Status: Complete Updated: 2026-03-15 Context: Codecov reports 75% line coverage; local coverage.sh measured 81.7% (20,973/25,669). The discrepancy likely comes from CI running without the filter_coverage.py step applied before upload, or Codecov including header-only code in its denominator. Regardless, 80%+ on Codecov is the target.
Summary
Closing the gap requires ~1,300–1,900 additional lines covered. The plan combines three strategies:
- Add LCOV exclusions for legitimately untestable code (~500–800 lines)
- Add unit tests for the largest untested ECS systems (~2,000+ lines coverable)
- Add tests for AttributeParser.cc, the single largest untested file (~2,462 lines)
Coverage Gap Inventory
Largest Untested Source Files (by line count)
| Rank | File | Lines | Category |
| 1 | svg/parser/AttributeParser.cc | 2,462 | Parser (no dedicated test) |
| 2 | archived full-Skia renderer source | 2,437 | Renderer (covered by integration tests) |
| 3 | svg/renderer/RendererTinySkia.cc | 1,697 | Renderer (covered by integration tests) |
| 4 | svg/renderer/RenderingContext.cc | 1,260 | Renderer (no dedicated test) |
| 5 | svg/components/animation/AnimationSystem.cc | 1,376 | ECS system (no test) |
| 6 | svg/components/layout/LayoutSystem.cc | 855 | ECS system (thin test: 262 lines) |
| 7 | svg/components/filter/FilterSystem.cc | 663 | ECS system (no test) |
| 8 | svg/renderer/TerminalImageViewer.cc | 577 | Tool (has test) |
| 9 | svg/components/shape/ShapeSystem.cc | 511 | ECS system (no test) |
| 10 | svg/components/paint/PaintSystem.cc | 334 | ECS system (no test) |
Untested ECS Component Systems (0 test files)
| System | Lines | Priority |
| AnimationSystem.cc | 1,376 | High |
| FilterSystem.cc | 663 | High |
| ShapeSystem.cc | 511 | High |
| PaintSystem.cc | 334 | Medium |
| ShadowTreeSystem.cc | 254 | Medium |
| StyleSystem.cc | 238 | Medium |
| ResourceManagerContext.cc | 233 | Medium |
| TextSystem.cc | 230 | Medium |
| RectComponent.cc | 115 | Low |
| EllipseComponent.cc | 98 | Low |
| CircleComponent.cc | 88 | Low |
| StopComponent.cc | 89 | Low |
| FilterPrimitiveComponent.cc | 96 | Low |
| Total | 4,325 | |
Untestable Code (candidates for LCOV_EXCL)
| Category | Files | Est. Lines | Action |
| operator<< debug formatters | 97 files with operator<< | ~300 | LCOV_EXCL_LINE on ostream operators that exist only for test/debug output |
| Signal handler | FailureSignalHandler.cc | 224 | LCOV_EXCL_START/STOP the entire file |
| CLI tool entry points | xml_tool.cc, svg_parser_tool.cc, renderer_tool.cc | 399 | Already excluded by being *_tool.cc; verify they're not in coverage denominator |
| Renderer backends (partially) | archived full-Skia source, RendererTinySkia.cc | ~200 | LCOV_EXCL on error-only paths, platform-specific fallbacks |
| ElementType.cc operator<< | 69 lines of switch-case for element names | 60 | LCOV_EXCL_LINE per case |
Estimated impact: ~500–800 lines removed from denominator or marked as covered.
Implementation Plan
Phase A: LCOV Exclusion Audit (est. +2–3% coverage)
Quick wins that reduce the denominator without writing new tests.
- A1: Mark FailureSignalHandler.cc with LCOV_EXCL — 224 lines of signal handling that cannot be exercised in unit tests.
- A2: Test operator<< debug formatters — Add unit tests that exercise operator<< for enum types and data structures. Focus on donner/svg/core/*.h enum formatters and donner/css/ type formatters. Use EXPECT_THAT(stream.str(), ...) patterns.
- A3: Test renderer error/fallback paths — Add unit tests that exercise error returns in the archived full-Skia source and RendererTinySkia.cc (e.g., failed SkSurface creation, null canvas checks, invalid dimensions, OOM-like conditions).
