The unicode-linebreak crate treats the pipe character '|' as a break opportunity (BA/AL class). This causes ShapeSpan::build to split text like '|>' into separate ShapeWords. When these words are shaped independently, the font shaping engine cannot form ligatures that cross the word boundary.
This patch manually checks for the '|>' sequence during segmentation and skips the break opportunity, ensuring they remain in the same shaping run.
Added a unit test 'ligature_segmentation' to verify that '|>' remains a single word.
* Use HarfRust for shaping
* Replace ttf-parser with skrifa entirely
* Fix clippy lints
* Add shape plan cache
* Bump harfrust and skrifa
* Fix no_std build
* Simplify the shape plan cache
* Please the paperclip
* Cache font ID with plan
* Tune shape plan cache for "BiDi Processing" bench
- Fix string formatting with modern interpolation syntax
- Improve Debug implementation with finish_non_exhaustive()
- Fix function placement in shape.rs to avoid items_after_statements warning
- Use more idiomatic Rust patterns (map_or_else, next_back)
- Clean up conditional imports in vi.rs
- Convert multiple methods to `const` functions for optimization and consistency
- Introduce `core_maths` for enhanced no-std compatibility
- Update `Cargo.toml` for the new optional dependency and feature adjustments
- Skip expensive grapheme iteration for simple ASCII words
- 32% performance improvement for code-like text
- 5-7% improvement for general ASCII content
- No regressions for Unicode text processing
- Conservative ASCII detection maintains correctness
* Variable font support
Here's a pretty naïve solution for variable fonts.
The iterator doesn't use the match keys' weight, but instead tries to
get the requested ideal weight, if the font is variable, otherwise it is
ignored and the actual (non-variable) weight is used. This is because I
didn't implement finding variable weight support for match keys; doing
so would be impossible without parsing TTF files when matching and I
didn't want to add that potentially expensive infrastructure if not
entirely necessary.
This is a breaking change, and I'm open for ideas on how to fix that
if it's an issue.
* cargo fmt
* Add variable font example to rich-text example
We want byte indices, not code point indices. Fixes a panic when
selecting the mixed-language text at the bottom of the rich-text example
when the shaping mode is set to "Basic".
The ShapePlanCache was added to improve performance when shaping the same
strings over and over. However, it never had the ability to be trimmed
and when it was moved to FontSystem, this created a permanently growing
allocation. It is recommended to instead use the shape-run-cache feature
which supports trimming if it is desired to have higher performance for
repeated shaping, at the cost of manually specifying when to trim.
