HTML -> MD: colspan is breaking pipe tables

[marph91]

I got a HTML table

<table>


<tr>
<th class="" colspan="2" rowspan="1"><b>Subject</b></th>
<th class=""><b>Grade</b></th>
</tr>
<tr>
<td class="" colspan="1" rowspan="2">Physics</td>
<td class="">Practical</td>
<td class="">A</td>
</tr>
<tr>
<td class="">Theory</td>
<td class="">B+</td>
</tr>

</table>

and convert it by

echo '<table><tr><th class="" colspan="2" rowspan="1"><b>Subject</b></th><th class=""><b>Grade</b></th></tr><tr><td class="" colspan="1" rowspan="2">Physics</td><td class="">Practical</td><td class="">A</td></tr><tr><td class="">Theory</td><td class="">B+</td></tr></table>' | ./bin/pandoc -f html -t markdown_strict+pipe_tables-raw_html

Expected Result (Pandoc 3.6.1)

| **Subject** |           | **Grade** |
|-------------|-----------|-----------|
| Physics     | Practical | A         |
|             | Theory    | B+        |

Actual Result (Pandoc 3.8)

[TABLE]

Checking the Changelog

I couldn't find anything for pipe tables, but maybe this change in 3.7.0 is related:

Markdown writer:

• Handle row/colspans in grid tables, and expand cells when it isn’t possible to lay them out without breaking string of non-whitespace.

[jgm]

Pipe tables don't have any way to represent colspans. We'll use a grid table or raw HTML, if either is supported, but failing that we are out of options and just say [TABLE].

It looks like behavior may have been different in 3.6 -- which seems to approximate the table without colspans.

I think commit 01257ef is the relevant one.
In particular this part

-       | hasSimpleCells &&
+       | hasSimpleCells,
+         not hasColRowSpans,
          isEnabled Ext_pipe_tables opts -> do
            rawHeaders <- padRow <$> mapM (blockListToMarkdown opts) headers
            rawRows <- mapM (fmap padRow . mapM (blockListToMarkdown opts))

[jgm]

I can confirm that removing the guard not hasColRowSpans restores the original behavior, where pipe tables are used but material is crammed into non-spanned cells. @tarleb do you think we should revert to this behavior?

[marph91]

I can confirm that removing the guard not hasColRowSpans restores the original behavior, where pipe tables are used but material is crammed into non-spanned cells. @tarleb do you think we should revert to this behavior?

That would be great. In my experience, colspan and rowspan were always mapped to the expected pipe table cells. And even if the data would be misaligned, it's better than no table at all.

[tarleb]

It might make sense to keep that check, but maybe we can add another fallback clause that uses pipe tables if they are enabled, even if the table is not simple? I.e., after the HTML fallback, but before the ultimate [TABLE] fallback?

Edit: after checking the code, my suggestion would be to switch the raw HTML / last pipe table case and to remove the not hasColRowSpans guard.

[jgm]

If we switch these as you suggest, if I understand correctly, then tables with colspans would be rendered as pipe tables without colspans and not as HTML tables with colspans? This seems non-ideal, because we are getting a less correct representation of the table.

Instead I'd suggest adding a new pipe table case after the HTML case, dropping the check for col/rowspans.

An approximate pipe table is better than no table at all, but it's not better than an HTML table if the format supports HTML.

[jgm]

I implemented the above suggestion. The test case proposed above now passes.

[marph91]

Sounds well. Thanks!