Hold shaping to HarfBuzz's own test suite, case by case

text/harfbuzz-suite.py reads a HarfBuzz release's test/shape/data
(in-house, text-rendering-tests, aots: fonts and inputs, many added when a
font or a shaper once went wrong) and shapes every case both ways with the
new vega-symbols shape-cases: it compiles a font's section for the scripts
its cases need (guessed as HarfBuzz guesses them), then compares each run
with HarfBuzz's, glyph for glyph and unit for unit. The oracle is HarfBuzz
run live, not the recorded outputs, which depend on options Vega has no
counterpart for (font functions, sizes, output formats). Cases asking for
what Vega does not do (user features, variations, languages, context,
cluster levels, buffer flags, vertical text) and fonts of other systems are
left out and counted by why.

text/harfbuzz-suite.known lists what differs today, and why: AAT, the
character map built from what HarfBuzz lists rather than how it maps
(format 13, symbol maps into the private use area), variation selectors,
Arabic fallback with shared glyphs, kern format 2, Zawgyi, and one run
without a plan. Any other difference fails the run; a listed case that
agrees is named. make check-harfbuzz HARFBUZZ_SRC=... runs it (make check
too, when given): 5929 cases, 5199 compared, 1355 the same.

0 files changed, 0 insertions, 0 deletions