- Use int16 for the cover array (values max ±8192, fits int16).
Halves cover array footprint; NEON loads 8×int16 and widens.
- When cover doesn't cancel at row end, compute the constant alpha
byte once and memset the output tail instead of filling, sweeping,
and clearing the working arrays.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Cast through void* before __m128i* to silence Clang's -Wcast-align.
The _mm_loadu/storeu intrinsics handle unaligned access correctly;
the warning is spurious.
Also document hb_raster_image_t and hb_raster_draw_t typedefs in
hb-raster.h.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Document all hb_raster_image_t and hb_raster_draw_t functions with
GTK-Doc style comments. Add SECTION header for hb-raster. Fix
hb_cairo_font_face_set/get_raster() doc comments to use XSince and
standard Return value format.
Also add hb-raster section to harfbuzz-sections.txt and include
hb-raster.xml in harfbuzz-docs.xml Integration API chapter.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The prefix sum of cover propagates across the entire row, but the
sparse x_min/x_max tracking only covered cells directly written by
edges. When the right edge falls on the image boundary (out of
bounds), the sweep stopped at the left edge column, leaving interior
pixels at zero.
Fix by extending the sweep to width-1 when cover_accum is non-zero
after the prefix sum, filling the extended range with the constant
accumulated cover value. For well-formed closed shapes where edges
cancel, x_end == x_max and there is no extra work.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Mirror the NEON vectorized sweep for x86-64 using pure SSE2 baseline
intrinsics (no SSSE3/SSE4.1 required). Emulates abs, min, and
multiply-by-255 with basic SSE2 ops.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Split the per-row sweep into a sequential prefix-sum pass followed by
a vectorized alpha computation pass. The NEON path processes 8 pixels
per iteration using vabsq/vminq/vmlaq and narrows to bytes in-place.
Significant speedup at large font sizes where wide scanlines dominate.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Precompute delta_fy = 64 * total_dy / total_dx once per edge per row,
then accumulate instead of dividing per column.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace dx/dy fields with a 16.16 fixed-point slope precomputed at
edge creation time. The two integer divisions per edge per scanline
in edge_sweep_row become multiply+shift. Same struct size.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Maintain a monotonically-advancing start index that skips past edges
whose yH has fallen below the current row. This avoids re-scanning
the dead prefix of the sorted edge array on every scanline.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Use hb_qsort() instead of qsort() to avoid function-pointer overhead
- Use >> 13 instead of / 8192 for alpha computation (unsigned shift)
- Track per-row x_min/x_max of touched cells; only sweep and clear
that range instead of memset'ing the full width every row
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace the multi-sample tile rasterizer (NEON / SSE4.1 / scalar variants)
with an analytic-coverage approach that computes exact area and cover
contributions per pixel cell per edge. This eliminates the tile bookkeeping
(tile_offsets, tile_edge_indices) and SIMD-specific code paths in favour of
a simpler, portable scanline sweep that yields higher quality output.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Instead of rendering the hb-raster image at subpixel resolution
(x_scale = font_size * 64 units/em, causing Cairo to resample 64×),
render at device-pixel resolution using the scale_matrix from the
scaled font.
The rasterizer transform is set to (px_x/x_scale, 0, 0, px_y/y_scale)
where px_x = scale_matrix.xx = font_size in device pixels per em.
The image is then placed with cairo_scale(1/px_x, -1/px_y) to map
raster pixels 1:1 to device pixels with Y-flip.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When enabled, hb_cairo_render_glyph() rasterizes via hb_raster and
blits the A8 coverage image using cairo_mask_surface(), leaving the
existing Cairo source undisturbed.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Renders every glyph in a font through the rasterizer.
Args: font-file [font-size-px] [output-dir]
font-size defaults to upem; output-dir writes per-glyph PGMs when given.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Renders shaped text into a single auto-sized A8 PGM raster image,
written to stdout or --output-file. Mirrors util/hb-draw in structure.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Renders glyph IDs 2-5 from fonts/adwaita.ttf at upem scale, writes
PGMs to /tmp/ for visual inspection, and asserts non-zero coverage.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fix out-of-bounds pixel write in rasterize_tile: border tiles that
extend past the image edge were writing px-x_origin beyond the row
stride, wrapping into the start of the next row and overwriting
correctly-rasterized pixels with zero. Pass px1/py1 (exclusive
bounds) to rasterize_tile and clip the pixel loops accordingly.
