Repository navigation
SVG rendering completeness: preserveAspectRatio, clipPath/mask, patterns, image, full filter primitives, CSS styling, feBlend - #29
Merged
Conversation
… support
- Implement preserveAspectRatio parsing ('[defer] <align> [<meetOrSlice>]', all 10
aligns, meet/slice) via a shared ComputePreserveAspectRatioFit helper reused by:
- root <svg> viewBox fitting (default unchanged: implicit case already was
'xMidYMid meet', confirmed via a dedicated pixel-for-pixel regression test)
- symbol/use viewBox-fit (new capability, resolving use/symbol width/height with
a use -> symbol -> viewport fallback chain)
- marker's own explicit preserveAspectRatio attribute only; the pre-existing
default (no-attribute) marker fit is preserved byte-for-byte unchanged
- Add viewport-relative percentage-based geometry resolution through the existing
ParseGeometryCoordinate/GetFloatAttribute/GetOptionalFloat choke point, so
existing finite/magnitude guards still apply after resolution:
- horizontal/vertical attributes resolve against current viewport width/height
- axis-agnostic attributes (r, stroke-width, stroke-dasharray) resolve against
the diagonal basis sqrt(w^2+h^2)/sqrt(2)
- font-size resolves against the parent element's own cascaded font-size
- stroke-miterlimit remains the sole exception (a unitless ratio, not a length)
- current viewport threaded via new RenderState.ViewportWidth/ViewportHeight
- Descope gradient userSpaceOnUse percentage plumbing (judgment call #5) as a
documented, bounded follow-up simplification; unrelated gradient coordinate
resolution (GetGradientCoordinateOrDefault) is untouched and unaffected.
- Update companion docs: design (svg-codec.md ViewBox Fitting Policy rewrite,
new Percentage-Based Geometry Resolution section, marker/use/symbol behavior),
reqstream (ViewBoxFitting rewrite, PercentageGeometryRejected ->
PercentageGeometryResolved, new MarkerPreserveAspectRatio/SymbolViewBoxFitting
requirements, all linked into codecs.yaml/canvas-net.yaml child chains with
reqstream --enforce confirming zero orphans), verification scenarios, and the
stale user-guide preserveAspectRatio claim.
- Add/repurpose tests: an 18-row root align/meetOrSlice Theory, an align=none
stretch test, an explicit default-vs-xMidYMid-meet pixel-identity regression,
symbol/use viewBox-fit tests, a marker explicit-preserveAspectRatio slice test
(meet is provably pixel-identical to the default path once refX/refY
ref-anchoring is applied, so slice is the only visibly distinguishable case),
and a full percentage-geometry matrix (rect/circle/line/gradient/stroke-width/
font-size/stroke-dasharray), repurposing the two prior percentage-rejection
facts into percentage-resolution proofs rather than deleting their coverage.
All 4446 tests pass across net8.0/net9.0/net10.0; build.ps1 and lint.ps1 both
clean; reqstream --enforce reports zero orphaned requirements.
