Conversation
Bring the static numbers in README.md back in sync with the released binary: startup-banner example (v9.0.0 → v9.2.0, 14/76 → 25/110), preset tool counts (core 14→25, review 19→42, full 80+→110, etc.), "20 quality checks" → 21, "LIP v2.0" → "LIP v2.3+", and drop the "v9.0" prefix on the Compound Operations heading. The npm version shield is dynamic so no change needed there. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Upstream renamed the crate; CKB follows. Mechanical rename only — no functional change to existing call sites; the Go API surface is identical. - Vendor path third_party/cartographer/mapper-core/cartographer/ → third_party/nyx-navigator/ (flattened; stale Python tooling and examples no longer carried). - Build tag cartographer → navigator. make build / make test resolve build-navigator / test-navigator. make build-fast (no Rust toolchain) unchanged. - Static lib libcartographer.a → libnavigator.a; FFI prefix cartographer_* → navigator_*; header cartographer.h → navigator.h; sync script scripts/sync-cartographer.sh → scripts/sync-nyx-navigator.sh. - Go package internal/cartographer → internal/navigator; error type CartographerError → NavigatorError. - ckb status backend label "cartographer" → "navigator". New FFI surfaces available for follow-up wiring (not yet exposed): navigator_poll_changes, navigator_doc_index, navigator_doc_context, navigator_query_docs. The navigator_render_architecture FFI shape is unchanged but accepts new format values (sequence, class, quadrant, er, ascii, html, plus cross-file sequence with an entry-point) — these can flow through getArchitecture / renderArchitecture without further bridge work. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Content edits to renamed files: build tag (//go:build navigator), package declaration (package navigator), C.cartographer_* → C.navigator_* calls in bridge.go, CFLAGS / LDFLAGS to third_party/nyx-navigator, header include navigator.h, and corresponding string updates in the sync script. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Three small fixes that surfaced from test-driving the navigator rename: 1. review_coupling.go: hidden-coupling pass only filtered the *missing* side for noise; when the *changed* side was itself a noise file (.gitignore, CHANGELOG.md, etc.) every co-change still got flagged. Added the symmetric filter; also added third_party/ to the noise prefix list (vendored code, same intent as vendor/ and node_modules/); and capped findings at 3 per changed file to match the regular pass. On the navigator rename PR: 779 → 43 hidden-coupling findings (95% reduction). 2. impact_outgoing_test.go: panic from "send on closed channel" race in the LIP fake. Cleanup closed reqC while connection-handler goroutines were still in flight on `reqC <- req`. Added a sync.WaitGroup for handlers and a tracked list of accepted conns; cleanup now closes the conns first, drains the handlers, then closes reqC. Verified clean across 20× stress under -tags navigator. 3. wide_result_budget_test.go: maxGetHotspotsBytes 10000 → 13000. Under -tags navigator each hotspot is augmented with up to 3 co-change partners from navigator.GitCochange (~1.6KB extra at limit=20). The 10000 budget was set when the augmentation didn't exist — the augmentation is intentional value-add, so accommodate it rather than truncate. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…p nyx naming Re-vendors the Rust crate from upstream CodeCartographer 1.3.3 (was the interim nyx-navigator 1.1.0 fork) and renames the CKB side back to `cartographer`: - third_party/nyx-navigator -> third_party/cartographer - internal/navigator -> internal/cartographer (package cartographer, CartographerError, status backend id "cartographer") - build tag navigator -> cartographer; FFI navigator_* -> codecartographer_*; header navigator.h -> codecartographer.h; -lnavigator -> -lcode_cartographer - scripts/sync-nyx-navigator.sh -> scripts/sync-cartographer.sh - Makefile/.gitignore updated; Makefile runs the localize script via `bash` so a re-vendor (rsync drops the exec bit) doesn't break `make build` Bundles the upstream go.mod-aware resolver fix, which eliminated fabricated dependency edges that leaked to users through getArchitecture, getBlastRadius, renderArchitecture, getModuleContext, analyzeImpact, prepareChange, and the review layers/arch-health checks. On CKB itself: edges 996 -> 769 (~23% fabricated), cycles 22 -> 0, bridges 244 -> 187, health 38 -> 71. Also pulls upstream 1.2-1.3 features: six new languages (Java, C#, Ruby, Kotlin, Swift, PHP) with call graphs and class diagrams, symbol-aware BM25 search, scope-qualified reach_symbol, orientation-first ranking, and O(N) import resolution. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Re-vendors the upstream edge-confidence change: every GraphEdge carries a `resolution` tag (exact | suffix | fuzzy), now exposed on cartographer.GraphEdge. Cycle detection and the "will create a cycle" prediction ignore fuzzy edges, so false cycles can't rest on a low-confidence stem guess — a safety net for the languages that still resolve heuristically, matching the go.mod backstop already in place for Go. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…endor) Re-vendors the upstream resolver extension: extensionless imports that carry directory structure (Python `from app.services import auth`, monorepo `components/Button`) resolve against the full path — module files and package dirs, preferring __init__.py / index.* entry points — instead of the last stem. Unresolvable qualified imports are external (no edge). Includes the Python extractor fix that preserves import keywords so dotted paths normalize. Cross-package same-name collisions (two `auth.py`) and third-party imports no longer fabricate edges, extending the go.mod precision to Python and pathed TS/JS. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Re-vendors the resolver change: JS/TS relative imports (`./bar`, `../models/user`) resolve against the source file's directory (path-precise, `..` walks up, package dirs use index.*); unresolved relatives are dangling refs (no edge). `export … from './x'` re-exports are now captured. Completes the standalone-resolver story alongside Go (go.mod) and Python (qualified dotted paths). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…re-vendor) Re-vendors the change that computes all structural metrics (bridges, cycles, god modules, layer violations, blast radius) on non-fuzzy edges only, so getArchitecture/getBlastRadius/analyzeImpact/prepareChange and the review layers/arch-health checks can't surface a finding built on a fabricated edge. CKB side: review's PR-split clustering (addCartographerEdges) skips fuzzy edges so a split recommendation isn't driven by a low-confidence dependency. No change to CKB's own numbers (Go edges are all exact/suffix); the effect is on polyglot targets CKB analyzes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…llback (re-vendor) Re-vendors the scaling work: parallel edge resolution and parallel betweenness centrality in rebuild_graph (the latter was the dominant serial cost at scale), plus an O(1) directory-suffix fallback (by_dir_last index) replacing the old O(dirs)-per-import scan. Betweenness node order is sorted before sampling so bridge/god-module counts are deterministic. Measured on a 40k-file tree: 56s -> 12.7s (~4.4x); Godot 16s -> ~2s (~8x), same results. CKB uses the parallel FFI path. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… (re-vendor) Re-vendor Cartographer with the betweenness rewrite. Centrality was 98% of a graph rebuild; it's now index-based + parallel (bit-identical across core counts) and topology-keyed cached so non-import edits skip the Brandes pass. Godot via the FFI path: full rebuild 3,197ms -> 76ms (~42x), identical output (3182 bridges / 68 god-modules / 0 cycles). The cache is the incremental win for watch/serve sessions and scales with repo size. FFI-tagged query tests green; full -tags cartographer build clean. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… (re-vendor) Re-vendor Cartographer with directory rollup and the resolution determinism fixes. Rollup folds the file-level graph to directory granularity and runs the full structural analysis on subsystems instead of files — on Godot, 4336 files / 3182 bridges becomes 160 subsystems / 139 bridges with actionable per- subsystem hints. Available via rebuild_graph_rolled_up(depth). Also picks up two determinism fixes (sorted resolver candidates + strongest- resolution edge dedup) that make the graph fully reproducible run-to-run. FFI-tagged query tests green; full -tags cartographer build clean. File-level metrics unchanged (3182 bridges / 60.0 on Godot). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
….0.0 + fixes)
Re-vendor CodeCartographer at 4.0.0 plus the very-large-repo crash fixes. On a
full Linux-kernel checkout (~64k C/H files) the vendored engine aborted the
whole process — including the CKB host via the FFI — through two unbounded
recursions (recursive tree-sitter walkers on deeply nested macro-generated C,
and recursive petgraph tarjan_scc). Parse now runs on a 256 MB-stack pool and
cycle detection uses an iterative Tarjan.