Also: drop unused ex0 in pass 2 of tile assignment; move test-raster
from src/ to test/api/ and convert it to the GLib g_assert/hb_test_add
framework.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New library harfbuzz-raster provides a fast CPU coverage rasterizer for
glyph outlines. Three public types: hb_raster_image_t (ref-counted pixel
buffer), hb_raster_draw_t (outline rasterizer consuming hb_draw_funcs_t
callbacks), and a stub for hb_raster_paint_t (v2).
Key design points:
- Nonzero winding rule
- 26.6 fixed-point edge representation
- 8-sample supersampling per pixel (HB_RASTER_SAMPLES constant)
- 16×16 tile culling with CSR edge lists
- SIMD inner loops: NEON (aarch64) and SSE4.1 (x86); scalar fallback
- Bézier flattening (quadratic + cubic) via de Casteljau, max depth 16
- Auto-size from edge bounding box or caller-supplied extents
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add bool exact=false parameter to all extend() overloads. When true,
the allocation is exact (no growth headroom), useful for copy
constructors. Pass exact=true from the Iterable copy constructor.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
extend(Iterable) called alloc(hb_len(iter)) which only allocated
space for the new elements, ignoring existing length. When the
vector already had content, push_has_room would write past the
allocated buffer.
Fix: alloc(length + hb_len(iter)).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* [buffer] Fix a typo and doc string
* [buffer] Add merge_grapheme_clusters / merge_out_grapheme_clusters
Cluster merges serve two distinct purposes that require different level
gates:
- Grapheme grouping: combining marks merged into their base cluster.
Correct gate: IS_GRAPHEMES (levels 0 and 3).
New functions: merge_grapheme_clusters / merge_out_grapheme_clusters.
- Monotonicity maintenance: clusters merged to prevent non-monotone
output after ligature formation, reordering, etc.
Correct gate: IS_MONOTONE (levels 0 and 1).
Existing functions: merge_clusters / merge_out_clusters.
Previously both paths shared merge_clusters_impl's single IS_MONOTONE
gate, so GRAPHEMES (level 3) was incorrectly blocked from merging
grapheme clusters. This also caused a regression in the Thai and
Hangul shapers for MONOTONE_CHARACTERS (level 1): when their explicit
== MONOTONE_GRAPHEMES call-site guards were removed in #5069, the
impls' IS_MONOTONE gate no longer protected them.
Gates are now in the inline wrappers in hb-buffer.hh; the _impl
functions are gate-free. hb_form_clusters, CoreText grapheme
detection, and the Thai/Hangul shapers are updated to call the
grapheme-semantic variants. Fixes#5750.
Fixes https://github.com/harfbuzz/harfbuzz/issues/5751
Fixes https://github.com/harfbuzz/harfbuzz/issues/5750
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* [thai] Adjust grapheme merging logic
* [hangul] Tweak grapheme merging
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Add CFF2 to CFF1 converter for font instantiation
When instantiating a CFF2 variable font to a static instance, automatically
convert the output to CFF1 format instead of keeping it as CFF2. This is
appropriate since CFF2's variation features are no longer needed in static
fonts, and CFF1 is more widely supported.
The converter integrates with the CFF2 subsetter and triggers automatically
when pinned=true (i.e., when instantiating with normalized coordinates).
Converts table structure:
- CFF2 inline TopDict → CFF1 TopDict INDEX
- Adds Name INDEX and String INDEX
- Creates CID-keyed font with ROS operator (Adobe-Identity-0)
- Generates identity CID charset
- Removes variation operators (vstore, vsindex, blend)
- Reuses desubroutinized charstrings and subroutines
Focus is on structural conversion; charstring encoding details like
endchar/return operators and width encoding can be added incrementally.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Fix CFF2 to CFF1 converter table tag and serialization
Updates the CFF2 to CFF1 converter to properly output CFF format:
- Use plan->add_table() to output 'CFF ' tag instead of 'CFF2'
- Return false from subset() to signal CFF2 table not needed
- Use passed serializer instead of creating new one
- Change String INDEX to use push/copy/pop_pack pattern
Known issue: Output has 5 extra bytes before CFF header that need
investigation of serializer head/tail allocation patterns.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Fix CFF2 to CFF1 serialization order and header version
Fixes the INDEX serialization order and version field initialization:
- Reorder serialization to match CFF1 pattern: CFF header, then Name INDEX,
then Top DICT INDEX. This eliminates extra bytes before the header.
- Fix version field by using cff->min_size instead of OT::cff1::min_size
for nameIndex offset. This resolves the issue where major version was
written as 0x00 instead of 0x01.