- Add clip-path=/mask= support for SVG elements/groups (userSpaceOnUse and objectBoundingBox unit modes for clipPathUnits/maskUnits, separate maskContentUnits for mask content coordinate system). - clip-path implemented as offscreen alpha-buffer coverage intersection (no native clip-region primitive exists on the Canvas/Surface abstraction); union of clip children falls out of shared coverage buffer rasterization. - mask implemented via luminance (0.2125R+0.7154G+0.0721B, SVG default) attenuation of an offscreen-rendered mask-content buffer, reusing the existing filter SourceGraphic-buffer/compositing infrastructure. - Ordering fixed for combined filter + clip-path/mask cases: clip then mask now apply to pre-filter content before the filter chain runs, matching spec order geometry -> fill/stroke -> clip -> mask -> filter -> group opacity. - Resource safety: clip/mask offscreen buffer charges reuse the existing shared FilterWorkBudget; new flat MaxClipPathShapesPerClipPath (256) cap (silent truncation) on clipPath children, which are rasterized directly and cannot cycle. Fixed a genuine pre-existing stack-overflow bug where elementDepth was not incremented before recursing into mask content. - Scoped out (documented in code + design doc): nested clip/mask/filter on a clipPath's own children or on the clipPath/mask element itself, use/g/line as clip-path children, alpha mask mode, href/xlink:href inheritance for clipPath/mask elements. - 21 new SvgCodecTests cases: clipPath unit modes, multi-shape union, mask luminance semantics, mask region/content unit variants, combined filter+clip/mask ordering, budget-exhaustion fallback for both clip-path and mask sharing the same FilterWorkBudget, 256-shape cap truncation, and regressions confirming pre-existing filter-only and plain-shape rendering are unaffected. - Updated design/verification docs and reqstream requirement chain (CanvasNet-Lib-SvgClipPathMaskRendering -> CanvasNet-Codecs-SvgClipPathMaskRendering -> 4 new unit requirements), 0 orphans confirmed via reqstream --lint.
RenderShapeEffectsPipeline/RenderGroupWithEffects each charged exactly one region-area work unit against the shared FilterWorkBudget before allocating the primary content buffer, regardless of how many of clip-path/mask/filter actually applied - so a combined clip+mask (or clip/mask+filter) element allocated up to 3x the memory it was charged for. Extract a shared ComputeEffectsPipelineWorkUnits helper used by both call sites: it charges one region-area unit for the content buffer (scaled by the filter's own primitive count when a filter applies, unchanged from before), plus one additional flat region-area unit for each of clip-path/mask that applies, covering ApplyClipPath's own coverage buffer and ApplyMask's own maskSource buffer. Updated the two pre-existing single-effect budget-exhaustion tests' expected shape counts to match the corrected per-application charge, and added a new regression test combining clip-path and mask together to prove the combined charge (not just one effect's own charge) correctly exhausts the shared cumulative budget. Clarified the design doc's resource-safety discussion accordingly.
…urceRef, flaky timing test
- Add SvgCodec.Patterns.cs implementing <pattern> element resolution: href/xlink:href inheritance chain with cycle detection, patternUnits and patternContentUnits (objectBoundingBox/userSpaceOnUse) matching existing gradient/clipPath precedent, patternTransform composition, viewBox/preserveAspectRatio reuse from Phase 1, tile rendering to an offscreen buffer with per-pixel inverse-transform wrap sampling and coverage-multiply compositing clipped to the actual fill/stroke path. - Thread pattern-aware params through RenderShape/RenderFill/RenderStroke and dispatch pattern paints before existing ResolvePaint/gradient path; gradients, solid colors, and fill=url(#nonexistent) fallback behavior are unaffected. - Bound tile-buffer allocation via Surface.MaxDimension and existing FilterWorkBudget charging; bound pattern-of-pattern recursion via the existing MaxElementDepth guard. - Add 18 deterministic tests covering patternUnits x patternContentUnits (4 combinations), patternTransform, viewBox/preserveAspectRatio, href inheritance (including cycles), fill+stroke usage, nonexistent/wrong-type pattern references, and resource-safety behavior (no wall-clock timing assertions). - Update design docs, reqstream requirements (linked into codecs.yaml children chain, 0 orphans confirmed), verification docs, and the SysML2 svg-codec unit sourceRef list for the new source file.
Add CanvasNet-Lib-SvgPatternRendering system-level requirement, linking down to the previously-orphaned CanvasNet-Codecs-SvgPatternRendering subsystem requirement, matching the structure of its sibling filter/clip-path-mask/marker/gradient system-level requirements.
