0.1.1

.gitignore

@@ -3,3 +3,4 @@ out/
 dist/
 *.vsix
 OpenSAGE/
+debug.log

README.md

@@ -4,19 +4,28 @@
 
 ## 功能
 
-- **语法高亮**：在普通 XML 高亮之上叠加领域标记（`$DEFINE` 常量、`inheritFrom`、`xai:joinAction`、结构标签）。
+- **语法高亮**：在普通 XML 高亮之上叠加领域标记（`$DEFINE` 常量、`inheritFrom`、`xai:joinAction`、结构标签）；XML 语法异常（如未闭合引号）期间由语义 token 兜底，标签/属性/值着色不中断。
 - **自动补全**：
-  - 元素名：按当前父元素的 XSD 模型补全子元素；顶层资产（`AssetDeclaration` 内）补全 `GameObject`、`WeaponTemplate` 等 99+ 类型。
+  - 元素名：按当前父元素的 XSD 模型补全子元素；顶层资产（`AssetDeclaration` 内）补全 `GameObject`、`WeaponTemplate` 等 295 种类型。
   - 属性名：必填属性优先，附带类型/文档/默认值；自动提示 `xai:joinAction` 与 `xmlns:xai`。
   - 属性值：
     - 引用型属性（如 `CommandSet`、`Weapon`）按 `xas:refType` 补全对应类型的资产 ID（**同名 ID 只补全匹配类型**）；
     - `inheritFrom` 补全可继承的资产 ID；
-    - 枚举（如 `Include type`）、布尔值、`$DEFINE` 常量；
+    - 枚举与位标志列表（如 `Include type`、`LocomotorTemplate@Surfaces`、`KindOf`；列表值支持空格后继续补全下一项）；
+    - 布尔值、`$DEFINE` 常量；
     - `<Include source>` 补全可解析的 `DATA:` / `ART:` / `AUDIO:` 与项目相对路径。
-- **悬停提示**：元素/属性显示 XSD 文档、类型、必填/默认值；引用值显示定义位置；`$DEFINE` 显示值与定义位置。
+- **悬停提示**：元素/属性显示 XSD 文档、类型、必填/默认值；引用值显示定义位置；`$DEFINE` 显示值与定义位置；`Include source` / `xi:include href` 显示解析后的目标文件；`xi:include` 元素与属性给出 XInclude 说明。
-- **引用导航**：从引用值（`CommandSet="..."`、`Weapon="..."`、`inheritFrom`）跳转到定义（严格按引用类型过滤）；`Ctrl+点击` Include 打开目标文件；Find All References 搜索整个工作区；文档大纲列出顶层资产与 `$DEFINE`。
+- **引用导航**：从引用值（`CommandSet="..."`、`Weapon="..."`、`inheritFrom`）跳转到定义（严格按引用类型过滤，候选由 `ra3modxml.definitionMode` 控制：`all` 列出 mod + 原版、`project-only` 优先项目内定义）；`Ctrl+点击` Include / `xi:include href` 打开目标文件；Find All References 搜索整个工作区；文档大纲列出顶层资产与 `$DEFINE`。
-- **错误检查**：XML 格式错误、未知元素/属性、顶层资产缺 `id`、重复 ID、未解析引用（含类型不匹配）、Include 找不到、`$DEFINE` 未定义。
+- **错误检查**：XML 格式错误、未知元素/属性（`xi:` 等外来命名空间不误报）、顶层资产缺 `id`、重复 ID、未解析引用（含类型不匹配）、Include / 嵌套 `xi:include` 目标找不到、`$DEFINE` 未定义。
 - **manifest 支持**：`<Include type="reference">` 指向的 `static/global/audio.manifest`（SDK `builtmods`）会被解析，manifest 中的原版资产 ID 可用于补全/悬停/导航/诊断。
+- **美术资产（`.w3x`）**：`W3X.xml` / `ART:` include 链中的 `.w3x` 模型文件会被
+  索引（`W3DContainer` / `W3DMesh` / `W3DHierarchy` 等顶层资产），因此
+  `Model@Name`、`Hierarchy`、`Mesh` 等引用可以解析、悬停与跳转。超大模型
+  （几十 MB 的顶点/三角形数据）采用浅扫描——只提取顶层资产记录、不建 DOM 树，
+  结果在 workspace 级缓存并跨重建复用，保存文件触发的重建不会重读未变化的模型文件。
+- **大项目性能**：索引记录（资产 / Define / Include / 行号）与 include 解析结果
+  跨重建缓存，保存触发的重建零 stat、零重读（Corona 实测约 2 秒）；DOM 树只按需
+  保留并设元素预算，避免内存膨胀。
 
 ## 使用
 
@@ -32,6 +41,7 @@
 | `ra3modxml.indexSageXml` | `true` | 是否索引 SDK 的 `SageXml` 原版源码 |
 | `ra3modxml.reportUnresolvedReferences` | `warning` | 未解析引用诊断级别（`warning`/`information`/`none`） |
 | `ra3modxml.diagnoseUnknownElements` | `true` | 是否报告未知元素/属性（自定义 XSD 项目可关闭） |
+| `ra3modxml.definitionMode` | `all` | 跳转候选：`all` 列出 mod 定义与原版定义（mod 优先）；`project-only` 仅在项目内已有定义时直接跳转 mod 定义 |
 | `ra3modxml.additionalDataSearchPaths` | `[]` | 追加的 `DATA:` 搜索目录 |
 
