Fix JSON array symbol hierarchy

DEVELOPER_GUIDE.md

@@ -1523,7 +1523,7 @@ Extractor strategy by language surface:
 | Scala | Uses a separate block-call pass for `name { ... }` / `name { x => ... }` forms so idiomatic calls such as `foreach {}`, `Try {}`, and `synchronized {}` stay visible. |
 | Common Lisp / Racket | Use a lightweight S-expression scanner that masks strings, line comments, and `#| ... |#` block comments before extracting definitions and function ranges. |
 | HTML | Uses a dedicated character-level state machine instead of the regex pattern loop. It walks tag openers, quoted/unquoted attribute values including multi-line values, and masks `<script>` / `<style>` / `<textarea>` / `<title>` bodies plus `<!-- ... -->` comments so attribute-lookalike strings inside those regions do not leak phantom symbols. |
-| JSON / JSON Lines | JSON emits `object`, `array`, `property`, and bounded primitive-array `value` symbols with indexed paths. `.jsonl` and `.ndjson` parse each non-empty physical line independently, prefix symbols with a stable zero-based record path such as `[0].result.path`, and emit repository-local path references from each valid record without flattening malformed neighbors. |
+| JSON / JSON Lines | JSON emits `object`, `array`, `property`, and bounded primitive-array `value` symbols with indexed paths. Array indexes attach directly to the parent path (`command_cases[0]`, not `command_cases.command_cases[0]`), and object/array parent kinds preserve the hierarchy used by outline depth. Root arrays start at paths such as `[0]`. `.jsonl` and `.ndjson` parse each non-empty physical line independently, prefix symbols with a stable zero-based record path such as `[0].result.path`, and emit repository-local path references from each valid record without flattening malformed neighbors. |
 | TOML / repository metadata | TOML tables and keys, EditorConfig sections and keys, Git/Docker ignore rules, Git attribute rules/attributes, and `.rules` blocks/keys are emitted as bounded structural symbols. References are limited to repository-local paths or globs; remote URLs, absolute filesystem paths, and parent traversal are suppressed. |
 | Windows application manifests | Manifest element paths, assembly identities, execution levels, and supported-OS values remain structural symbols. Dependent assembly identities emit `dependency` references, while local `file`, `codeBase`, and probing paths emit `project_reference` edges. |
 | XML / NuGet.config | Generic XML emits bounded element and attribute paths. NuGet.config additionally promotes package sources, source mappings, signature validation mode, trusted signer names, certificate fingerprints, and `allowUntrustedRoot` values to semantic `property` symbols with `nuget.*` subkinds. |
@@ -4875,7 +4875,7 @@ LIMIT 20;
 | Scala | `name { ... }` / `name { x => ... }` 形式を拾う専用の block-call path があり、`foreach {}`、`Try {}`、`synchronized {}` のような慣用的な呼び出しも graph から消えないようにしています。 |
 | Common Lisp / Racket | 文字列、行 comment、`#| ... |#` block comment を mask してから definition と function range を抽出する軽量な S-expression scanner を使います。 |
 | HTML | 汎用の正規表現 loop を使わず、専用の文字単位 state machine で tag opener、引用符付き/なし attribute value（複数行値を含む）、`<script>` / `<style>` / `<textarea>` / `<title>` body、`<!-- ... -->` comment を扱い、attribute 名に似た body 内文字列から phantom symbol が漏れないようにします。 |
-| JSON / JSON Lines | JSON は `object`、`array`、`property` と、上限付きの primitive-array `value` symbol を index 付き path で出力します。`.jsonl` と `.ndjson` は空でない物理行を個別に parse し、`[0].result.path` のような安定した 0 始まり record path を付けます。各有効 record から repository-local path reference を出し、不正な隣接 record の内容は平坦化しません。 |
+| JSON / JSON Lines | JSON は `object`、`array`、`property` と、上限付きの primitive-array `value` symbol を index 付き path で出力します。配列 index は親 path に直接連結し（`command_cases.command_cases[0]` ではなく `command_cases[0]`）、object / array の親 kind を保持して outline の depth に階層を反映します。root array は `[0]` のような path から始まります。`.jsonl` と `.ndjson` は空でない物理行を個別に parse し、`[0].result.path` のような安定した 0 始まり record path を付けます。各有効 record から repository-local path reference を出し、不正な隣接 record の内容は平坦化しません。 |
 | TOML / repository metadata | TOML の table / key、EditorConfig の section / key、Git / Docker ignore rule、Git attribute の rule / attribute、`.rules` の block / key を上限付き structural symbol として出力します。reference は repository-local な path / glob に限定し、remote URL、絶対 filesystem path、親 directory traversal は抑止します。 |
 | Windows application manifest | manifest element path、assembly identity、execution level、supported OS value を structural symbol として維持します。依存 assembly identity は `dependency` reference、local な `file` / `codeBase` / probing path は `project_reference` edge を出力します。 |
 | XML / NuGet.config | 汎用 XML は上限付きの element / attribute path を出力します。NuGet.config ではさらに package source、source mapping、署名検証モード、trusted signer 名、証明書 fingerprint、`allowUntrustedRoot` の値を `nuget.*` subkind 付きの semantic `property` symbol にします。 |

