The 1.41 branch cut left the parser cache split in three: we have a cache name for legacy content and *two different* parser caches for parsoid content, depending on whether the content is generated via API call or read views.
We'd like to clean this up for the 1.41 release and have a single parsoid parser cache. The primary patch to accomplish this is:
but it has a number of followups to correct issues found after initial deploy. This task will track those to make sure they also get backported to 1.41.
- https://gerrit.wikimedia.org/r/c/mediawiki/core/+/966317 (trivial typos in comments but will ensure we don't get spurious conflicts)
- https://gerrit.wikimedia.org/r/c/mediawiki/core/+/966525 (fixes CI issue in DiscussionTools, may have production implications)
- https://gerrit.wikimedia.org/r/c/mediawiki/core/+/966578 (fixes issues with RESTBase redirects)
- https://gerrit.wikimedia.org/r/c/mediawiki/services/parsoid/+/966558 (fixed CI issues in Parsoid, no production implications)
- https://gerrit.wikimedia.org/r/967296 ( For Internal REST API users, make revid handling lenient )
T347632: Remove ParserCache hack added to support merging ParsoidOutputAccess and ParserOutputAccess without cold cache misses is also related but I had added a separate tracking task for that when we first added it. But, adding it here for completeness since that change is also part of the ParserCache unification work.
T349310: DOMException: Invalid Character Error is somewhat related, since this bug started appearing when zhwiki pages were passed through Parsoid, but given the fact that this is related to specific bad wikitext found on WMF wikis, the patch for this (a bump in wikimedia/remex-html) is probably not strictly needed for 1.41.