Widthdom · Widthdom · Aug 9, 2026 · Aug 8, 2026 · Aug 8, 2026 · Aug 8, 2026
diff --git a/DEVELOPER_GUIDE.md b/DEVELOPER_GUIDE.md
diff --git a/TESTING_GUIDE.md b/TESTING_GUIDE.md
diff --git a/changelog.d/unreleased/+allocation-free-bounded-regex-enumeration.changed.md b/changelog.d/unreleased/+allocation-free-bounded-regex-enumeration.changed.md
@@ -0,0 +1,22 @@
+---
+category: changed
+affected:
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+  - src/CodeIndex/Indexer/BoundedRegex.cs
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.cs
+  - src/CodeIndex/Indexer/References/Languages/SqlReferenceExtractor.AlterTargets.cs
+  - src/CodeIndex/Indexer/References/Languages/SqlReferenceExtractor.Sources.cs
+  - src/CodeIndex/Indexer/References/Languages/SqlReferenceExtractor.StatementState.cs
+  - src/CodeIndex/Indexer/References/Support/LanguageReferenceExtractionSupport.Utilities.cs
+  - tests/CodeIndex.Tests/BoundedRegexTests.cs
+  - tests/CodeIndex.Tests/PerformanceTests.cs
+---
+
+## English
+
+- **Lazy regex match enumeration no longer allocates iterator wrappers on indexing hot paths** — instance-regex scans now use concrete value enumerables and enumerators through reference extraction and SQL helpers. Direct scans and already-full reference caps avoid iterator and interface-boxing allocations while preserving timeout diagnostics, zero-length and `\G` progression, explicit and right-to-left start positions, early termination, and LINQ compatibility.
+
+## 日本語
+
+- **indexing の hot path で lazy regex match 列挙の iterator wrapper allocation を解消しました** — instance regex の走査は、reference extraction と SQL helper を通して concrete な value enumerable / enumerator を使います。direct scan と既に満杯の reference cap は iterator および interface boxing allocation を避けながら、timeout diagnostic、zero-length と `\G` の進行、明示的および right-to-left の開始位置、早期終了、LINQ 互換性を維持します。
diff --git a/changelog.d/unreleased/+avoid-path-casing-self-invalidation.fixed.md b/changelog.d/unreleased/+avoid-path-casing-self-invalidation.fixed.md
@@ -0,0 +1,15 @@
+---
+category: fixed
+affected:
+  - TESTING_GUIDE.md
+  - src/CodeIndex/Cli/PathCasing.cs
+  - tests/CodeIndex.Tests/PathCasingTests.cs
+---
+
+## English
+
+- **Exact-case path comparisons no longer mutate directories during indexing** — path equality now resolves ordinal matches and impossible unequal-length matches, while parent-boundary checks accept exact-case matches and descendants, before either operation probes filesystem case sensitivity. Empty nested workspace members therefore keep their discovery snapshot stable instead of receiving a temporary case-probe entry after traversal.
+
+## 日本語
+
+- **大小文字まで一致する path 比較で index 中の directory を変更しないよう修正しました** — filesystem の case sensitivity を probe する前に、path equality は ordinal 一致と長さの異なる不一致を確定し、親境界の判定は大小文字まで一致する同一 path と descendant を受理します。これにより、空の nested workspace member は走査後に一時 case-probe entry を作られず、discovery snapshot を安定して維持します。
diff --git a/changelog.d/unreleased/+csharp-reference-graph-facts.internal.md b/changelog.d/unreleased/+csharp-reference-graph-facts.internal.md
@@ -0,0 +1,17 @@
+---
+category: internal
+affected:
+  - src/CodeIndex/Database/DbWriter.References.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceGraphRefreshScope.cs
+  - tests/CodeIndex.Tests/DatabaseTests.cs
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+---
+
+## English
+
+- **C# graph finalization reuses materialized scalar facts** — full, scoped, and retained reference-graph refreshes now compute C# reference and symbol arity, receiver, constructor, and value-type facts once per applicable row, avoiding repeated managed SQLite callbacks for every candidate in large graphs.