 ### 命令
@@ -57,15 +67,25 @@ npm run package          # 生成可安装的 .vsix
 src/
   extension.ts            激活入口与 provider 注册
   workspace.ts            项目检测、索引生命周期、状态栏
-  language/xmlParser.ts   带源码偏移的轻量 XML 解析器
+  settings.ts             配置读取（sdkPath、definitionMode 等）
-  language/context.ts     补全上下文分析
+  language/
-  model/schemaModel.ts    XSD 模型运行时（由 tools 生成 JSON 驱动）
+    xmlParser.ts          带源码偏移的轻量 XML 解析器（容错、行尾恢复）
+    context.ts            补全上下文分析
+    typeContext.ts        上下文感知元素类型解析
+    semanticTokens.ts     语义 token 兜底高亮（纯 TS）
+  model/
+    schemaModel.ts        XSD 模型运行时（schema-model.json / asset-types.json 由 tools 生成）
   indexer/
     includeResolver.ts    Include 路径解析（纯 TS，移植 check_duplicate_ids.py）
     manifestParser.ts     .manifest 二进制解析（移植 OpenSAGE ManifestFile.cs）
+    fileScanner.ts        目录扫描与 Include source 候选
     refs.ts               引用目标解析（按引用类型过滤）
-    indexer.ts            工作区索引器（后台、缓存、增量重建）
+    shallowScan.ts        .w3x 等大体积美术资产顶层浅扫描（纯 TS，不建 DOM）
-  features/               completion / hover / navigation / diagnostics
+    records.ts            每文件紧凑索引记录（资产/Define/Include/xi + 行号）
+    caches.ts             跨重建持久缓存（DocumentCache / IndexRecordsCache /
+                          IncludeResolveCache）
+    indexer.ts            工作区索引器（后台、缓存、记录驱动重建）
+  features/               completion / hover / navigation / diagnostics / semanticTokens
 syntaxes/                 TextMate 注入语法
 tools/                    XSD → 模型、AssetType 枚举提取
 ```
@@ -76,4 +96,5 @@ tools/                    XSD → 模型、AssetType 枚举提取
 
 - 领域说明与需求：`docs/requirements.md`
 - 调研与设计决策：`docs/plan.md`
+- 问题分析与修复记录：`docs/analysis-issues.md`
 - Manifest 格式参考：OpenSAGE `src/OpenSage.Game/Data/StreamFS/ManifestFile.cs`（本仓库 `OpenSAGE/` 子目录，commit `d45d361`）

debug.log

@@ -1 +0,0 @@
-[0801/032923.150:ERROR:third_party\crashpad\crashpad\util\win\registration_protocol_win.cc:108] CreateFile: 拒绝访问。 (0x5)

docs/analysis-issues.md

@@ -240,3 +240,525 @@ CommandSet 引用仍能解析到 LogicCommandSet 定义（无回归）
 - AttachTest 实机：三个原始场景全部按预期（Locomotor 命中 LocomotorTemplate、_SKN 命中 W3DContainer、cannon 双候选均可精确定位）。
 
 `.vsix` 已重新打包（13:33）。D 盘恢复后照旧可补跑 GenEvoTest / Corona 回归。
+
+---
+
+## 九、问题分析（第四轮，2026-08-01）：模块 `id` 被误报为未解析引用
+
+### 问题：`id="ModuleTag_Draw"` 误报 `Unresolved reference`
+
+**现象**：AttachTest `Allied Vehicle\Guardian Tank\GameObject.xml` 第 45 行
+
+```xml
+<TruckDraw
+    id="ModuleTag_Draw"
+    ...>
+```
+
+报 `Unresolved reference "ModuleTag_Draw" (not found in the current index)`，
+hover 同时显示 `No matching definition of the expected declared type...`。
+但该 id 是 TruckDraw（GameObject 模块）自身的标识，只在所属 `<GameObject />` 内部有效，
+此处就是定义处，全局资产索引中不存在（也不应存在）它的定义。
+
+**根因（两层叠加）**：
+
+1. **模型生成器丢失属性级 `xas:refType`**：`ModuleData@id` 在 XSD 中声明为
+
+   ```xml
+   <xs:complexType name="ModuleData" xas:isPolymorphic="true">
+     <xs:attribute name="id" type="Poid" xas:refType="ModuleData" />
+   </xs:complexType>
+   ```
+
+   refType 写在 `<xs:attribute>` 节点上，而 `tools/xsd-to-model.mjs` 只从 simple type
+   描述符读取 refType，属性级声明被丢弃。`Poid` 本身带 `xas:isWeakRef="true"`（“管线对象
+   ID”），于是该 id 在模型里变成 `{ type: "Poid", refType: null, isRef: true }`——一个
+   “无类型引用”，导致**所有继承自 ModuleData 的模块类型（以及 ObjectFilter、MapObject、
+   GameScript、AIStateTactic 等共 430 个类型）的 `id` 都被当作全局引用检查**。
+
+2. **`id` 的语义是“定义点”而非“引用”**：即便 refType 正确，嵌套元素（模块、nugget、
+   地图对象）的 `id` 也只是其局部标识，检查全局 unresolved 必然误报。反过来，XSD 里确有一类
+   真正引用其他资产类型的 `id`（如 `RoadObject@id` 为 `AssetReference` + `xas:refType="Road"`），
+   这类检查必须保留。
+
+**验证**：
+
+- 用未修改的生成器对当前 SDK XSD 重新生成模型，与仓库内模型逐字段一致（0 差异），
+  证明修复后重新生成的 diff 只落在属性级 refType 上，无无关噪音。
+- 修复后 `W3DTruckDrawModuleData@id` → `refType: ModuleData`；
+  `isReferenceAttributeOfType(...) === false`；空索引下 `TruckDraw@id` 不再产生诊断；
+  全文件扫描 0 个 `id` 误报，61 处真实引用属性（`inheritFrom`、`CommandSet`、`Side`、
+  `Locomotor`、`TrackMarks` 等）行为不变。
+
+**修复**：
+
+1. **生成器**（`tools/xsd-to-model.mjs`）：`collectAttributes` 改为
+   `attr["@_refType"] ?? desc?.refType`（属性级优先、simple type 兜底），重新生成
+   `schema-model.json`——共恢复 444 处 refType（含继承传播），`isRef` 与其他字段零变化。
+   附带收益：`Locomotor → LocomotorTemplate`、`Armor → ArmorTemplate`、
+   `ThingTemplate → GameObject` 等此前被当作“无类型引用”的属性恢复真实类型
+   （第三轮中 Locomotor“无 refType”的结论实为该生成器 bug 的误判）。
+2. **局部引用规则**（`src/indexer/refs.ts` 新增 `isLocalReferenceAttribute`）：
+   - `id`：无 refType，或 refType 与元素自身类型兼容（`isAssignableTo`，如
+     `W3DTruckDrawModuleData → ModuleData`）→ 定义点，不做全局引用检查；
+     refType 指向不同类型（`RoadObject@id → Road`）→ 保留真实引用检查；
+   - 非 `id` 且类型为 `Poid` 的属性（`ModuleId`、`AutoResolveBody`、`SoundRef`、
+     `AttachModuleId`…）→ 管线局部引用，全局索引无法判定，不检查。
+   `isReferenceAttributeOfType` 与 `resolveReferenceTargetsForType` 同步使用该守卫，
+   诊断 / hover / 跳转 / 补全行为一致。
+3. **补全**（`src/features/completion.ts`）：`id` 与 Poid 属性不再按 refType 提供
+   全局资产补全（模块 id 是局部的，全局资产列表是错误建议）。
+
+**测试（31 → 37，全部通过）**：
+
+- `refs.test.mjs`：TruckDraw 实景结构（GameObject → Draws → TruckDraw）下 `id` 不再是
+  引用且不解析；`RoadObject@id` 仍是引用并能解析到 Road；`AttachModuleId` 等 Poid 属性
+  不误报；Locomotor 改为严格类型引用（同名 GameObject 不再匹配）；
+- `schemaModel.test.mjs`：`ModuleData@id` / `MapObject@id` / `ThingTemplate` /
+  `RoadObject@id` / `AttachModuleId` 的属性级 refType 断言；
+- 回归：原有 31 个用例全部保持通过。
+
+> **后续可做**：GameObject 内模块 id 的“局部作用域”解析——`AttachModuleId`、
+> `ModuleId` 等模块引用指向同一 GameObject 内的兄弟模块，但部分引用（如武器上的
+> `AttachModuleId`）目标 GameObject 跨文件无法静态确定，本轮先统一不检查；待局部
+> 作用域建模落地后再启用这些引用的解析与诊断。
+
+---
+
+## 十、问题分析（第五轮，2026-08-01）：`xi:include` 的 `href`/`xpointer` 误报未知属性
+
+### 问题
+
+AttachTest `Allied Vehicle\Guardian Tank\GameObject.xml` 第 249 行附近：
+
+```xml
+<xi:include
+    href="DATA:Includes/HeadlightDraw2.xml"
+    xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:HeadlightDraw2/child::*)"/>
+```
+
+报两条 `Unknown attribute "href" / "xpointer" for <include>`（`unknown-attribute`），
+hover 同时显示 `Unknown attribute for this element.`。
+
+这个元素属于 **W3C XInclude 命名空间**（`xmlns:xi="http://www.w3.org/2001/XInclude"`），
+并不是 EA `uri:ea.com:eala:asset` XSD 的一部分。同一行在第二轮“问题 C”处理过
+（嵌套 `xi:include` 的索引与导航），但那轮没有覆盖 unknown-attribute 诊断，属于遗留缺口。
+
+### 根因
+
+诊断的属性校验没有像元素校验那样排除外来命名空间：
+
+- 元素校验已有 `!el.name.startsWith("xi:")` 守卫（所以 `<include>` 本身不报 unknown element）；
+- 属性校验只跳过 `xmlns*` / `xai:` / `xi:` 前缀的属性名，而 `href`、`xpointer` 是不带
+  前缀的普通属性名；
+- `<xi:include>` 解析类型为 null（XSD 模型不含该元素），knownAttrs 为空 → 任何属性
+  都被判为 unknown。
+
+### 修复
+
+1. `schemaModel` 新增两个纯函数：
+   - `isXsdElementName`：`xi:` 前缀元素不属于 EA XSD 模型；
+   - `isXsdAttributeName`：EA XSD 属性不带命名空间前缀，带前缀（`xai:`、`xi:`、
+     `xlink:`、`xml:`、`xsi:`、`xmlns:*`）的都是命名空间机制，不做 schema 校验。
+2. `diagnostics`：`xi:` 前缀元素整体跳过 schema 校验（元素与属性都不再误报）；
+   前缀属性名统一跳过。
+3. `hover`：`xi:include` 元素/属性给出 XInclude 说明；`href` 值悬停像
+   `<Include source>` 一样解析目标文件（Ctrl+点击跳转此前已可用）。
+
+### 验证
+
+- 真实文件全量扫描：0 未知元素、0 未知属性（修复前 `href`/`xpointer` 两条必现）；
+- 新增测试：`isXsdElementName` / `isXsdAttributeName` 断言；`xi:include` 解析类型为
+  null 且不参与校验；全量 39/39 通过。
+
+### 后续（架构方向，待确认）
+
+用户提出“先展开 `xi:include`（类比 C++ 宏展开），再处理 mod XML 解析”。该方向与第二轮
+遗留的“虚拟合并”开放项一致，设计要点：
+
+- 构建**逻辑树**而非文本拼接：把目标文件选中内容（`xpointer` 子集）作为子节点拼入父
+  元素，节点保留源文件与原始偏移，避免文本级拼接导致的偏移断裂；
+- 展开范围：`xi:include` 与 EA `<Include type="all">`（内容合并）；`instance` /
+  `reference` 是可见性 / 编译产物语义，不拼树；`inheritFrom` + `joinAction` 是属性级
+  继承合并，不是宏展开；
+- 收益：跨 include 的上下文类型解析、包含内容的结构校验、以及后续“GameObject 内模块
+  id 局部作用域”（HeadlightDraw2 的模块也是该 GameObject 的模块）；
+- 风险：include 环 / 深度限制、大文件性能、`xpointer` 仅支持现有子集形式
+  （`/n:Name/child::*`）。
+
+---
+
+## 十、问题分析（第五轮，2026-08-01）：`xi:include` 的 `href` / `xpointer` 被误报为未知属性
+
+### 问题
+
+同一 GameObject.xml 第 246–249 行：
+
+```xml
+<!-- include Headlight draw module. -->
+<xi:include
+    href="DATA:Includes/HeadlightDraw2.xml"
+    xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:HeadlightDraw2/child::*)"/>
+```
+
+报两条 `Unknown attribute "href" / "xpointer" for <include>`（unknown-attribute），
+hover 显示 `Unknown attribute for this element.`。
+
+**与第二轮的关系**：第二轮“问题 C”处理的正是同一行的嵌套 `xi:include`——但那一轮修的是
+**索引器**（嵌套 include 不再被静默忽略、缺失目标产生 include-not-found、目标内容进索引），
+本轮这处 **unknown-attribute 诊断**是当时未覆盖的遗留问题。
+
+### 根因
+
+`xi:include` 属于 W3C XInclude 命名空间（`http://www.w3.org/2001/XInclude`），
+**不是 RA3 XSD（`uri:ea.com:eala:asset`）定义的元素**：
+
+- 未知元素检查已通过 `el.name.startsWith("xi:")` 跳过，所以没有 unknown-element 误报；
+- 但属性检查没有同类守卫：`xi:include` 解析类型为 null → `knownAttrs` 为空 →
+  `href`、`xpointer` 两个非 `xi:` 前缀的属性名全部落入 unknown-attribute 分支。
+
+复现证据（真实文件）：
+
+```
+xi:include found: true | parent: Draws
+resolved element type: null
+known attribute names: (none)
+attr href: known=false -> would flag unknown-attribute: true
+```
+
+### 修复
+
+1. **模型层新增命名空间守卫**（`schemaModel.ts`）：
+   - `isXsdElementName(name)`：`xi:` 前缀（XInclude）等外来命名空间元素不属于 XSD 模型；
+   - `isXsdAttributeName(name)`：EA XSD 属性一律无前缀，带前缀的属性
+     （`xai:`、`xi:`、`xlink:`、`xml:`、`xsi:`、`xmlns:*`）都是命名空间机制，不做未知属性校验。
+2. **诊断**（`diagnostics.ts`）：外来命名空间元素的未知元素/未知属性检查整体跳过
+   （`href`、`xpointer` 不再误报）；属性名带前缀的一律跳过校验（比原先只跳过
+   `xmlns`/`xai:`/`xi:` 更完整）。
+3. **hover**（`hover.ts`）：`xi:include` 的元素/属性悬停显示 XInclude 说明；
+   `xi:include@href` 与 `Include@source` 一样显示解析后的目标文件（与第二轮已可用的
+   Ctrl+点击跳转对齐）。
+
+**测试（37 → 39，全部通过）**：
+
+- `schemaModel.test.mjs`：`isXsdElementName` / `isXsdAttributeName` 判定
+  （`xi:include` 非 XSD 元素；`href`/`xpointer` 是合法属性名形态；
+  `xai:joinAction`、`xlink:href`、`xmlns:xi` 等带前缀属性不校验）；
+- `refs.test.mjs`：GameObject → Draws → `xi:include` 实景结构解析类型为 null、
+  元素被判定为外来命名空间，`href`/`xpointer` 不会进入未知属性分支；
+- 实机复验：整份 GameObject.xml 0 个 unknown-attribute 残留。
+
+> **架构讨论（用户提议）**：把 XML 先“宏展开”成不含 `xi:include` 的版本再解析。
+> 这与 BAB 编译时的实际行为一致（`defaultscript.cs` 把整个 Mod 合并成一份大 XML），
+> 也是实现“GameObject 内模块 id 局部作用域解析”（第四轮遗留）的正确地基——展开后一个
+> GameObject 连同 include 进来的兄弟模块都在同一棵树里，`AttachModuleId` 等模块引用才能
+> 静态判定。设计备忘（现状 / 逻辑树方案 / 展开范围 / 落地点 / 检查清单）已整理在
+> `docs/plan.md` 第六节，等待确认后作为下一阶段实现。
+
+---
+
+## 十一、问题分析（第六轮，2026-08-01）：`Surfaces="` 未闭合引号导致枚举补全失效
+
+### 现象
+
+在 `<Locomotor ...>` 的起始标签里输入 `Surfaces="`（引号尚未闭合）时：
+
+1. 光标处不出 `LocomotorSurfaceBitFlags` 的枚举补全（GROUND、WATER 等）；
+2. 整个文件高亮退化（XML 变成“不合法”的观感），直到补上第二个引号才恢复。
+
+### 根因（两个独立缺陷叠加）
+
+**A. 上下文分析不认未闭合的引号（`src/language/context.ts`）**
+
+复现证据（编译产物直接执行）：
+
+```
+闭合引号:  <Locomotor id="x" Surfaces="GROUND">…
+ctx.kind = attribute-value, attr = Surfaces, valuePrefix = "GROUND"
+
+未闭合引号: <Locomotor id="x" Surfaces="GROUND>…
+Surfaces: quoteStart=27, quoteEnd=-1   ← 解析器吞掉整个文件
+ctx.kind = attribute-name              ← 补全走错分支
+```
+
+`analyzeStartTag` 判断属性值上下文的条件是 `offset >= quoteStart && offset <= quoteEnd`，
+未闭合时 `quoteEnd = -1` 永远不成立，于是回退成 attribute-name。
+
+**B. 模型生成器不支持 `xs:list`（`tools/xsd-to-model.mjs`）**
+
+XSD 中 `LocomotorSurfaceBitFlags` 是：
+
+```xml
+<xs:simpleType name="LocomotorSurfaceBitFlags">
+  <xs:list itemType="Surface"></xs:list>
+</xs:simpleType>
+```
+
+而 `Surface` 才是真正带 11 个枚举值（GROUND、WATER、CLIFF、AIR…）的类型。生成器
+只读取 `restriction.enumeration`，list 层把枚举全部丢掉——所以**即使引号闭合，模型里
+该属性也没有任何候选值**。影响面：SDK XSD 共 79 个 `xs:list` 简单类型、317 处属性
+声明使用它们（`KindOfBitFlags`、`ObjectStatusBitFlags`、`WeaponFlagsBitFlags`、
+`ModelConditionBitFlags`、`BuildPlacementTypeBitFlags` 等），展开到继承后的模型条目
+共 890 个属性受影响。
+
+### 关于高亮丢失
+
+这是 TextMate XML 语法对“未闭合字符串”的正常行为：后续内容被当作字符串吞掉，直到
+遇到下一个引号或 EOF。与插件注入语法无关（注入部分只有 `$DEFINE`、`inheritFrom` 等
+少量规则），**即使改成 LSP 也不会自动消失**。真正的解法是语义 token
+（`DocumentSemanticTokensProvider`），本次未实施，列为可选后续。
+
+### 修复（第 1–4 项）
+
+1. **解析器行尾恢复**（`src/language/xmlParser.ts`）：起始标签扫描到 EOF 且引号仍未
+   闭合时，把标签在第一个换行处截断并继续解析主循环。未闭合引号只影响当前行，后面
+   的元素照常进入解析树，补全 / hover / 诊断不中断；解析错误仍照常上报
+   （`Unterminated start tag`）。
+2. **未闭合引号上下文**（`src/language/context.ts`）：`quoteEnd < 0` 时，
+   `offset >= quoteStart` 即视为 attribute-value 上下文，`valuePrefix` 照常取引号后
+   到光标处文本。
+3. **模型支持 `xs:list`**（`tools/xsd-to-model.mjs`）：`resolveTypeDescriptor` 解析
+   `xs:list` 的 `itemType`（属性形式或内联 simpleType），继承其枚举值 / refType /
+   isRef / allowsDefine，新增 `isList` 标记；重新生成 `schema-model.json`
+   （`LocomotorSurfaceBitFlags` 恢复 11 个枚举值，`ModelConditionBitFlags` 457 个
+   值与 XSD 一致）。`AttributeInfo` / `SimpleTypeInfo` 接口同步新增 `isList`。
+4. **多值补全按“最后一段”过滤**（`src/features/completion.ts` +
+   `context.splitListValuePrefix`）：list 属性只取当前空格段做前缀过滤，替换范围只
+   覆盖该段——`Surfaces="GROUND ` 之后输入 `W` 也能提示 WATER / WALL_RAILING，而不是
+   用整段前缀匹配失败。
+
+### 验证
+
+- 未闭合引号复现场景：仅报 1 条 `Unterminated start tag`，`<Other/>` 等后续元素仍被
+  解析；`ctx.kind = attribute-value`、`attr = Surfaces`、`valuePrefix = "GROUND"`。
+- 模拟补全过滤：前缀 `G` → `GROUND`；前缀 `W` → `WATER, WALL_RAILING`；
+  `GROUND ` 后输入 `W` → `WATER, WALL_RAILING`。
+- `GameObject@KindOf`：`isList=true`、284 个枚举值。
