Testing and Validation

Unit tests

FontManager tests: load TTF/OTF/WOFF1, verify stbtt_fontinfo initialization, test cascade fallback, verify font metrics.
TextEngine helper tests: verify chunking, text-anchor, textLength, lengthAdjust, and baseline-shift calculations without real fonts.
TextBackend tests: verify shaping, metrics, and outline extraction for both base and full backends.
Glyph outline tests: extract outlines for known glyphs, verify path geometry matches expected curves.

Golden image tests

Add text-specific SVG test cases to renderer_tests:
- Basic Latin text.
- Multi-span positioning (dx/dy).
- text-anchor (start/middle/end).
- @font-face with embedded WOFF1 font.
- @font-face with WOFF2 font (requires text_full).
- Stroke + fill text.
- Per-glyph rotation.
Resvg text test subset: most e-text-* and e-tspan-* tests now pass with tight thresholds. Primary remaining failures are in e-textPath-* and a-font-size-*.

Feature-gated test skipping

{"text_basic.svg", Params::RequiresFeature(Feature::Text)},

{"text_woff2_font.svg", Params::RequiresFeature(Feature::TextFull)},

CI runs all feature combinations to ensure both enabled and disabled paths work.

Backend parity

Text golden images differ between Skia and TinySkia at the base tier because Skia uses its own internal layout (with GPOS access) while TinySkia uses stb_truetype (kern-table only).
At the text_full tier, shaping output is identical between backends since both use HarfBuzz, and golden images should be much closer.
TinySkia renders glyph outlines as vector paths (no hinting), so small-size text will look slightly different from Skia's hinted rasterization regardless of shaping tier.

Current Snapshot (2026-04-04)

Current base-tier (text, TinySkia) resvg status for the most relevant text slices:

Slice	Current status	Notes
e-text-*	30/30 enabled tests passing	12 disabled: mostly text-full-only or explicit skips
e-tspan-*	23/24 enabled tests passing	Active failure: e-tspan-030; 7 disabled
a-text-decoration-*	All enabled tests passing	Includes custom golden for 019
a-lengthAdjust-*	All enabled tests passing	spacingAndGlyphs now stretches glyphs along text direction
a-dominant-baseline-*	Enabled coverage passing at default threshold	Coverage is still thin
a-letter-spacing-*	All enabled tests passing in base tier	Arabic case requires text-full; 3 disabled/UB cases remain
a-textLength-*	All enabled tests passing	a-textLength-008 still disabled
a-writing-mode-*	All enabled tests passing	10 disabled mixed-script / rotate / dx/dy cases remain
e-textPath-*	Still disabled	Largest intentional feature gap

Release-significant gaps from the current snapshot:

e-tspan-030 is the main remaining active base-tier failure.
e-textPath-* is still intentionally out of the enabled suite.
BiDi and advanced mixed-script vertical text remain explicitly deferred.

Historical Failure Analysis (2026-03-30)

Historical snapshot of resvg test suite failures for the base config (TinySkia backend). 73 tests fail at the default threshold (100 max pixels).

Note: Previous counts (42) were artificially low because the a-font suite used a 17,000px default threshold that masked failures. The true count with the standard 100px threshold is 73.

Failure Summary

Category	Failing	Disabled	Pixel Range	Severity
textPath	34	8	112–13,209	Medium-High
Font properties	31	2	990–6,265	High
Writing mode	3	3	3,286–14,562	Medium
tspan	3	6	2,159–6,324	Medium
Letter spacing	1	0	3,055	Low
Text length	1	0	13,790	Medium

Font size rendering – RESOLVED

All 8 previously-failing font-size tests now pass. The 2 previously-disabled tests (004, 020) are now enabled and passing. See Font size gap analysis for details.

What was fixed (2026-03-30): Font-size is now resolved to absolute pixels during the CSS cascade (StyleSystem::computePropertiesInto), using the parent's computed font-size for relative units. Previously, em/ex/% units resolved against a hardcoded 16px default. Additionally, larger/smaller keywords are now parsed (converted to 120%/83.3% of parent).

Previously all 8 tests involved CSS font-size with relative units (em, ex, %, larger/smaller keywords) that must resolve against the parent element's computed font-size.