changelog.d/unreleased/4874.fixed.md

@@ -0,0 +1,23 @@
+---
+category: fixed
+issues:
+  - 4874
+affected:
+  - src/CodeIndex/Indexer/Symbols/SymbolExtractor.StructuredData.cs
+  - src/CodeIndex/Indexer/Symbols/SymbolExtractor.Contracts.cs
+  - src/CodeIndex/Database/DbSymbolReader.Outline.cs
+  - tests/CodeIndex.Tests/DatabaseTests.cs
+  - tests/CodeIndex.Tests/IndexCommandRunnerTests.cs
+  - tests/CodeIndex.Tests/SymbolExtractorRepositoryMetadataTests.cs
+  - tests/CodeIndex.Tests/SymbolExtractorTests.cs
+  - tests/CodeIndex.Tests/QueryCommandRunnerInspectTests.cs
+  - DEVELOPER_GUIDE.md
+---
+
+## English
+
+- **JSON arrays now preserve their paths and hierarchy (#4874)** — array indexes attach directly to their parent path, root arrays are extracted, and object/array parent metadata keeps `symbols`, `outline`, and `definition` results at the correct depth without duplicated names such as `command_cases.command_cases[0]`.
+
+## 日本語
+
+- **JSON 配列の path と階層を正しく保持するようになりました (#4874)** — 配列 index を親 path に直接連結し、root array を抽出するとともに、object / array の親 metadata を保持することで、`symbols`、`outline`、`definition` が `command_cases.command_cases[0]` のような重複名を出さず正しい depth を返すようになりました。

src/CodeIndex/Database/DbSymbolReader.Outline.cs

@@ -30,6 +30,9 @@ public partial class DbReader
         }
 
         var startColumnOrderSql = GetSymbolColumnSql("start_column", "CAST(2147483647 AS INTEGER)");
+        var structuredHierarchyOrderSql = lang is "json" or "jsonl"
+            ? "LENGTH(CAST(s.name AS BLOB)) ASC,"
+            : string.Empty;
         var includeReferenceCountSql = includeReferenceCounts && _hasReferencesTable;
         var referenceCountSql = includeReferenceCountSql
             ? "CASE WHEN COALESCE(symbol_defs.definition_sites, 0) = 1 THEN COALESCE(symbol_rank.reference_count, 0) ELSE 0 END"
@@ -96,6 +99,7 @@ FROM symbols s
             WHERE s.file_id = @fileId
             ORDER BY s.line ASC,
                      {startColumnOrderSql} ASC,
+                     {structuredHierarchyOrderSql}
                      s.kind COLLATE BINARY ASC,
                      s.name COLLATE BINARY ASC,
                      s.id ASC";
@@ -104,6 +108,7 @@ FROM symbols s
             SqliteCommandPolicy.AddNullableText(symCmd, "@lang", lang);
 
         var symbols = new List<OutlineSymbol>();
+        var isJsonStructuredData = lang is "json" or "jsonl";
         using (var reader = symCmd.ExecuteTrackedReader())
         {
             while (reader.TrackedRead())
@@ -123,15 +128,18 @@ FROM symbols s
                     Signature = GetNullableString(reader, 7),
                     ContainerKind = GetNullableString(reader, 8),
                     ContainerName = containerName,
-                    Path = BuildOutlineSymbolPath(containerQualifiedName ?? containerName, name),
+                    Path = BuildOutlineSymbolPath(
+                        containerQualifiedName ?? containerName,
+                        name,
+                        isJsonStructuredData),
                     Visibility = GetNullableString(reader, 11),
                     ReturnType = GetNullableString(reader, 12),
                     ReferenceCount = GetNullableInt32(reader, 13),
                 });
             }
         }
 
