Status: In Progress (Round 1 complete, Round 2 in progress) Updated: 2026-04-07 Tracking: v1.0 milestone

Summary

After the filter support merge, Codecov reports 74.3% line coverage (down from ~82% pre-merge). The drop comes from ~10K new lines in the filter pipeline and vendored tiny-skia filter library. This plan restores coverage to 80%+ and charts a path to the v1.0 target of 90%.

Round 1 Results (PR #478, merged)

Excluded third_party/ from coverage via codecov.yml and filter_coverage.py
Added 228 new tests across 8 new test files and 5 expanded files (~4,600 lines)
Codecov impact: 74.3% → 74.6% (+0.3%) — third_party exclusion not yet reflected in codecov baseline; local measurement shows 71.5% lines with --config=latest_llvm
Note: codecov.yml must be on the default branch (main) to take effect on project-level coverage; it landed with PR #478

Remaining gap

~6% from undertested production code in renderer backends, text engine, and rendering context. The largest uncovered files are RendererTinySkia.cc (2,241 lines), the archived full-Skia renderer source (4,232 lines), TextEngine.cc (1,715 lines), and RenderingContext.cc (1,148 lines).

Current State

Coverage by subsystem (estimated)

Subsystem	Source Lines	Test Lines	Ratio	Status
XML parser	1,890	2,294	121%	Excellent
CSS parser/cascade	3,500	5,000+	143%	Excellent
Core/base utilities	245	5,449	2225%	Excellent
SVG parsers	4,133	3,381	82%	Good
ECS systems	3,500	3,200	91%	Good
Text engine	1,715	292	17%	Critical gap
Renderer backends	6,473	~850	13%	Critical gap
RendererDriver	2,069	456	22%	Critical gap
RenderingContext	1,148	286	25%	Major gap
FilterGraphExecutor	788	298	38%	Moderate gap
SVG element classes	~900	~1,200	133%	Good (but 12 types untested)
Third-party (vendored)	~15,000	0	0%	Excluded from metrics

Untested SVG element types

These element types have no dedicated *_tests.cc file:

Element	Source Lines	Notes
SVGSVGElement	78	Root container — viewBox, viewport
SVGImageElement	73	External image loading
SVGUseElement	71	Shadow tree instantiation
SVGTextPathElement	43	Text along paths
SVGStopElement	68	Gradient stops
SVGLineElement	57	Basic shape
24× SVGFe*Element	~14 each	Filter primitive stubs

Implementation Plan

Phase 0: Exclude third-party from coverage metrics

Remove vendored third_party/ code from the LCOV coverage denominator. This code is tested via the resvg integration test suite, not unit tests.

Update filter_coverage.py to skip LCOV records for third_party/ paths
Verify coverage increases by ~3–4% after exclusion

Add codecov.yml with ignore: ["third_party/**"] for redundancy

Expected impact: 74% → ~77–78%. Landed in PR #478.

Phase 1: Text engine tests (est. +2–3%)

The text engine is the largest gap in first-party code. TextEngine.cc (1,715 lines) has only 292 lines of tests.

Per-character positioning — Test x/y/dx/dy/rotate attribute lists with multi-span text, verify glyph positions match expected coordinates.

letter-spacing / word-spacing — Test CSS property application in both TextLayout (stb_truetype) and TextShaper (HarfBuzz) paths.

text-anchor — Test start/middle/end alignment with known text widths.

dominant-baseline / alignment-baseline — Test vertical positioning for different baseline values.

textPath sampling — Test Path::pointAtArcLength() with known paths, verify glyph rotation along curves.

textLength / lengthAdjust — Test spacing vs spacingAndGlyphs stretch modes.

text-decoration — Test underline/overline/line-through positioning.

Font matching — Test family name resolution, weight matching, style matching via FontManager with registered @font-face rules.

WOFF/WOFF2 — Test font data loading paths (already partially covered by WoffParser_tests).

Phase 2: Renderer tests (est. +2–3%)

The renderer backends (RendererTinySkia.cc: 2,241 lines, archived full-Skia source: 4,232 lines) are primarily tested via golden image comparisons. Add targeted unit tests for specific code paths.

RendererDriver paint resolution — Test fill/stroke paint resolution for solid colors, gradients, patterns, and context-fill/context-stroke. Verify correct paint parameters are passed to the backend.

