websem: close the SVG <feConvolveMatrix> rows - #110
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Warning Review limit reachedNext included review available in 20 minutes. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Plus Run ID: ⛔ Files ignored due to path filters (86)
📒 Files selected for processing (16)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Hosted-x86 exact-gate follow-up The first hosted Cargo run found two fixture-only one-channel differences:
Etiology: both controls unnecessarily sat on architecture-sensitive fractional raster surfaces. The blur source mixed three colors through an sRGB Gaussian kernel; the mask used Commit 87a1ab4 replaces those constructions without tolerance: the blur control now uses a single-color boundary source in the default filter color space, and the mask control uses binary white/black geometry. Both replacements were recaptured through the hash-pinned Chromium module, rendered through strict and best-effort Gate sensitivity remains explicit. Forcing nonzero Gaussian blur from transparent-edge to repeated-edge sampling makes the repaired blur cell fail on 806 pixels at maximum channel delta 102. Restoring the production path returns the complete 741-cell gate to green. The binary target-mask candidate differs from its no-mask control by 444 pixels at maximum channel delta 202. No renderer code, tolerance, baseline rule, score, or FLIP record changed in this follow-up. |
|
Hosted-x86 exact-gate follow-up 2 The full Cargo job exercised a second test binary after the hosted Etiology: the binary mask removed the original fractional luminance surface, but its source still carried unrelated Commit 3624ed9 makes the source opaque. Its Chromium 149 scratch matrix was twice deterministic and proves the claimed branch directly:
All nine scratch candidates render exactly against Chromium through both strict and best-effort |
1 participant
Verdict
CLOSE/SPLIT.
This rung admits the complete measured static Chromium behavior of
<feConvolveMatrix>and closes nine checklist rows: the element plusbias,divisor,edgeMode,kernelMatrix, convolutionorder,preserveAlpha,targetX, andtargetY.kernelUnitLengthremains open for its lighting-primitive applicability. Shared graph, region, interpolation, resource, and dynamics rows remain open.Measured behavior
Four twice-deterministic Chromium 149.0.7827.55 scratch matrices covered grammar, invalid states, graph routing, effect order, precision, and source classes. Every candidate also rendered through both actual n0 admissions and was compared by pixels.
The admitted contract carries rectangular kernels through 256 coefficients, kernel reversal, ordered divisor arithmetic, bias, asymmetric targets, all three edge modes, alpha preservation, both filter color spaces, all established graph inputs and result reuse, hard regions, primitive units, safe mappings, source geometry, target effects,
<use>, andviewBox.The corpus gains 40 convolution cells plus one blur
edgeModebrowser-drop cell. It now contains 741 Chromium-baked cells plus 10 sampled frames; the filter estate has 380 cells. Three precision/arithmetic patrols move the named refusal register to 146 rows:All 41 new cells are exact-byte controls with no new tolerance. No conformance score was produced or inspected, and this PR takes no FLIP action.
Sensitive gate
I temporarily removed the required kernel reversal.
just gaterejected 11 convolution cells; the focused reversal cell changed 230 pixels at maximum channel delta 202, and broad failures reached 2,814 pixels at maximum delta 250. Restoring the reversal returned the complete gate to green.The first hosted-x86 run then exposed two fixture-only delta-1 rounding surfaces: an sRGB three-color blur source and a fractional-luminance mask. Commit 87a1ab4 replaced those constructions without tolerance. A later full Cargo run found seven residual delta-1 pixels at an unrelated
fill-opacity=".58"overlap that survived in the mask cell; commit 3624ed9 removes that fractional source too. Forcing the repaired blur from transparent-edge to repeated-edge sampling rejects it on 806 pixels at maximum delta 102. The opaque binary-mask control is identical to explicit filter-then-mask nesting, differs from reversed effect order on 75 pixels at maximum delta 99, and differs from no mask on 444 pixels at maximum delta 204. The complete 741-cell gate is exact on ARM and hosted x86 in both the dedicated host gate and the full Cargo test binary.Independent verification
No Workflow runner is exposed in this environment, so
.agents/workflows/verify-rung.jscould not be invoked through its required tool. I reproduced both roles manually and independently:kernelUnitLengthsplit, source grammar, and every documentation claim. It found two should-fixes; both were applied.Verification
just bake: Chromium 149.0.7827.55 verified all 741 primitive oraclesjust gate: 741 exact gates green locally on ARMn0 host gateand full Cargo reftest binary: 741 exact gates green on x86just status: generated capability view freshcargo test --workspace --exclude grida-canvas-wasmcargo clippy --no-deps --workspace --exclude grida-canvas-wasm -- -D warningscargo fmt --all -- --checkpnpm fmt:checkpnpm exec oxlint --deny-warnings