- Add <image> element rendering (SvgCodec.Image.cs): data: URI base64 raster decode reusing existing PNG/JPEG/BMP/TIFF/GIF codecs - Reuse Phase 1 preserveAspectRatio helper for placement-rect fitting - Non-tiled per-pixel inverse-transform compositing (Phase 3 pattern precedent, without tiling/wrap) - Dispatch through the shared element-rendering pipeline so clip-path/mask/filter/opacity apply automatically - Security: non-data hrefs (external file/URL) and nested SVG-in-image (data:image/svg+xml or .svg refs) are explicit, documented no-ops; no filesystem/network access is added - Resource/DoS safety: FilterWorkBudget charged with region area + base64 length before decode; decoded dimensions bounded by each raster codec's existing Surface.MaxDimension check - Add 15 new tests: PNG/JPEG data URIs, preserveAspectRatio variants, clip-path/mask/filter/opacity on <image>, malformed/truncated base64, external href no-op, nested-SVG no-op, deterministic oversized-budget test (no wall-clock timing) - Update companion artifacts: design doc, verification doc, sysml2 sourceRef, and requirement chain at unit/subsystem/system levels (system-level link in docs/reqstream/canvas-net.yaml verified this time, addressing the gap missed in Phase 3)
…unsupported embedded raster images RenderImageWithEffects's raster-decode catch filter only handled FormatException, InvalidDataException, and ArgumentOutOfRangeException. PngCodec.Load throws UnsupportedImageFeatureException (derives from IOException, not InvalidDataException) for well-formed but unsupported raster features such as Adam7-interlaced PNG, so an embedded interlaced PNG <image> href propagated an uncaught exception all the way out of SvgCodec.Load, aborting the entire SVG render instead of skipping just that element per this codec's documented tolerant per-element no-op contract. Widened the catch filter to also catch IOException, so any recognized- but-unsupported raster feature (from any sibling raster codec, not just PNG) is tolerantly skipped. Checked JPEG/BMP/TIFF/GIF: none currently define a similar unsupported-feature exception type - their unsupported- feature cases are already thrown as InvalidDataException, which was already caught. Added a regression test embedding PngSuite's own interlaced basi0g01.png fixture as a data:image/png;base64 href, proving SvgCodec.Load completes without throwing and that sibling elements before/after the skipped image still render correctly.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
The kernel coefficient lookup in ApplyConvolveMatrixPremultiplied and ApplyConvolveMatrixPreservingAlpha performed plain correlation (no rotation) instead of true convolution per the SVG Filter Effects spec's kernelMatrix(orderX-J-1, orderY-I-1) term. This produced spec-incorrect output for any non-point-symmetric kernel. - Flip the kernel-matrix index in both call sites in SvgCodec.Filters.Convolution.cs; sampleX/sampleY offset math is unchanged (already spec-correct). - Correct SvgCodec_Load_FeConvolveMatrixEdgeDetectKernel_ProducesExpectedEdgeResponse and SvgCodec_Load_FeConvolveMatrixEdgeModeWrap_SamplesAcrossOppositeEdge assertions to the spec-correct hand-derived expected pixel values. No requirement/design/verification/review-set changes needed - this is a pure correctness fix to already-documented behavior, not new user-visible behavior.
Add four tests covering blue/alpha channel selectors, a negative scale (direction reversal), and a larger-magnitude scale, bringing feDisplacementMap test depth in line with its 7 sibling filter primitives (only R/G selectors and one scale value were previously covered).