-        PopulateOutlineDepths(symbols);
+        PopulateOutlineDepths(symbols, isJsonStructuredData);
         ApplyQueryOutputSignatureLimits(symbols);
         PopulateOutlineDisplayNames(symbols, lang);
 
@@ -172,16 +180,34 @@ private static string BuildOutlineDisplayName(OutlineSymbol symbol, string? lang
             : symbol.Name;
     }
 
-    private static string BuildOutlineSymbolPath(string? containerQualifiedName, string name)
+    private static string BuildOutlineSymbolPath(
+        string? containerQualifiedName,
+        string name,
+        bool allowArrayIndexBoundary)
     {
         if (string.IsNullOrWhiteSpace(containerQualifiedName))
             return name;
 
-        return name.StartsWith(containerQualifiedName + ".", StringComparison.Ordinal)
+        return IsQualifiedOutlineChildPath(containerQualifiedName, name, allowArrayIndexBoundary)
             ? name
             : $"{containerQualifiedName}.{name}";
     }
 
+    private static bool IsQualifiedOutlineChildPath(
+        string containerPath,
+        string childPath,
+        bool allowArrayIndexBoundary)
+    {
+        if (!childPath.StartsWith(containerPath, StringComparison.Ordinal)
+            || childPath.Length <= containerPath.Length)
+        {
+            return false;
+        }
+
+        var boundary = childPath[containerPath.Length];
+        return boundary == '.' || (allowArrayIndexBoundary && boundary == '[');
+    }
+
     private static bool IsCallableOutlineSymbol(string kind)
     {
         return kind is "function" or "operator" or "method" or "constructor";
@@ -327,15 +353,27 @@ private static string SimplifyParameterForOutline(string parameter, string? lang
         return string.Join(" ", parts.Take(parts.Length - 1));
     }
 
-    private static void PopulateOutlineDepths(List<OutlineSymbol> symbols)
+    private static void PopulateOutlineDepths(
+        List<OutlineSymbol> symbols,
+        bool isJsonStructuredData)
     {
         var depthCache = new Dictionary<int, int>();
         var activeStack = new HashSet<int>();
         for (var i = 0; i < symbols.Count; i++)
-            symbols[i].Depth = GetOutlineDepth(symbols, i, depthCache, activeStack);
+            symbols[i].Depth = GetOutlineDepth(
+                symbols,
+                i,
+                depthCache,
+                activeStack,
+                isJsonStructuredData);
     }
 
-    private static int GetOutlineDepth(List<OutlineSymbol> symbols, int index, Dictionary<int, int> depthCache, HashSet<int> activeStack)
+    private static int GetOutlineDepth(
+        List<OutlineSymbol> symbols,
+        int index,
+        Dictionary<int, int> depthCache,
+        HashSet<int> activeStack,
+        bool isJsonStructuredData)
     {
         if (depthCache.TryGetValue(index, out var cachedDepth))
             return cachedDepth;
@@ -347,20 +385,35 @@ private static int GetOutlineDepth(List<OutlineSymbol> symbols, int index, Dicti
         var depth = 0;
         if (!string.IsNullOrEmpty(symbol.ContainerName))
         {
-            var parentIndex = FindOutlineContainerIndex(symbols, index, symbol.ContainerName, symbol.ContainerKind);
+            var parentIndex = FindOutlineContainerIndex(
+                symbols,
+                index,
+                symbol.ContainerName,
+                symbol.ContainerKind,
+                isJsonStructuredData);
             if (parentIndex >= 0)
-                depth = GetOutlineDepth(symbols, parentIndex, depthCache, activeStack) + 1;
+                depth = GetOutlineDepth(
+                    symbols,
+                    parentIndex,
+                    depthCache,
+                    activeStack,
+                    isJsonStructuredData) + 1;
         }
 
         activeStack.Remove(index);
         depthCache[index] = depth;
         return depth;
     }
 
