When the subset serializer deduplicates a LigatureSet that is reachable from several coverage entries of one LigatureSubst, pop_pack() links all of those entries to a single LigatureSet vertex. The LigatureSubst splitter numbers ligatures per coverage entry and then edits the LigatureSet vertices in place (move_children, compact_liga_set, shrink), which assumes one vertex per entry, so the shared entries silently lose their ligatures. #6088 removed the duplicate_shared_liga_sets pre-pass that used to guard this, on the assumption that as_mutable_table()'s copy-before-mutate made it redundant; but that only un-shares a child shared with *other* parents, not one reached by several links from the same parent, so the bug came back. Restore the pre-pass, but make it linear in both time and memory, since a naive restore is O(n^2) on a subtable whose n coverage entries all share one set: - Memory: LigatureSet::subset()/Ligature::subset() each add one virtual link to Coverage ("ensure Coverage is packed after this") per subset call, and pop_pack()'s merge_virtual_links appends them to the deduped vertex without collapsing, so the shared set accumulates n redundant virtual links. duplicate(copy_table) copies that whole vector into every per-entry clone, i.e. O(n^2). unshare_duplicate_children() collapses each set's duplicate virtual links to one (dedup_virtual_links, fixing the dropped links' incoming-edge counts) before cloning, so each clone copies O(1), and does the whole un-share in a single pass over the parent's links instead of an O(n) remap-and-rescan per entry. - Time: the split's analysis passes (compute_split_points, total_number_ligas, liga_counts) and clone_range() each did an O(n) index_for_offset() per entry. Precompute the ligatureSet index -> object id map once per split and pass it to the analysis passes, and in clone_range() move each run of fully-contained sets with a single move_children() (and look moved sets up through the new, small subtable) instead of a move_child()/as_mutable_table() per entry. The un-share runs after the analysis passes: the deep copies share the same Ligatures and keep the same ligature.len, so the counts and split points are unchanged. Testing: the K86 correctness reproducer now loses 0 of 8400 ligatures (was lossy). The 1P5QWD69 reproducer (32759 entries sharing one set, 4 split subtables) drops from ~24 s / an OOM at ~12 GB to ~0.65 s CPU and ~137 MB peak RSS; instruction count vs entry count is linear (doubling n multiplies work by ~1.8x, not ~4x), and peak RSS is O(n). Adds test_resolve_with_liga_split_shared_liga_set to test-repacker. The full meson test suite (including the subset and subset+repack suites) passes. Discovered by Anthropic's OSS Scanner, as vulnerabilities ANT-2026-K86LK9JR (dropped ligatures) and ANT-2026-1P5QWD69 (quadratic DoS). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
HarfBuzz
HarfBuzz started as a text shaping engine but has grown into a
full font platform — the ffmpeg of text shaping. It primarily
supports OpenType, but also Apple Advanced Typography.
HarfBuzz shapes the majority of text on modern screens.
HarfBuzz is optimized for robustness, correctness, and performance — in that order. Achieve all.
Try it live at harfbuzz-world.cc — an interactive playground for shaping, subsetting, rasterization, vector output, and GPU rendering, all running in your browser.
Here is a quick map of its components:
Core libraries
| Library | Description |
|---|---|
| libharfbuzz | Text shaping, draw API, paint API. Highly configurable (see CONFIG.md). Optional integration backends compiled in: hb-ft (FreeType), hb-coretext (macOS), hb-uniscribe (Windows), hb-directwrite (Windows), hb-gdi (Windows), hb-glib, hb-graphite2. |
| libharfbuzz-subset | Font subsetting and variable-font instancing. |
Auxiliary libraries
| Library | Description |
|---|---|
| libharfbuzz-icu | ICU Unicode integration. |
| libharfbuzz-cairo | Cairo rendering integration. |
| libharfbuzz-gobject | GObject/GI bindings. |
Rendering libraries
| Library | Description |
|---|---|
| libharfbuzz-raster | Glyph rasterization to bitmaps, including color fonts. Uses hb-draw and hb-paint. |
| libharfbuzz-vector | Glyph output to vector formats (currently SVG), including color fonts. Uses hb-draw and hb-paint. |
| libharfbuzz-gpu | Encodes glyph outlines for GPU rasterization (Slug algorithm). Provides shader sources in GLSL, WGSL, MSL, and HLSL. Live demo. |
Notable missing feature: font hinting (including autohinting) is not implemented. For hinted rasterization, use FreeType or Skrifa.