Phase 6 of the SVG filter roadmap. Adds spec-exact implementations: - SvgCodec.Filters.Lighting.cs: shared Sobel surface-normal estimation (9 kernel variants: interior + 4 edges + 4 corners), feDistantLight / fePointLight / feSpotLight resolution (including pointsAt/limitingConeAngle/ specularExponent falloff), feDiffuseLighting (Lambertian), and feSpecularLighting (Blinn-Phong, alpha = max(R,G,B) of the specular color). - SvgCodec.Filters.Turbulence.cs: the SVG spec's own reference Perlin noise algorithm (permutation table init, noise2/turbulence, turbulence vs fractalNoise remap, per-channel gradient offsets, baseFrequency/numOctaves/ seed). stitchTiles=stitch tolerantly falls back to noStitch (documented). - SvgCodec.Core.cs: MaxTurbulenceOctaves = 32 cap, enforced via InvalidDataException, charged through the existing FilterWorkBudget pattern used by feConvolveMatrix/feMorphology. - SvgCodec.Filters.cs: dispatch cases for the 3 new primitives; work-unit accounting for feTurbulence numOctaves; trimmed the no-op primitive list to feBlend only (the sole remaining unimplemented primitive). - SvgCodecTests.cs: 16 new facts covering distant/point/spot light x diffuse/specular hand-computed values, specular alpha != 1 proof, the spec's own PRNG conformance test vector (1043618065), determinism, turbulence-vs-fractalNoise remapping, numOctaves variation/cap, and stitch-tiles tolerant fallback. Verified via full regression run: all 1574 pre-existing tests pass unchanged, plus all 16 new tests (1590 total). The InkscapeFilters.svg fixture's hard-coded pixel assertion was independently re-verified against real rendering output and requires no change.
- docs/design/canvas-net/codecs/svg-codec.md: new prose paragraphs describing the shared Sobel surface-normal estimation, the three light-source element types, the diffuse/specular formulas (including the Sa=max(R,G,B) specular alpha subtlety), and the feTurbulence reference-algorithm-fidelity approach (naming the Park-Miller PRNG conformance test vector); trimmed the no-op primitive list to feBlend only; extended the resource-safety-caps sentence with MaxTurbulenceOctaves. - docs/reqstream/canvas-net/codecs/svg-codec.yaml: added CanvasNet-Codecs-SvgCodec-FeDiffuseLighting, -FeSpecularLighting, and -FeTurbulence unit requirements; extended FilterPrimitiveResourceSafety's title/justification/tests for the new numOctaves cap. - docs/reqstream/canvas-net/codecs.yaml: linked the 3 new unit ids into CanvasNet-Codecs-SvgFilterRendering's children/title/tests. - docs/reqstream/canvas-net.yaml: updated CanvasNet-Lib-SvgFilterRendering's title/tests to name the three new primitives. - docs/verification/canvas-net/codecs/svg-codec.md: 3 new subsections mirroring the existing FeMorphology/FeConvolveMatrix style; extended the FilterPrimitiveResourceSafety subsection for the new numOctaves cap test. - docs/sysml2/model/canvas-net/codecs/svg-codec.sysml: appended the 2 new source file paths to the unit's sourceRef comment. - .reviewmark.yaml: added the 2 new source files under CanvasNet-Codecs-SvgCodec's review paths, tagged '# phase 6 filter partial'.
…erms to cspell - docs/verification/canvas-net/codecs/svg-codec.md: fixed a markdown-fixer line-wrap artifact from fix.ps1 (a mid-sentence dash was reflowed onto its own line, forming an accidental list item); reworded with parentheses instead to avoid the ambiguity. - .cspell.yaml: added the genuine technical terms introduced by this phase (Sobel, Lambertian, Blinn, Phong, Perlin, PRNG/Prng, Schrage) to the project word list.
…view findings
- SvgCodec.cs class remarks: filter primitive no-op paragraph listed
feTurbulence, feDiffuseLighting, feSpecularLighting, feColorMatrix,
feConvolveMatrix, feComponentTransfer, feMorphology, feDisplacementMap,
feImage, feTile, and feDropShadow as unimplemented no-op passthroughs.
All are now implemented (Phases 5-6). Rewrote the paragraph to state
that only feBlend remains a tolerant no-op passthrough, matching the
wording already used by EvaluateFilterChain's XML doc remarks, and
added the feImage MaxUseDepth recursion-guard note already documented
in docs/design/canvas-net/codecs/svg-codec.md.
