Optimizations

.codegraphignore

@@ -0,0 +1,11 @@
+# Repo-local ignores for codegraph indexing.
+# This file uses .gitignore-like patterns and is loaded by codegraph in addition
+# to config defaults.
+
+.codegraph/**
+
+codegraph
+codegraph.exe
+codegraph.sqlite*
+.gotelemetry/**
+.claude/**

.github/workflows/release.yml

@@ -23,7 +23,7 @@ jobs:
       - name: Run tests
         run: go test ./...
 
-  build-and-release:
+  build:
     needs: test
     runs-on: ubuntu-latest
     strategy:
@@ -48,6 +48,7 @@ jobs:
           - goos: windows
             goarch: arm64
             archive_ext: zip
+
     steps:
       - name: Check out repository
         uses: actions/checkout@v4
@@ -93,12 +94,80 @@ jobs:
             sha256sum "${dist}/${archive_base}.tar.gz" > "${dist}/${archive_base}.tar.gz.sha256"
           fi
 
-      - name: Publish release assets
-        uses: softprops/action-gh-release@v2
+      - name: Upload build artifacts
+        uses: actions/upload-artifact@v4
         with:
-          generate_release_notes: true
-          files: |
+          name: release-${{ matrix.goos }}-${{ matrix.goarch }}
+          path: |
             dist/*.tar.gz
             dist/*.tar.gz.sha256
             dist/*.zip
             dist/*.zip.sha256
+
+  release:
+    needs: build
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+
+    steps:
+      - name: Check out repository
+        uses: actions/checkout@v4
+
+      - name: Download all build artifacts
+        uses: actions/download-artifact@v4
+        with:
+          path: dist
+
+      - name: Flatten downloaded artifacts
+        shell: bash
+        run: |
+          set -euo pipefail
+          mkdir -p release-dist
+          find dist -type f \( -name "*.tar.gz" -o -name "*.tar.gz.sha256" -o -name "*.zip" -o -name "*.zip.sha256" \) -exec cp {} release-dist/ \;
+
+      - name: Extract release notes from CHANGELOG.md
+        shell: bash
+        run: |
+          set -euo pipefail
+          version="${GITHUB_REF_NAME}"
+          section_header="## ${version} -"
+
+          awk -v header="$section_header" '
+            BEGIN { found=0 }
+            $0 ~ /^## / && found { exit }
+            index($0, header) == 1 { found=1 }
+            found { print }
+          ' CHANGELOG.md > release-notes-full.md
+
+          if ! grep -q "^## ${version} -" release-notes-full.md; then
+            echo "Could not find release notes section for tag ${version} in CHANGELOG.md"
+            exit 1
+          fi
+
+          tail -n +2 release-notes-full.md > release-notes.md
+
+      - name: Create or update GitHub release
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        shell: bash
+        run: |
+          set -euo pipefail
+
+          if gh release view "${GITHUB_REF_NAME}" >/dev/null 2>&1; then
+            gh release edit "${GITHUB_REF_NAME}" \
+              --title "${GITHUB_REF_NAME}" \
+              --notes-file release-notes.md
+          else
+            gh release create "${GITHUB_REF_NAME}" \
+              --title "${GITHUB_REF_NAME}" \
+              --notes-file release-notes.md
+          fi
+
+      - name: Upload release assets
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        shell: bash
+        run: |
+          set -euo pipefail
+          gh release upload "${GITHUB_REF_NAME}" release-dist/* --clobber

.gitignore

@@ -21,3 +21,4 @@ codegraph.exe
 codegraph.sqlite
 codegraph.sqlite-shm
 codegraph.sqlite-wal
+OPTIMIZATIONS.md

CHANGELOG.md

@@ -1,97 +1,99 @@
 # Changelog
 
-# Release v1.0.9 - 16-04-2026
+## v1.1.0 - 27-04-2026
 
-Improved Node.js repo indexing stability by hard-skipping common generated/tooling directories (for example node_modules and .next), refining default excludes, and clarifying ignore override behavior.
+A performance, correctness, and operability release focused on faster indexing, schema-backed edge resolution, streaming exports, and tighter DB lifecycle handling.
 
