See pull request: https://github.com/fonttools/fonttools/pull/2326
The new module `otlLib.optimize.gpos` provides `compact` functions that
can reduce the file size of GPOS PairPos lookups by splitting subtables
in a smart way to avoid storing zero-valued pairs.
The compaction methods are called from `otlLib.builder` and
`varLib.merger` so that static and variable fonts can benefit from the
optimization at compile time.
The new module `otlLib.optimize` is also executable, to allow running
the optimization on existing fonts.
The optimization is a trade-off because on the one hand it can reduce
significantly the byte size of the GPOS table (up to 50% in random
Google Fonts) but on the other hand it adds to the compilation time and
may make fonts very slightly bigger once compressed to WOFF2 (because
WOFF2 doesn't mind about zero values and compresses them very well).
As such, the optimization is off by default, and you can activate it by
providing the environment variable `FONTTOOLS_GPOS_COMPACT_MODE=5` for
example (values from 0 = off to 9 = max file size savings, but many more
subtables).
Previously otlLib was generating None if the values themselves were
empty even if the value format was non-empty. This happened to work
for compiling to binary since the compiler handles Value=None.
But this was confusing varLib.merger module (as in when building
variable fonts from such otlLib-built master GSUB/GPOS tables, without
roundtripping to OTF/TTF binary first), because in varLib.merger,
a None means "this master doesn't provide that info; skip it"; whereas
in a PairPos table a None as generated by otlLib simply meant "all
values are zero", which is different from "this master doesn't
provide this value".
This fixes that, such that ufo2ft can build variable-font without
saving masters to binary.
Part of https://github.com/googlefonts/ufo2ft/pull/486
It's a bit gross, but I'm blaming OpenType for that. I'm doing this now because otherwise it would be even more repetitive once we start adding format 1 Rule and Ruleset subtables.
Currently unused but will be super helpful for optimizing the formats of contextual lookups.
* Splits the rules of a class contextual lookup based on explicit subtable breaks
* Returns various properties on the ruleset to help determine appropriate layout format.
* (More properties, such as "touched glyphs", planned - will be added when needed.)
