Vertical, unescaped pipes (|) cut off comment content

[secretorchard]

Similar to #212, the usage of a vertical pipe in Twinkle warnings & reports cuts off everything after it in a message.

For example, adding this is a test of the twinkle warn-er with |link= parameters to the "Optional message" box in a warning renders as shown in this diff: https://en.wikipedia.org/w/index.php?title=User_talk:Staraction&diff=prev&oldid=1325391029.

I believe this is because the pipes are rendered as parameters in the template, instead of as plaintext.

[NovemLinguae]

Indeed. Sounds like we need a templateEscape() function somewhere that replaces = with {{=}} and | with {{!}}.

[siddharthvp]

Maybe could be done by using Morebits.unbinder to hide all templates and links, and then regex-replacing pipes with {{!}}. But we would be hoping that no nested templates are used. As the comments in #212 imply, a complete fix is quite complex.

[NovemLinguae]

There's a js wikitext parser with AST now. https://github.com/bhsd-harry/wikiparser-node

We're experimenting with it in one of my user scripts. NovemLinguae/UserScripts#269

It completely solves issues with nowiki tags, nested templates, and triple braces.

Maybe it would make sense to introduce it to twinkle, maybe not. The downside is having to include a giant library.

[siddharthvp]

wikiparser-node is also used in MediaWiki (in CodeMirror), but I don't think it's available in way that gadgets can use. Maybe we can convince upstream to expose it as an RL module. We can then load it on-demand when comment text is submitted (in csd/prod/xfd/warn).

Otherwise, we could use a simpler parser like in Morebits.wikitext.parseTemplate, which handles pipe-containing elements (templates and links) without going to the extent of handling everything. That's a lot smaller than a 155 kb library, but it's still a fair bit of code which may not be worth it as comments aren't realistically going to include nested templates.

[bhsd-harry]

There's a js wikitext parser with AST now. https://github.com/bhsd-harry/wikiparser-node

To be more specific, the library for template parsing is wikiparser-template which is about 36 kb after minimization.

[siddharthvp]

uh, seems we were overthinking this.

As this bug is for the warn module where the reason is part of a substed template, all we have to do is use Morebits.string.formatReasonText which replaces | with {{subst:!}}. This correctly handles all 3 cases - pipe being in a link, template or being free-form. PROD, XFD and RPP are already dealing with pipes this way. (This doesn't work for CSD as the reason is part of a transcluded template.)

Sounds like we need a templateEscape() function somewhere that replaces = with {{=}} and | with {{!}}.

There's no need to escape = as we always use numbers (eg. |1=) for unnamed parameters.

[secretorchard]

For what it's worth, this bug exists in reporting to noticeboards as well. I noticed it when reporting a user to SPI, in this diff: https://en.wikipedia.org/w/index.php?title=Wikipedia:Sockpuppet_investigations/Sahilsambyal&diff=prev&oldid=1325119668.

[siddharthvp]

Ok, SPI also uses substed templates. I updated the PR to cover that as well.