-## Changes
+### Highlights
+- **Indexing performance:** Repo-wide change detection is dramatically lighter — narrower DB projections, the existing-files load now overlaps the filesystem walk, per-row payloads are slimmed, and content-hash reads are deferred to the slow path. No-op repo updates do far less work and allocate substantially less.
+- **Edge resolver scaling:** Slash-suffix, dot-tail2, and the dominant 2-dot dot-suffix paths are now schema-backed indexed joins instead of repo-wide symbol scans, removing large constant-factor and quadratic costs on bigger graphs (the 2-dot dot-suffix case is roughly three orders of magnitude faster on the large-scale benchmark). A steady-state guard short-circuits resolver work when no unresolved edges remain. Cross-file edge resolution on update runs stays path-scoped instead of going repo-wide.
+- **Write-path allocations:** Symbol, edge, and batch insert SQL plus IN-placeholder strings are cached; edge source selection uses span/binary-search; residual allocations in the multi-file batch path are trimmed; tokenization allocations are reduced.
+- **Streaming exports:** JSON and DOT exports stream through paged store calls. Peak memory is now O(page) instead of O(repo) for the unbounded paths, and the bounded-page CLI uses the same primitives.
+- **Watch & update correctness:** Repo include/exclude is applied consistently; the dirty-file queue is crash-safe via claim/delete; chmod-only and directory-create events are ignored; `--jsonl` output is stable across `index`, `watch`, and `doctor`.
+- **DB lifecycle hardening:** Per-version migrations run on a single connection under `BEGIN IMMEDIATE`; pragmas apply across the pool with a unified driver-aware busy-retry policy; `doctor` uses the same DSN/pragma path; `clean` and `doctor` add ANALYZE, WAL checkpoint, incremental vacuum, and a `--deep` integrity check.
+- **Query correctness fixes:** `RelatedTests(file)` is correctly scoped via `target_file_id`; symbol lookup is deterministic; deleted-file graph rows are purged and cross-file references nullified, including ghost `test_links` rows pointing at deleted files; duplicate token-stat counting in batched writes is fixed; `FindDeadCode` is faster via dedicated indexes.
+- **Operability:** New `index_smoke` runner produces compact perf-diff output; repo artifacts default to `.codegraph/` with legacy fallback; the CLI gains a command registry, per-command help, and canonical query command names with backward-compatible aliases; benchmarks capture `--sqlite-profile` and host context for reproducible perf comparisons.
 
-### Fixes
-- **Indexer:** Established a strict skip policy for common Node.js generated directories (e.g., `node_modules`, `.next`, `.nuxt`). These are now hardcoded and enforced early during filesystem traversal.
-- **Indexer:** Clarified ignore override behavior; hardcoded skips are now non-overridable via negation patterns in `.codegraphignore` to ensure predictable indexer performance.
-- **Config:** Centralized default exclude patterns to maintain consistency across the CLI and indexer.
-- **SQL Hardening:** Added explicit bounds and safety checks for all path-filtering SQL queries.
+## v1.0.9 - 16-04-2026
 
-## Upgrade Notes
-- No required migrations or configuration changes.
+Improved Node.js repo indexing stability by hard-skipping common generated/tooling directories (for example `node_modules` and `.next`), refining default excludes, and clarifying ignore override behavior.
 
-## v1.0.7 - 27-03-2026
+### Changed
+- **indexer:** Enforce strict early skips for common Node.js generated directories; hardcoded skips are non-overridable via `.codegraphignore` negations. (#8)
+- **config:** Centralize default exclude patterns to keep CLI/indexer behavior consistent. (#8)
 
 ### Fixed
+- **sql:** Add explicit bounds/safety checks for path-filtering queries. (#8)
+- **build/release:** Carry through `CGO_ENABLED=0` + tree-sitter cross-compilation fixes and release diagnostics. (#4, #5, #6)
 
-- Restored release cross-compilation by splitting tree-sitter adapters behind `//go:build cgo` and using heuristic parsers in `CGO_ENABLED=0` builds.
+### Docs
+- **readme/changelog:** Update Node.js support status and release notes. (#8)
 
-## v1.0.6 - 26-03-2026
+## v1.0.8 - 27-03-2026
 
 ### Fixed
-
-- Stopped sending JSON-RPC responses to MCP notifications (`notifications/initialized` and other `notifications/*` methods), which violated the protocol and caused strict clients to fail on connect.
-- Changed tool schema `"required": null` to omit the field when empty, fixing JSON Schema validation failures in strict MCP clients.
-- Removed non-standard `structuredContent` field from tool call responses to conform to the MCP spec.
-- Routed unhandled-method logging through the configured stderr writer instead of Go's default logger.
+- **build:** Restore `CGO_ENABLED=0` cross-compilation by splitting tree-sitter adapters behind `//go:build cgo` and using heuristic parsers in no-cgo builds. (#5)
 
 ### Changed
+- **ci:** Add release build diagnostics to improve cross-platform release debugging. (ci/workflow)
+
+## v1.0.7 - 27-03-2026
 
-- `NewServer` now accepts an `io.Writer` for error output, giving callers control over diagnostic logging.
+### Fixed
+- **mcp:** Tighten MCP protocol compliance for stricter clients. (#3)
 
 ### Docs
+- Add `v1.0.6` changelog entry. (docs)
 
-- Added Claude Code MCP setup section to README with `.mcp.json` examples.
-- Added missing `list_scans` and `latest_scan_errors` to the MCP tools list in README.
-- Added one-line descriptions to all 14 MCP tools in README.
+## v1.0.6 - 26-03-2026
+
+### Fixed
+- **mcp:** Stop sending JSON-RPC responses to notifications; fix JSON Schema `required` handling; remove non-standard fields; route unhandled-method logging via configured stderr writer. (mcp)
+
+### Changed
+- `NewServer` accepts an `io.Writer` for error output, giving callers control over diagnostic logging. (mcp)
+
+### Docs
+- Add Claude Code MCP setup examples; add missing tools to MCP docs list; add short tool descriptions. (readme)
 