- Add String INDEX with "Adobe" and "Identity" strings (SIDs 391, 392)
required by the ROS operator in CID-keyed fonts.
The CFF header now correctly starts at offset 0 with version 1.0.
Remaining issue: FDArray offset parsing error needs investigation.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Improve FDArray serialization in CFF2 to CFF1 converter
Change FDArray serialization to use inferred template types and avoid
preprocessor issues with the likely() macro when using template syntax.
Progress:
- CFF header now at offset 0 with correct version 1.0 ✓
- String INDEX with ROS strings (Adobe, Identity) ✓
- ROS operator with correct SIDs ✓
- Serialization order matches CFF1 pattern ✓
Remaining issue:
- FDArray INDEX structure needs investigation - fontTools reports
"unpack requires a buffer of 4 bytes" when reading FDArray offsets.
May be related to INDEX offSize or data alignment.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* WIP: Add CFF2 to CFF1 converter for font instantiation
When a CFF2 variable font is fully instantiated (all axes pinned),
convert it to CFF1 format. This is work in progress.
Key changes:
- Add serialize_cff2_to_cff1() function to convert instantiated CFF2 to CFF1
- Use cff2_private_dict_op_serializer_t to instantiate Private DICT blends
- Add CFF1 table structures (Name INDEX, String INDEX for ROS, etc.)
- Create CID-keyed fonts with ROS operator (Adobe-Identity-0)
- Use subsetter's serializer with end_serialize() to resolve links
Current status:
- Private DICT instantiation works (blends evaluated correctly)
- CharStrings are properly flattened (verified: no blend operators remain)
- CFF1 table structure created but CharStrings INDEX validation fails in OTS
- Need to investigate CharStrings serialization format issue
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Add FDSelect and FDArray to Top DICT for CID-keyed fonts
CFF1 CID-keyed fonts require FDSelect and FDArray operators in the
Top DICT, even when there's only one Font DICT. CFF2 makes FDSelect
optional when there's a single FD, but CFF1 always requires it.
This commit:
- Always creates an FDSelect for CID-keyed CFF1 fonts (even with 1 FD)
- Explicitly serializes FDSelect and FDArray operators in Top DICT
- Avoids duplicating these operators when processing CFF2 Top DICT
Fixes font rendering - glyphs now have proper paths.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Use FDSelect3 (range-based) instead of FDSelect0
When creating a default FDSelect for single-FD CID fonts, use the
compact range-based FDSelect3 format instead of FDSelect0.
For a font with 254 glyphs mapping to FD 0:
- FDSelect0: 254 bytes (1 byte per glyph)
- FDSelect3: 8 bytes (format + nRanges + range + sentinel)
Saves 246 bytes (~1KB in this test font).
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Update documentation with implementation status
Documents:
- What's implemented and working
- Known limitations (FontBBox, width, stack depth)
- Key conversion steps
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* Copy FontBBox from head table
Instead of calculating from scratch, copy the font bounding box
directly from the head table (xMin, yMin, xMax, yMax).
Before: FontBBox value="0 0 0 0"
After: FontBBox value="-175 -250 1178 899"
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [CFF2->CFF1] Add width optimization and encoding
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [CFF2->CFF1] Port O(n) width optimization algorithm
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [cff2-to-cff1] Add command capture infrastructure for specialization
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [cff2-to-cff1] Implement CharString specialization with stack depth control
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [cff2-to-cff1] Add generalization phase for stack safety
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [cff2-to-cff1] Update documentation to reflect completed features
Removed outdated limitations that have been implemented:
✓ FontBBox (now copied from head table)
✓ Width handling (O(n) optimization + CharString encoding)
✓ Stack depth control (generalize→specialize with maxstack=48)
Added new features to implementation status:
- CharString specialization
- Width optimization and encoding
- Stack depth control
Remaining limitations are now accurate (only CID-keyed, no curve
specialization/peephole optimizations).
* [cff2-to-cff1] Remove CID-keyed 'limitation' from docs
CID-keyed output is by design, not a limitation:
- Variable fonts (CFF2) are instantiated to CID-keyed CFF1
- We explicitly add ROS operator to make output CID-keyed
- This is the correct behavior for instantiated fonts
Changed section from 'KNOWN LIMITATIONS' to 'FUTURE ENHANCEMENTS'
to better reflect that curve/peephole optimizations are optional
improvements, not missing functionality.
