Widthdom · Widthdom · Jul 27, 2026 · Jul 27, 2026 · Jul 27, 2026 · Jul 27, 2026
diff --git a/DEVELOPER_GUIDE.md b/DEVELOPER_GUIDE.md
@@ -1125,6 +1125,8 @@ every non-inspect command must reject that cursor family before execution.
 
 `symbol_references.reference_kind` stores raw extractor labels. Default call-graph surfaces (`callers`, `callees`, inspect/analyze caller and callee bundles, and their JSON/MCP fields) expose the canonical public vocabulary `call`, `instantiate`, and `subscribe`. The primary `reference_kind`, `reference_kinds`, and `reference_kind_counts` keys use that same vocabulary. Use `--raw-kinds` on `callers` / `callees`, or `references --kind <raw-kind>`, when debugging raw extractor output.
 
+`ReferenceRecord.SpanLength` and `symbol_references.span_length` persist the physical matched-token width rather than deriving it from the resolved symbol name; this matters for constructor-chain tokens such as `base`, `super`, and `this`. `DbReader.GetCallees` preserves that span while aggregating counts: it selects the smallest `(line, column_number)` among rows with a stored column, exposes that 1-based pair as `first_line` / nullable `first_column`, and carries the same row's nullable width as `first_length`; `reference_count` remains the independent aggregate. When every contributing legacy row has `column_number IS NULL`, the reader retains the minimum line and a null column. A migrated row can also retain a column with a null span length. CLI/MCP location adapters degrade either case without fabricating precision.
+
 Reference extraction deduplicates only within the same indexed file and language context. When adding extractor paths, include the file id and language hint in shared `seen` keys so same line/column/name edges from polyglot workspaces do not collapse across Java, Rust, C#, SQL, or other language-specific normalization contexts.
 
 | Raw kind | Logical graph kind | Notes |
@@ -4307,6 +4309,8 @@ inspect 以外の各 command は実行前にその cursor family を拒否しな
 
 `symbol_references.reference_kind` には extractor が出力した raw label を保存する。既定の call-graph 表示（`callers`、`callees`、inspect/analyze の caller / callee bundle、および JSON/MCP フィールド）は、公開 canonical 語彙 `call`、`instantiate`、`subscribe` を返す。primary `reference_kind`、`reference_kinds`、`reference_kind_counts` の key はすべて同じ語彙を使う。raw extractor 出力を調べる場合は、`callers` / `callees` の `--raw-kinds`、または `references --kind <raw-kind>` を使う。
 
+`ReferenceRecord.SpanLength` と `symbol_references.span_length` は、解決後の symbol 名から導出せず、物理的に一致した token 幅を永続化する。これは `base`、`super`、`this` のような constructor-chain token で重要になる。`DbReader.GetCallees` は count を集約しながらその span を保持し、列が保存された row のうち最小の `(line, column_number)` を選び、その 1-based 座標を `first_line` / nullable な `first_column`、同じ row の nullable な幅を `first_length` として公開し、`reference_count` は独立した集約値のままにする。寄与する legacy row がすべて `column_number IS NULL` の場合は最小行と null 列を保持する。移行済み row では列があっても span 長が null の場合がある。CLI/MCP の location adapter はどちらの場合も精度を捏造せず劣化させる。
+
 | Raw kind | Logical graph kind | 備考 |
 |---|---|---|
 | `call` | `call` | 直接実行される呼び出しエッジ。 |

diff --git a/USER_GUIDE.md b/USER_GUIDE.md
@@ -1658,6 +1658,8 @@ By default, `callers` and `callees` return only executable call, construction, a
 
 `callers` and `callees` rank results by weighted structural importance by default: `instantiate` edges count as `3.0`, direct `call` edges as `1.0`, and event `subscribe` edges as `0.1`. This keeps factory or composition-root constructor use from being buried under noisy event subscriptions. Use `--rank-by count` to restore raw `reference_count` ordering, or `--rank-by kind` to group `instantiate`, `call`, then `subscribe` before count. JSON rows keep the raw `reference_count` and add `reference_kind_counts` plus `reference_weight_score` so consumers can re-rank without losing the source counts; MCP structured rows expose the same data as `referenceKindCounts` and `referenceWeightScore`.
 