+- 单元测试 **39 → 50 全部通过**（新增 `test/context.test.mjs` 与带 vscode stub 的
+  `test/completion.test.mjs` 集成用例；xmlParser 新增未闭合引号恢复 / EOF 用例；
+  schemaModel 新增 list 枚举与 `isList` 用例）。
+- `tsc` / `esbuild` 构建通过。
+
+> 补充：真实文件中该场景的元素名是 `<LocomotorTemplate ...>`（AttachTest
+> `Locomotor.xml` 实测，`Surfaces="GROUND CRUSHABLE_OBSTACLE"` 正是多值 list）；
+> SDK XSD 中没有名为 `Locomotor` 的元素，补全集成测试按真实写法夹具。
+
+### 后续可做（未列入本次）
+
+- `AssetIdList` 等“任意资产 ID 列表”的引用语义建模（list 补全框架已就绪，但这类
+  属性在 XSD 里没有 refType，需要另行定义过滤规则）。
+
+> 语义 token 兜底高亮已在第七轮实现（见下节）。
+
+---
+
+## 十二、问题分析（第七轮，2026-08-01）：语义 token 兜底高亮
+
+### 目标
+
+未闭合引号期间 TextMate 把后续内容当字符串吞掉、整个文件高亮退化，这是 XML 语法
+固有的行为（任何 XML 编辑器皆然），也无法靠注入 grammar 修复。本轮的解法是语义
+token：文档出现解析错误时，由插件用自己的容错解析树继续给标签 / 属性 / 值着色。
+
+### 设计
+
+- **纯 TS 核心**（`src/language/semanticTokens.ts`）：`buildSemanticTokenRanges(doc, text)`
+  把解析树转换成按位置排序的 `{ line, startChar, length, tokenType }`；token 类型只用
+  标准 `type` / `property` / `string`，所有主题自带配色，无需额外贡献样式。
+  - 元素名：起始标签与闭合标签各一个 `type` token；
+  - 属性名：`property` token；
+  - 属性值：`string` token，闭合时含两端引号，未闭合时从开引号到行尾恢复点
+    （如 `"GROUND>`）。
+- **provider**（`src/features/semanticTokens.ts`）：`parseXml` 后若
+  `doc.errors.length === 0` 直接返回空——合法文件观感与纯 TextMate 完全一致；
+  有解析错误时才用 `SemanticTokensBuilder` 编码输出。
+- **注册**：`extension.ts` 对 `xml` 语言注册 `DocumentSemanticTokensProvider`，
+  legend 与 provider 共用同一实例。
+
+### 验证
+
+- malformed（`Surfaces="GROUND>` 未闭合）：`AssetDeclaration` / `LocomotorTemplate` /
+  `<Other/>` 标签名、`id` / `Surfaces` 属性名、`"x"` 与 `"GROUND>` 值均有 token，
+  且按位置升序排列；
+- 合法文档：返回空 token 数组；
+- 单元测试 **50 → 53 全部通过**（新增 `test/semanticTokens.test.mjs`，纯函数 + vscode
+  stub 的 provider 集成）；`tsc` 通过。
+
+### 边界与取舍
+
+- 语义 token 只在解析报错时启用，且使用主题对 `type` / `property` / `string` 的默认
+  配色，可能与 TextMate XML 配色略有差异——只在打字过程中出现，可接受；
+- 未实现 `provideDocumentSemanticTokensEdits`（delta 版本），每次全量计算；单文件
+  解析在 KB 级，开销可忽略。
+
+---
+
+## 十三、问题分析（第八轮，2026-08-02）：`.w3x` 美术资产未被索引（AUGunship_SKN）
+
+### 问题
+
+AttachTest `Harbinger Gunship\GameObject.xml` 中 `<Model Name="AUGunship_SKN"/>`
+报两条错误：
+
+1. Problems 面板诊断：`Unresolved reference "AUGunship_SKN" (not found in the current index)`
+   （`unresolved-reference`，来自 `features/diagnostics.ts`）；
+2. 悬停提示：`No matching definition of type BaseRenderAssetType in the current index
+   (may exist in a compiled manifest or vanilla data).`（来自 `features/hover.ts`）。
+
+两条是同一缺失定义的两种呈现（诊断 + hover），不是两个独立 bug。
+`W3DContainer:AUGunship_SKN` 确实由 mod 定义——但定义在
+`Harbinger Gunship\W3X\AUGUNSHIP_SKN.w3x` 里，而索引器只解析 `.xml` / `.manifestxml`。
+
+### 关键事实（实测）
+
+1. **`.w3x` 是文本 XML**：文件头即 `<?xml ...?>` + `<AssetDeclaration>`，内容为
+   `<W3DContainer id="AUGUNSHIP_SKN" Hierarchy="AUGUNSHIP_SKL">` + `<SubObject>` 子树。
+   附带 UTF-8 BOM（抽样 292 个 Corona w3x：0 个 UTF-16，40 个带 BOM）。
+2. **w3x 通过 `<Include type="all">` 链进索引**：`Mod.xml → … → W3X.xml → W3X/*.w3x`，
+   也常用 `ART:xxx.w3x`（SDK 根、项目 `Art` 等搜索路径）。索引器此前只把 w3x 登记为
+   文件（`stream.files`），从不解析，因此其顶层资产不在 `assetsById` 中。
+3. **类型匹配本来是对的**：`Model@Name` 的 refType 是 `BaseRenderAssetType`，
+   `W3DContainer → BaseRenderAssetType` 可赋值（`isAssignableTo` = true）。缺的只是定义。
+4. **manifest / SageXml 支持早已存在**（第二轮/第三轮），但该资产是 mod 自己的 w3x，
+   不在 `builtmods/*.manifest` 也不在 SageXml——hover 的 "may exist in a compiled manifest
+   or vanilla data" 只是通用兜底文案。
+
+### 规模调查（Corona，D: 盘已连接）
+
+- **3788 个 w3x，共 2.64 GB**；最大 22.8 MB；163 个超过原 4 MB 解析上限，11 个超 10 MB。
+- 大文件结构符合"建模软件导出"模式：顶层是少量固定 W3D 资产
+  （`W3DHierarchy` + 若干 `W3DMesh` / `W3DContainer` / `W3DCollisionBox`，
+  有的还带 `<Includes>` 引用 `ART:*.xml`）；体积大头是 `W3DMesh` 内的
+  `Vertices/V`、`Normals/N`、`TexCoords/T`、`Triangles/T` 等 `maxOccurs="unbounded"`
+  数值元素。例如 21.7 MB 的 `CBRefinery_BLD.W3X` 只有 22 个顶层记录，
+  Vertices+Triangles 块占约 12 MB（55%）。
+- 全量建 DOM 的代价（实测 6.3 MB Aegis 文件）：407 ms、193,651 个元素、
+  **堆内存 +109 MB（约 17 倍文本体积）**；22 MB 文件外推约 1.5 s + ~380 MB/份。
+  浅扫描（只取顶层记录）：6.3 MB ~200 ms、22 MB ~600 ms，保留内存近似为零。
+- **读整个文件不可避免，但建树不是**：浅扫描仍然是线性扫描全文（要知道顶层元素边界
+  就必须扫完），只是不分配子节点对象——所以"事后优化"完全有意义，且是必要项。
+
+### 修复
+
+1. **新增浅扫描模块** `src/indexer/shallowScan.ts`（纯 TS）：
+   单次线性扫描，只提取顶层元素 `name + id`（含精确 offset）、顶层 `<Includes>`
+   的 `Include@type/source`、任意层级 `<xi:include>` 的 `href/xpointer`、`<Defines>`
+   常量；注释 / CDATA / DOCTYPE / PI 整体跳过；属性解析兼容引号内 `>` 与 `/`。
+2. **索引模式三分**：`.xml`/`.manifestxml` 全量解析（4 MB 上限不变）；
+   `.w3x` 一律浅扫描；未知扩展名先嗅探文件头（512 字节，BOM/空白后以 `<` 开头且无
+   NUL 字节 → 按 XML 浅扫描，否则按二进制仅登记）。w3x 自身的 `<Includes>` 与
+   嵌套 `xi:include` 会继续被 walk（BAB 语义）。
+3. **持久缓存**：`DocumentCache` / `ShallowScanCache` 移到 `src/indexer/caches.ts`，
+   由 `ModWorkspace` 持有并传入每次新建的 `ModIndexer`；读取时按 `mtimeMs + size`
+   校验，未变化不重读。这是 w3x 方案在 Corona 上可用的前提（否则每次保存都重读 2.6 GB）。
+4. **BOM 剥离**：`xmlParser` 新增 `stripBom()`，全量解析与浅扫描前统一剥离，
+   保证第一行偏移与编辑器一致。
+5. **Include source 补全候选**：DATA 目录与项目相对路径候选加入 `.w3x`
+   （`fileScanner`），`W3X/xxx.w3x`、`ART/xxx.w3x` 可补全。
+6. 索引报告/状态栏新增 `shallowScannedFiles` / `shallowCacheHits` 统计。
+
+### 验证
+
+- 单元测试 **53 → 63 全绿**：新增 `test/shallowScan.test.mjs`（顶层资产 / Includes /
+  xi:include / Defines / 引号内 `>` / CDATA / 未闭合标签 / 数值载荷不产生记录）；
+  indexer 新增 w3x 链、`.w3d` 嗅探、二进制 `.dds` 跳过、跨重建缓存命中、
+  BOM 偏移断言；xmlParser 新增 `stripBom` 用例。
+- AttachTest 实机：
+  - 首次构建 1.2 s，浅扫 62 个文件；`Model@Name=AUGunship_SKN` →
+    `W3DContainer @ …\W3X\AUGUNSHIP_SKN.w3x:3`；`Hierarchy=AUGunship_SKL`、
+    `AUGunship_FP` 同样解析；资产数 35,546 → 35,607。
+  - 第二次构建 354 ms，0 次重扫，62 次缓存命中。
+- Corona 实机（D: 盘）：
+  - 首次构建 241 s：8,976 个文件、**浅扫 4,829 个**、资产 64,868
+    （manifest 35,322）、3 个流、183 个 Define。
+  - 第二次构建 38 s：0 次重扫、4,829 次缓存命中、资产数一致。
+  - `CBREFINERY_BLD` 正确解析到 `W3DHierarchy @ cb/CBRefinery_BLD.w3x:16` 与
+    `W3DContainer @ …:776412`。
+
+### 边界与后续
+
+- 首次建索引较慢（Corona ~4 分钟，机械盘）：读 2.6 GB 是下限，浅扫描本身 ~30-90 s；
+  后续可做并行扫描或把 w3x 顶层记录做成独立小缓存文件。
+- 第二次构建仍有 ~38 s（主要是对 ~9k 文件逐文件 stat + 重建 Map，机械盘随机读）；
+  后续可做"文件清单 + stat 快照"级缓存。
+- 浅扫描对 w3x 不做 XSD 校验（编辑器特性不注册 w3x 语言）；若用户手动把 `*.w3x`
+  关联为 xml，完整解析仍会发生在打开的文档上（VS Code 自身行为）。
+- UTF-16 XML 的 w3x 会被嗅探判定为二进制（文件头含 NUL）；实测生态中不存在，
+  如遇可再扩展解码。
+
+---
+
+## 十四、问题分析（第九轮，2026-08-02）：Corona 重建性能与内存
+
+### 目标与基线
+
+第八轮后 Corona 的实测基线：
+
+| 场景 | 耗时 |
+|---|---|
+| 首次全量建索引 | ~180-290s（机械盘波动） |
+| 信任二次构建（保存触发） | 8-21s |
+| 强制 reindex | 26-38s |
+| 首建后保留堆 | ~2.5GB（原因不明） |
+
+### 调查 1：二次构建为何仍有 8-21s
+
+插桩 `fs.statSync` 后发现：**每次构建（含信任重建）都会执行约 11 万次同步 statSync**，
+全部来自 `resolveSource` 的 include 目标存在性检查（BAB 搜索路径逐 base `statSync`）。
+机械盘上这就是 10-20s 的来源——即使文件内容全部命中缓存，include 解析仍在逐路径 stat。
+
+修复：新增 `IncludeResolveCache`（workspace 级、跨重建复用）：
+- 键 = 当前文件目录 + source 字符串；命中直接返回，不 stat；
+- 内容编辑不影响"文件是否存在"，因此保存触发的重建**不清除**解析缓存；
+- 文件创建/删除（watcher `onDidCreate` / `onDidDelete`）与 `ra3modxml.reindex`
+  （强制校验）清除缓存；
+- `reference` 的 manifest 查找（`builtmods/*.manifest` 存在性）同样缓存；
+- 配置变更（搜索路径 / builtmods 目录）时清除。
+
+### 调查 2：首建后保留堆 2.5GB 是什么
+
+逐级清空测量（构建 → 清 DOM 缓存 → 清 records → 清 walker → 清索引 Map）：
+
+- `DocumentCache`（元素预算 1M、64 条）：实际只保留 64 条 / 1788 个元素 ≈ 1MB；
+- `IndexRecordsCache`（8976 个文件的紧凑记录）：约 11MB；
+- 目录 walker：约 4MB；
+- 全部索引 Map（manifests + assets + assetsById + defines + files + streams +
+  candidates + diagnostics）：约 75MB；
+- 全部清空并强制 GC 后，堆回到基线（0MB 保留）。
+
+结论：2.5GB 是**构建期可回收垃圾**（60MB XML 的 DOM 瞬态 + 2.6GB w3x 扫描文本/行映射
+的分配压力），不是常驻泄漏；VSCode 正常 GC 压力下会回收。扩展常驻内存约为
+基线 + ~100MB。顺带修复了一个潜在常驻风险：w3x 的 `LineMap`（Corona 全量约 700MB）
+不再随 `ShallowScanCache` 保留，浅扫描结果以"带行号的紧凑记录"形式缓存。
+
+### 其他改动
+
+1. **`IndexRecordsCache` 取代 DOM 依赖的重建**：新增 `src/indexer/records.ts`，
+   每个文件解析时提取紧凑索引记录（顶层资产 / Define / Include / xi:include +
+   1-based 行号），跨重建缓存；信任重建完全不接触 DOM。
+2. **`DocumentCache` 双重淘汰**：条数（LRU）+ 元素预算（超预算先淘汰最大树），
+   把 DOM 常驻内存封顶；DOM 只服务于按需特性（跳转精确范围等）。
+3. **候选目录扫描并行化**（`collectSourceCandidates` 各目录 `Promise.all`）。
+4. **阶段计时**：`stats.candidatesMs` / `stats.walkMs` / `stats.resolveCalls` /
+   `stats.resolveCacheHits` 进入索引报告，便于后续定位耗时。
+5. 版本 0.1.0 → **0.1.1**。
+
+### 实测（Corona，D: 机械盘）
+
+| 场景 | 优化前 | 优化后 |
+|---|---|---|
+| 首次全量建索引 | ~180-290s | ~250s（含 2.6GB w3x 读取+浅扫描、~7.6 万次冷 statSync） |
+| 信任二次构建 | 8-21s | **2.0s**（statSync 0 次，resolveHits 15,333） |
+| 强制 reindex | 26-38s | ~5s（保留解析缓存时）；显式命令会清解析缓存，约 15-25s |
+| 首建后保留堆 | ~2.5GB（疑为泄漏） | 确认是可回收垃圾；常驻 ~100MB |
+
+单元测试 **63 → 73 全绿**：新增 `records.test.mjs`（DOM→记录提取、浅扫描→记录）、
+`caches.test.mjs`（元素预算淘汰、LRU、records/resolve 缓存），indexer 测试补充
+resolve 缓存命中与"信任重建 0 次重解析"断言。
+
+### 遗留
+
+- 首次全量建索引仍受限于机械盘读 2.6GB + 冷 statSync（~4 分钟）；后续可考虑
+  并行浅扫描（worker）或把 w3x 顶层记录持久化到磁盘缓存。
+- `ra3modxml.reindex` 出于正确性会清空解析缓存（可能 ~15-25s）；如接受 watcher
+  可靠性可改为保留。

docs/plan.md

@@ -1,6 +1,6 @@
 # 调研结论与实施计划（已按最新代码同步更新）
 
-> 说明：本文档随实现演进持续同步。最近一次同步（2026-08-01）对齐了实现过程中新增的模块与设计变更：BAB 精确搜索路径、manifest 类型/ID 推导、上下文感知元素类型、无类型引用、精确跳转范围、嵌套 `xi:include`、注入式语法高亮等。
+> 说明：本文档随实现演进持续同步。最近一次同步（2026-08-01）对齐了实现过程中新增的模块与设计变更：BAB 精确搜索路径、manifest 类型/ID 推导、上下文感知元素类型、属性级 refType / Poid 局部引用（`id` 定义点）、精确跳转范围、嵌套 `xi:include`、注入式语法高亮等。
 
 ## 一、调研结论（带证据）
 
@@ -26,7 +26,8 @@
 - 根元素 `AssetDeclaration`：`Tags` / `Includes` / `Defines` + **295 个顶层资产元素**（含内联声明）。
 - 每个元素名对应一个 `complexType`，子元素用 `xs:sequence` / `xs:choice` 定义，属性用 `xs:attribute` 定义；复杂类型通过 `xs:extension` 继承（如 `BaseInheritableAsset` 提供 `inheritFrom`）。
 - `Includes/Ref.xsd` 定义了大量带 `xas:refType="<资产类型>"` 的引用类型（如 `CommandSet` 引用 `LogicCommandSet`）→ 补全/导航按引用类型过滤的依据。
-- `XmlEdit:Default` 提供默认值；`xs:enumeration` 提供枚举值；`xas:isRef="true"` 无 `refType` 时为**无类型引用**（匹配任意已声明资产）。
+- `XmlEdit:Default` 提供默认值；`xs:enumeration` 提供枚举值。`xas:refType` 可声明在 simple type 上，也可声明在 `<xs:attribute>` 节点上（模型生成器两者都读、属性级优先）。
+- `Poid`（"Pipeline Object Id"，`xas:isWeakRef="true"`）表示**管线局部标识**：`id` 属性定义元素自身（如 `ModuleData@id` → refType `ModuleData`）；`ModuleId`、`AutoResolveBody`、`SoundRef` 等 Poid 属性引用同一资产/子树内的模块、子对象、材质——它们都不对全局资产索引做 resolved 判定。
 - **同名元素在不同父节点下类型不同**（如 `<Weapon>` 在武器槽下是 `WeaponSlot_WeaponData`，在别处可能是 `WeaponRef`）→ 需要上下文感知解析。
 
 ### 4. 领域特有约定
@@ -60,7 +61,7 @@
 - **核心与编辑器解耦**：`language/`、`model/`、`indexer/` 为纯 TS 模块（不 import `vscode`），可单测与复用（呼应 P1 需求 7）。
 - **esbuild 打包**，产物 `dist/extension.js`。
 - **XSD → JSON 模型**：开发期工具 `tools/xsd-to-model.mjs` 把 821 个 XSD 解析成 `schema-model.json`（元素树、属性、文档、枚举、引用类型映射），随插件发布；运行时不再解析 XSD。
-- **运行时 XML 解析**：自研带源码偏移的轻量解析器 `language/xmlParser.ts`（标签/属性/值均记录起止偏移，容错解析以支持输入中的补全与诊断）。`fast-xml-parser` 仅用于开发期 XSD 生成。
+- **运行时 XML 解析**：自研带源码偏移的轻量解析器 `language/xmlParser.ts`（标签/属性/值均记录起止偏移，容错解析以支持输入中的补全与诊断；未闭合引号在行尾恢复，避免吞掉整个文档）。`fast-xml-parser` 仅用于开发期 XSD 生成。
 - **AssetType 哈希表**：`tools/extract-asset-types.mjs` 从 OpenSAGE `AssetType.cs` 提取 `asset-types.json`；哈希未知时以 manifest 名称前缀推导类型。
 
 ### 架构
@@ -74,6 +75,7 @@ src/
     xmlParser.ts       带源码偏移的轻量 XML 解析器（格式错误定位、容错）
     context.ts         补全上下文分析（元素名/属性名/属性值/内容）
     typeContext.ts     上下文感知元素类型解析（resolveElementType 沿解析树逐层解析）
+    semanticTokens.ts  语义 token 兜底高亮（纯 TS：标签/属性/值范围，仅 malformed 时启用）
   model/
     schemaModel.ts     schema-model.json 的类型/属性/子元素查询 + 类型名规范化（纯 TS）
     schema-model.json  由 tools/xsd-to-model.mjs 生成
@@ -83,36 +85,85 @@ src/
     manifestParser.ts  .manifest 二进制解析 + 类型/ID 推导（纯 TS）
     fileScanner.ts     目录遍历缓存 + Include source 候选收集
     refs.ts            引用目标解析（按 refType / isRef / inheritFrom 过滤，纯 TS）
-    indexer.ts         工作区索引器（资产/Define/流/manifest 合并，LRU 解析缓存）
+    shallowScan.ts     大体积美术资产（.w3x 等）顶层浅扫描（纯 TS，不建 DOM）
+    records.ts         每文件紧凑索引记录（资产/Define/Include/xi + 行号）
+    caches.ts          跨重建持久缓存（DocumentCache / IndexRecordsCache /
+                       IncludeResolveCache）
+    indexer.ts         工作区索引器（资产/Define/流/manifest/w3x 合并）
     types.ts           共享类型
   features/
-    completion.ts      补全 provider（元素/属性/值，上下文感知）
+    completion.ts      补全 provider（元素/属性/值，上下文感知；xs:list 多值按当前段过滤）
     hover.ts           hover provider
     navigation.ts      定义/引用/文档链接/大纲
     diagnostics.ts     实时诊断
+    semanticTokens.ts  语义 token provider（文档有解析错误时接管着色）
 syntaxes/
   ra3modxml.tmLanguage.json   注入 source.xml 的领域高亮（纯注入，不替换 XML 主语法）
 tools/
-  xsd-to-model.mjs       XSD → schema-model.json
+  xsd-to-model.mjs       XSD → schema-model.json（含 xs:list：继承 item 枚举/引用语义，isList 标记）
   extract-asset-types.mjs OpenSAGE AssetType.cs → asset-types.json
 test/
   fixtures/minimod      样例 Mod（include 各种情形、同名 ID、嵌套 xi:include、manifest 回退）
-  *.test.mjs            8 个测试文件（xmlParser / includeResolver / manifestParser /
+  *.test.mjs            11 个测试文件（xmlParser / context / completion / semanticTokens /
-                        indexer / schemaModel / refs / typeContext / manifestTypes）
+                        includeResolver / manifestParser / indexer / schemaModel / refs /
+                        typeContext / manifestTypes）
 ```
 
