https://gerrit.wikimedia.org/r/#/c/91425/ will add a linkprefixcharset entry in the site info instead of the occasionally-broken linkprefix regexp. We should use that whenever it is available, but fall back to the old-style linkprefix regexp for backwards-compat.
This might also implicitly fix bug 52855.
Version: unspecified
Severity: normal
https://gerrit.wikimedia.org/r/#/q/Iaa7eaa44,n,z already changed the regexp syntax already, which breaks parsing of wikis with link prefix in production and rt testing. Test case:
node roundtrip-test.js --prefix ar 'مرس،_إب'
So we'll have to add handling for the new regexp syntax in the interim to avoid polluting the cache with broken link prefix handling.
Change 106106 had a related patch set uploaded by GWicke:
Fix link prefix and trail regexp handling