+
+## 日本語
+
+- **C# graph finalization が materialize 済み scalar fact を再利用します** — full / scoped / retained の reference-graph refresh は、C# reference と symbol の arity、receiver、constructor、value-type fact を対象 row ごとに1回だけ計算し、巨大 graph の candidate ごとに managed SQLite callback を反復する処理を避けるようになりました。
diff --git a/changelog.d/unreleased/+csharp-reference-graph-identity-facts.internal.md b/changelog.d/unreleased/+csharp-reference-graph-identity-facts.internal.md
@@ -0,0 +1,17 @@
+---
+category: internal
+affected:
+  - src/CodeIndex/Database/DbWriter.References.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceGraphRefreshScope.cs
+  - tests/CodeIndex.Tests/DatabaseTests.cs
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+---
+
+## English
+
+- **C# graph finalization reuses materialized type and constructor identities** — full, scoped, and retained reference-graph refreshes now build project/file-local type identities and ranked constructor-owner identities once per applicable symbol, replacing repeated candidate-side string construction and range scans with TEMP primary-key lookups.
+
+## 日本語
+
+- **C# graph finalization が materialize 済み type / constructor identity を再利用します** — full / scoped / retained の reference-graph refresh は project / file-local type identity と順位付き constructor-owner identity を対象 symbol ごとに1回だけ構築し、candidate 側で反復していた文字列生成と range scan を TEMP 主キー lookup に置き換えます。
diff --git a/changelog.d/unreleased/+defer-reference-indexes.changed.md b/changelog.d/unreleased/+defer-reference-indexes.changed.md
@@ -0,0 +1,16 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Database/ReferenceSecondaryIndexBulkLoadGuard.cs
+  - src/CodeIndex/Database/ReferenceSecondaryIndexSql.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.FullScan.cs
+  - src/CodeIndex/Mcp/McpToolHandlers.Indexing.Execution.cs
+---
+
+## English
+
+- **Fresh and rebuilt indexes persist large reference sets without maintaining every query index row by row** — CLI indexing now defers reference query and graph secondary indexes inside its outer transaction and restores them once before graph finalization. Empty-database MCP indexing uses the same optimization with recoverable cleanup, while maintenance indexes, rollback safety, read repair, and the final database schema remain unchanged.
+
+## 日本語
+
+- **fresh / rebuild index が、大量の reference を query index ごとに逐次更新せず永続化するようになりました** — CLI indexing は外側 transaction 内で reference query / graph 用 secondary index を遅延し、graph finalization 前に1回だけ復元します。空 database から始める MCP indexing も recoverable cleanup 付きで同じ最適化を使い、保守用 index、rollback safety、read repair、最終 database schema は従来どおり維持します。
diff --git a/changelog.d/unreleased/+marker-gated-reference-extraction.changed.md b/changelog.d/unreleased/+marker-gated-reference-extraction.changed.md
@@ -0,0 +1,20 @@
+---
+category: changed
+affected:
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.CoreCallReferences.cs
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.CoreTypeReferences.cs
+  - src/CodeIndex/Indexer/References/Languages/
+  - src/CodeIndex/Indexer/References/Support/
+  - tests/CodeIndex.Tests/ReferenceExtractorMarkerGateTests.cs
+  - tests/CodeIndex.Tests/ReferenceExtractorPerformanceBudgetTests.cs
+---
+
+## English
+
+- **Large multi-language indexes avoid impossible per-line reference regex work** — reference extraction now checks cheap syntax markers before entering expensive call, type, framework, and markup patterns across C#, Java, Kotlin, JavaScript/TypeScript, Go, Dockerfile, GraphQL, HTML, Markdown, XAML/XML, Terraform, JSON, and GitHub Actions. Stateful multi-line parsing and syntax without the shared markers retain their existing references, while markerless source lines complete with substantially less regex work.