- A4: Verify tool files excluded — Confirm *_tool.cc and *_benchmark.cc files are not included in the coverage denominator by the CI workflow. If they are, exclude them in the Bazel coverage invocation.
Phase B: ECS System Tests (est. +3–5% coverage)
Unit tests for the component systems that form the core of the rendering pipeline. These are currently exercised only indirectly via integration tests.
- B1: ShapeSystem_tests.cc (511 lines to cover) — Test shape computation for rect, circle, ellipse, line, polyline, polygon. Verify computed path splines match expected geometry. Test attribute parsing → computed shape pipeline.
- B2: PaintSystem_tests.cc (334 + 245 lines) — Test gradient resolution (linear, radial), pattern tiling, stop color interpolation, paint inheritance. Include StopComponent, GradientComponent, LinearGradientComponent, RadialGradientComponent, PatternComponent.
- B3: StyleSystem_tests.cc (238 lines) — Test CSS cascade computation, property inheritance, specificity resolution, and !important handling.
- B4: FilterSystem_tests.cc (663 + 96 lines) — Test filter graph construction, primitive chaining, filter region computation. Include FilterPrimitiveComponent.
- B5: ShadowTreeSystem_tests.cc (254 lines) — Test <use> element shadow tree instantiation, attribute inheritance into shadow trees, cycle detection.
- B6: AnimationSystem_tests.cc (1,376 lines) — Test timing model (begin/end/dur/ repeatCount), value interpolation, animation sandwich composition, freeze/remove behavior. This is the largest single untested file.
- B7: TextSystem_tests.cc (230 lines) — Test text chunk formation, bidirectional algorithm setup, text measurement.
- B8: Expand LayoutSystem_tests.cc (855 lines, only 262 tested) — Add tests for viewport computation, viewBox transforms, preserveAspectRatio, nested SVG layout.
Phase C: AttributeParser Tests (est. +2–4% coverage)
- C1: AttributeParser_tests.cc — AttributeParser.cc is 2,462 lines with no dedicated test file. It handles parsing of all SVG presentation attributes. Create a comprehensive test file covering:
- Paint attributes (fill, stroke, opacity)
- Transform attributes
- Filter-related attributes
- Text attributes
- Geometry attributes (x, y, width, height, r, cx, cy, etc.)
- Gradient/pattern attributes
- Error handling for malformed values
Phase D: Secondary Coverage Gaps (est. +1–2% coverage)
Lower-priority files that contribute to the long tail.
- D1: RenderingContext_tests.cc (1,260 lines) — Test render tree instantiation, computed style resolution, layer decomposition setup.
- D2: SVG element tests — Add tests for the largest untested elements: SVGTextPositioningElement (143), SVGSVGElement (78), SVGImageElement (73), SVGUseElement (71), SVGStopElement (66).
- D3: CSS type tests — Add tests for ComponentValue.cc (54), ComplexSelector.cc (72), PseudoClassSelector.cc (80).
- D4: Property parsing tests — Add tests for PresentationAttributeParsing.cc (50), PropertyParsing.cc (121), RxRyProperties.cc (33).
Priority Order
To maximize coverage gain per effort:
- Phase A (LCOV exclusions) — Fastest, no test code needed, +2–3%
- Phase B1–B3 (Shape/Paint/Style) — Core systems, moderate complexity, +2–3%
- Phase C (AttributeParser) — Single largest gap, +2–4%
- Phase B4–B6 (Filter/Shadow/Animation) — More complex systems, +1–2%
- Phase D (Secondary gaps) — Long tail cleanup, +1–2%
Conservative estimate: Phases A + B1–B3 + C should bring coverage from 75% to ~82%.
Running Coverage
# Full coverage run (requires genhtml + Java)
./tools/coverage.sh
# Scoped to specific module (faster iteration)
./tools/coverage.sh //donner/svg/components/...
# Quiet mode for CI
./tools/coverage.sh --quiet
After each phase, verify with ./tools/coverage.sh and check the HTML report at coverage-report/index.html.
Validation
- Coverage ≥80% on Codecov after PR merge
- No regressions in existing tests
- New tests follow project conventions (gTest/gMock, *_tests.cc naming)
- LCOV exclusions only applied to genuinely untestable code (not to hide gaps)
1.15.0