '<figure><img src="foo" alt="bar"><figcaption><span>baz</span></figcaption></figure>' \
  | pandoc -f html -t markdown
![baz](foo)

[jgm]

HTML figures without a corresponding figcaption should fall back to the alt text of the img (as was the behaviour prior to #3813) rather than having no caption at all.

I'm not sure why. In the HTML page as displayed by a browser, there will be no caption. So, why should pandoc parse this as a figure with a caption?

[davidar]

For the same reason this happens:

$ echo '<p>baz</p><p><img src="foo" alt="bar"></p><p>baz</p>' | pandoc -f html -t markdown
baz

![bar](foo)

baz

$ echo '<p>baz</p><p><img src="foo" alt="bar"></p><p>baz</p>' | pandoc -f html -t markdown \
  | pandoc
<p>baz</p>
<figure>
<img src="foo" alt="bar" /><figcaption>bar</figcaption>
</figure>
<p>baz</p>

[jgm]

A better fix for the round-trip failure you point to would be making the markdown output from the first step something like

![bar](foo)<!-- nofig -->

to avoid having it interpreted as an implicit figure.

[davidar]

Basically all I'm saying is that wrapping an img in figure shouldn't result in the alt text being completely ignored, I'm not sure what the best way to handle that would be.

[mb21]

I'm not sure why. In the HTML page as displayed by a browser, there will be no caption.

I changed the pull request accordingly.

I guess parsing the alt text when there is no figcaption will have to wait for #3177

[gwern]

I disagree about the need for waiting for #3177 (which has been 4 years and shows no particular sign of being solved soon). Pandoc already supports the necessary construct in the form of attributes, and already correctly parses and deals with img tags with alt attributes... just only in the Markdown->HTML direction:

$ echo '![](foo.jpg){alt="bar"}'                              | pandoc -f markdown -t html
<p><img src="foo.jpg" alt="bar" /></p>
$ echo '<p><img src="foo.jpg" alt="bar" /></p>' | pandoc -f html -t markdown
![bar](foo.jpg)

And of course, Pandoc even solves figures to some degree by treating it as a HTML literal... but only if you parse the HTML as Markdown, as parsing the HTML as HTML delivers worse results, erasing attributes and classes:

$ echo '<figure><img src="foo.jpg" alt="bar" /></figure>' | pandoc -f html -t markdown
![](foo.jpg)
$ echo '<figure><img src="foo.jpg" alt="bar" /></figure>' | pandoc -f markdown -t markdown
 \```{=html}
<figure>
 \```
 \`<img src="foo.jpg" alt="bar" />`{=html}
 \```{=html}
</figure>
 \```

The HTML reader just needs to be improved to handle alt/title/class attributes on imgs,

[tarleb]

waiting for #3177 (which has been 4 years and shows no particular sign of being solved soon).

We aim to tackle this in a GSoC project this summer, see #7184.

[jgm]

We could easily change the reader so that the alt was used as the caption when no caption exists. But I'm not sure this is the right way forward; it would mean that a figure without a caption became a figure with a caption after round-tripping html -> markdown -> html.