Test	Pixels	SVG Pattern
a-font-size-003	34,397	font-size="5ex" with parent font-size="12"
a-font-size-006	34,308	font-size="larger" then font-size="80%" chain
a-font-size-007	39,492	font-size="300%" (no explicit parent)
a-font-size-012	23,675	parent font-size="0", child font-size="50%"
a-font-size-014	26,994	font-size="3em" with parent font-size="12"
a-font-size-015	18,603	font-size="200%" with parent font-size="20"
a-font-size-016	26,994	font-size="3em" on root <svg> element
a-font-size-017	18,603	font-size="5ex" on root <svg> element

See Font size gap analysis for root cause details and fix plan.

Font properties (31 failures)

31 a-font-* tests fail at the standard 100px threshold. These were previously hidden by a 17,000px default max-mismatched-pixels override on the entire a-font prefix. Grouped by unimplemented feature:

Sub-category	Failing	Pixel Range	Implementation Status
Generic font-family mapping	10	990–5,375	Not implemented
font-weight (bold variants)	7	2,719–5,470	Partial
font-style	3	2,375	Not implemented
font-stretch	3	3,950	Not implemented
font-variant	2	5,570	Not implemented
font-size keywords	3	1,010–4,699	Not implemented
Residual rendering diffs	3	4,134–6,265	Rendering engine difference

See Font property gap analysis for details.

textPath rendering (34 failures)

All e-textPath-* tests are commented out in resvg_test_suite.cc (lines 999–1082) and run with the default 100px threshold, so all 34 fail. Tests 006 and 016 pass (startOffset overflow and link-to-rect respectively).

Severity	Tests	Pixel Range
Near-passing (<500px)	034, 026, 005, 020	112–396
Low (500–2,000px)	001, 014, 037, 029, 028, 039, 002, 003, 004, 015, 009, 038	626–1,651
Medium (2,000–5,000px)	011, 022, 024, 017, 018, 019, 023, 031, 021, 012, 010, 025	2,113–4,900
High (>5,000px)	013, 030, 027, 036, 035, 032	5,242–13,209

See textPath gap analysis for categorization and fix plan.

Writing mode (3 failures)

Test	Pixels	SVG Pattern
a-writing-mode-010	14,562	writing-mode="tb" + text-anchor="middle", font-size 64
a-writing-mode-013	3,286	writing-mode="tb" with mixed scripts (Japanese + English + Arabic)
a-writing-mode-020	10,599	writing-mode="tb" with dy="-105" on second tspan

See writing mode gap analysis for root cause details.

tspan positioning (3 active failures + 6 disabled)

Test	Pixels	SVG Pattern
e-tspan-006	2,159	Nested <tspan fill="green">Text <tspan font-weight="bold">Text</tspan></tspan>
e-tspan-024	3,599	A<tspan font-weight="bold">V</tspan>A + text-anchor="middle"
e-tspan-028	6,324	<tspan>T<tspan font-size="80">ex</tspan></tspan>t

Disabled: e-tspan-011 (xml:space preserve whitespace-only nodes), 016 (nested rotate), 020/021/022 (SVG 2 mask/clip-path/filter), 026 (BIDI + gradient).

See tspan gap analysis for phase status.

Letter spacing (1 failure)

Test	Pixels	SVG Pattern
a-letter-spacing-009	3,055	letter-spacing="5" with Arabic script + text-anchor="middle"

Root cause: letter-spacing interaction with RTL/Arabic shaping and text-anchor centering. Requires BIDI support to fix properly.

Text length (1 failure)

Test	Pixels	SVG Pattern
a-textLength-008	13,790	textLength on both <text> (500) and <tspan> (50, 120) elements

Root cause: textLength glyph adjustment when specified on both parent and child elements. The parent's textLength should distribute remaining space after child tspans have applied their own adjustments.

Changes since 2026-03-29

10 font-size tests newly passing (8 fixed + 2 enabled):

Test	Before	What was fixed
a-font-size-003	34,397	Font-size cascade: ex + parent resolution
a-font-size-004	Skipped	Enabled: nested percentage now works
a-font-size-006	34,308	larger/smaller keyword parsing
a-font-size-007	39,492	Font-size cascade + golden override (resvg wrong)
a-font-size-012	23,675	Font-size cascade: % of zero parent
a-font-size-014	26,994	Font-size cascade: em + parent resolution
a-font-size-015	18,603	Font-size cascade: % + parent resolution
a-font-size-016	26,994	Font-size cascade + golden override (resvg wrong)
a-font-size-017	18,603	Font-size cascade + golden override (resvg wrong)
a-font-size-020	Skipped	Enabled: nested percentage now works

