mirror of
https://github.com/hanzoai/font.git
synced 2026-11-02 23:42:52 +00:00
zeekay
e738240139
The Arabic table now holds a third companion, and every companion must draw
every letter the Arabic-script catalogs in @hanzo/i18n write ("needs":
Arabic, Persian, Dari, Urdu, Pashto, Sindhi, Sorani Kurdish, Uyghur,
Shahmukhi Punjabi). arabic.py refuses a face that does not. Measured
against those catalogs, Cairo, Readex Pro and Rubik each miss letters
Pashto, Sindhi, Kurdish or Uyghur write; Noto Kufi, Noto Sans and IBM Plex
Sans Arabic miss none.
Plex ships as IBM published it, one woff2 per weight, byte for byte: the
OFL reserves "Plex", and a subset is a Modified Version that may not carry
it. So a face is either cut (the Noto pair, as before) or whole, and a
face lists its files. The default must still be one variable file over
Zen's 100-900, since a static face under 'Zen' wins the weight match for
Latin; Plex is offered under its own name only.
The publish check reads the files to expect from the table.
Claude-Session: https://claude.ai/code/session_015XYKLjZtDTJEnMo4eK3AFH