tree b42a7002902be622cd853b87f48508f8bdd89e9b
parent 0f15dc05d9332bb9dc71bd9b70bc737da2dc33c5
author Kirill Chibisov <contact@kchibisov.com> 1580086473 +0300
committer GitHub <noreply@github.com> 1580086473 +0300
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJeLjTJCRBK7hj4Ov3rIwAAdHIIAGc7SQvuBN07BNQ7Eq0MSBV/
 Obq8KOZvhPiiKBBfOs/g4tXQXzyXe0vuTmOrCBSgN2bSvBHgDClBWomnxzootFNu
 s6NfSzH/WGjMM03QBx1bL/VUH1CPhfj0CT90DyXdwpN3ZnkWMI8N+hstsEA0+UXW
 1xa4oFGYzDFq7Ax+BSMSDEnQ16v3pI/seS5w/r6jGfqdTW2W0vLeI8hF2Oc9ciCL
 H5MZUGjkFtg8ijo11kkVzJfzzs/VPYwVdr1f4M1U1KWhKiPEaYKgnoNFE0X4wB42
 vzewiKfN+1/XGz0aXw/gxoE1sVB3MslZeSYfntz6SN88oUZQubW7mc+pythCvMM=
 =PUpL
 -----END PGP SIGNATURE-----
 

Rework Fontconfig fallback to use cached list from font_sort

Previous implementation was querying Fontconfig using `charset` in a pattern,
which was leading to unpredictable fallbacks in some cases, since Fontconfig
was picking the font with the most coverage for a given charset, regardless of
user configuration. Moreover all fallback was based on font_match which is
extremely slow for such performance sensitive task as a fallback, so alacritty
had a hard times on vtebench's unicode-random-write.

The new approach is to use some internal fallback list from font_sort
and iterate over it to get a proper fallback font, since it matches the
following example query from `fc-match`:

`fc-match -s "monospace:pixelsize=X:style=Y"

That being said it's more intuitive for users to setup their system Fontconfig
fallback, and also most applications are doing similar things. Moreover the new
implementation uses internal caches over Fontconfig API when possible and
performs font matches only once during load of requested font with font_sort,
which leads to dramatically improved performance on already mentioned
vtebench's unicode-random-write.

Fixes #3176.
Fixes #3134.
Fixes #2657.
Fixes #1560.
Fixes #965.
Fixes #511.