Changes since 2026-03-21

7 tspan tests newly passing:

Test	Before	After	What was fixed
e-tspan-009	2,983	21	xml:space preserve-to-default boundary handling
e-tspan-010	5,597	6	xml:space default-to-preserve boundary handling
e-tspan-014	4,373	2	dy positioning on sibling tspan
e-tspan-025	2,773	10	Gradient fill on tspan (objectBoundingBox)
e-tspan-027	16,873	0	Per-character y-coordinates on nested tspan
e-tspan-029	7,191	0	Rotate + display:none tspan index consumption
e-tspan-030	5,478	379	Gradient + preserved spaces + text-decoration

2 tspan tests improved but still failing:

Test	Before	After	Improvement
e-tspan-024	16,722	3,599	78% reduction (cross-tspan bold shaping)
e-tspan-028	12,001	6,324	47% reduction (mixed font-size baseline)

tspan Gap Analysis and Fix Plan

3 actively failing + 6 disabled e-tspan-* tests. Phases B, C, D (partial), F, H completed since 2026-03-21.

Phase status

Phase	Description	Status	Tests
A	Text chunk splitting	Partial (006 still fails)	~~005~~, 006, ~~031~~
B	Whitespace/xml:space	COMPLETE	~~009~~, ~~010~~, 011 (disabled)
C	dy positioning	COMPLETE	~~014~~
D	Rotate inheritance	Partial (029 fixed, 016 disabled)	~~017~~, ~~029~~, 016 (disabled)
E	Cross-tspan bold	Improved (16,722 to 3,599)	024
F	Per-char coordinates	COMPLETE	~~027~~
G	Mixed font-size	Improved (12,001 to 6,324)	028
H	Gradient bbox	COMPLETE	~~025~~, ~~030~~
I	SVG 2 features	Not started	020, 021, 022 (disabled)
J	BIDI	Not started	026 (disabled)

Phase A: Text chunk splitting – Partial

Fixed tests: e-tspan-005 (basic mixed content), e-tspan-031 (triple-nested whitespace).

Still failing: e-tspan-006 (2,159px). Nested <tspan fill="green">Text <tspan font-weight="bold">Text</tspan></tspan> – deeply nested tspan with fill inheritance and font-weight change. The inner content positioning is off when style changes cascade through multiple nesting levels.

Phase B: Whitespace and xml:space – COMPLETE

Fixed: e-tspan-009 (preserve-to-default boundary), e-tspan-010 (default-to-preserve boundary).

Remaining disabled: e-tspan-011 (three-level preserve/default/preserve nesting). The XML parser strips whitespace-only text nodes before a child tspan in default mode, losing content that should be preserved when the child switches back to preserve mode.

Phase C: dy positioning – COMPLETE

Fixed: e-tspan-014 (dy on sibling tspan).

Phase D: Rotate inheritance – Partial

Fixed: e-tspan-017 (simple rotate on child tspan), e-tspan-029 (rotate + display:none index consumption).

Remaining disabled: e-tspan-016 (1,764px). Complex nested rotate lists across 4 levels of tspan nesting with partial overrides. The root text's rotate list should apply globally, with child rotate lists overriding for their character range.

Phase E: Cross-tspan bold (e-tspan-024, 3,599px)

A<tspan font-weight="bold">V</tspan>A with text-anchor="middle". The bold "V" uses NotoSans-Bold.ttf which has different advance widths, and text-anchor="middle" amplifies the positioning error. Improved from 16,722px after per-span font resolution was added.

Remaining gap: Cross-font kerning between the regular "A" and bold "V" is not applied. The total text width differs from the reference, causing the centered position to shift.

Phase F: Per-character coordinates – COMPLETE

Fixed: e-tspan-027 (per-character y-coordinates with nested tspan override).

Phase G: Mixed font-size (e-tspan-028, 6,324px)

<tspan>T<tspan font-size="80">ex</tspan></tspan>t where font-size changes from 48 to 80 mid-text. Improved from 12,001px after per-span font resolution. Remaining gap: baseline alignment when font-size changes – the dominant baseline of the larger font should align with the parent's baseline.

Phase H: Gradient bbox – COMPLETE

Fixed: e-tspan-025 (gradient fill objectBoundingBox), e-tspan-030 (gradient + preserved spaces + text-decoration).