+
+## 日本語
+
+- **大規模な multi-language index で成立しない行単位 reference regex の実行を回避しました** — C#、Java、Kotlin、JavaScript/TypeScript、Go、Dockerfile、GraphQL、HTML、Markdown、XAML/XML、Terraform、JSON、GitHub Actions の call、type、framework、markup pattern を処理する前に、安価な構文 marker を確認します。stateful な複数行解析と共有 marker を持たない構文の既存 reference は維持しつつ、marker のない source 行に対する regex work を大幅に減らします。
diff --git a/changelog.d/unreleased/+mixed-filesystem-marker-scope.fixed.md b/changelog.d/unreleased/+mixed-filesystem-marker-scope.fixed.md
@@ -0,0 +1,17 @@
+---
+category: fixed
+affected:
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.ProjectMarkers.cs
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.ScanOrchestration.cs
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.Types.cs
+  - tests/CodeIndex.Tests/FileIndexerTests.cs
+  - TESTING_GUIDE.md
+---
+
+## English
+
+- **Project family scopes now honor case policy changes inside mixed filesystems** — the completed scan snapshot keeps case-only marker directories distinct under case-sensitive children while still resolving aliases under case-insensitive children, without returning to live marker enumeration or adding filesystem probes.
+
+## 日本語
+
+- **混在 filesystem 内で case policy が変わる場合も project family scope が正しくなりました** — 完了した scan snapshot は case-sensitive child 配下の大小文字だけが異なる marker directory を分離しつつ、case-insensitive child 配下の alias は同一 scope として解決し、live marker 列挙への後退や filesystem probe の追加も行いません。
diff --git a/changelog.d/unreleased/+parallel-authoritative-csharp-updates.changed.md b/changelog.d/unreleased/+parallel-authoritative-csharp-updates.changed.md
@@ -0,0 +1,24 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Cli/CliFlagSchema.cs
+  - src/CodeIndex/Cli/EnvironmentVariableInventory.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.FullScan.Interruption.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.Update.FileLoop.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.Update.FilePersistence.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.Update.ParallelExtraction.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.WorkItems.cs
+  - src/CodeIndex/Indexer/Hooks/PostExtractionHooks.cs
+  - tests/CodeIndex.Tests/IndexCommandRunnerUpdateTests.cs
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+---
+
+## English
+
+- **Large authoritative C# scoped updates now extract files concurrently within bounded ordered windows** — eligible static-interface workspace updates reuse a fixed worker pool, cap each window at twice the worker count, and keep SQLite persistence, hooks, readiness, byte accounting, and progress on one target-ordered consumer. Three file-stat barriers, nullable language re-detection, serial probe/hook/filter fallbacks, phase-aware cancellation and stall handling, and source-contract candidate ordering preserve the existing update semantics while reducing extraction time on large C# workspaces.
+
+## 日本語
+
+- **大規模な authoritative C# scoped update が、上限付き ordered window 内で file extraction を並列実行するようになりました** — static-interface workspace の対象 update は固定 worker pool を再利用し、各 window を worker 数の2倍までに制限します。SQLite persistence、hook、readiness、byte accounting、progress は target 順の single consumer に維持します。3段階の file-stat barrier、nullable language re-detection、probe / hook / filter の serial fallback、phase-aware な cancellation / stall 処理、source-contract candidate の順序制御により既存の update semantics を保ちながら、大規模 C# workspace の extraction 時間を短縮します。
diff --git a/changelog.d/unreleased/+prune-reference-indexes.changed.md b/changelog.d/unreleased/+prune-reference-indexes.changed.md
@@ -0,0 +1,19 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Database/ReferenceSecondaryIndexSql.cs
+  - src/CodeIndex/Database/ReferenceSecondaryIndexBulkLoadGuard.cs
+  - src/CodeIndex/Database/DbContext.cs
+  - src/CodeIndex/Database/DbContext.ReadMigrations.cs
+  - src/CodeIndex/Database/DbReader.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceGraphRefreshScope.cs
+  - src/CodeIndex/Database/DbWriter.References.cs
+---
+
+## English
+
+- **Large reference graphs rebuild fewer redundant index pages** — canonical schema creation, read migration, and bulk-load recovery now retire six single-column name/container indexes already covered by retained composite prefixes. The former all-row folded mutual-recursion index is replaced by a partial index containing only unresolved callable edges, while folded and legacy NOCASE exact queries continue to use retained composite indexes and older databases are pruned on open.