 ### 关键设计决策
 
 1. **语言激活范围**：不劫持 `*.xml`。通过 `workspaceContains:**/Data/Mod.xml`、`**/*.babproj` 激活；语法高亮为**纯注入** grammar（不声明 `language`，避免覆盖内置 XML 语法）。
 2. **索引范围与默认值**：索引“项目 Data + additionalmaps + 沿 include 可达的 SageXml 原版源码”；SDK 路径默认 `C:\Apps\RA3-MODSDK-X`（可配置）。`reference` include 解析为 `builtmods` 下对应 manifest（惰性解析、按文件缓存），manifest 缺失/无效时回退到占位 XML。
+   **美术资产（.w3x）**：`<Include type="all">` / `ART:` 指向的 `.w3x`（及内容嗅探为
+   XML 的未知扩展名文件）按其顶层资产入库（`W3DContainer` / `W3DMesh` /
+   `W3DHierarchy` / `W3DCollisionBox` 等），使 `Model@Name`、`Hierarchy`、`Mesh`
+   等引用可解析；大模型文件**浅扫描**（不建 DOM），结果缓存在 workspace 级、
+   跨重建复用（详见设计决策 14）。
 3. **manifest 资产建模**：类型优先用哈希表，未知时从名称前缀推导；可引用 ID 取最后冒号段；类型名统一走大小写规范化（`W3dContainer` ↔ `W3DContainer`），类型匹配严格遵循 XSD 继承链。
 4. **上下文感知元素类型**：同名元素按父元素类型解析（`resolveElementType` 沿解析树逐层 `childTypeOf`，失败回退全局映射），保证 `<Weapon>` 等元素的属性/引用判定正确。
-5. **引用判定与解析**：`refType` 或 `isRef` 均视为引用；带 `refType` 时严格按类型过滤（同名 ID 不串类型）；无类型引用匹配任意声明 ID；`inheritFrom` 按可继承类型过滤。
+5. **引用判定与解析**：`refType` 或 `isRef` 均视为引用；带 `refType` 时严格按类型过滤（同名 ID 不串类型）；`inheritFrom` 按可继承类型过滤。**局部作用域例外**（`isLocalReferenceAttribute`）：`id` 是元素自身的定义点——无 refType 或 refType 与自身类型兼容时不检查、不解析（`RoadObject@id→Road` 这类跨类型 id 引用保留检查）；Poid 类型属性是管线局部引用，全局索引无法判定，不检查、不解析。
 6. **重复 ID 诊断**：与 `check_duplicate_ids.py` 一致——SageXml 不参与冲突判定，mod 覆盖原版视为正常。
 7. **未解析引用诊断**：按设置严重级别报告（默认 warning）；类型不匹配时给出明确文案（"有同名 ID 但类型不匹配"）。`definitionMode` 设置控制跳转候选：`all`（mod + 原版全部列出，mod 优先）或 `project-only`。
 8. **跳转精度**：XML 定义跳转到 `id` 属性值的精确 Range；manifest 定义映射到源码文件（如 SageXml）时也在文件内精确定位；找不到再回退到记录行。
 9. **嵌套 `xi:include`**：任意层级处理——目标缺失产生诊断、目标存在则纳入索引；根级 `xpointer` 容器内容按顶层资产索引。
 10. **性能**：索引在后台执行；解析结果 LRU 缓存（约 64 个文档）；文件保存后防抖全量重建（1.5s），重建期间的新请求标记脏并在完成后重跑；状态栏显示进度与统计。
+11. **`xs:list` 建模与多值补全**：list 简单类型继承 itemType 的枚举 / refType / isRef /
+    allowsDefine 并标记 `isList`（`LocomotorSurfaceBitFlags`、`KindOfBitFlags` 等 79 个
+    类型、317 处属性声明受益）；补全只对“最后一个空格段”过滤，替换范围只覆盖当前段，
+    支持 `Surfaces="GROUND ` 之后继续输入 `W` 提示 `WATER`。
+12. **未闭合引号的行尾恢复**：起始标签扫描到 EOF 且引号未闭合时，在第一个换行处截断
+    标签并继续解析，未闭合只影响当前行（仍上报 `Unterminated start tag`），后续元素
+    的补全 / hover / 诊断不中断。
+13. **语义 token 兜底高亮**：TextMate 对未闭合引号会把后续内容当字符串吞掉（任何
+    XML 编辑器皆然）；扩展注册 `DocumentSemanticTokensProvider`，仅当解析报错时用
+    语义 token 覆盖标签名 / 属性名 / 属性值（标准 token 类型 `type` / `property` /
+    `string`，主题自带配色）。合法文件返回空，观感与纯 TextMate 完全一致。
+14. **大体积美术资产浅扫描 + 跨重建持久缓存**（第八轮，2026-08-02）：
+    - 实测 Corona：3788 个 w3x / 2.64 GB，163 个超过 4 MB，最大 22.8 MB；大文件是
+      `W3DHierarchy` + 若干 `W3DMesh`（`Vertices/V`、`Triangles/T` 等 unbounded 数值
+      载荷），顶层记录通常只有几个到二十几个。
+    - 全量解析内存放大约 17 倍（6.3 MB 文本 → +109 MB DOM），不可接受；`scanXmlShallow`
+      单次线性扫描只提取顶层 `name+id`、`<Includes>`、`<xi:include>`、`<Defines>`，
+      22 MB 文件 ~600 ms、保留内存≈0。
+    - 索引按扩展名三分：`.xml`/`.manifestxml` 全量解析（4 MB 上限不变）；`.w3x` 浅扫描；
+      未知扩展名嗅探文件头（`<` 开头、无 NUL）决定按 XML 浅扫描或二进制登记。
+    - `DocumentCache` / `ShallowScanCache` 由 `ModWorkspace` 持有，每次重建传入新的
+      `ModIndexer`；按 `mtimeMs + size` 校验，未变化不重读。Corona 第二次构建
+      w3x 重扫数为 0（4,829 次缓存命中）。
+    - 读取整个文件不可避免（顶层边界需要全量扫描），但建 DOM 不是；优化的是
+      "不分配子节点对象"与"跨重建不重读"，两者叠加后方案可行。
+15. **重建零 stat + 记录驱动索引**（第九轮，2026-08-02，v0.1.1）：
+    - 插桩发现每次重建（含信任重建）都有约 11 万次同步 `statSync`
+      （`resolveSource` 的 include 存在性检查），机械盘上占 10-20s；
+      `IncludeResolveCache` 按（目录 + source）缓存解析结果，内容编辑不清、
+      创建/删除文件与强制 reindex 才清。信任重建 statSync 降到 0。
+    - `IndexRecordsCache` 缓存每文件紧凑索引记录（顶层资产 / Define / Include /
+      xi:include + 1-based 行号），信任重建完全不接触 DOM；`DocumentCache`
+      双重淘汰（条数 LRU + 元素预算，超预算先淘汰最大树）把 DOM 常驻内存封顶。
+    - w3x 缓存不再保留 LineMap（Corona 全量约 700MB），浅扫描直接产出带行号的记录。
+    - 实测：Corona 信任二次构建 21s → **2.0s**（statSync 0）；首建后 2.5GB
+      堆保留确认为构建期可回收垃圾，常驻 ~100MB；强制 reindex ~5-25s。
+    - 候选目录扫描并行化；`stats` 新增 `candidatesMs` / `walkMs` / `resolveCalls` /
+      `resolveCacheHits` 供索引报告定位耗时。
 