* [CFF2->CFF1] Add HB_SUBSET_FLAGS_DOWNGRADE_CFF2 flag
Make the CFF2→CFF1 converter opt-in via the HB_SUBSET_FLAGS_DOWNGRADE_CFF2
flag (disabled by default). This gives users control over whether variable
font instantiation produces CFF2 (default) or CFF1 (for compatibility).
Changes:
- Add HB_SUBSET_FLAGS_DOWNGRADE_CFF2 flag to hb-subset.h
- Gate CFF2→CFF1 conversion on flag check in hb-subset-cff2.cc
- Add --downgrade-cff2 command-line option to hb-subset tool
The flag enables converting instantiated variable fonts from CFF2 to CFF1,
providing compatibility with older renderers that don't support CFF2.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [CFF2->CFF1] Deduplicate CFF1Font string constant
Define CFF1_DEFAULT_FONT_NAME constant once in hb-subset-cff2-to-cff1.hh
and use it in both the plan creation and serialization code.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* [CFF2->CFF1] Fix swapped horizontal/vertical logic in specializer
The specialization logic for rmoveto/rlineto to h/v variants was backwards:
- When dx=0 and dy!=0, movement is VERTICAL (vmoveto/vlineto), not horizontal
- When dy=0 and dx!=0, movement is HORIZONTAL (hmoveto/hlineto), not vertical
This was causing completely garbled outlines in converted fonts. The fix
corrects the operator selection to match the actual movement direction.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Fix the following build error with HB_DISABLE_DEPRECATED:
In file included from harfbuzz\src\harfbuzz.cc:14:
harfbuzz\src\hb-directwrite.cc(468,1): error : no previous prototype
for function 'hb_directwrite_font_get_dw_font'
[-Werror,-Wmissing-prototypes]
468 | hb_directwrite_font_get_dw_font (hb_font_t *font)
| ^
hb_directwrite_font_get_dw_font() is guarded with
HB_DISABLE_DEPRECATED only in the header file, but not in the .cc.
Signed-off-by: Alexander Lobakin <alobakin@mailbox.org>
Fixed three bugs found during code review:
1. Division by zero in normalize_pt(): Added check to return {0, 0}
when normalizing a zero-length vector, preventing NaN values in
sweep gradient rendering.
2. Memory leak in get_color_stops(): Now properly frees old memory
before reallocating and checks for malloc failure to prevent
null pointer dereference.
3. Integer division in helper_cairo_create_context(): Fixed missing
decimal point causing integer division (br / 255) instead of
floating-point division, which resulted in incorrect alpha
channel calculations for color blending.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Position PNG images at bottom-left of glyph box for correct rendering.
Uses y_bearing + height to account for negative height values in glyph extents.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
PNG color-fonts (like Apple Color Emoji's sbix table) were not being
rendered by hb-draw. There were two issues:
1. probe_paint_funcs had no callbacks registered, so HarfBuzz couldn't
detect PNG glyphs as color glyphs. Fixed by adding a simple
probe_paint_image callback that returns true.
2. PNG images were using negative height directly in SVG <image>
elements, which doesn't work properly. Fixed by wrapping PNG images
in a <g> element with a transform that handles coordinate system
flip, similar to how SVG embedded images are handled.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Removes the --quiet/-q option that output raw path data only.
This simplifies the codebase by eliminating the write_raw_paths()
function and associated logic.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This increases the precision of scale and transform values in SVG output
for better accuracy in rendering.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Renames --ned option to --flat for better clarity. The option flattens
geometry and disables reuse of glyph definitions in SVG output.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Uses std::unordered_map to cache hb_glyph_extents_t by glyph ID, avoiding
redundant hb_font_get_glyph_extents() calls when the same glyph appears
multiple times (e.g., 'e' in "speed").
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
When both --logical and --ink are specified, the bounding box is computed as
the union of logical and ink extents, providing maximum coverage. This matches
the behavior of specifying both flags in hb-view.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Adds --logical (default) and --ink options to control bounding box calculation.
--logical uses font logical extents (ascender/descender), --ink uses actual
glyph ink extents for a tighter bounding box. Default behavior is --logical.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Adds --margin option similar to hb-view but with default of 0 instead of 16.
Accepts 1-4 space-separated numbers (top right bottom left, with CSS-style
fallback logic). Expands SVG viewBox and dimensions accordingly.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Previously foreground only worked for color fonts. Now it works for all fonts
by wrapping content in a compact <g fill="..."> element that applies to all
child elements. Supports RGB and RGBA colors with transparency.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>