Phase I: SVG 2 tspan features (disabled: 020, 021, 022)

SVG 2 allows mask, clip-path, and filter on <tspan>. Requires rendering each tspan as a compositing layer with isolation. Not implemented.

Phase J: BIDI (disabled: 026)

Bidirectional text reordering with gradient tspan crossing script boundary. Shared root cause with e-text-035. Requires UAX#9 bidi algorithm implementation.

textPath Gap Analysis

34 tests fail when run with the default 100px threshold. All overrides in resvg_test_suite.cc lines 999–1082 are commented out. 8 additional tests are disabled with Params::Skip().

Passing tests (2)

Test	Feature	Why it passes
e-textPath-006	startOffset=9999	Text is entirely past path end, nothing renders
e-textPath-016	Link to <rect> (SVG 2)	Falls back to inline rendering

Category 1: Core glyph positioning (001–005)

Basic text-on-path with various startOffset values. These are the foundation for all other textPath tests.

Test	Pixels	Feature
001	774	Basic cubic Bezier path
002	1,086	startOffset=30 (absolute)
003	993	startOffset=5mm (length unit)
004	985	startOffset=10% (percentage)
005	284	startOffset=-100 (negative)

Root cause: Arc-length parameterization accuracy. The Path::parameterForArcLength binary search may have insufficient precision for cubic Bezier segments, leading to systematic glyph placement offsets. The consistent 774–1,086px range (except 005 at 284) suggests a small but consistent positioning error.

Fix: Increase precision of arc-length computation or switch to Gauss-Legendre quadrature for cubic Bezier arc length. Test 005 is near-passing and may be fixed by the same change.

Category 2: Path geometry variants (009, 020, 024, 026, 027, 034, 037–039)

Different path geometries exercise edge cases in path walking.

Test	Pixels	Feature
009	1,506	Two separate textPath elements in one text
020	396	Closed path (arc with Z)
024	3,763	Path with subpaths (M L M L)
026	237	Path with ClosePath (triangle, M L L Z)
027	6,608	Simple polygon (M L Z)
034	112	Arc command (M A)
037	716	Transform on group outside referenced path
038	1,651	Big letter-spacing on arc path
039	954	Subpaths + startOffset

Root cause: Mixed issues. Tests 020, 026, 034 are near-passing (<500px), suggesting closed paths and arcs mostly work. Test 027 (6,608px) has a simple M L Z path where the ClosePath segment may not be handled correctly. Test 024 (3,763px) involves subpath handling – text should follow only the first subpath unless startOffset pushes past it.

Fix priority: Medium. Near-passing tests (020, 026, 034) may be fixed by Category 1 improvements. Tests 024, 027 need specific ClosePath/subpath logic.

Category 3: Text positioning on path (013–015, 022–023)

Per-character x/y/dx/dy coordinates when text is on a path.

Test	Pixels	Feature
013	5,312	x/y coordinates on parent <text> element
014	770	x/y coordinates on <textPath> element
015	1,455	Very long text (overflow past path end)
022	3,005	tspan with absolute x/y position inside textPath
023	4,209	tspan with relative dx/dy inside textPath

Root cause: Absolute x/y positioning should map to distance-along-path rather than canvas coordinates. Relative dx/dy should offset from current path position. Test 015 needs overflow handling (hide glyphs past path end).

Fix priority: Medium-High. Affects text with tspan children on paths.

Category 4: Text properties on path (010, 019, 021, 028–032)

Various text properties that need special handling on text-on-path.

Test	Pixels	Feature
010	4,624	Nested textPath elements
019	4,183	text-anchor (middle/end)
021	4,505	writing-mode=tb (vertical Japanese)
028	827	text-decoration (underline)
029	626	rotate attribute on text
030	5,242	Complex combination (writing-mode + letter-spacing + baseline-shift + startOffset)
031	4,353	letter-spacing
032	13,209	baseline-shift

Root cause: Each property interacts with path layout differently:

text-anchor must center/end-align along path length, not canvas X.
baseline-shift offsets perpendicular to the path tangent.
text-decoration (underline) should follow the path curve.
writing-mode=tb changes the glyph orientation paradigm on path.
letter-spacing adds to distance-along-path between glyphs.

Fix priority: Low-Medium. These are property interactions on top of core positioning.

Category 5: Mixed content (011, 012, 025)

Text elements with both textPath and tspan children.