 ## 三、实施步骤
 
@@ -121,24 +172,114 @@ test/
 3. [x] 生成模型：`schema-model.json`（XSD，295 顶层元素 / 1851 类型）与 `asset-types.json`（79 个 AssetType 哈希）。
 4. [x] 纯 TS 核心：include 解析、manifest 解析、XML 解析封装、索引器、引用解析。
 5. [x] 功能层：补全、hover、导航、诊断、大纲、高亮 grammar。
-6. [x] 单测（fixture Mod，31 个用例全绿）+ 编译 + `vsce package` 打包（ra3-mod-xml-0.1.0.vsix，约 259KB）。
+6. [x] 单测（fixture Mod，73 个用例全绿：含 xs:list 枚举、未闭合引号恢复、list 多值
-7. [x] 在 AttachTest / GenEvoTest / Corona 上做冒烟验证，并按真实项目反馈修复问题（详见 `docs/analysis-issues.md` 三轮分析）。
+   分段、带 vscode stub 的补全集成、语义 token 兜底）+ 编译 + `vsce package` 打包
+   （ra3-mod-xml-0.1.1.vsix，约 499KB）。
+7. [x] 在 AttachTest / GenEvoTest / Corona 上做冒烟验证，并按真实项目反馈修复问题（详见 `docs/analysis-issues.md` 八轮分析）。
+8. [x] w3x 美术资产索引（第八轮）：新增 `shallowScan.ts` / `caches.ts`，w3x 与内容嗅探
+   XML 走浅扫描，Include source 补全候选加入 w3x，BOM 剥离，缓存跨重建持久化；
+   AttachTest 报错场景（`AUGunship_SKN`）修复，Corona 首次/二次构建实测验证。
+9. [x] Corona 性能与内存优化（第九轮，v0.1.1）：`records.ts` 记录驱动索引、
+   `IncludeResolveCache` 零 stat 重建、DOM 元素预算淘汰、w3x LineMap 移除、
+   候选并行扫描、阶段计时；Corona 信任重建 2.0s；确认首建 2.5GB 为可回收垃圾。
 
 ## 四、验证结果（实测）
 
 | 项目 | 规模 | 索引耗时 | 资产数 | 说明 |
 |---|---|---|---|---|
-| AttachTest | 71 文件 | ~0.6s | 35,546（manifest 35,322） | 2 个流（static + mapmetadata） |
+| AttachTest | 88 文件 | ~1.2s | 35,607（manifest 35,322，w3x 浅扫 62） | 2 个流（static + mapmetadata）；二次构建 354ms / 62 缓存命中 |
-| GenEvoTest | 24 文件 | ~1.8s | 35,392（manifest 35,322） | 项目 ID（alliedmcv 等）正确收录 |
+| GenEvoTest | 66 文件 | 首次 ~2.6s / 二次 ~0.4s | 35,502（manifest 35,322，w3x 浅扫 38） | 2 个流、73 个 Define；二次构建 0 重扫 / 38 缓存命中 |
-| Corona | 3,448 文件（+ 非 XML 资产路径） | ~54.5s | 55,305（manifest 35,322） | 3 个流、183 个 Define、0 诊断 |
+| Corona | 8,976 文件 | 首次 ~250s / 信任二次 ~2s / 强制 ~5-25s | 64,868（manifest 35,322，w3x 浅扫 4,829） | 3 个流、183 个 Define、0 诊断；二次构建 statSync 0、resolveHits 15,333 |
 
-单元测试覆盖：XML 解析（自闭合/容错/偏移）、include 解析（BAB 顺序、SDK 根优先于 SageXml）、manifest 二进制解析（合成 v5 样本、类型/ID 推导）、索引器（资产/Define/流/缺失 include/嵌套 xi:include）、XSD 模型（上下文类型、`childTypeOf`、大小写规范化）、引用过滤（`Weapon="X"` 只跳 `WeaponTemplate`、无类型引用、`Side="Allies"` 命中 manifest 的 `PlayerTemplate`）。
+单元测试覆盖：XML 解析（自闭合/容错/偏移/未闭合引号行尾恢复）、补全上下文（未闭合引号仍为 attribute-value、list 多值分段）、补全集成（vscode stub 下 `LocomotorTemplate@Surfaces` 未闭合引号枚举补全、空格后第二段过滤与替换范围）、语义 token（标签/属性/值范围、合法文档返回空、malformed 返回兜底 token）、include 解析（BAB 顺序、SDK 根优先于 SageXml）、manifest 二进制解析（合成 v5 样本、类型/ID 推导）、索引器（资产/Define/流/缺失 include/嵌套 xi:include）、XSD 模型（上下文类型、`childTypeOf`、大小写规范化、属性级 refType、外来命名空间判定、`xs:list` 枚举继承与 `isList` 标记）、引用过滤（`Weapon="X"` 只跳 `WeaponTemplate`、模块 `id` 定义点、Poid 局部引用、`xi:include` 不校验、`Side="Allies"` 命中 manifest 的 `PlayerTemplate`）。
 
-> 注：`D:\Mods\CoronaMod` 位于移动硬盘，当前未连接；GenEvoTest / Corona 的回归需在 D: 盘可用时补跑（用户会另行通知）。
+> 注：D: 盘移动硬盘已恢复连接；Corona 已在第八 / 九轮按上述新数据回归。
 
 ## 五、假设与开放问题
 
 - 假设 SDK 路径默认 `C:\Apps\RA3-MODSDK-X`（与 prompts 一致），可在设置中修改。
 - 假设补全/导航以“文本语义分析”为主，不做完整 XSD 校验（BAB 才是权威校验器）。
 - 开放：是否发布到 VS Code Marketplace（需要 publisher）——本期先保证本地 `vsce package` 可安装。
-- 开放：嵌套 `xi:include` 内容的"虚拟合并"进父文档（用于父文档内的补全/诊断感知被内联内容）——当前仅保证目标文件可索引、可导航、缺失可诊断。
+- 开放：**“宏展开”式虚拟合并**（用户提议，方向已确认）：解析前先把 `xi:include`
+  （以及顶层 `<Include type="all">`、`inheritFrom` 继承合并）展开成不含 include 的
+  文档树，再对展开后的树做 XSD 校验、补全与诊断——与 BAB 编译时把整个 Mod 合并成
+  一份大 XML 的行为一致。展开树需携带**来源追溯**（错误/跳转仍定位到原始文件），
+  并处理 xpointer 子集解析与 include 循环。当前仅做到：目标文件可索引、可导航、
+  缺失可诊断；`xi:include` 本身不参与 XSD 校验（第五轮）。
+  该功能也是“GameObject 内模块 id 局部作用域解析”（第四轮遗留）的地基。
+  详细设计备忘见下一节。
+
+## 六、include 展开设计备忘（2026-08-01，待实施）
+
+> 目的：集中记录 include 处理相关的现状、结论与设计，下次遇到 include 问题时从这里继续，
+> 并在实施后把结果回写本节。
+
+### 1. 现状（截至第五轮，已实现）
+
+| 能力 | 状态 |
+|---|---|
+| `<Include type="all">` / `instance` 递归索引（顶层资产、流、Define） | 已实现（indexer `walk`） |
+| `reference` → builtmods manifest 解析 / 缺失回退占位 XML | 已实现 |
+| 嵌套 `xi:include`（任意层级）：目标可索引、缺失报 `include-not-found`、Ctrl+点击跳转、`href` hover 解析目标 | 已实现（第二轮 + 第五轮） |
+| `xi:include` 及其属性不参与 XSD 校验（外来命名空间守卫 `isXsdElementName` / `isXsdAttributeName`） | 已实现（第五轮） |
+| include 目标内容“虚拟合并”进父文档的逻辑树 | **未实现**（本文档主题） |
+
+### 2. 已确认的方向
+
+先展开成不含 include 的文档树（类比 C++ 宏展开），再对展开后的树做 mod XML 解析。
+BAB（`defaultscript.cs`）编译时正是这样把整个 Mod 合并成一份大 XML 的。
+
+### 3. 关键设计决策：逻辑树拼接，不做文本拼接
+
+- **不要**把 include 目标展开成文本再整体重新解析：源码偏移会断裂，诊断 / 跳转 / hover /
+  补全全部无法映射回原始文件。
+- **要做**的是：解析器逐文件解析（现状不变）；展开器把目标文件选中节点按 `xpointer`
+  子集挂进父元素，节点保留各自的源文件与原始偏移（来源追溯）。后续分析跑在逻辑树上，
+  范围映射按节点 `sourceFile` 回到对应文件的 lineMap。
+- 现有 `parseXml` 已记录标签 / 属性 / 值的起止偏移，`XmlElement` 结构可直接复用；拼接时
+  用浅拷贝节点壳并重建 parent 链，避免破坏目标文件缓存树自身的 parent 指针。
+
+### 4. 展开范围
+
+| 构造 | 拼入逻辑树 | 理由 |
+|---|---|---|
+| `xi:include` | ✅ | 内容并入父元素（HeadlightDraw2 场景） |
+| EA `<Include type="all">` | ✅ | BAB“内容合并”，等价于复制进来 |
+| `type="instance"` | ❌ | 只影响编译可见性；拼树会把 BaseVehicle 的顶层资产错误塞进当前文档 |
+| `type="reference"` | ❌ | manifest 编译产物，无文本内容 |
+| `inheritFrom` + `xai:joinAction` | 单独一轮 | 元素级继承深合并（Replace/Remove），不是宏展开 |
+
+### 5. 落地位置与接入点
+
+- 新纯模块（与编辑器解耦，呼应 P1）：输入 `(parse 树, resolveSource 回调, readDocument 回调)`，
+  输出逻辑树（root + elements 扁平列表，沿用 diagnostics 的遍历形态）。
+- 接入：diagnostics / hover / navigation / completion 目前各自 `parseXml(text)`；
+  改为 parse 后过 expander 取逻辑树；范围映射按节点 `sourceFile` 选对应文件的 lineMap。
+- 按需展开（当前打开文档）+ 按文件缓存（复用 indexer `readDocument` 的 LRU）；
+  环 / 深度守卫复用现有 `visitedAll` 与深度限制思路（建议最大深度 64）。
+- `xpointer`：仅支持 mod 实际使用的 `xmlns(n=...) xpointer(/n:Name/child::*)` 子集
+  （现有 `findXPointerContainer` 正则已覆盖）；完整 XPath 暂不支持，遇到新形式先记录到本节。
+
+### 6. 后续收益（承接第四 / 五轮遗留）
+
+- **GameObject 内模块 id 局部作用域**：展开后 include 进来的兄弟模块（HeadlightDraw2）
+  与本体模块同树，`AttachModuleId` / `ModuleId` / `AutoResolveBody` 等 Poid 引用
+  才能静态解析与诊断（第四轮遗留）；
+- 跨 include 的上下文类型解析与结构校验；
+- 顶层 `type="all"` 合并后，当前文档视角的重复 id / 引用诊断更接近 BAB 结果。
+
+### 7. 风险与边界
+
+- 大文件性能（Corona 约 7500 文件 / 38MB）：只展开当前文档的可达链，不全局展开；
+- include 环 / 深度：visited 集合 + 最大深度；
+- 被包含内容的诊断上报位置：建议按节点源文件 URI 上报（与偏移一致），包含点的 `href`
+  上只报“缺失 / 无法解析 / 环”类问题；
+- `inheritFrom` 深合并（`xai:joinAction` 的 Replace/Remove 语义）单独设计，别与宏展开混在一轮。
+
+### 8. 下次遇到 include 问题的检查清单
+
+1. 现象发生在哪一层：索引（indexer）、诊断（diagnostics）、导航 / hover、还是补全？
+2. 现状能力是否已覆盖（见第 1 节表格）；
+3. 涉及内容是否跨文件（需要逻辑树）还是本文件内（当前解析树即可）；
+4. 若要展开：先实现第 5 节的纯模块与单测（fixture 增加 HeadlightDraw2 场景），再接入 feature；
+5. 把新结论回写本节与 `docs/analysis-issues.md`。

docs/requirements.md

@@ -70,6 +70,15 @@ XML 之间的组织靠 `<Include>` 标签，共有三种语义：
   - `TypeId` 哈希 → 类型名的映射来自 OpenSAGE 的 `AssetType` 枚举（本工作区可提取）；
   - 资产名与源文件名各自存放在独立的空字符结尾字符串缓冲区中。
 