 ## v1.0.5 - 21-03-2026
 
 ### Fixed
+- Default to a repo-local SQLite DB (while continuing to recognize legacy locations) and exclude repo DB artifacts from indexing. (config/store)
+
+### Changed
+- Treat prior global `db_dir` default as legacy so existing installs fall forward safely. (config)
+
+## v1.0.4 - 18-03-2026
 
-- Switched the default database location to per-repository `codegraph.sqlite` and excluded `codegraph.sqlite*` from indexing so repo-local DB mode works on repeated `index` and `serve` runs.
-- Treated the previous global `db_dir` default as a legacy value so existing installs fall forward to repo-local DB behavior without manual config edits.
-- Updated Codex MCP setup guidance and examples to use `config.toml` with `startup_timeout_sec = 60`.
+### Docs
+- README update to include graph/export usage. (docs)
 
 ## v1.0.3 - 18-03-2026
 
-### Changed
+### Added
+- **cli:** `watch`, `benchmark`, `config init`, `clean`, `doctor --fix`, and `--jsonl` output for long-running/indexing workflows. (cli)
+- **mcp/query:** Query commands + tools, including offset pagination and supported-languages introspection. (mcp)
+- **parser:** Heuristic adapters for major languages plus a Python adapter. (parser)
+- **export:** Include symbols + edges in graph exports; support export streaming. (export)
 
-- Simplified core code paths in `internal/store`, `internal/mcp`, `internal/cli`, and parser adapters while preserving behavior.
-- Reduced duplicated row-scanning and pagination parsing logic to improve maintainability.
-- Cached MCP tool definition payload construction for lower repeated allocation overhead on `tools/list`.
+### Changed
+- **indexer/scan:** `.codegraphignore` negation patterns; per-language scan coverage; best-effort parse policy; batched metadata writes and scoped edge resolution. (indexer)
+- **performance:** Parallelize indexing and reduce allocation/IO overhead; improve watcher flush/coalescing; add scan phase timings; SQLite/store tuning. (perf)
 
 ### Notes
+- **licensing:** Relicensed under FSL-1.1 to prevent commercial reselling. (license)
 
-- This release is focused on code quality, readability, and safe internal optimization with no intended user-facing breaking changes.
+## v1.0.2 - 18-03-2026
 
-## v1.0.2 - 2026-03-18
-
-### Highlights
-
-- Added `watch --jsonl`, benchmark improvements, and `config init` workflow updates.
-- Added/expanded query, export, MCP, and parser capabilities (including Python and heuristic multi-language adapters).
-- Improved indexing and store performance with batching, scoped edge resolution, and scan/stat tuning.
-- Added cleaner local maintenance workflows (`clean`, doctor improvements, and setup/path guidance updates).
+### Fixed
+- Installation hardening + README updates to unblock `go install` workflows. (install/docs)
 
 ## v1.0.1 - 18-03-2026
 
-### Fixes
-
-- Corrected Go module path from `github.com/example/localcodegraph` to `github.com/isink17/codegraph`.
-- Updated internal imports to match the published module path.
-- Updated install docs with actual module path and Go install/PATH troubleshooting guidance.
+### Fixed
+- Correct Go module path to `github.com/isink17/codegraph`; align imports and install docs accordingly. (install/docs)
 
 ## v1.0.0 - 18-03-2026
 
 Initial public release of `codegraph`.
-
-### Highlights
-
-- Added a local-first Go CLI for installation, indexing, updates, stats, graph export, doctor checks, watch mode, and MCP serving.
-- Added a SQLite-backed repository graph with explicit migrations and incremental file hashing.
-- Added a parser abstraction with a working Go parser adapter and a clean seam for future Tree-sitter adapters.
-- Added MCP stdio support with generic tools for indexing, symbol lookup, call graph navigation, impact analysis, related test discovery, semantic search, and graph stats.
-- Added release automation for macOS, Linux, and Windows GitHub release artifacts.
-- Added agent-oriented documentation for Codex-style clients, Gemini CLI, and Claude-compatible MCP configuration examples.
-- Added initial automated tests for install flow, platform paths, incremental indexing, and MCP `graph_stats`.
-
-### Notes
-
-- This release targets local-first usage on macOS and Linux first, with Windows kept feasible by design.
-- The parser subsystem is intentionally shaped for Tree-sitter-backed adapters, while the first shipping implementation uses the Go standard AST for reliability and simple installation.

README.md

@@ -359,6 +359,8 @@ dist/
 *.generated.go
 ```
 
+Note: `.codegraph/` is always skipped by the indexer (repo-local config/artifacts).
+
 ---
 
 ## Architecture

examples/README.md

@@ -42,13 +42,14 @@ If `codegraph` is not on `PATH`, replace `command` and `args` with a form that l
 By default, `codegraph` stores its database in the served repository as:
 
 ```text
-codegraph.sqlite
+.codegraph/codegraph.sqlite
 ```
 
 If you do not want to commit the local database, ignore:
 
 ```text
-codegraph.sqlite
-codegraph.sqlite-shm
-codegraph.sqlite-wal
+.codegraph/
+
+# Legacy location (still recognized if present):
+codegraph.sqlite*
 ```