Test	Pixels	Feature
011	2,113	Text before + textPath + tspan after
012	4,524	Multiple textPath + tspan interspersed
025	4,900	Invalid textPath element mid-content

Root cause: When <text> has both <textPath> and <tspan> children, text after the textPath should continue at the end position of the path or revert to normal positioning.

Fix priority: Low.

Category 6: Edge cases and transforms (017, 018, 035, 036)

Test	Pixels	Feature
017	4,025	No href on textPath (should render as inline text)
018	4,044	Invalid href (points to <text>, should render inline)
035	8,292	dy with tiny coordinates + scale(100) transform
036	7,334	Transform attribute on referenced <path> element

Root cause: Tests 017 and 018 test fallback behavior when textPath has no valid path reference. Tests 035 and 036 involve transform application – the referenced path's transform must be applied to the path geometry before glyph placement.

Fix priority: Low.

Disabled tests (8)

Test	Reason
007	Not implemented: method="stretch"
008	Not implemented: spacing="auto"
033	UB: baseline-shift + rotate combination
040	Not implemented: filter on textPath
041	Not implemented: side="right" (SVG 2)
042	Not implemented: path attribute (SVG 2)
043	Not implemented: path + xlink:href (SVG 2)
044	Not implemented: invalid path + href (SVG 2)

textPath fix plan

Phase 1 – Re-enable tests: Uncomment the 34 tests in resvg_test_suite.cc with per-test thresholds matching current pixel diffs. This prevents regressions.

Phase 2 – Core positioning (fixes ~5 tests from Category 1): Improve arc-length parameterization precision in Path. Target: tests 001–005 under threshold.

Phase 3 – Path geometry (fixes ~5 tests from Category 2): Handle ClosePath segments correctly (Z contributes path length back to start). Handle subpaths (text follows first subpath only). Target: tests 020, 026, 027, 034.

Phase 4 – Positioning on path (fixes ~5 tests from Category 3): Map absolute x/y to distance-along-path. Map dx/dy to path offset. Add overflow handling (hide past path end). Target: tests 013–015, 022–023.

Phase 5 – Properties on path (fixes ~8 tests from Category 4): Implement text-anchor along path length, baseline-shift perpendicular to tangent, text-decoration following path curve. Target: tests 019, 028–032.

Phase 6 – Mixed content + edge cases (fixes ~7 tests from Categories 5–6): Text continuation after textPath, fallback for invalid refs, transform application. Target: tests 011, 012, 017, 018, 025, 035, 036.

Font Size Gap Analysis – RESOLVED

Status: All 8 failures resolved (2026-03-30). 2 previously-disabled tests (004, 020) also enabled and passing.

What was fixed

Two changes in a single commit:

Font-size cascade resolution (StyleSystem::computePropertiesInto): After the CSS cascade, font-size is now resolved to absolute pixels using the parent's computed font-size. This fixes em, ex, and percentage units. A dedicated PropertyRegistry::resolveFontSize() method handles the font-size-specific percentage semantics (% of parent font-size, not viewBox).
larger/smaller keyword parsing (PropertyRegistry.cc font-size parser): These relative keywords are now parsed and converted to 120% / 83.3% respectively, which the cascade resolution then computes correctly.

Results

Test	Before	Root Cause Fixed
003	34,397	ex + parent font-size
004	Skipped	Nested percentage
006	34,308	larger keyword + chain
007	39,492	% resolved against viewBox; resvg golden overridden
012	23,675	% of zero
014	26,994	em + parent font-size
015	18,603	% + parent font-size
016	26,994	em on root element; resvg golden overridden
017	18,603	ex on root element; resvg golden overridden
020	Skipped	Nested percentage

Tests 007, 016, 017 use custom golden overrides (WithGoldenOverride) — the resvg goldens are incorrect (confirmed via Chrome). Our font-size resolution is correct: 300% of 16px = 48px, 3em * 16px = 48px, 5ex * 0.5 * 16px = 40px.

Previous root cause analysis (for reference)

Relative unit resolution: Font-size values like 3em, 200%, 5ex, and larger are relative to the parent element's computed font-size. The previous implementation resolved these against a hardcoded 16px default instead of the actual parent's computed value, and percentages resolved against the viewBox diagonal instead of the parent font-size.

