Shape the Universal Shaping Engine's scripts as HarfBuzz does

The USE shaper, from HarfBuzz 10.2's source, for Sinhala, Tibetan,
Javanese, Balinese, Mongolian and the rest of its eighty-odd scripts
(and Indic fonts with the new dev3-style tags). The syllable machine
reads past CGJs and ZWNJs that marks follow; rphf goes on each
syllable's first glyphs and records the repha it made, pref records its
pre-base glyph, with substitution marks cleared between; a repha moves
before the first post-base glyph and pre-base vowels move back. Scripts
that join like Arabic take Arabic's forms (Mongolian's free variation
selectors their base's), the others join syllable to syllable. Marks
lose their advances before GPOS.

text/syllabic.py carries HarfBuzz's compressed USE table over as it is,
compiles the USE grammar with the others (100,000 random runs against
Ragel's machine: no syllable differs), and checks at compile time that
the categories syllabic.h names are the machines' own.

Noto Sans Sinhala, Javanese, Balinese, Mongolian, Noto Serif Tibetan,
FreeSerif's Sinhala and Unifont's five shape as HarfBuzz does at 5,000
runs a plan; all installed fonts (1,748 plans) and 21 Noto fonts too.

1 files changed, 6 insertions, 4 deletions

diff --git a/text/shape.h b/text/shape.h
index 7f02729..bd38d96 100644
--- a/text/shape.h
+++ b/text/shape.h
@@ -6,8 +6,8 @@
 
 /* Shaping (docs/design/004-text.md): a run of characters (one script, one
  * direction) to glyphs and their positions, as HarfBuzz 10 shapes it with
- * its default, Arabic, Hebrew, Hangul, Indic, Khmer and Myanmar shapers,
- * glyph for glyph and unit for unit.
+ * its default, Arabic, Hebrew, Hangul, Indic, Khmer, Myanmar and Universal
+ * (USE) shapers, glyph for glyph and unit for unit.
  *
  * What a font's shaping needs travels in a symbol file (its shaping
  * section), written by the build tool:
@@ -48,7 +48,7 @@ enum { VEGA_SHAPING_HEADER = 80, VEGA_PLAN_BYTES = 144, VEGA_STEP_BYTES = 8, VEG
        VEGA_PLAN_STAGES = 5 };
 enum { VEGA_SHAPE_LTR = 1, VEGA_SHAPE_RTL = 2 };
 enum { VEGA_SHAPER_DEFAULT, VEGA_SHAPER_ARABIC, VEGA_SHAPER_HEBREW, VEGA_SHAPER_HANGUL, VEGA_SHAPER_INDIC,
-       VEGA_SHAPER_KHMER, VEGA_SHAPER_MYANMAR };
+       VEGA_SHAPER_KHMER, VEGA_SHAPER_MYANMAR, VEGA_SHAPER_USE };
 /* Plan flags: GPOS runs; mark advances are zeroed after it (by GDEF); glyph
  * classes come from Unicode (the font has none); the font positions marks
  * (a 'mark' feature); automatic fractions; the kern table kerns; marks are
@@ -66,6 +66,7 @@ enum { VEGA_MASK_RPHF = VEGA_MASK_SHAPER, VEGA_MASK_PREF, VEGA_MASK_BLWF, VEGA_M
        VEGA_MASK_PSTF, VEGA_MASK_INDIC_INIT };                                /* Indic */
 enum { VEGA_MASK_KHMER_PREF = VEGA_MASK_SHAPER, VEGA_MASK_KHMER_BLWF, VEGA_MASK_KHMER_ABVF, VEGA_MASK_KHMER_PSTF,
        VEGA_MASK_KHMER_CFAR };                                                /* Khmer */
+enum { VEGA_MASK_USE_RPHF = VEGA_MASK_INIT + 1 };         /* USE: that, and Arabic's for its joining forms */
 /* The stages a plan names: Indic's, whose features decide a consonant's
  * position by whether they would substitute it. */
 enum { VEGA_STAGE_RPHF, VEGA_STAGE_PREF, VEGA_STAGE_BLWF, VEGA_STAGE_PSTF, VEGA_STAGE_VATU };
@@ -75,7 +76,8 @@ enum { VEGA_INDIC_OLD_SPEC = 1 };
 /* Step flags; pauses, which the shaper tells apart. */
 enum { VEGA_STEP_AUTO_ZWNJ = 1, VEGA_STEP_AUTO_ZWJ = 2, VEGA_STEP_RANDOM = 4, VEGA_STEP_PER_SYLLABLE = 8 };
 enum { VEGA_STEP_PAUSE = 0xFFFF };
-enum { VEGA_PAUSE_SYLLABLES = 1, VEGA_PAUSE_INITIAL_REORDERING, VEGA_PAUSE_FINAL_REORDERING, VEGA_PAUSE_CLEAR_SYLLABLES };
+enum { VEGA_PAUSE_SYLLABLES = 1, VEGA_PAUSE_INITIAL_REORDERING, VEGA_PAUSE_FINAL_REORDERING, VEGA_PAUSE_CLEAR_SYLLABLES,
+       VEGA_PAUSE_CLEAR_SUBSTITUTED, VEGA_PAUSE_RECORD_RPHF, VEGA_PAUSE_RECORD_PREF };
 
 typedef struct { const unsigned char *p; uint32_t size; } VegaTable;
 /* Which glyphs may be in a set, as three 64-bit patterns of glyph bits