rusttype

• Major rework to use crates ttf-parser & ab_glyph_rasterizer to respectively read and render OpenType .oft format fonts.
• Remove dependencies approx, stb_truetype & ordered-float along with in-crate rasterization code.
• Strip back some non-vital API functionality.
  • Remove support for .standalone() variants which are sparsely used.
  • Remove some functions that didn't immediately translate to ttf-parser. Please raise issues to re-add any you relied on via the new stack.

• Remove arrayvec dependency.
• Add Default implementations for geometry structs.

• Update crossbeam-utils -> 0.7.
• Update libm -> 0.2.1.

• Update arrayvec -> 0.5.

• Support no-std usage by disabling the new default feature std and using new features libm-math and has-atomics. The gpu_cache module/feature requires std.

• Use semver trick to re-export rusttype 0.8 with default-features on.

yanked

• gpu_cache: Add CacheBuilder::align_4x4 method which forces texture updates to align to 4x4 pixel boxes.
• gpu_cache: Disable multithread code and remove dependencies on wasm32.

• GlyphIter and LayoutIter provide the lifetime of the font data.

• gpu_cache: Cache::cache_queued now returns CachedBy for successes which can allow callers to tell that the texture cache has been re-ordered.

• Add fn PositionedGlyph::set_position
• gpu_cache: Update crossbeam-deque -> 0.7, use Injector for minor rasterization performance boost.

• gpu_cache: Update crossbeam-utils -> 0.6

• Update ordered-float -> 1

• Fix PositionedGlyph::pixel_bounding_box() size inconsistencies at different positions with identical sub-pixel positions.

• Rework CacheBuilder to use methods to allow non-breaking additions in future. New style is Cache::builder().dimensions(512, 512).build().
• Support multithreaded rasterization in the gpu_cache. This yields significant improvements in worst case performance when more than 1 CPU core is available. Thrashing, resizing & population benchmarks are ~3x faster on a 4-core Haswell. Multithreading is enabled by default in environments with more than a single core, but can be explicitly disabled using Cache::builder().multithread(false).
• Remove all deprecated API.
• Add Debug implementations for Font, Glyph, ScaledGlyph & PositionedGlyph
• Add and improve documentation + examples.

4-core Haswell benchmark comparison

name                                    0.6.4 ns/iter    0.7 ns/iter     diff ns/iter   diff %  speedup 
cache::multi_font_population            8,239,309        2,570,034         -5,669,275  -68.81%   x 3.21 
cache_bad_cases::moving_text_thrashing  21,589,054       6,691,719        -14,897,335  -69.00%   x 3.23 
cache_bad_cases::resizing               15,162,054       4,607,499        -10,554,555  -69.61%   x 3.29 

• Re-export rusttype 0.7 non-breaking main API, while keeping the current version of the gpu_cache module.

• Add CacheBuilder::rebuild & Cache::to_builder methods.
• gpu_cache: Only rasterize & upload after queue has successfully fit in cache producing a 1.16-1.29x speedup in resizing & thrashing benchmarks.

• Documentation clarifications
• Avoid depending on unused dependency default-features

• Add From<&AsRef<[u8]>> for SharedBytes.
• Optimise gpu_cache hashing to improve benchmark performance by ~30%.

• Optimise rasterizer removing internal hashing. Improves draw benchmark performance by 11-91%.

• Add gpu cache glyph padding option to fix texture bleeding from other glyphs when using interpolated texture coordinates near edges. Use CacheBuilder to construct a Cache that makes use of padding.
• Inlining performance improvements.

• Fix tree removal on row clear (gpu_cache).

• Improve gpu_cache performance ~6-17%

• Fix NLL migration warning

• Made the API more convenient (courtesy of @mitchmindtree, @I1048576).
• Fixes for the examples (@I1048576)
• Removed the dependency on ndarray (@I1048576)

Fixed issue #8

Fixed issue #7