+**补充（美术资产 `.w3x` 索引）**：
+- `W3X.xml` / `ART:` include 链中的 `.w3x` 是文本 XML（建模工具导出，BAB 同样按
+  XML 编译），其顶层资产（`W3DContainer` / `W3DMesh` / `W3DHierarchy` 等）应参与
+  补全、悬停、导航与诊断——`Model@Name`、`Hierarchy`、`Mesh` 等引用依赖这些定义；
+- 大模型文件（实测 Corona 最大 22.8 MB，顶点/三角形数据占大头）采用**顶层浅扫描**
+  （不建 DOM 树），结果在 workspace 级缓存并跨重建复用，避免每次保存都重读整个
+  美术资产目录（Corona 约 2.6 GB）；索引记录与 include 解析结果同样跨重建缓存，
+  保存触发的重建零 stat、零重读（Corona 实测约 2 秒）。
+
 ### P1：非近期目标（本期不做，但预留扩展点）
 
 6. **高效搜索**：Mod 项目巨大（Corona 约 7500 个 XML、38MB）时直接全文搜索很慢，需要一个高效的 XML 内容索引机制。

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "ra3-mod-xml",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "ra3-mod-xml",
-      "version": "0.1.0",
+      "version": "0.1.1",
       "license": "MIT",
       "dependencies": {
         "fast-xml-parser": "^4.5.0"

package.json

@@ -2,7 +2,7 @@
   "name": "ra3-mod-xml",
   "displayName": "RA3 Mod XML",
   "description": "Red Alert 3 Mod XML tooling: syntax highlighting, completions, reference navigation and diagnostics for SAGE/BinaryAssetBuilder XML.",
-  "version": "0.1.0",
+  "version": "0.1.1",
   "publisher": "ra3-mod-xml",
   "license": "MIT",
   "engines": {
@@ -57,7 +57,7 @@
             "none"
           ],
           "default": "warning",
-          "description": "Severity for attribute references (CommandSet=..., inheritFrom=...) that cannot be resolved in the index. IDs that only exist in compiled manifests cannot be resolved yet."
+          "description": "Severity for attribute references (CommandSet=..., inheritFrom=...) that cannot be resolved in the index. The index covers mod XML sources, SDK SageXml sources and compiled manifests."
         },
         "ra3modxml.diagnoseUnknownElements": {
           "type": "boolean",

src/extension.ts

@@ -9,6 +9,10 @@ import {
   Ra3ReferenceProvider,
 } from "./features/navigation";
 import { Ra3Diagnostics } from "./features/diagnostics";
+import {
+  Ra3SemanticTokensProvider,
+  RA3_SEMANTIC_TOKENS_LEGEND,
+} from "./features/semanticTokens";
 
 const XML_SELECTOR: vscode.DocumentSelector = [{ language: "xml" }];
 
@@ -55,6 +59,13 @@ export function activate(context: vscode.ExtensionContext): void {
       new Ra3DocumentSymbolProvider(),
     ),
   );
+  context.subscriptions.push(
+    vscode.languages.registerDocumentSemanticTokensProvider(
+      XML_SELECTOR,
+      new Ra3SemanticTokensProvider(),
+      RA3_SEMANTIC_TOKENS_LEGEND,
+    ),
+  );
 
   const diagnostics = new Ra3Diagnostics(ws);
   context.subscriptions.push(diagnostics);
@@ -98,18 +109,24 @@ export function activate(context: vscode.ExtensionContext): void {
   context.subscriptions.push(
     vscode.workspace.onDidSaveTextDocument((doc) => {
       if (doc.languageId !== "xml") return;
+      ws.invalidate(doc.uri.fsPath);
       ws.scheduleRebuild();
       void diagnostics.update(doc);
     }),
   );
   context.subscriptions.push(
     vscode.workspace.onDidChangeConfiguration((e) => {
-      if (e.affectsConfiguration("ra3modxml")) ws.scheduleRebuild();
+      if (e.affectsConfiguration("ra3modxml")) {
+        // Search paths / builtmods locations may have changed: cached include
+        // resolutions and manifest lookups are no longer valid.
+        ws.invalidateExistence();
+        ws.scheduleRebuild();
+      }
     }),
   );
 
   context.subscriptions.push(
-    vscode.commands.registerCommand("ra3modxml.reindex", () => ws.rebuild()),
+    vscode.commands.registerCommand("ra3modxml.reindex", () => ws.rebuild(true)),
   );
   context.subscriptions.push(
     vscode.commands.registerCommand("ra3modxml.openIndexReport", () => {
@@ -124,7 +141,7 @@ export function activate(context: vscode.ExtensionContext): void {
       void vscode.window.showInformationMessage(
         `RA3 Mod XML index\n` +
           `Project: ${s.projectDir}\n` +
-          `Files: ${s.indexedFiles} (${s.parsedFiles} parsed)\n` +
+          `Files: ${s.indexedFiles} (${s.parsedFiles} parsed, ${s.shallowScannedFiles} shallow-scanned, ${s.shallowCacheHits + s.recordsCacheHits} cache hits)\n` +
           `Assets: ${s.assetCount} (${s.manifestAssetCount} from ${s.manifestFiles} manifests)\n` +
           `Defines: ${s.defineCount} · Streams: ${s.streams} · Candidates: ${s.sourceCandidates}\n` +
           `Indexed in ${(s.elapsedMs / 1000).toFixed(1)}s`,

src/features/completion.ts

@@ -1,8 +1,13 @@
 import * as vscode from "vscode";
 import { parseXml, type XmlElement } from "../language/xmlParser";
-import { analyzeContext, type CompletionContext } from "../language/context";
+import {
+  analyzeContext,
+  splitListValuePrefix,
+  type CompletionContext,
+} from "../language/context";
 import { resolveElementType } from "../language/typeContext";
 import * as model from "../model/schemaModel";
+import { isLocalReferenceAttribute } from "../indexer/refs";
 import type { ModWorkspace } from "../workspace";
 import type { ModIndex, AssetDef } from "../indexer/types";
 
@@ -177,12 +182,31 @@ export class Ra3CompletionProvider implements vscode.CompletionItemProvider {
     const el = ctx.element;
     const attr = ctx.attr;
     if (!el || !attr) return [];
-    const prefix = ctx.valuePrefix;
+    const rawPrefix = ctx.valuePrefix;
 
-    const endOffset = attr.quoteEnd > attr.valueEnd ? attr.valueEnd : document.offsetAt(position);
+    const attrName = attr.name.toLowerCase();
+    const elType = resolveElementType(el);
+    const attrInfo = model
+      .attributesOfType(elType)
+      .find((a) => a.name.toLowerCase() === attrName);
+
+    // xs:list values (bit flags such as Surfaces="GROUND WATER") are
+    // whitespace-separated: only the token currently being edited is used for
+    // filtering, and the replacement range covers that token instead of the
+    // whole value.
+    const seg = attrInfo?.isList
+      ? splitListValuePrefix(rawPrefix)
+      : { token: rawPrefix, start: 0 };
+    const prefix = seg.token;
+
+    const valueStartOffset =
+      attr.valueStart >= 0 ? attr.valueStart : document.offsetAt(position);
+    const endOffset =
+      attr.quoteEnd > attr.valueEnd ? attr.valueEnd : document.offsetAt(position);
+    const rangeStart = valueStartOffset + seg.start;
     const valueRange = new vscode.Range(
-      document.positionAt(attr.valueStart),
+      document.positionAt(rangeStart),
-      document.positionAt(Math.max(attr.valueStart, endOffset)),
+      document.positionAt(Math.max(rangeStart, endOffset)),
     );
 
     const make = (
@@ -200,7 +224,6 @@ export class Ra3CompletionProvider implements vscode.CompletionItemProvider {
     };
 
     const isInclude = el.name === "Include";
-    const attrName = attr.name.toLowerCase();
 
     // Include type / source
     if (isInclude && attrName === "type") {
@@ -217,16 +240,17 @@ export class Ra3CompletionProvider implements vscode.CompletionItemProvider {
       );
     }
 
-    const elType = resolveElementType(el);
-    const attrInfo = model
-      .attributesOfType(elType)
-      .find((a) => a.name.toLowerCase() === attrName);
-
     // inheritFrom: same element type first, then everything.
     if (attrName === "inheritfrom") {
       return this.assetIdItems(idx, el.name, null, prefix, make);
     }
 
+    // `id` attributes are definitions and Poid attributes are pipeline-local
+    // references; offering global asset ids for them would be wrong.
+    if (attrInfo && isLocalReferenceAttribute(elType, attr.name)) {
+      return [];
+    }
+
     if (attrInfo?.refType) {
       return this.assetIdItems(idx, null, attrInfo.refType, prefix, make);
     }

src/features/diagnostics.ts

@@ -115,8 +115,12 @@ export class Ra3Diagnostics {
         }
       }
 
+      // Elements outside the EA asset namespace (e.g. XInclude <xi:include>)
+      // are not part of the RA3 XSD model; skip their validation entirely.
+      const isXsdElement = model.isXsdElementName(el.name);
+
       // Unknown element.
-      if (settings.diagnoseUnknownElements && !el.name.startsWith("xi:")) {
+      if (settings.diagnoseUnknownElements && isXsdElement) {
         const knownType = model.elementTypeName(local);
         if (!knownType) {
           diags.push(
@@ -131,12 +135,15 @@ export class Ra3Diagnostics {
       }
 
       // Attributes.
+      if (isXsdElement) {
         const elType = resolveElementType(el);
         const knownAttrs = model.attributesOfType(elType);
         const knownNames = new Set(knownAttrs.map((a) => a.name));
         for (const attr of el.attrs) {
           const aName = attr.name;
-        if (aName.startsWith("xmlns") || aName.startsWith("xai:") || aName.startsWith("xi:")) {
+          // Namespace declarations and prefixed attributes (xai:, xi:,
+          // xlink:, xml:, xsi:, xmlns:*) are not defined by the EA XSD.
+          if (aName.startsWith("xmlns") || !model.isXsdAttributeName(aName)) {
             continue;
           }
           if (settings.diagnoseUnknownElements && !knownNames.has(aName)) {
@@ -164,6 +171,7 @@ export class Ra3Diagnostics {
             diags,
           );
         }
+      }
 
       // Include-specific checks.
       if (local === "Include") {

src/features/hover.ts

@@ -29,7 +29,7 @@ export class Ra3HoverProvider implements vscode.HoverProvider {
     // Attribute name.
     for (const attr of el.attrs) {
       if (offset >= attr.nameStart && offset <= attr.nameEnd) {
-        return this.attributeHover(elType, attr.name);
+        return this.attributeHover(el, elType, attr.name);
       }
     }
     // Attribute value.
@@ -47,6 +47,14 @@ export class Ra3HoverProvider implements vscode.HoverProvider {
   }
 
   private elementHover(name: string): vscode.Hover | null {
+    if (name.startsWith("xi:")) {
+      const md = new vscode.MarkdownString();
+      md.appendCodeblock(`<${name}>`, "xml");
+      md.appendMarkdown(
+        "XInclude element (W3C XInclude namespace) — not part of the RA3 XSD model.",
+      );
+      return new vscode.Hover(md);
+    }
     const type = model.elementTypeName(name);
     const info = type ? model.typeInfo(type) : undefined;
     const md = new vscode.MarkdownString();
@@ -68,7 +76,11 @@ export class Ra3HoverProvider implements vscode.HoverProvider {
     return new vscode.Hover(md);
   }
 
-  private attributeHover(elementType: string | null, attrName: string): vscode.Hover | null {
+  private attributeHover(
+    el: { name: string },
+    elementType: string | null,
+    attrName: string,
+  ): vscode.Hover | null {
     const attrs = model.attributesOfType(elementType);
     const attr = attrs.find((a) => a.name === attrName);
     const md = new vscode.MarkdownString();
@@ -78,6 +90,12 @@ export class Ra3HoverProvider implements vscode.HoverProvider {
         md.appendMarkdown(`Namespace/instance attribute.`);
         return new vscode.Hover(md);
       }
+      if (!model.isXsdElementName(el.name)) {
+        md.appendMarkdown(
+          `XInclude attribute (W3C XInclude namespace) — not part of the RA3 XSD model.`,
+        );
+        return new vscode.Hover(md);
+      }
       md.appendMarkdown("Unknown attribute for this element.");
       return new vscode.Hover(md);
     }
@@ -117,7 +135,10 @@ export class Ra3HoverProvider implements vscode.HoverProvider {
     }
 
