Ongoing focus areas
- Client support (VE, Flow, etc.)
- Parsoid HTML read views
- New directions (stable ids, incremental parse, etc.)
Identify and fix the most prominent remaining semantic roundtripping diffs.
Specifically, at least 99.95% of the 160K test pages roundtrip (wikitext -> HTML -> wikitext) without semantic errors in full roundtrip testing (which translates to significantly higher accuracy using the selective serializer in production). We want to nail down our functionality wrt semantic roundtripping and provide a reasonable metric that indicates this accuracy. Because of wikitext markup errors and edge cases, 100% is not a realistic goal.
Getting to this point not only builds confidence in Parsoid, but also enables us to advocate for other fundamental work improving wikitext.
Doing this will require categorizing semantic diffs (WIP @ https://www.mediawiki.org/wiki/Parsoid/Round-trip_testing/Diffs), improving our testing infrastructure to eliminate false positive semantic diffs (see T95258, T94861, T89628). This will help identify real errors that need fixing.
Other goals based on VE/RESTBase priorities
- Move inlined data-mw to its own attribute (T78676)
- Provide API end points for section editing (T94890 has initial ideas based on discussions between RESTBase, VE, and Parsoid)
- Support switching between HTML and wikitext editing
- Handle large / pathological pages on which Parsoid is currently timing out (T75412, T88915)