[subbu@earth:~/work/wmf/parsoid] cat /tmp/wt
{|
|-
{|
| x
|}
a
b
c
[subbu@earth:~/work/wmf/parsoid] parse.js --normalize=parsoid < /tmp/wt
<table>
<tbody>
<tr class="mw-empty-elt"></tr>
</tbody>
</table>
<table>
<tbody>
<tr>
<td>x</td>
</tr>
</tbody>
</table>
<p>a b c</p>Unclosed tables are not that uncommon as a syntactical error category. So, we should probably try to fix this.
| Project | Branch | Lines +/- | Subject | |
|---|---|---|---|---|
| mediawiki/services/parsoid | master | +184 -138 | Pare down some things in paragraph wrapping |
So, we should probably try to fix this.
By "this", I'm going to assume you mean the paragraph wrapping of the abc. The rest of the render looks right to me.
It has "pseudo-dom-info" via table stacks .. and obviously, that goes haywire with unclosed tables. So, unclear how this can actually be fixed in a reasonable manner ... besides moving p-wrapping to the DOM (which keeps popping up every once in a while).
Alternatively, we can throw up our hands and say this is broken wikitext and behavior is undefined => we won't try to make it always do the right thing. But, if possible, we should recover from it.
I'm going to see if the table stacks actually match up with the categories grouped in core,
https://github.com/wikimedia/mediawiki/commit/1f907d500a257a8af6b7d67072843a1ab4d3becf
Notes to self ...
Parsoid's p-wrapper seems to at least get the block on a line part of it right, but the table stacks / hasOpenHTMLPTag don't exactly mimic the php parser's $inBlockElem. For example,
<p> hi </p>
parses as,
<p class="mw-empty-elt" data-parsoid='{"stx":"html"'></p>
<p>
<br/>
hi</p>
<p></p>Change 436847 had a related patch set uploaded (by Arlolra; owner: Arlolra):
[mediawiki/services/parsoid@master] [WIP] Pare down some things in paragraph wrapping
Change 436847 merged by jenkins-bot:
[mediawiki/services/parsoid@master] Pare down some things in paragraph wrapping