- Renamed test SvgTurbulencePrng_10000thValueFromSeedOne_MatchesSpecPublishedTestVector
to SvgCodec_TurbulenceTablesRandom_10000thValueFromSeedOne_MatchesSpecPublishedTestVector
to conform to the {ClassName}_{MethodUnderTest}_{Scenario}_{ExpectedBehavior}
naming convention (method under test is SvgCodec.TurbulenceTables.Random).
Updated the matching reference in SvgCodec.Filters.Turbulence.cs XML docs,
and all requirements traceability references in
docs/reqstream/canvas-net.yaml, docs/reqstream/canvas-net/codecs.yaml,
docs/reqstream/canvas-net/codecs/svg-codec.yaml, and
docs/verification/canvas-net/codecs/svg-codec.md.
- CssMatchWorkBudget now charges each stylesheet's precomputed MatchWorkPerElement (sum, per rule, of every comma-separated selector's own worst-case weight: 1 for its rightmost compound, +1 per further child-combinator segment, +MaxElementDepth per further descendant-combinator segment) instead of the bare Rules.Count, so the charge genuinely tracks real per-element matching CPU cost (selector-list length and descendant ancestor-walk cost), closing the gap that let a selectors-per-rule-heavy stylesheet drive ~29x real wall-clock cost for an identical charged budget value. - ParseRawRules/FindMatchingBrace no longer let an unterminated <style> rule swallow every later rule in the stylesheet: on failing to find a rule's own closing brace, the parser resynchronizes onto the next rule attempt (via the new FindSelectorRecoveryStart) rather than aborting the rest of the scan, so well-formed rules after a single malformed one are still parsed and applied. - Strengthened SvgCodec_Load_CssMalformedRule_SkipsOnlyThatRuleAndStillLoads to assert the specific, now-documented recovery behavior instead of hedging, and added regression tests proving the CSS match-work budget scales with selector-list/combinator-chain length rather than bare rule count. - Updated svg-codec.md's CSS malformed-input/resource-safety section to describe both fixes.
Task 1 - feBlend: - New SvgCodec.Filters.Blend.cs implementing ApplyFeBlend with in/in2 two-input resolution and all 14 CSS Compositing Level 1 modes (normal, multiply, screen, darken, lighten, color-dodge, color-burn, hard-light, soft-light, difference, exclusion, hue, saturation, color, luminosity). Non-separable modes use the spec's exact whole-RGB Lum/ClipColor/SetLum/ Sat/SetSat reference algorithm. Blending operates on straight color, composited via the CSS Compositing simple-alpha formula, mirroring CompositeFeOperator's premultiply/un-premultiply convention. - Wired case "feBlend" into EvaluateFilterChain's primitive switch. Task 2 - feComposite arithmetic bug fix: - ApplyFeComposite's "arithmetic" operator was previously unimplemented and fell through to a no-op passthrough, causing feDiffuseLighting's always-opaque alpha to propagate unchanged through filter50's chain (InkscapeFilters.svg), producing an opaque rectangular blob instead of following the source silhouette. - Added ApplyFeCompositeArithmetic implementing the premultiplied k1*i1*i2 + k2*i1 + k3*i2 + k4 formula (color and alpha), clamped to [0,1], per CSS Filter Effects spec (the one Porter-Duff-family operator defined on premultiplied values). - Also fixes the same class of bug in InkscapeFilters.svg's filter48, recomputing that fixture test's pixel expectations accordingly. Tests: 20 new hand-computed tests (11 feBlend separable modes via Theory, normal-equals-over cross-check, alpha-compositing, 2 non-separable luminosity cases, mode-fallback, 4 arithmetic-operator/regression tests). Repurposed FilterUnsupportedPrimitive test off feBlend onto a genuinely bogus primitive tag. Docs: updated design/verification docs, 3-level reqstream requirements, SysML2 sourceRef, .reviewmark.yaml, and user guide per the verified implementation plan in .agent-logs/planning-febend-arithmetic-alpha-bug-8f3c21e0.md.
