ttf-parser 0.6.2

ttf-parser

A high-level, safe, zero-allocation TrueType font parser.

Can be used as Rust and as C library.

Features

• A high-level API, for people who doesn't know how TrueType works internally. Basically, no direct access to font tables.
• A C API.
• Zero heap allocations.
• Zero unsafe.
• Zero dependencies.
• no_std/WASM compatible.
• Fast. See the Performance section.
• Stateless. No mutable parsing methods.
• Simple and maintainable code (no magic numbers).

Safety

• The library must not panic. Any panic considered as a critical bug and should be reported.
• The library forbids the unsafe code.
• No heap allocations, so crash due to OOM is not possible.
• All recursive methods have a depth limit.
• Technically, should use less than 64KiB of stack in worst case scenario.
• Most of arithmetic operations are checked.
• Most of numeric casts are checked.

Alternatives

It's very hard to compare different libraries, so we are using table-based comparison. There are roughly three types of TrueType tables:

• A table with a list of properties (like head, OS/2, etc.). If a library tries to parse it at all then we mark it as supported.
• A table that contains a single type of data (glyf, CFF (kinda), hmtx, etc.). Can only be supported or not.
• A table that contains multiple subtables (cmap, kern, GPOS, etc.). Can be partially supported and we note which subtables are actually supported.

Legend:

• ✓ - supported
• ~ - partial
• nothing - not supported

Notes:

1. stb_truetype outline parsing method is reentrant.
2. Very primitive.
3. Matching points are not supported.
4. type2 only. seac is not supported.
5. Depends on build flags.

Performance

TrueType fonts designed for fast querying, so most of the methods are very fast. The main exception is glyph outlining. Glyphs can be stored using two different methods: using Glyph Data format and Compact Font Format (pdf). The first one is fairly simple which makes it faster to process. The second one is basically a tiny language with a stack-based VM, which makes it way harder to process.

The benchmark tests how long it takes to outline all glyphs in the font.

Note: FreeType is surprisingly slow, so I'm worried that I've messed something up.

And here are some methods benchmarks:

test from_data_otf_cff2          ... bench:         775 ns/iter (+/- 75)
test outline_glyph_276_from_cff2 ... bench:         763 ns/iter (+/- 59)
test outline_glyph_276_from_cff  ... bench:         754 ns/iter (+/- 69)
test from_data_otf_cff           ... bench:         618 ns/iter (+/- 8)
test outline_glyph_276_from_glyf ... bench:         581 ns/iter (+/- 14)
test outline_glyph_8_from_cff2   ... bench:         451 ns/iter (+/- 27)
test from_data_ttf               ... bench:         400 ns/iter (+/- 9)
test family_name                 ... bench:         392 ns/iter (+/- 7)
test outline_glyph_8_from_cff    ... bench:         285 ns/iter (+/- 10)
test outline_glyph_8_from_glyf   ... bench:         252 ns/iter (+/- 8)
test glyph_name_276              ... bench:         220 ns/iter (+/- 2)
test glyph_index_u41             ... bench:          13 ns/iter (+/- 0)
test subscript_metrics           ... bench:           2 ns/iter (+/- 0)
test glyph_advance               ... bench:           2 ns/iter (+/- 0)
test glyph_side_bearing          ... bench:           2 ns/iter (+/- 0)
test glyph_name_8                ... bench:           1 ns/iter (+/- 0)
test ascender                    ... bench:           1 ns/iter (+/- 0)
test underline_metrics           ... bench:           1 ns/iter (+/- 0)
test strikeout_metrics           ... bench:           1 ns/iter (+/- 0)
test x_height                    ... bench:           1 ns/iter (+/- 0)
test units_per_em                ... bench:         0.5 ns/iter (+/- 0)
test width                       ... bench:         0.2 ns/iter (+/- 0)

family_name is expensive, because it allocates a String and the original data is stored as UTF-16 BE.

glyph_name_8 is faster than glyph_name_276, because for glyph indexes lower than 258 we are using predefined names, so no parsing is involved.

Error handling

ttf-parser is designed to parse well-formed fonts, so it does not have an Error enum. It doesn't mean that it will crash or panic on malformed fonts, only that the error handling will boil down to Option::None. So you will not get a detailed cause of an error. By doing so we can simplify an API quite a lot since otherwise, we will have to use Result<Option<T>, Error>.

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.