     // Include source.
-    if (el.name === "Include" && attrName === "source") {
+    if (
+      (el.name === "Include" && attrName === "source") ||
+      (el.name === "xi:include" && attrName === "href")
+    ) {
       const resolved = idx
         ? resolveSource(
             value,

src/features/semanticTokens.ts

@@ -0,0 +1,42 @@
+import * as vscode from "vscode";
+import { parseXml } from "../language/xmlParser";
+import { buildSemanticTokenRanges } from "../language/semanticTokens";
+
+const TOKEN_TYPES = ["type", "property", "string"] as const;
+
+export const RA3_SEMANTIC_TOKENS_LEGEND = new vscode.SemanticTokensLegend([
+  ...TOKEN_TYPES,
+]);
+
+/**
+ * Highlighting fallback for malformed XML.
+ *
+ * While the document is well-formed, the built-in TextMate XML grammar colors
+ * it as usual and this provider returns no tokens, so nothing changes. When
+ * parsing reports errors (e.g. an attribute value whose closing quote has not
+ * been typed yet), the TextMate structure is lost, and these semantic tokens
+ * keep element names, attribute names and values colored.
+ */
+export class Ra3SemanticTokensProvider
+  implements vscode.DocumentSemanticTokensProvider
+{
+  async provideDocumentSemanticTokens(
+    document: vscode.TextDocument,
+    _token: vscode.CancellationToken,
+  ): Promise<vscode.SemanticTokens> {
+    const text = document.getText();
+    const doc = parseXml(text);
+    if (doc.errors.length === 0) {
+      return new vscode.SemanticTokens(new Uint32Array(0));
+    }
+    const ranges = buildSemanticTokenRanges(doc, text);
+    const builder = new vscode.SemanticTokensBuilder(RA3_SEMANTIC_TOKENS_LEGEND);
+    for (const r of ranges) {
+      builder.push(
+        new vscode.Range(r.line, r.startChar, r.line, r.startChar + r.length),
+        r.tokenType,
+      );
+    }
+    return builder.build();
+  }
+}

src/indexer/caches.ts

@@ -0,0 +1,226 @@
+/**
+ * Caches shared by the indexer.
+ *
+ * The workspace owns one instance of each cache and passes them into every
+ * ModIndexer, so a rebuild (which creates a fresh indexer) does not re-read
+ * files whose stat (mtime/size) is unchanged. This is what makes indexing
+ * large art-asset corpora (Corona: ~3800 .w3x files, 2.6 GB) practical:
+ * after the first build, save-triggered rebuilds only re-scan files that
+ * actually changed.
+ */
+
+import { resolve } from "node:path";
+import type { IndexedFile, ParsedFile } from "./types";
+import type { IndexRecords } from "./records";
+import type { ResolveResult } from "./includeResolver";
+
+/** Case-insensitive absolute path key. */
+export function normKey(path: string): string {
+  return resolve(path).toLowerCase();
+}
+
+/**
+ * LRU cache for fully parsed XML documents.
+ *
+ * Parse trees of huge mods can be memory-heavy (~17x the source text), so
+ * retention is bounded twice: by entry count (least-recently-used eviction)
+ * and by a total element budget (largest trees are evicted first). Evicted
+ * entries are re-read from disk on demand.
+ */
+export class DocumentCache {
+  private map = new Map<string, ParsedFile>();
+  private totalElements = 0;
+
+  constructor(
+    private capacity = 64,
+    private elementBudget = 2_000_000,
+  ) {}
+
+  get(path: string): ParsedFile | undefined {
+    const key = normKey(path);
+    const hit = this.map.get(key);
+    if (!hit) return undefined;
+    this.map.delete(key);
+    this.map.set(key, hit);
+    return hit;
+  }
+
+  /** Number of cached documents (for diagnostics). */
+  get size(): number {
+    return this.map.size;
+  }
+
+  /** Total elements held by cached parse trees (for diagnostics). */
+  get elements(): number {
+    return this.totalElements;
+  }
+
+  set(parsed: ParsedFile): void {
+    const key = normKey(parsed.file.path);
+    const prev = this.map.get(key);
+    if (prev) this.totalElements -= elementCount(prev);
+    this.map.delete(key);
+    this.map.set(key, parsed);
+    this.totalElements += elementCount(parsed);
+    this.evict();
+  }
+
+  invalidate(path: string): void {
+    const key = normKey(path);
+    const prev = this.map.get(key);
+    if (prev) this.totalElements -= elementCount(prev);
+    this.map.delete(key);
+  }
+
+  clear(): void {
+    this.map.clear();
+    this.totalElements = 0;
+  }
+
+  private evict(): void {
+    while (this.map.size > this.capacity || this.totalElements > this.elementBudget) {
+      if (this.map.size === 0) break;
+      if (this.map.size > this.capacity) {
+        // Over capacity: drop the least recently used entry.
+        const oldest = this.map.keys().next().value;
+        if (oldest === undefined) break;
+        this.remove(oldest);
+      } else {
+        // Over the element budget: drop the largest tree, which frees the
+        // most memory per eviction.
+        let largestKey: string | undefined;
+        let largest = -1;
+        for (const [key, value] of this.map) {
+          const n = elementCount(value);
+          if (n > largest) {
+            largest = n;
+            largestKey = key;
+          }
+        }
+        if (largestKey === undefined || largest <= 0) break;
+        this.remove(largestKey);
+      }
+    }
+  }
+
+  private remove(key: string): void {
+    const prev = this.map.get(key);
+    if (prev) this.totalElements -= elementCount(prev);
+    this.map.delete(key);
+  }
+}
+
+function elementCount(parsed: ParsedFile): number {
+  return parsed.parse?.elements.length ?? 0;
+}
+
+export interface IndexRecordsCacheEntry {
+  stat: IndexedFile["stat"];
+  records: IndexRecords;
+  /** "shallow" for art-asset scans (.w3x), "full" for parsed XML. */
+  kind: "shallow" | "full";
+}
+
+/**
+ * Cache for per-file index records (top-level assets, defines, includes,
+ * xi:include targets). Records are tiny compared to DOM trees or line maps of
+ * multi-megabyte model files, so the capacity comfortably covers a whole mod
+ * (Corona: ~9k files) and rebuilds never re-read unchanged files.
+ */
+export class IndexRecordsCache {
+  private map = new Map<string, IndexRecordsCacheEntry>();
+
+  constructor(private capacity = 16384) {}
+
+  get(path: string): IndexRecordsCacheEntry | undefined {
+    const key = normKey(path);
+    const hit = this.map.get(key);
+    if (!hit) return undefined;
+    this.map.delete(key);
+    this.map.set(key, hit);
+    return hit;
+  }
+
+  /** Number of cached record sets (for diagnostics). */
+  get size(): number {
+    return this.map.size;
+  }
+
+  set(path: string, entry: IndexRecordsCacheEntry): void {
+    const key = normKey(path);
+    this.map.delete(key);
+    this.map.set(key, entry);
+    if (this.map.size > this.capacity) {
+      const oldest = this.map.keys().next().value;
+      if (oldest !== undefined) this.map.delete(oldest);
+    }
+  }
+
+  invalidate(path: string): void {
+    this.map.delete(normKey(path));
+  }
+
+  clear(): void {
+    this.map.clear();
+  }
+}
+
+/**
+ * Cache for Include/@source (and xi:include/@href) resolution results.
+ *
+ * Resolving a source performs synchronous statSync existence checks against
+ * every search base; a Corona build issues ~110k of them (tens of seconds on
+ * a mechanical drive). Content edits never change *existence*, so this cache
+ * survives content rebuilds and is only cleared when files are created or
+ * deleted (or on a forced reindex).
+ */
+export class IncludeResolveCache {
+  private map = new Map<string, ResolveResult>();
+  private manifestMap = new Map<string, string | null>();
+
+  constructor(private capacity = 262144) {}
+
+  get(key: string): ResolveResult | undefined {
+    const hit = this.map.get(key);
+    if (!hit) return undefined;
+    this.map.delete(key);
+    this.map.set(key, hit);
+    return hit;
+  }
+
+  set(key: string, result: ResolveResult): void {
+    this.map.delete(key);
+    this.map.set(key, result);
+    if (this.map.size > this.capacity) {
+      const oldest = this.map.keys().next().value;
+      if (oldest !== undefined) this.map.delete(oldest);
+    }
+  }
+
+  getManifest(key: string): string | null | undefined {
+    const hit = this.manifestMap.get(key);
+    if (hit === undefined) return undefined;
+    this.manifestMap.delete(key);
+    this.manifestMap.set(key, hit);
+    return hit;
+  }
+
+  setManifest(key: string, path: string | null): void {
+    this.manifestMap.delete(key);
+    this.manifestMap.set(key, path);
+    if (this.manifestMap.size > this.capacity) {
+      const oldest = this.manifestMap.keys().next().value;
+      if (oldest !== undefined) this.manifestMap.delete(oldest);
+    }
+  }
+
+  clear(): void {
+    this.map.clear();
+    this.manifestMap.clear();
+  }
+
+  /** Number of cached resolutions (for diagnostics). */
+  get size(): number {
+    return this.map.size;
+  }
+}

src/indexer/fileScanner.ts

@@ -1,7 +1,18 @@
 import { readdir, stat } from "node:fs/promises";
-import { join, relative, resolve } from "node:path";
+import { extname, join, relative, resolve } from "node:path";
 import type { FileWalker, SourceCandidate } from "./types";
 
+/**
+ * Extensions offered as Include/@source candidates in DATA directories:
+ * regular XML plus art-asset XML (.w3x) that mods include via W3X.xml hubs.
+ * ART/AUDIO directories already list every file.
+ */
+const DATA_CANDIDATE_EXTENSIONS = new Set([".xml", ".w3x"]);
+
+function isDataCandidate(path: string): boolean {
+  return DATA_CANDIDATE_EXTENSIONS.has(extname(path).toLowerCase());
+}
+
 /**
  * Recursive file list walker with a simple directory-mtime cache. Used to
  * enumerate candidate files for Include/@source completion. Directory scans
@@ -76,29 +87,32 @@ export async function collectSourceCandidates(
     out.push({ source, path: resolve(path), prefix, baseDir: resolve(baseDir) });
   };
 
-  for (const dir of dataDirs) {
+  // List directories in parallel; the walker caches each directory by its
-    const files = await walker.listFiles(dir);
+  // mtime, so rebuilds after the first are still cheap.
+  const [dataLists, artLists, audioLists] = await Promise.all([
+    Promise.all(dataDirs.map(async (dir) => ({ dir, files: await walker.listFiles(dir) }))),
+    Promise.all(artDirs.map(async (dir) => ({ dir, files: await walker.listFiles(dir) }))),
+    Promise.all(audioDirs.map(async (dir) => ({ dir, files: await walker.listFiles(dir) }))),
+  ]);
+
+  for (const { dir, files } of dataLists) {
     for (const f of files) {
-      if (!f.toLowerCase().endsWith(".xml")) continue;
+      if (!isDataCandidate(f)) continue;
       add(f, dir, "DATA");
     }
     if (samePath(dir, projectDataDir)) {
       // Also offer project-relative paths (what Mod.xml itself uses).
       for (const f of files) {
-        if (f.toLowerCase().endsWith(".xml")) add(f, projectDataDir, null);
+        if (isDataCandidate(f)) add(f, projectDataDir, null);
       }
     }
   }
 
-  for (const dir of artDirs) {
+  for (const { dir, files } of artLists) {
-    for (const f of await walker.listFiles(dir)) {
+    for (const f of files) add(f, dir, "ART");
-      add(f, dir, "ART");
-    }
-  }
-  for (const dir of audioDirs) {
-    for (const f of await walker.listFiles(dir)) {
-      add(f, dir, "AUDIO");
   }
+  for (const { dir, files } of audioLists) {
+    for (const f of files) add(f, dir, "AUDIO");
   }
 
   return out;

src/indexer/indexer.ts

@@ -11,13 +11,21 @@
  * outside the extension.
  */
 
-import { readFile, readdir, stat } from "node:fs/promises";
+import { open, readFile, readdir, stat } from "node:fs/promises";
+import type { Stats } from "node:fs";
 import { basename, dirname, extname, join, resolve } from "node:path";
-import { LineMap, parseXml, type XmlDocument, type XmlElement } from "../language/xmlParser";
+import {
+  LineMap,
+  parseXml,
+  stripBom,
+  type XmlDocument,
+  type XmlElement,
+} from "../language/xmlParser";
 import {
   buildSearchPaths,
   manifestPathForReference,
   resolveSource,
+  type ResolveResult,
   type SearchPaths,
 } from "./includeResolver";
 import {
@@ -28,6 +36,14 @@ import {
 } from "./manifestParser";
 import { canonicalTypeName } from "../model/schemaModel";
 import { collectSourceCandidates } from "./fileScanner";
+import { DocumentCache, IncludeResolveCache, IndexRecordsCache, normKey } from "./caches";
+import type { IndexRecordsCacheEntry } from "./caches";
+import { scanXmlShallow } from "./shallowScan";
+import {
+  extractIndexRecords,
+  recordsFromShallow,
+  type IndexRecordXi,
+} from "./records";
 import type {
   AssetDef,
   DefineDef,
@@ -42,54 +58,35 @@ import type {
 const MAX_DEPTH = 300;
 /** Files above this size are never parsed (safety against binary blobs). */
 const MAX_PARSE_BYTES = 4 * 1024 * 1024;
-/** Only these extensions are treated as XML documents. */
-const XML_EXTENSIONS = new Set([".xml", ".manifestxml"]);
-
-function normKey(path: string): string {
-  return resolve(path).toLowerCase();
-}
-
 /**
- * LRU cache for parsed documents. The parse trees of huge mods can be
+ * Fully parsed XML documents. `.xml` / `.manifestxml` files are small enough
- * memory-heavy, so only a bounded number of recent documents is retained;
+ * that a full DOM is affordable.
- * evicted entries are re-read from disk on demand.
  */
-export class DocumentCache {
+const FULL_XML_EXTENSIONS = new Set([".xml", ".manifestxml"]);
-  private map = new Map<string, ParsedFile>();
+/**
+ * XML documents whose top-level structure is all the index needs (art-asset
+ * files exported by modeling tools, e.g. .w3x). They are shallow-scanned so
+ * multi-megabyte vertex/triangle payloads never become a DOM.
+ */
+const SHALLOW_XML_EXTENSIONS = new Set([".w3x"]);
+/** Bytes peeked when deciding whether an unknown extension is XML text. */
+const SNIFF_BYTES = 512;
 