- introduction.md: remove image/pattern from the out-of-scope list (both are fully implemented - pattern in Phase 3, image in Phase 4) and add them to the supported-feature narrative alongside filter/clipPath/mask - svg-codec.md: update the 'In-scope subset' filter bullet to list all 17 currently-implemented fe* primitives instead of the stale 5-item Phase 5/6-era list, reconciled against the accurate narrative section elsewhere in the same document
…ction Fix stale filter primitive enumeration in introduction.md's SvgCodec narrative (feFlood/feGaussianBlur/feOffset/feComposite/feMerge only), found during Phase 8 final formal-review recheck as inconsistent with the just-corrected svg-codec.md design doc. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
There was a problem hiding this comment.
Copilot review overview
🟡 Changes recommended
Unresolved moderate findings remain across CSS behavior, filter correctness and tolerance, and requirements traceability.
Review effort: Lite
Findings: 5
Open (5)
What changed in this PR
Expands SVG rendering support with CSS styling, aspect-ratio handling, patterns, images, masking, and advanced filters.
Changes:
- Adds CSS parsing, selectors, cascading, and fixtures.
- Implements additional geometry, paint servers, images, and filter primitives.
- Updates documentation, requirements traceability, tests, and review configuration.
| File | Summary |
|---|---|
test/DemaConsulting.CanvasNet.Tests/SvgFixtures/README.md |
Registers SVG fixtures. |
test/DemaConsulting.CanvasNet.Tests/SvgFixtures/css-styling.svg |
Adds CSS styling coverage. |
test/DemaConsulting.CanvasNet.Tests/Codecs/SvgFixtureTests.cs |
Adds fixture assertions. |
test/DemaConsulting.CanvasNet.Tests/CanvasNetTests.cs |
Adds integration coverage. |
src/DemaConsulting.CanvasNet/Codecs/SvgCodec.cs |
Updates SVG integration and documentation. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Text.cs |
Adds percentage geometry handling for text. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Shapes.cs |
Adds percentage-based shape geometry. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.PreserveAspectRatio.cs |
Implements aspect-ratio fitting. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.PathRender.cs |
Adds pattern rendering. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Parsing.cs |
Applies root fitting behavior. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Gradients.cs |
Adds bounds-based paint mapping. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.Tile.cs |
Implements feTile. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.Morphology.cs |
Implements feMorphology. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.FeImage.cs |
Implements feImage. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.DropShadow.cs |
Implements feDropShadow. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.DisplacementMap.cs |
Implements displacement mapping. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.Convolution.cs |
Implements convolution filters. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.ComponentTransfer.cs |
Implements component-transfer filters. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.ColorMatrix.cs |
Implements color-matrix filters. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Filters.Blend.cs |
Implements blend-mode evaluation. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Css.Selectors.cs |
Parses and matches CSS selectors. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Css.Parser.cs |
Parses CSS rules and declarations. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Core.cs |
Adds filter and resource safety limits. |
src/DemaConsulting.CanvasNet/Codecs/Svg/SvgCodec.Attributes.cs |
Adds percentage parsing. |
README.md |
Updates SVG capability documentation. |
docs/user_guide/introduction.md |
Documents expanded SVG support. |
docs/sysml2/model/canvas-net/codecs/svg-codec.sysml |
Updates SysML traceability. |
docs/reqstream/canvas-net/codecs.yaml |
Updates subsystem requirements and verification. |
docs/reqstream/canvas-net.yaml |
Updates system requirements and verification. |
.reviewmark.yaml |
Registers reviewed files. |
.cspell.yaml |
Adds technical vocabulary. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