Coordinate transforms — Test viewBox → viewport → device pixel transform chain. Test nested SVG coordinate resolution.

Clip path application — Test clip path intersection with various geometries.

Mask rendering — Test mask-on-mask composition, mask luminance computation.

Isolated layer compositing — Test opacity, mix-blend-mode layer bracketing.

RenderingContext state — Test render tree instantiation, dirty flag fast path, computed component creation.

Phase 3: Filter executor tests (est. +1%)

Expand FilterGraphExecutor_tests.cc (298 lines) to cover more graph topologies.

Multi-node chains — 3+ nodes with named buffer routing.

Color space transitions — Graphs mixing sRGB and linearRGB primitives.

Filter region edge cases — Oversized regions, zero-size regions, rotated regions.

feImage fragment refs — Test same-document element references via the executor.

CSS shorthand functions — Test blur(), brightness(), grayscale(), etc. through the filter graph construction path.

Error/edge paths — Empty graph, invalid inputs, missing named buffers.

Phase 4: SVG element tests (est. +1–2%)

Add dedicated test files for untested element types.

SVGSVGElement_tests.cc — viewBox parsing, preserveAspectRatio, width/height attribute handling, nested SVG viewport computation.

SVGImageElement_tests.cc — href attribute, image loading callbacks, aspect ratio.

SVGUseElement_tests.cc — href resolution, shadow tree creation, attribute inheritance, cycle detection.

SVGTextPathElement_tests.cc — href, startOffset, method, side, spacing attributes.

SVGStopElement_tests.cc — offset, stop-color, stop-opacity attributes.

SVGLineElement_tests.cc — x1, y1, x2, y2 attributes, computed path.

Phase 5: Secondary gaps (est. +1%)

Lower-priority files with moderate gaps.

FontMetadata.cc (136 lines) — Font metadata extraction tests.

RendererUtils.cc — Utility functions used by both backends.

Property resolution edge cases — CSS shorthand expansion, inheritance across shadow tree boundaries, !important in presentation attributes.

Incremental invalidation paths — Test dirty flag propagation, selective recomputation in StyleSystem, fast path verification.

Phase 6: Long-tail coverage (stretch goal: 90%)

After Phases 0–5 achieve 80%+, these items push toward 90%:

Renderer backend error paths — OOM conditions, invalid surface creation, corrupt image data handling.

Parser error recovery — Malformed SVG/CSS input that triggers specific error paths. Leverage fuzzer corpus for inspiration.

Rare element combinations — Nested patterns, recursive use references, filter on clipPath content, mask with gradient fill.

Platform-specific paths — macOS vs Linux font loading differences, Skia-specific fallbacks.

Debug/diagnostic code — operator<< formatters, verbose logging paths. Mark genuinely untestable paths with LCOV_EXCL.

DOM mutation edge cases — appendChild to self, removeChild of non-child, replaceChild with the same node, setAttribute with empty value.

FE element attribute tests — All 24 filter element types have stub implementations (~14 lines each). Add attribute roundtrip tests for each.

Coverage Targets

Milestone	Target	Strategy
Immediate (Phase 0)	77–78%	Exclude third_party
v0.5 release	80%+	Phases 0–2
v1.0 release	90%	All phases

Priority Order

Phase	Impact	Effort	Priority
Phase 0 (third_party exclusion)	+3–4%	1 hour	Immediate
Phase 1 (text engine)	+2–3%	3–5 days	High
Phase 2 (renderer)	+2–3%	3–5 days	High
Phase 3 (filter executor)	+1%	1–2 days	Medium
Phase 4 (SVG elements)	+1–2%	2–3 days	Medium
Phase 5 (secondary gaps)	+1%	2–3 days	Low
Phase 6 (stretch to 90%)	+5–8%	2–3 weeks	v1.0

Running Coverage

# Full coverage run (Linux CI only — requires --config=latest_llvm)
./tools/coverage.sh
 
# View HTML report
open coverage-report/index.html
 
# Scoped to a module (faster iteration)
./tools/coverage.sh //donner/svg/components/text/...

Validation

Coverage ≥80% on Codecov after Phases 0–2
No test regressions
New tests use project conventions (donner_cc_test, gTest/gMock)
Each new test file has a corresponding fuzzer where applicable
Test across all build configs: default, --config=text-full, legacy full-Skia configuration