Future improvement: ex unit accuracy

The exUnitInEm ratio is still hardcoded at 0.5 (the CSS fallback). Measuring the actual font's x-height from the OS/2 sxHeight field or the 'x' glyph bounding box would improve accuracy for ex-unit tests, but the remaining diffs are small enough to be within threshold.

Font Property Gap Analysis

31 a-font-* tests fail. The core issue is that only font-family (lookup by name), font-size, and font-weight (numeric/bold/normal) are fully wired through to rendering. Several font properties are either not parsed, not used during font selection, or missing key feature support.

Implementation status

Property	Parsed	Stored	Used in rendering	Missing
font-family	Yes	SmallVector<RcString>	Yes (name lookup)	Generic family mapping
font-size	Yes	Lengthd	Yes	Absolute-size keywords
font-weight	Yes	int (100–900)	Yes (TextShaper only)	bolder/lighter keywords
font-style	No	Unparsed	No	Entire property
font-stretch	No	Unparsed	No	Entire property
font-variant	No	Unparsed	No	Entire property

1. Generic font-family mapping – NOT IMPLEMENTED (10 tests)

Generic families (serif, sans-serif, monospace, cursive, fantasy) are parsed and stored as literal strings but FontManager::findFont() has no mapping from generic names to actual fonts. All generic families render with the fallback font (Public Sans).

Test	Pixels	What's tested
family-001	3,854	serif
family-002	2,123	sans-serif
family-003	2,727	cursive
family-004	5,375	fantasy
family-005	3,924	monospace
family-007	1,271	Named font Source Sans Pro (not in @font-face)
family-008	1,271	Font list: 'Source Sans Pro', Noto Sans, serif
family-009	5,228	Fallback from Invalid family
family-010	990	Fallback: Invalid, Noto Sans
family-011	3,729	Bold sans-serif

What's needed: Add a generic-family-to-font mapping in FontManager. The mapping could be configurable, with defaults: serif → Noto Serif, sans-serif → Noto Sans, monospace → Noto Sans Mono, cursive/fantasy → fallback. The resvg test suite bundles Noto fonts for this purpose.

Also: Tests 007–010 test the font-family fallback list. When the first family isn't available, the renderer should try subsequent families in order. The current FontManager::findFont() takes a single family string — it needs to accept the full family list and iterate.

2. font-weight: bold rendering – PARTIAL (7 tests)

Parsing supports normal (400), bold (700), and numeric 1–1000. TextShaper does weight-matched font selection via FontManager::findFont(family, weight). However:

bolder/lighter keywords not parsed (same issue as larger/smaller was for font-size — relative keywords need parent context to resolve).
Bold font variants not always available: Weight matching only works against registered @font-face rules. If no bold variant is registered, the fallback font (Public Sans Regular) is used regardless of weight.
TextLayout (stb_truetype path) ignores weight entirely — only TextShaper uses it.

Test	Pixels	What's tested
weight-002	5,470	bold / 700
weight-003	5,470	bolder (relative, unimplemented)
weight-004	5,470	bolder with clamping at 900
weight-005	5,470	bolder without parent
weight-006	5,470	lighter from parent 800
weight-009	2,735	Numeric 700
weight-012	2,719	Numeric 650

Passing weight tests for reference: weight-001 (normal/400), weight-007 (lighter clamped at 100), weight-008 (lighter without parent → 200), weight-010 (inherit from 400), weight-011 (invalid 1500 → fallback). These all resolve to weight 400 or below, which matches the fallback font.

What's needed:

Parse bolder/lighter as relative keywords and resolve during cascade (analogous to larger/smaller for font-size). bolder steps up: ≤300→400, ≤500→700, >500→900. lighter steps down: ≥800→400, ≥600→100, <600→100.
Register the test suite's Noto Sans Bold font so weight matching has a bold variant.
Optionally: add synthetic bold (wider stroke) as a fallback when no bold variant exists.

3. font-style – NOT IMPLEMENTED (3 tests)

font-style is listed in kValidPresentationAttributes but has no parser in kProperties. Values are stored in unparsedProperties and never used. Italic and oblique rendering is not supported.

Test	Pixels	What's tested
style-001	2,375	italic
style-002	2,375	oblique
style-003	2,375	inherit (italic from parent)

What's needed:

Add font-style to kProperties with parser (values: normal, italic, oblique).
Store as an enum in PropertyRegistry.
Add style parameter to FontManager::findFont() for italic variant selection.
Optionally: synthetic italic (shear transform) when no italic variant exists.