+Grouped `callees` rows preserve the earliest precise call site separately from the aggregate `reference_count`. CLI JSON exposes its 1-based `first_line`, nullable `first_column`, and persisted physical-token `first_length`; MCP uses the corresponding `firstLine`, `firstColumn`, and `firstLength` fields. This keeps resolved constructor-chain calls on their actual `base`, `super`, or `this` token. Compact and quickfix output use the same 1-based column, SARIF carries the same token-sized region, and LSP converts it to a zero-based, non-empty token range. If every contributing legacy reference lacks a column, the column remains `null`, quickfix reports column `0`, SARIF omits `endColumn`, and LSP returns a zero-width range at the start of the retained source line. If a migrated row retains its column but has no persisted span length, SARIF still omits `endColumn` and LSP returns a zero-width range at that column instead of inventing a token width.
+
 ### Outline a single file
 
 ```bash
@@ -4769,6 +4771,8 @@ cdidx callees AddToGitExclude --exclude-tests
 
 `callers` と `callees` は既定で構造的重要度の weighted 順に並びます。`instantiate` は `3.0`、直接 `call` は `1.0`、event `subscribe` は `0.1` として数えるため、factory や composition root の constructor 利用が大量の event subscription に埋もれにくくなります。従来どおり生の `reference_count` で並べたい場合は `--rank-by count`、reference kind を優先して `instantiate`、`call`、`subscribe` の順でまとめたい場合は `--rank-by kind` を使ってください。JSON の各行は生の `reference_count` を維持し、`reference_kind_counts` と `reference_weight_score` も追加で返します。MCP structured row では同じ情報を `referenceKindCounts` と `referenceWeightScore` として返すため、consumer 側で再ランキングできます。
 
+集約された `callees` 行は、集約値の `reference_count` とは別に、最初の精密な call site を保持します。CLI JSON は 1-based の `first_line`、nullable な `first_column`、永続化された物理 token 幅の `first_length` を返し、MCP は対応する `firstLine`、`firstColumn`、`firstLength` を使います。そのため、解決後の constructor-chain call も実際の `base`、`super`、`this` token を指します。compact / quickfix も同じ 1-based 列を使い、SARIF は同じ token 幅の region、LSP は 0-based の非空 token range に変換します。寄与する legacy reference の全てで列が欠けている場合は列を `null` のまま保ち、quickfix は列 `0`、SARIF は `endColumn` を省略し、LSP は保持した source line の先頭に zero-width range を返します。移行済み row に列はあるものの永続化 span 長がない場合も、SARIF は `endColumn` を省略し、LSP はその列に zero-width range を返すため、token 幅を捏造しません。
+
 ### 1ファイルのアウトラインを見る
 
 ```bash

diff --git a/changelog.d/unreleased/4841.fixed.md b/changelog.d/unreleased/4841.fixed.md
@@ -0,0 +1,28 @@
+---
+category: fixed
+issues:
+  - 4841
+affected:
+  - src/CodeIndex/Models/ReferenceRecord.cs
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.ReferenceRecords.cs
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.PrimaryConstructors.cs
+  - src/CodeIndex/Indexer/References/ReferenceExtractor.CoreCallReferences.cs
+  - src/CodeIndex/Indexer/References/Support/JvmMethodReferenceExtractor.cs
+  - src/CodeIndex/Indexer/Hooks/PostExtractionHookMutationMaterializer.cs
+  - src/CodeIndex/Database/DbContext.SchemaInitialization.cs
+  - src/CodeIndex/Database/DbWriter.ReferenceSql.cs
+  - src/CodeIndex/Database/DbWriter.References.cs
+  - src/CodeIndex/Database/DbReader.GraphQueries.cs
+  - src/CodeIndex/Models/QueryResults.cs
+  - src/CodeIndex/Cli/QueryCommandRunner.Graph.cs
+  - src/CodeIndex/Cli/QueryCommandRunner.Locations.cs
+  - src/CodeIndex/Mcp/McpToolHandlers.Graph.cs
+---
+
+## English
+
+- **`callees` now preserves the first physical call-site span in every location output (#4841)** — following the graph-location fidelity work in #4417, reference rows persist the matched token width and aggregated callee rows expose its deterministic 1-based column and length separately from `reference_count`; quickfix, LSP, SARIF, JSON, compact, and MCP output point at the called token—including resolved `base` / `super` / `this` constructor chains—while legacy rows without complete spans degrade explicitly.
+
+## 日本語
+
+- **`callees` がすべての location 出力で最初の物理 call-site span を保持するようになりました (#4841)** — #4417 の graph location fidelity 対応に続き、reference row は一致した token 幅を永続化し、集約された callee 行は `reference_count` とは別に決定的な 1-based 列と長さを公開します。quickfix、LSP、SARIF、JSON、compact、MCP は、解決済みの `base` / `super` / `this` constructor chain を含めて実際の呼び出し token を指し、完全な span を持たない legacy row は明示的に劣化します。
diff --git a/src/CodeIndex/Cli/ProjectionFieldRegistry.cs b/src/CodeIndex/Cli/ProjectionFieldRegistry.cs
@@ -239,9 +239,7 @@ private static ProjectionCommandFieldSchema CreateCallGraphSchema(string command
         var resultFields = isCallerCommand
             ? GetJsonFieldNames<CallerResult>()
             : GetJsonFieldNames<CalleeResult>();
-        var compactFields = isCallerCommand
-            ? new[] { "file", "line", "column" }
-            : ["file", "line"];
+        string[] compactFields = ["file", "line", "column"];
         return Create(
             command,
             compactFields,
@@ -253,9 +251,8 @@ private static ProjectionCommandFieldSchema CreateCallGraphSchema(string command
                     "reference_extraction_limits", "reference_graph_complete",
                     "reference_extraction_cap_hits")
                     .Alias("file", "path")
-                    .Alias("line", "first_line");
-                if (isCallerCommand)
-                    builder.Alias("column", "first_column");
+                    .Alias("line", "first_line")
+                    .Alias("column", "first_column");
             });
     }
 