Verified: Health("/tmp/linux") via cgo completes without crashing (new
regression test in internal/cartographer, skips if /tmp/linux absent). FFI
C-ABI unchanged (no bridge changes). FFI-tagged query tests green; full
-tags cartographer build clean. Directly de-risks the very-large-C++-repo case.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Re-vendor the walker depth guard: a hard recursion ceiling (cap 50,000; the
Linux kernel's measured real peak is 3,348) as a belt against adversarial or
machine-generated deep nesting, on top of the 256 MB parse-pool stack. Prevents
a pathological single file from aborting the CKB host via the FFI. Verified:
Health("/tmp/linux") via FFI completes without crashing.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds a `cartographer` check to `ckb doctor` so a user can confirm the SCIP-free structural fast tier (dependents, blast-radius, rollup) is actually linked into the binary. CGO-free npm/Homebrew builds warn with a `make build` fix; source builds report the linked engine version. Bumps version 9.2.0 → 9.3.0 (version.go, npm, sarif testdata) and cuts the [Unreleased] CHANGELOG section as [9.3.0], covering the large-repo FFI crash fixes, depth guard, 4.0.0 re-vendor, parallel betweenness, directory rollup, and this doctor check. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
🟢 Change Impact Analysis
Blast Radius: 0 modules, 0 files, 0 unique callers 📝 Changed Symbols (82)
Recommendations
Generated by CKB |
NFR Tests ✅ 39 unchangedComparing PR against main branch (dynamic baseline). Regressions: 0 ✅ Thresholds: WARN ≥ +5% • FAIL ≥ +10% All scenarios
* = new scenario, compared against static baseline |
CKB Analysis
Risk factors: Large PR with 88 files • High churn: 29725 lines changed • Touches 11 hotspot(s) 👥 Suggested: @lisa.welsch1985@gmail.com (11%), @talantyyr@gmail.com (5%), @lisa@tastehub.io (10%)
🎯 Change Impact Analysis · 🟢 LOW · 82 changed → 0 affected
Symbols changed in this PR:
Recommendations:
💣 Blast radius · 0 symbols · 3 tests · 0 consumersTests that may break:
🔥 Hotspots · 11 volatile files
📦 Modules · 2 at risk
📊 Complexity · 9 violations
💡 Quick wins · 10 suggestions
📚 Stale docs · 197 broken references
Generated by CKB · Run details |
CKB Review: 🟡 WARN — 59/10088 files (+29725 changes) · 19 modules ·
Top Risks
Findings (6 actionable, 24 informational)
✂️ Suggested PR Split (13 clusters)
... and 3 more clusters Code HealthNew files: 1 (avg health: 90) Estimated review: not feasible as a single PR (88 files, 29725 lines, 13 clusters) Reviewers: lisa.welsch1985 (11%) · talantyyr (5%) · lisa (10%) |
The analyzeImpact LIP staleness check read the annotation key `lip:nyx-agent-lock` and reported "an active nyx.code agent". Debrand: key → `lip:agent-lock`, message → "an active agent". The LIP writer side is being updated in lockstep to the same generic key. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
1 participant
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
Cuts 9.3.0. Hardens the vendored Cartographer graph engine for very large C/C++ trees, re-vendors it at 4.0.0, parallelizes the hot paths, and adds a
ckb doctorcheck so users can confirm the SCIP-free fast tier is linked.Highlights
tarjan_sccis iterative, and walkers carry a hard depth ceiling (cap 50,000; measured real peak 3,348). Verified with an FFI regression test.ckb doctorcartographer check — reports whether the fast structural tier is linked (CGO-free builds warn with amake buildhint).Testing
-tags cartographerbuild clean; FFI-tagged tests green.getArchitecturereturned a module-level result via the fast tier with no SCIP index, no crash.Version
9.2.0 → 9.3.0(version.go, npm/package.json, sarif testdata, CHANGELOG).🤖 Generated with Claude Code