4. font-stretch – NOT IMPLEMENTED (3 tests)

font-stretch is listed in kValidPresentationAttributes but has no parser. Not used in font selection.

Test	Pixels	What's tested
stretch-001	3,950	narrower
stretch-002	3,950	inherit (extra-condensed from parent)
stretch-003	3,950	extra-condensed

What's needed:

Add font-stretch to kProperties with keyword parser (ultra-condensed through ultra-expanded, plus narrower/wider relative keywords).
Store as numeric percentage (50%–200%) in PropertyRegistry.
Add stretch parameter to FontManager::findFont().
Lower priority than font-style since condensed/expanded variants are less common.

5. font-variant – NOT IMPLEMENTED (2 tests)

font-variant is listed in kValidPresentationAttributes but has no parser. No small-caps support.

Test	Pixels	What's tested
variant-001	5,570	small-caps
variant-002	5,570	inherit (small-caps from parent)

What's needed:

Add font-variant to kProperties with parser (values: normal, small-caps).
For small-caps: either use OpenType smcp feature (requires HarfBuzz / text-full), or synthesize by rendering lowercase chars at ~70% scale with uppercase glyphs.

6. font-size: absolute-size keywords – NOT IMPLEMENTED (3 tests)

larger/smaller are now parsed, but absolute-size keywords (xx-small through xx-large) are not.

Test	Pixels	What's tested
size-005	4,411	All named values (xx-small through xx-large) + larger/smaller + %
size-008	1,010	xx-large without parent
size-010	4,699	Parent font-size="0", child absolute font-size="40" on tspan

What's needed: Add keyword parsing for xx-small (9px), x-small (10px), small (13px), medium (16px), large (18px), x-large (24px), xx-large (32px). Convert to Lengthd(Npx) at parse time.

Note: size-010 may be a different issue — font-size="40" is already absolute, so the failure may be related to tspan font-size override handling when the parent is zero.

7. Residual rendering diffs (3 tests)

These font-size tests pass within the old 17,000px threshold but fail at 100px. The diffs (4,134–6,265px) are from font rendering engine differences (glyph outlines, hinting) rather than missing features.

Test	Pixels	What's tested
size-007	6,265	300% without parent
size-016	6,138	3em on root <svg>
size-017	4,134	5ex on root <svg>

These use font-family="Noto Sans" which is available, and the font-size resolution is now correct. The remaining diff is inherent to the stb_truetype vs reference renderer glyph rasterization. These should be given per-test thresholds.

Writing Mode Gap Analysis

3 failures in the base config. 3 additional tests disabled (016, 017, 019).

Failures

Test	Pixels	SVG Pattern	Root Cause
010	14,562	writing-mode="tb" + text-anchor="middle", font-size 64	text-anchor centering uses horizontal advances instead of vertical
013	3,286	writing-mode="tb" with Japanese + English + Arabic	Script-specific rotation heuristic incorrect for Arabic in vertical mode
020	10,599	writing-mode="tb" with dy="-105" on second tspan	dy applied along Y axis regardless of writing mode; should be inline direction (X in vertical-rl)

Root cause analysis

010 – text-anchor in vertical mode: The text-anchor adjustment should operate along the Y axis (block direction for vertical-rl writing mode), centering the text vertically. Current implementation computes the centering offset using horizontal advance widths.

013 – mixed-script vertical: In vertical writing mode, CJK characters are rendered upright while Latin characters are rotated 90 degrees clockwise. The stb_truetype backend (base config) uses a heuristic based on Unicode code point range (< 0x2E80 = rotate) which doesn't handle Arabic script correctly. This test passes in text-full config where HarfBuzz provides proper vert GSUB feature access.

020 – dy axis in vertical mode: dy in vertical writing mode should shift along the inline direction (X axis in vertical-rl). Current implementation applies dy to Y regardless of writing mode. The dy="-105" should move the second tspan leftward, not upward.

Fix plan

Phase 1: Fix text-anchor calculation in TextLayout::layout() to use vertical advances when writing-mode is vertical. (Fixes 010.)

Phase 2: Fix dx/dy axis mapping based on writing mode. (Fixes 020.)

Phase 3: Improve mixed-script vertical rotation heuristic for Arabic, or accept as a base-config limitation resolved by text-full. (Fixes 013 in base config, already passes in text-full.)