The generic assigns swap sans-serif, serif and monospace for fonts
with no CJK glyphs, so tagged Chinese text ends up on whatever Noto
CJK variant the charset scan finds first, usually the Japanese one.
Fold the many Chinese tags down to one per variant and prepend the
matching family ahead of the assigns.
* Fix Arabic falling back to Nastaliq in Chromium and Electron apps
* Cover the widened patterns too, and leave Urdu in Nastaliq
The weak binding only won the last-resort race for patterns fontconfig
had not widened. Anything whose family expands through the latin and
nonlatin alias chains -- Arial, Helvetica, Verdana, sans-serif, and so
the bulk of real CSS -- still resolved Arabic to Kufi, a display face
that suits body text no better than Nastaliq did. A strong binding wins
those too. Charset matching outranks family matching either way, so a
font that does not cover the codepoint still cannot be pulled in: Latin,
monospace, emoji, Nerd Font glyphs and CJK all resolve as before.
The untargeted rule also captured Urdu, which #6322 had deliberately
left alone by scoping itself to lang=ar. Urdu is conventionally set in
Nastaliq, so name it for lang=ur ahead of the Naskh append. Appended
rather than prepended, or it would displace the family the app asked for
and render Latin text in an Urdu locale as Nastaliq.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Mb1gPTJ6s7QyQZck7Z6F9v
---------
Co-authored-by: David Heinemeier Hansson <david@hey.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
ZacharyZhang-NY