+
+## 日本語
+
+- **大規模なreference graphで冗長なindex pageの再構築を削減しました** — canonical schema作成、read migration、bulk-load recoveryは、保持済みcomposite prefixで代替できるname/container単一カラムindex 6本を退役させます。全rowを保持していた旧folded mutual-recursion indexは未解決のcallable edgeだけを含むpartial indexへ置き換え、folded/legacy NOCASEのexact queryは保持済みcomposite indexを引き続き利用し、旧databaseもopen時にpruneします。
diff --git a/changelog.d/unreleased/+reuse-project-marker-scan.changed.md b/changelog.d/unreleased/+reuse-project-marker-scan.changed.md
@@ -0,0 +1,17 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.ProjectMarkers.cs
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.DirectoryEnumeration.cs
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.ScanOrchestration.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.FullScan.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.DryRun.cs
+---
+
+## English
+
+- **Full indexing no longer walks large directory trees again for project-family metadata** — authoritative scans now collect C#, VB, F#, and MSBuild marker fingerprints during normal discovery and reuse a complete marker-directory snapshot for per-file family scopes. Fingerprint limits, ignore boundaries, incomplete-scan warnings, and live fallbacks remain fail-closed.
+
+## 日本語
+
+- **full indexing が project-family metadata のために巨大な directory tree を再走査しなくなりました** — authoritative scan は通常 discovery 中に C#、VB、F#、MSBuild の marker fingerprint を収集し、complete な marker-directory snapshot を file ごとの family scope に再利用します。fingerprint 上限、ignore 境界、不完全 scan warning、live fallback は引き続き fail-closed です。
diff --git a/changelog.d/unreleased/+single-pass-content-facts.changed.md b/changelog.d/unreleased/+single-pass-content-facts.changed.md
@@ -0,0 +1,18 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Indexer/Scanning/FileContentLoader.cs
+  - src/CodeIndex/Indexer/Scanning/NormalizedContentFacts.cs
+  - src/CodeIndex/Indexer/Scanning/ChunkSplitter.cs
+  - src/CodeIndex/Indexer/Scanning/FileIndexer.ContentValidation.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.WorkItems.cs
+  - src/CodeIndex/Mcp/McpToolHandlers.Indexing.Execution.cs
+---
+
+## English
+
+- **Large multi-language files are normalized, analyzed, chunked, and validated with one shared content scan** — full scans, scoped updates, dry runs, and MCP indexing now carry normalized line, size, conflict-marker, replacement-character, FTS-token, and chunk-boundary facts through every language-independent path instead of rediscovering them in each consumer. Short files avoid unused boundary/token tracking, and high-ratio invalid UTF-8 input no longer retains one replacement-line entry per damaged line.
+
+## 日本語
+
+- **大規模な multi-language file の正規化・解析・chunk 分割・validation を1回の共有 content scan にまとめました** — full scan、scoped update、dry-run、MCP indexing は、正規化後の行数、size、conflict marker、replacement character、FTS token、chunk 境界の facts を全言語共通経路で引き回し、consumer ごとの再検出を省きます。短い file では不要な境界 / token 追跡を避け、高比率の invalid UTF-8 input では破損行ごとの replacement-line entry を保持しません。
diff --git a/changelog.d/unreleased/+stage-reference-index-restoration.internal.md b/changelog.d/unreleased/+stage-reference-index-restoration.internal.md
@@ -0,0 +1,25 @@
+---
+category: internal
+affected:
+  - src/CodeIndex/Database/ReferenceSecondaryIndexSql.cs
+  - src/CodeIndex/Database/ReferenceSecondaryIndexBulkLoadGuard.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceGraphRefreshScope.cs
+  - src/CodeIndex/Database/DbWriter.References.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.FullScan.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.Update.cs
+  - src/CodeIndex/Mcp/McpToolHandlers.Indexing.Execution.cs
+  - tests/CodeIndex.Tests/ReferenceSecondaryIndexBulkLoadGuardTests.cs
+  - tests/CodeIndex.Tests/IndexCommandRunnerReferenceIndexBulkLoadTests.cs
+  - tests/CodeIndex.Tests/McpServerToolsCallTests.cs
+  - tests/CodeIndex.Tests/DatabaseTests.cs
+  - DEVELOPER_GUIDE.md
+  - TESTING_GUIDE.md
+---
+
+## English
+
+- **Reference-index restoration is staged around graph finalization** — large CLI and MCP bulk loads now keep query-only reference indexes absent through identity resolution, restore only three reverse-edge indexes for mutual-recursion evaluation, and rebuild the remaining query indexes afterwards. Active scoped graph refreshes are promoted to full plans before deferral, while cancellation, rollback, recovery, heartbeat, and SQLite `changes()` contracts remain intact.
