rustybuzz

rustybuzz is an attempt to incrementally port harfbuzz's shaping algorithm to Rust. But while harfbuzz does a lot of things (shaping, subsetting, font properties querying, etc.), rustybuzz is strictly a shaper.

You can use it already, since we are simply linking the harfbuzz statically. And we're testing rustybuzz against the harfbuzz's test suite.

The final goal is to produce exactly the same results as harfbuzz.

The current progress can be found at CHANGELOG.md

Embedded harfbuzz version: 2.7.0

Major changes

• Subsetting removed.
• Malformed fonts will cause an error. HarfBuzz uses fallback/dummy shaper in this case.
• Most of the TrueType and Unicode handling code moved into separate crates.
• rustybuzz doesn't interact with any system libraries and must produce exactly the same results on all OS'es and targets.

Roadmap

• [ ] Port OpenType tables:
  • [ ] GDEF (ttf-parser already supports it, but we have to integrate it with rustybuzz)
  • [ ] GPOS
  • [ ] GSUB
• [ ] Port Apple tables:
  • [ ] Make macos tests OS-independent (ultra hard).
  • [ ] ankr (easy)
  • [ ] feat
  • [ ] kern (ttf-parser already supports it, but we have to integrate it with rustybuzz)
  • [ ] kerx
  • [ ] mort
  • [ ] morx
  • [ ] trak
• [ ] Port text normalization (hb-ot-shape-normalize.cc, easy).
• [ ] Port fallback shaper (hb-ot-shape-fallback.cc, easy).
• [ ] Port shaper logic (hb-ot-shape.cc).
• [ ] Remove C++ sources.

Performance

At the moment, performance itsn't that great. We're 1.5-2x slower than harfbuzz. Mainly because we have a lot of FFI glue code, which doesn't work well with inlining. And we have to constanly convert between C++ and Rust types (like char for example). And also, because rustybuzz doesn't support shaping plan caching at the moment.

See benches/README.md for details.

Prior work

This is mine yet another attempt to port harfbuzz to Rust. The previous attempt can be found at rustybuzz-old.

This time I'm focusing on delivering a well-tested bindings for a stripped-down harfbuzz fork with some parts reimplemented in Rust.

The problem is that harfbuzz code is very interconnected and it's hard to swap out random parts. Also, despite having almost 1800 tests, there are still a lot of white spots which I will try to fill first. Since it's way too easy to miss some important edge-case.

License

rustybuzz is licensed under the MIT.

harfbuzz is licensed under the Old MIT