-    private static int FindOutlineContainerIndex(List<OutlineSymbol> symbols, int childIndex, string containerName, string? containerKind)
+    private static int FindOutlineContainerIndex(
+        List<OutlineSymbol> symbols,
+        int childIndex,
+        string containerName,
+        string? containerKind,
+        bool isJsonStructuredData)
     {
         var child = symbols[childIndex];
-        var expectedContainerPath = GetOutlineContainerPath(child);
+        var expectedContainerPath = GetOutlineContainerPath(child, isJsonStructuredData);
         for (var i = childIndex - 1; i >= 0; i--)
         {
             var candidate = symbols[i];
@@ -375,32 +428,46 @@ private static int FindOutlineContainerIndex(List<OutlineSymbol> symbols, int ch
             }
             if (candidate.Line > child.Line)
                 continue;
-            if (IsOutlineContainerMatch(candidate, child.Line))
+            if (IsOutlineContainerMatch(candidate, child.Line, isJsonStructuredData))
                 return i;
         }
 
         return -1;
     }
 
-    private static string? GetOutlineContainerPath(OutlineSymbol symbol)
+    private static string? GetOutlineContainerPath(
+        OutlineSymbol symbol,
+        bool allowArrayIndexBoundary)
     {
         if (string.IsNullOrWhiteSpace(symbol.Path) || string.IsNullOrWhiteSpace(symbol.Name))
             return null;
 
         var suffix = "." + symbol.Name;
-        return symbol.Path.EndsWith(suffix, StringComparison.Ordinal)
-            ? symbol.Path[..^suffix.Length]
-            : null;
+        if (symbol.Path.EndsWith(suffix, StringComparison.Ordinal))
+            return symbol.Path[..^suffix.Length];
+
+        return !string.IsNullOrWhiteSpace(symbol.ContainerName)
+            && string.Equals(symbol.Path, symbol.Name, StringComparison.Ordinal)
+            && IsQualifiedOutlineChildPath(
+                symbol.ContainerName,
+                symbol.Name,
+                allowArrayIndexBoundary)
+                ? symbol.ContainerName
+                : null;
     }
 
-    private static bool IsOutlineContainerMatch(OutlineSymbol candidate, int childLine)
+    private static bool IsOutlineContainerMatch(
+        OutlineSymbol candidate,
+        int childLine,
+        bool isJsonStructuredData)
     {
         if (candidate.StartLine <= childLine && candidate.EndLine >= childLine)
             return true;
 
-        // file-scoped namespaces have no body range, so they do not enclose children by lines
-        // even though they are the correct logical container.
-        return candidate.Kind == "namespace"
+        // File-scoped namespaces and structured-data containers have no body range, so they
+        // do not enclose children by lines even though their exact paths identify the parent.
+        return (candidate.Kind == "namespace"
+                || (isJsonStructuredData && candidate.Kind is "object" or "array" or "record"))
             && candidate.BodyStartLine == null
             && candidate.BodyEndLine == null
             && candidate.Line <= childLine;

src/CodeIndex/Indexer/Symbols/SymbolExtractor.Contracts.cs

@@ -20,6 +20,8 @@ public static partial class SymbolExtractor
     public const int ApplicationManifestContractVersion = 3;
     public const int MarkdownContractVersion = 3;
     public const int DependencyLockContractVersion = 3;
+    public const int JsonContractVersion = 3;
+    public const int JsonLinesContractVersion = 3;
 
     private static readonly string[] ExplicitReferenceGraphContractLanguages =
         ["crystal", "groovy", "tcl", "prolog", "ambiguous_pl"];
@@ -68,11 +70,13 @@ public static int GetContractVersion(string? lang)
             "crystal" or "groovy" or "tcl" => DynamicReferenceGraphContractVersion,
             "prolog" or "ambiguous_pl" => PrologReferenceGraphContractVersion,
             "ada" or "ambiguous_m" or "cython" or "d" or "julia" or "matlab" or "nim" or "objc" => ScientificNativeGraphContractVersion,
-            "config" or "dockerignore" or "editorconfig" or "gitattributes" or "gitignore" or "jsonl" or "toml" => RepositoryMetadataContractVersion,
+            "config" or "dockerignore" or "editorconfig" or "gitattributes" or "gitignore" or "toml" => RepositoryMetadataContractVersion,
             "app_manifest" => ApplicationManifestContractVersion,
             "markdown" => MarkdownContractVersion,
             "dependency_lock" => DependencyLockContractVersion,
-            "cmake" or "dependency_manifest" or "graphql" or "html" or "json" or "justfile" or "msbuild" or "solution" or "yaml" => ExpandedLanguageContractVersion,
+            "json" => JsonContractVersion,
+            "jsonl" => JsonLinesContractVersion,
+            "cmake" or "dependency_manifest" or "graphql" or "html" or "justfile" or "msbuild" or "solution" or "yaml" => ExpandedLanguageContractVersion,
             _ => DefaultContractVersion,
         };
     }