For simplified builds, amalgamated sources are available:
harfbuzz.cc (just libharfbuzz), harfbuzz-subset.cc (just
libharfbuzz-subset), or harfbuzz-world.cc (everything, driven
by a custom hb-features.h). For a live in-browser playground
plus a worked example of the world.cc single-file build, see
harfbuzz-world.cc.
Command-line tools
| Tool | Description |
|---|---|
| hb-shape | Shape text and display glyph output. |
| hb-view | Render shaped text to an image. |
| hb-subset | Subset and optimize fonts. |
| hb-info | Display font metadata. |
| hb-raster | Render glyphs to bitmap images. |
| hb-vector | Render glyphs to vector formats (SVG). |
| hb-gpu | Interactive GPU text rendering. |
The canonical source tree and bug trackers are available on github. Both development and user support discussion around HarfBuzz happen on github as well.
For license information, see COPYING.
API stability
The API that comes with hb.h will not change incompatibly. Other, peripheral,
headers are more likely to go through minor modifications, but again, we do our
best to never change API in an incompatible way. We will never break the ABI.
The API and ABI are stable even across major version number jumps. In fact, current HarfBuzz is API/ABI compatible all the way back to the 0.9.x series. If one day we need to break the API/ABI, that would be called a new library.
As such, we bump the major version number only when we add major new features, the minor version when there is new API, and the micro version when there are bug fixes.
Documentation
For user manual as well as API documentation, check: https://harfbuzz.github.io
Download
Tarball releases and Win32/Win64 binary bundles are available on the github releases page.
Development
For build information, see BUILD.md.
For custom configurations, see CONFIG.md.
For testing and profiling, see TESTING.md.
For using with Python, see README.python.md. There is also uharfbuzz.
For cross-compiling to Windows from Linux or macOS, see README.mingw.md.
To report bugs or submit patches please use github issues and pull-requests.
Developer documents
To get a better idea of where HarfBuzz stands in the text rendering stack you may want to read State of Text Rendering 2024. Here are a few presentation slides about HarfBuzz over the years:
- 2026 – HarfBuzz at 20!
- 2016 – Ten Years of HarfBuzz
- 2014 – Unicode, OpenType, and HarfBuzz: Closing the Circle
- 2012 – HarfBuzz, The Free and Open Text Shaping Engine
- 2009 – HarfBuzz: the Free and Open Shaping Engine
More presentations and papers are available on behdad's website. In particular, the following studies are relevant to HarfBuzz development:
- 2025 – AAT layout caches
- 2025 – OpenType Layout lookup caches
- 2025 – Introducing HarfRust
- 2025 – Subsetting
- 2025 – Caching
- 2025 –
hb-decycler - 2022 –
hb-iter - 2022 – A C library written in C++
- 2022 – The case of the slow
hb-ft>h_advancefunction - 2022 – PackTab: A static integer table packer
- 2020 – HarfBuzz OT+AAT "Unishaper"
- 2014 – Building the Indic Shaper
- 2012 – Memory Consumption
Name
HarfBuzz /hærfˈbɒːz/
From Persian حرف (Harf: letter) and باز (Buzz: open). Transliteration of the Persian calque for OpenType.
As a noun: The Open Source text shaping engine.
As an adjective: Insincerely talkative; glib. A nod to the GNOME project where HarfBuzz originates from.
The logo shows حرفباز in the IranNastaliq font, on a Damascus steel background.
Background: Originally there was this font format called TrueType. People and companies started calling their type engines all things ending in Type: FreeType, CoolType, ClearType, etc. And then came OpenType, which is the successor of TrueType. So, for my OpenType implementation, I decided to stick with the concept but use the Persian translation. Which is fitting given that Persian is written in the Arabic script, and OpenType is an extension of TrueType that adds support for complex script rendering, and HarfBuzz is an implementation of OpenType text shaping.
Users
HarfBuzz is used in Android, Chrome, ChromeOS, Firefox, Flutter, GNOME, GTK+, KDE, Qt, LibreOffice, OpenJDK, XeTeX, Adobe Photoshop, Illustrator, InDesign, Microsoft Edge, Amazon Kindle, PlayStation, Godot Engine, Unreal Engine, Figma, Canva, QuarkXPress, Scribus, smart TVs, car displays, and many other places.