…t-combinator backtracking, and CSS cascade coverage for filter/clip-path/mask/clip-rule/stop-color/stop-opacity - Add missing feBlend overlay mode (Overlay(Cb,Cs) = HardLight(Cs,Cb) per CSS Compositing Level 1), the one blend mode omitted from Phase 8's original 14. - Fix CSS selector matching: descendant-combinator segments previously committed to the nearest matching ancestor without backtracking, causing selectors like 'g > .a .target' to under-match when the nearest .a wasn't the one satisfying the child combinator. Now backtracks recursively through every candidate ancestor. Updated CssMatchWorkBudget's cost model from an additive to a multiplicative (saturating) formula to correctly bound the new backtracking algorithm's real worst-case cost. - Route filter/clip-path/mask/clip-rule/stop-color/stop-opacity through the same 3-tier CSS cascade chokepoint (ResolveElementProperty) already used for fill/stroke/font/marker properties, so stylesheet rules and inline style declarations now actually apply to these previously cascade-bypassed, XML-attribute-only properties. Added a per-element style context cache to avoid double-charging the CSS budget across the multiple call sites that now resolve properties on the same element. Investigated 2 further Copilot findings (feColorMatrix/feComponentTransfer allegedly needing premultiplied-alpha handling) and determined both are false positives: this codebase's Surface.PremultiplyAlpha docs state straight/unassociated alpha is the representation every codec uses, and the SVG Filter Effects spec explicitly defines both primitives to operate on non-premultiplied values - the existing straight-byte implementation is already correct per spec and per this codebase's own architecture. Test count: 4899 -> 4926, all passing. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Member
Author
|
Thanks for catching this - fixed in ca525dc. Presentation properties (\filter, \clip-path, \mask, \clip-rule, \stop-color, \stop-opacity) now route through the same 3-tier CSS cascade chokepoint (\ResolveElementProperty) already used for fill/stroke/font/marker properties. Added regression tests proving a stylesheet-only rule (no XML attribute) now correctly applies \filter/\clip-path/\mask. Also added a per-element style-context cache so the extra call sites don't double-charge the CSS work budget. |
This was referenced Oct 5, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Full SVG rendering-completeness roadmap: 8 phases implemented on a shared long-lived branch, each independently code-reviewed and formally reviewed locally (per cost-saving workflow - see phase notes below) before this single consolidated PR.
Phases
<pattern>paint server, tile-wrap math,patternUnits/patternContentUnits/patternTransform.<image>element - raster embedding viadata:URIs (externalhrefis a documented safe no-op; no filesystem/network access).feColorMatrix(all types),feComponentTransfer,feMorphology,feConvolveMatrix,feDisplacementMap,feTile,feDropShadow,feImage.feDiffuseLighting/feSpecularLighting(feDistantLight/fePointLight/feSpotLight),feTurbulence(spec-exact Perlin noise).<style>element / selector support - parser, selector matching, 3-tier cascade (presentation attribute < stylesheet rule < inlinestyle).feBlend(all 14 CSS Compositing Level 1 blend modes) + afeComposite operator="arithmetic"fix (was a silent no-op, causing filter chains combining lighting + arithmetic composites to render as opaque rectangular blobs instead of following the source shape's silhouette).Verification
build.ps1(0 warnings) andlint.ps1(markdownlint-cli2, cspell, yamllint, dotnet format, reqstream, versionmark, reviewmark, sysml2tools) before and after every fix.code-reviewpass (several found real defects via executed reproduction cases and hand/spec-verified numeric checks, not just static inspection) and aformal-reviewpass (compliance/traceability/documentation rigor), with fixes re-reviewed until clean.docs/reqstream/canvas-net.yaml) and subsystem (docs/reqstream/canvas-net/codecs.yaml) levels for every new requirement; SysML2 modelsourceRefkept current.SvgFixtures/InkscapeFilters.svg, 59 distinct filter definitions) compared against ImageMagick's reference rendering of the same file.Test plan
pwsh ./build.ps1- 4899/4899 tests passing, 0 warningspwsh ./lint.ps1- cleancode-reviewandformal-reviewpasses per phase, all approvedCo-authored-by: Copilot 223556219+Copilot@users.noreply.github.com