perf: index label route clearance candidates

[seekskyworld]

Problem and value

Large diagrams currently compare every relationship label against every route to measure clearance. A conservative grid index removes distant candidates before the existing exact segment calculation. It preserves diagnostic ordering, distances and segment identities, with full-scan fallbacks for large routes, unselective queries and unsafe grid coordinates.

The prior non-terminating-loop review is addressed: all four grid coordinates must be safe integers before either nested loop. Unsafe routes enter the global candidate set; unsafe label queries use the full route list. No dependency, authored geometry, schema, validation threshold or version changes.

Integration and cost

Original bca9add69604be0f38292eb07472bdf5fcdff509 history is preserved. Current head 86ebff230f6ab11f1ccfc6be863e7a9456ae64b7 integrates dev 2d87fec2e555f1dd266c4fbb2a866bc744034343, including #449, #416 and the shared browser gate #440. The merge retains dev's large-receipt minimum-clearance regression alongside the indexing tests.

The index is private to one shared clearance function, with bounded per-route indexing and deterministic ordering. Its benefit is strongest for sparse, larger diagrams; dense/unselective inputs can still require the original full comparison. This is not a claim that overall diagram authoring is hundreds of times faster.

Verification

• On combined source 5440dbb, geometry/layout/render-output suites: 201 passed, 0 failed, 0 skipped.
• A seeded 2,000-case differential comparison against 128a8aa produces identical complete results, including hit order, distances and segment indexes. Cases include negative and unsafe coordinates, duplicate identities, diagonal/multisegment routes, varied rectangles, zero/negative/nonfinite/unselective thresholds. The clearance implementation is unchanged by the subsequent dev sync/package commit, so this evidence is reused with that scope.
• Same-machine Node 22.23.2 sparse fixture, 3,000 routes and 3,000 labels, two ABBA cycles: baseline 3266–3304 ms; candidate 8.57–10.58 ms. Both return the same empty hit set. This measures only the synthetic clearance stage, not real end-to-end generation or a universal speedup.
• bc26a22 geometry + package gates (unchanged production payload after fix(ci): share required browser regressions with releases #440 sync): 82 passed, 0 failed, 2 explicit platform/version skips. Canonical ZIP freshness is checked. Golden renderer/source/package examples remain byte-identical; no new visual acceptance is claimed because the generated diagrams are unchanged.
• After syncing fix(ci): share required browser regressions with releases #440, current-head shared-gate/geometry tests pass 66/66 with no skips. The only intervening package-manifest change adds a repository-only npm script, excluded from the canonical ZIP.
• Full final-head remote CI must pass before merge. Local focused results do not substitute for the Node matrix, browser/WebM and platform gates.

Generated artifacts

Rebuilt only archify.zip with official Node 22.23.2 from tracked combined source. Renderer examples remain byte-identical, so Gallery/proof/Viewer outputs need no regeneration. main remains unchanged; this PR targets dev integration only.

[tt-a1i]

Reviewed at exact head f7dc11958a220d41c0a2cd0a7ff36685483688d2. The spatial broad-phase is a valuable optimization, and I verified that ordinary finite-coordinate behavior remains identical in a 2,000-case differential check. However, two merge blockers remain:

1. Unsafe grid coordinates can make index construction or lookup loop forever. cellCount may still be a safe integer even when x1/x2 or y1/y2 is not; at that magnitude, x += 1 or y += 1 no longer changes the value. Route indexing should put such routes in the global set, while label queries should fall back to the full route list.
2. archify.zip was not regenerated. The source geometry.mjs and the packaged copy have different SHA-256 hashes, so source users receive the optimization while ZIP installs keep the old implementation. Please rebuild the canonical archive with Node 22 after the code fix.

Please also commit regression coverage for unsafe finite coordinates and the unselective-threshold fallback. The PR currently has no GitHub checks and is behind main; after updating it, the exact new head and package freshness need to pass CI.

[seekskyworld]

Addressed review blockers in latest head bca9add:- Guard route and label grid indices with Number.isSafeInteger; unsafe finite coordinates and unselective query bounds fall back to the full route list, preventing non-progressing x += 1/y += 1 loops.- Added regression tests for unsafe finite coordinates and unselective thresholds.- Rebased onto latest origin/main (5de7275).- Rebuilt canonical archify.zip with Node 22.23.2.Verification: npm test 987 passed, 31 skipped (1,018 total); targeted geometry tests 55/55; 200 randomized differential checks remain identical.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Advanced

Run ID: 735d255b-836d-4982-92a4-7db4e03cec99

📥 Commits

Reviewing files that changed from the base of the PR and between 2d87fec and 86ebff2.

⛔ Files ignored due to path filters (1)

• archify.zip is excluded by !**/*.zip

📒 Files selected for processing (2)

• archify/renderers/shared/geometry.mjs
• archify/test/geometry.test.mjs

---

📝 Summary

Reviewed base 72c750b and head 86ebff2. The head includes route-candidate indexing and regression tests, plus later integration changes across 211 files. The geometry change preserves exact clearance results and uses bounded grid queries with safe full-scan fallbacks. Compatibility impact is limited to internal performance behavior. Author-reported validation is reused evidence from its original revision: 987 passing tests, 31 skips, 55/55 geometry tests, and 200 differential checks. No CI result was observed at the checked head. Browser or perceptual acceptance is not established by these static results.

Walkthrough

collectLabelRouteClearance now indexes routes in 128px spatial cells and checks only candidates near each label. Large routes, unsafe coordinates, and unselective queries use fallback candidates. A zero threshold returns immediately. New tests cover ordering, fallback paths, extreme coordinates, long routes, and performance with 3,000 routes.

Estimated code review effort: 3 (Moderate) | ~25 minutes

---

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.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[tt-a1i]

Re-reviewed current head 86ebff2 on dev2d87fec. The unsafe-coordinate loop finding is fixed and the package is current. The private spatial index retains the exact clearance calculation and deterministic ordering; its fallbacks bound index construction without changing authored geometry.

Verified 2,000 seeded differential cases, 201 focused geometry/layout/checker tests, current-head 66 gate/geometry tests, package freshness and unchanged golden output. The sparse 3,000-route benchmark shows a substantial stage-level improvement; no end-to-end generation speedup is inferred. Final-head CI35134509724 passed all Node, shared-browser/WebM, package and Windows gates. No blocking finding remains; ready for dev integration.