+
+## 日本語
+
+- **reference index の復元を graph finalization の前後に段階化しました** — 大規模な CLI / MCP bulk load は identity resolution 中も query-only reference index を外したままにし、mutual-recursion 評価には reverse-edge 用3本だけを復元して、残りの query index はその後に再構築します。active な scoped graph refresh は遅延前に full plan へ昇格し、cancellation、rollback、recovery、heartbeat、SQLite `changes()` の各契約を維持します。
diff --git a/changelog.d/unreleased/+streamline-reference-persistence.changed.md b/changelog.d/unreleased/+streamline-reference-persistence.changed.md
@@ -0,0 +1,16 @@
+---
+category: changed
+affected:
+  - src/CodeIndex/Cli/IndexCommandRunner.FullScan.cs
+  - src/CodeIndex/Cli/IndexCommandRunner.Update.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceSql.cs
+  - src/CodeIndex/Database/DbWriter.References.cs
+---
+
+## English
+
+- **Large multi-language reference batches perform less index maintenance, parameter binding, and tuple hashing** — existing-database full scans that cross the established FTS bulk-load threshold now defer reference-query indexes transactionally, while scoped updates do the same recoverably only after at least 64 targets reach 60% of indexed files. Reference inserts also write the normalized legacy context column as a SQL `NULL` literal and carry materialized reference-line IDs through compact ordinal arrays, avoiding one binding and a repeated file/line/context dictionary lookup per edge across fresh and replacement paths. Small updates retain every query index and avoid rebuild overhead.
+
+## 日本語
+
+- **大規模な multi-language reference batch の index maintenance、parameter binding、tuple hashing を削減しました** — 既存DBの full scan が既定の FTS bulk-load 閾値を超えた場合は reference query index を transaction 内で一時退避し、scoped update では64 target以上かつ indexed fileの60%以上の場合だけ recoverable に同じ処理を行います。reference insert は正規化済み legacy context column を SQL の `NULL` literal として書き込み、materialize 済み reference-line ID を compact な ordinal array で渡します。fresh / replacement の全経路で edge ごとの binding 1件と file / line / context dictionary の再 lookup を省きます。小規模 update は全 query index を維持し、再構築 overhead を避けます。
diff --git a/src/CodeIndex/Cli/CliFlagSchema.cs b/src/CodeIndex/Cli/CliFlagSchema.cs
@@ -670,7 +670,7 @@ private static IReadOnlyList<CliFlag> BuildAll()
             new() { Name = "--max-file-bytes", ValuePlaceholder = "<bytes>", Description = "Override the per-file indexing size limit", PrimaryCommands = Set("index") },
             new() { Name = "--max-symbols-per-file", ValuePlaceholder = "<n>", Description = "Skip file content, symbols, and references when one file emits too many symbols (max 50000)", PrimaryCommands = Set("index") },
             new() { Name = "--max-references-per-file", ValuePlaceholder = "<n>", Description = "Skip references when one file emits too many references (max 1000000)", PrimaryCommands = Set("index") },
-            new() { Name = "--parallelism", ValuePlaceholder = "<n>", Description = "Full-scan extraction worker count (default: CPU count capped at 8; explicit max: 16; also honors CDIDX_INDEX_PARALLELISM)", PrimaryCommands = Set("index") },
+            new() { Name = "--parallelism", ValuePlaceholder = "<n>", Description = "Full-scan and authoritative C# update extraction worker count (default: CPU count capped at 8; explicit max: 16; also honors CDIDX_INDEX_PARALLELISM)", PrimaryCommands = Set("index") },
             new() { Name = "--memory-trace", Description = "Include phase memory samples in index JSON output", PrimaryCommands = Set("index") },
             new() { Name = "--commits", ValuePlaceholder = "<commit-ref>", Description = "Update files changed in given git commits", PrimaryCommands = Set("index") },
             new() { Name = "--changed-between", ValuePlaceholder = "<old-ref> <new-ref>", Description = "Update files changed between two git refs", PrimaryCommands = Set("index") },