#font #db #css #truetype #ttf


A simple, in-memory font database with CSS-like queries

24 releases (12 breaking)

0.13.0 Feb 21, 2023
0.11.2 Jan 9, 2023
0.11.1 Dec 26, 2022
0.10.0 Nov 8, 2022
0.2.0 Jul 21, 2020

#50 in Text processing

Download history 6220/week @ 2022-11-26 7896/week @ 2022-12-03 7037/week @ 2022-12-10 6256/week @ 2022-12-17 4596/week @ 2022-12-24 6963/week @ 2022-12-31 6663/week @ 2023-01-07 6593/week @ 2023-01-14 7572/week @ 2023-01-21 8407/week @ 2023-01-28 10701/week @ 2023-02-04 10615/week @ 2023-02-11 9281/week @ 2023-02-18 9176/week @ 2023-02-25 8801/week @ 2023-03-04 7151/week @ 2023-03-11

36,127 downloads per month
Used in 62 crates (18 directly)

MIT license



Build Status Crates.io Documentation Rust 1.49+

fontdb is a simple, in-memory font database with CSS-like queries.


  • The database can load fonts from files, directories and raw data (Vec<u8>).
  • The database can match a font using CSS-like queries. See Database::query.
  • The database can try to load system fonts. Currently, this is implemented by scanning predefined directories. The library does not interact with the system API.
  • Provides a unique ID for each font face.


  • Advanced font properties querying.
    The database provides only storage and matching capabilities. For font properties querying you can use ttf-parser.

  • A font fallback mechanism.
    This library can be used to implement a font fallback mechanism, but it doesn't implement one.

  • Application's global database.
    The database doesn't use static, therefore it's up to the caller where it should be stored.

  • Font types support other than TrueType.

Font vs Face

A font is a collection of font faces. Therefore, a font face is a subset of a font. A simple font (*.ttf/*.otf) usually contains a single font face, but a font collection (*.ttc) can contain multiple font faces.

fontdb stores and matches font faces, not fonts. Therefore, after loading a font collection with 5 faces (for example), the database will be populated with 5 FaceInfo objects, all of which will be pointing to the same file or binary data.


The database performance is largely limited by the storage itself. We are using ttf-parser, so the parsing should not be a bottleneck.

On my machine with Samsung SSD 860 and Gentoo Linux, it takes ~20ms to load 1906 font faces (most of them are from Google Noto collection) with a hot disk cache and ~860ms with a cold one.

On Mac Mini M1 it takes just 9ms to load 898 fonts.


The library relies on memory-mapped files, which is inherently unsafe. But since we do not keep the files open it should be perfectly safe.

If you would like to use a persistent memory mapping of the font files, then you can use the unsafe Database::make_shared_face_data function.




~23K SLoC