diff --git a/src/CodeIndex/Cli/QueryCommandRunner.Graph.cs b/src/CodeIndex/Cli/QueryCommandRunner.Graph.cs
@@ -369,7 +369,7 @@ public static int RunCallees(string[] cmdArgs, JsonSerializerOptions jsonOptions
             {
                 if (TryWriteFormattedLocations(
                     options,
-                    results.Select(r => new FormattedLocation(r.Path, r.FirstLine, null, $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}")),
+                    results.Select(r => new FormattedLocation(r.Path, r.FirstLine, r.FirstColumn, $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}")),
                     jsonOptions))
                     return CommandExitCodes.Success;
                 if (options.OutputFormat == OutputFormatLsp)
@@ -379,12 +379,22 @@ public static int RunCallees(string[] cmdArgs, JsonSerializerOptions jsonOptions
                 }
                 if (options.OutputFormat == OutputFormatQf)
                 {
-                    WriteQuickfix(results.Select(r => (r.Path, r.FirstLine, 1, $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}")));
+                    WriteQuickfix(results.Select(r => (r.Path, r.FirstLine, r.FirstColumn ?? 0, $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}")));
                     return CommandExitCodes.Success;
                 }
                 if (options.OutputFormat == OutputFormatSarif)
                 {
-                    WriteSarif(results.Select(r => (r.Path, r.FirstLine, 1, $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}", r.ReferenceKind)), jsonOptions);
+                    WriteSarif(
+                        results.Select(r => new SarifLocation(
+                            r.Path,
+                            r.FirstLine,
+                            r.FirstColumn ?? 1,
+                            r.FirstColumn.HasValue && r.FirstLength.HasValue
+                                ? r.FirstColumn.Value + Math.Max(1, r.FirstLength.Value)
+                                : null,
+                            $"{r.CallerName ?? "<top-level>"} -> {r.CalleeName}",
+                            r.ReferenceKind)),
+                        jsonOptions);
                     return CommandExitCodes.Success;
                 }
                 foreach (var r in results)

diff --git a/src/CodeIndex/Cli/QueryCommandRunner.Locations.cs b/src/CodeIndex/Cli/QueryCommandRunner.Locations.cs
@@ -92,7 +92,18 @@ private static LspLocation ToLspLocation(CallerResult result)
         => BuildLspLocation(result.Path, result.FirstLine, Math.Max(1, result.FirstColumn), result.FirstLine, Math.Max(1, result.FirstColumn) + Math.Max(1, result.CalleeName.Length));
 
     private static LspLocation ToLspLocation(CalleeResult result)
-        => BuildLspLocation(result.Path, result.FirstLine, 1, result.FirstLine, 1);
+    {
+        var column = Math.Max(1, result.FirstColumn ?? 1);
+        var endColumn = result.FirstLength.HasValue
+            ? column + Math.Max(1, result.FirstLength.Value)
+            : column;
+        return BuildLspLocation(
+            result.Path,
+            result.FirstLine,
+            column,
+            result.FirstLine,
+            endColumn);
+    }
 
     private static void WriteLspLocations(IEnumerable<LspLocation> locations, JsonSerializerOptions jsonOptions)
     {

diff --git a/src/CodeIndex/Database/DbContext.ReadMigrations.cs b/src/CodeIndex/Database/DbContext.ReadMigrations.cs
@@ -157,6 +157,7 @@ file_id         INTEGER NOT NULL REFERENCES files(id) ON DELETE CASCADE,
                 reference_kind  TEXT,
                 line            INTEGER,
                 column_number   INTEGER,
+                span_length     INTEGER,
                 context         TEXT,
                 reference_line_id INTEGER REFERENCES reference_lines(id) ON DELETE SET NULL,
                 container_kind  TEXT,
@@ -166,6 +167,8 @@ is_mutual_recursion INTEGER NOT NULL DEFAULT 0
             )"));
         yield return ("EnsureColumn symbol_references.reference_line_id",
             () => EnsureColumn("symbol_references", "reference_line_id", "INTEGER REFERENCES reference_lines(id) ON DELETE SET NULL"));
+        yield return ("EnsureColumn symbol_references.span_length",
+            () => EnsureColumn("symbol_references", "span_length", "INTEGER"));
         yield return ("EnsureColumn symbol_references.is_self_reference",
             () => EnsureColumn("symbol_references", "is_self_reference", "INTEGER NOT NULL DEFAULT 0"));
         yield return ("EnsureColumn symbol_references.is_mutual_recursion",