-  constructor(private capacity = 64) {}
+type XmlMode = "full" | "shallow" | "binary";
-
-  get(path: string): ParsedFile | undefined {
-    const key = normKey(path);
-    const hit = this.map.get(key);
-    if (!hit) return undefined;
-    this.map.delete(key);
-    this.map.set(key, hit);
-    return hit;
-  }
-
-  set(parsed: ParsedFile): void {
-    const key = normKey(parsed.file.path);
-    this.map.delete(key);
-    this.map.set(key, parsed);
-    if (this.map.size > this.capacity) {
-      const oldest = this.map.keys().next().value;
-      if (oldest !== undefined) this.map.delete(oldest);
-    }
-  }
-
-  invalidate(path: string): void {
-    this.map.delete(normKey(path));
-  }
-
-  clear(): void {
-    this.map.clear();
-  }
-}
 
 export class ModIndexer {
   private searchPaths: SearchPaths;
-  private docs = new DocumentCache();
+  private docs: DocumentCache;
+  private recordsCache: IndexRecordsCache;
+  private resolveCache: IncludeResolveCache;
+  private scanCounters = {
+    shallowScannedFiles: 0,
+    shallowCacheHits: 0,
+    recordsCacheHits: 0,
+    resolveCacheHits: 0,
+    resolveCalls: 0,
+  };
+  private phase = { candidatesMs: 0, walkMs: 0 };
   private assets = new Map<string, Map<string, AssetDef[]>>();
   private assetsById = new Map<string, AssetDef[]>();
   private defines = new Map<string, DefineDef[]>();
@@ -106,42 +103,221 @@ export class ModIndexer {
     this.searchPaths = buildSearchPaths(opts.sdkDir, opts.projectDir, {
       DATA: opts.additionalDataSearchPaths,
     });
+    // Caches may be owned by the workspace so they survive rebuilds.
+    this.docs = opts.documentCache ?? new DocumentCache();
+    this.recordsCache = opts.recordsCache ?? new IndexRecordsCache();
+    this.resolveCache = opts.resolveCache ?? new IncludeResolveCache();
   }
 
-  /** Re-reads and caches a document; null when unreadable. */
+  /**
+   * Returns a document for indexing/navigation:
+   * - `.xml` / `.manifestxml` files are fully parsed (bounded by
+   *   MAX_PARSE_BYTES);
+   * - `.w3x` (and unknown-extension files whose content looks like XML) are
+   *   shallow-scanned, so huge model files never become a DOM;
+   * - everything else is registered as a file but never parsed.
+   *
+   * Cached entries are reused when the file stat is unchanged, which lets a
+   * workspace-owned cache survive rebuilds.
+   */
   async readDocument(path: string): Promise<ParsedFile | null> {
-    const hit = this.docs.get(path);
+    const key = normKey(path);
-    if (hit) return hit;
+    const trust =
+      this.opts.trustUnchanged === true && !this.opts.changedFiles?.has(key);
+
+    // Trusted fast path: a file that the watcher has not reported as changed
+    // is reused without any stat / content sniff / read at all. This is what
+    // makes save-triggered rebuilds cheap on huge corpora (Corona: ~9k files,
+    // ~2.6 GB of art assets on a mechanical drive).
+    if (trust) {
+      const rec = this.recordsCache.get(key);
+      if (rec) return this.recordsParsed(path, rec);
+      const cached = this.docs.get(key);
+      if (cached) {
+        this.files.set(key, cached.file);
+        return cached;
+      }
+    }
+
+    // Untrusted / cache miss: verify the stat against caches, then read.
     try {
-      const [st, text] = await Promise.all([stat(path), readFile(path, "utf8")]);
+      const st = await stat(path);
-      if (st.size > MAX_PARSE_BYTES) {
+      const rec = this.recordsCache.get(key);
+      if (rec?.stat && rec.stat.mtimeMs === st.mtimeMs && rec.stat.size === st.size) {
+        return this.recordsParsed(path, rec);
+      }
+      const hit = this.docs.get(key);
+      if (
+        hit?.file.stat &&
+        hit.file.stat.mtimeMs === st.mtimeMs &&
+        hit.file.stat.size === st.size
+      ) {
+        this.files.set(key, hit.file);
+        return hit;
+      }
+      const mode = await this.detectXmlMode(path);
+      if (mode === "shallow") return this.scanShallow(path, st);
+      if (mode === "binary") {
         const file: IndexedFile = { path: resolve(path), stat: { mtimeMs: st.mtimeMs, size: st.size } };
-        const parsed: ParsedFile = { file, parse: null, lineMap: null };
+        const parsed: ParsedFile = { file, parse: null, records: null, lineMap: null };
         this.docs.set(parsed);
-        this.files.set(normKey(parsed.file.path), file);
+        this.files.set(key, file);
         return parsed;
       }
+      if (st.size > MAX_PARSE_BYTES) {
+        const file: IndexedFile = { path: resolve(path), stat: { mtimeMs: st.mtimeMs, size: st.size } };
+        const parsed: ParsedFile = { file, parse: null, records: null, lineMap: null };
+        this.docs.set(parsed);
+        this.files.set(key, file);
+        return parsed;
+      }
+      const text = stripBom(await readFile(path, "utf8"));
+      const lineMap = new LineMap(text);
       const parse = parseXml(text);
+      const records = extractIndexRecords(parse, lineMap);
       const parsed: ParsedFile = {
         file: { path: resolve(path), stat: { mtimeMs: st.mtimeMs, size: st.size } },
         parse,
-        lineMap: new LineMap(text),
+        records,
+        lineMap,
       };
       this.docs.set(parsed);
-      this.files.set(normKey(parsed.file.path), parsed.file);
+      this.recordsCache.set(key, { stat: parsed.file.stat, records, kind: "full" });
+      this.files.set(key, parsed.file);
       return parsed;
     } catch {
       const parsed: ParsedFile = {
         file: { path: resolve(path), stat: null },
         parse: null,
+        records: null,
         lineMap: null,
       };
       this.docs.set(parsed);
-      this.files.set(normKey(parsed.file.path), parsed.file);
+      this.files.set(key, parsed.file);
       return parsed;
     }
   }
 
+  /**
+   * Shallow-scans a large art-asset XML document (no DOM built) and caches
+   * its compact index records. The transient LineMap used to compute record
+   * lines is discarded, so the cache never retains megabytes of line offsets
+   * for model files (Corona w3x alone would otherwise keep ~700 MB).
+   */
+  private async scanShallow(path: string, st: Stats): Promise<ParsedFile | null> {
+    const key = normKey(path);
+    try {
+      const text = stripBom(await readFile(path, "utf8"));
+      const lineMap = new LineMap(text);
+      const records = recordsFromShallow(scanXmlShallow(text), lineMap);
+      const parsed: ParsedFile = {
+        file: { path: resolve(path), stat: { mtimeMs: st.mtimeMs, size: st.size } },
+        parse: null,
+        records,
+        lineMap: null,
+      };
+      this.recordsCache.set(key, { stat: parsed.file.stat, records, kind: "shallow" });
+      this.files.set(key, parsed.file);
+      this.scanCounters.shallowScannedFiles++;
+      return parsed;
+    } catch {
+      return null;
+    }
+  }
+
+  /** Builds a lean ParsedFile from cached index records (no DOM / line map). */
+  private recordsParsed(path: string, entry: IndexRecordsCacheEntry): ParsedFile {
+    if (entry.kind === "shallow") this.scanCounters.shallowCacheHits++;
+    else this.scanCounters.recordsCacheHits++;
+    const file: IndexedFile = { path: resolve(path), stat: entry.stat };
+    this.files.set(normKey(path), file);
+    return { file, parse: null, records: entry.records, lineMap: null };
+  }
+
+  /**
+   * Reads a document and guarantees a DOM parse tree. Used only for
+   * root-level <xi:include> xpointer selection (rare), where the target's
+   * container children are needed.
+   */
+  private async readDom(path: string): Promise<ParsedFile | null> {
+    const key = normKey(path);
+    const cached = this.docs.get(key);
+    if (cached?.parse?.root) {
+      this.files.set(key, cached.file);
+      return cached;
+    }
+    try {
+      const st = await stat(path);
+      const hit = this.docs.get(key);
+      if (
+        hit?.parse?.root &&
+        hit.file.stat &&
+        hit.file.stat.mtimeMs === st.mtimeMs &&
+        hit.file.stat.size === st.size
+      ) {
+        this.files.set(key, hit.file);
+        return hit;
+      }
+      if (st.size > MAX_PARSE_BYTES) return null;
+      const text = stripBom(await readFile(path, "utf8"));
+      const lineMap = new LineMap(text);
+      const parse = parseXml(text);
+      const records = extractIndexRecords(parse, lineMap);
+      const parsed: ParsedFile = {
+        file: { path: resolve(path), stat: { mtimeMs: st.mtimeMs, size: st.size } },
+        parse,
+        records,
+        lineMap,
+      };
+      this.docs.set(parsed);
+      this.recordsCache.set(key, { stat: parsed.file.stat, records, kind: "full" });
+      this.files.set(key, parsed.file);
+      return parsed;
+    } catch {
+      return null;
+    }
+  }
+
+  /** Decides how a resolved include target should be consumed. */
+  private async detectXmlMode(path: string): Promise<XmlMode> {
+    const ext = extname(path).toLowerCase();
+    if (FULL_XML_EXTENSIONS.has(ext)) return "full";
+    if (SHALLOW_XML_EXTENSIONS.has(ext)) return "shallow";
+    return (await looksLikeXml(path)) ? "shallow" : "binary";
+  }
+
+  /**
+   * Resolves an include source through the cross-rebuild cache. Existence
+   * does not change on content edits, so trusted rebuilds pay zero statSync
+   * for include resolution (Corona does ~110k checks per build otherwise).
+   */
+  private resolveCached(source: string, currentDir: string | null): ResolveResult {
+    const key = `${normKey(currentDir ?? "")}\u0000${source}`;
+    const hit = this.resolveCache.get(key);
+    if (hit) {
+      this.scanCounters.resolveCacheHits++;
+      return hit;
+    }
+    this.scanCounters.resolveCalls++;
+    const result = resolveSource(source, currentDir, this.searchPaths);
+    this.resolveCache.set(key, result);
+    return result;
+  }
+
+  /** Cached manifest lookup for `reference` includes. */
+  private manifestPathCached(source: string): string | null {
+    const key = source.toLowerCase();
+    const hit = this.resolveCache.getManifest(key);
+    if (hit !== undefined) {
+      this.scanCounters.resolveCacheHits++;
+      return hit;
+    }
+    this.scanCounters.resolveCalls++;
+    const path = manifestPathForReference(source, this.opts.builtmodsDirs);
+    this.resolveCache.setManifest(key, path);
+    return path;
+  }
+
   /** Returns the cached parse if present (does not read from disk). */
   cachedDocument(path: string): ParsedFile | undefined {
     return this.docs.get(path);
@@ -155,6 +331,7 @@ export class ModIndexer {
       : null;
 
     // ── Streams ──
+    const walkStart = Date.now();
     const staticEntry = projectData ? join(projectData, "Mod.xml") : null;
     if (staticEntry) {
       const stream: StreamInfo = { name: "static", entry: staticEntry, files: new Set() };
@@ -183,8 +360,10 @@ export class ModIndexer {
         await this.walk(entry, "all", stream, 0);
       }
     }
+    this.phase.walkMs = Date.now() - walkStart;
 
     // ── Source completion candidates ──
+    const candidatesStart = Date.now();
     const dataDirs = [
       projectData ?? join(this.opts.projectDir, "Data"),
       join(this.opts.sdkDir, "SageXml"),
@@ -229,6 +408,7 @@ export class ModIndexer {
       ...sdkRootCandidates,
       ...this.sourceCandidates,
     ]);
+    this.phase.candidatesMs = Date.now() - candidatesStart;
 
     const manifestAssetCount = [...this.manifests.values()].reduce(
       (sum, m) => sum + m.assets.length,
@@ -253,6 +433,13 @@ export class ModIndexer {
         parsedFiles: [...this.files.values()].filter(
           (f) => f.stat != null && f.stat.size <= MAX_PARSE_BYTES,
         ).length,
+        shallowScannedFiles: this.scanCounters.shallowScannedFiles,
+        shallowCacheHits: this.scanCounters.shallowCacheHits,
+        recordsCacheHits: this.scanCounters.recordsCacheHits,
+        resolveCacheHits: this.scanCounters.resolveCacheHits,
+        resolveCalls: this.scanCounters.resolveCalls,
+        candidatesMs: this.phase.candidatesMs,
+        walkMs: this.phase.walkMs,
         assetCount: [...this.assets.values()].reduce((sum, byId) => sum + byId.size, 0),
         defineCount: this.defines.size,
         manifestFiles: this.manifests.size,
@@ -294,142 +481,135 @@ export class ModIndexer {
 
     stream.files.add(key);
 
-    // Binary assets (w3x/dds/...) are referenced but never parsed.
+    // readDocument returns compact index records for every indexable XML
-    if (!isXmlPath(path)) return;
+    // document (full parse or shallow scan), or a bare file registration
-
+    // for binary / unparseable targets.
     const parsed = await this.readDocument(path);
-    if (!parsed?.parse?.root) return;
+    if (!parsed) return;
-    const root = parsed.parse.root;
+    if (parsed.records) {
-
+      await this.applyRecords(parsed, stream, depth, mode === "instance");
-    for (const child of root.children) {
+      return;
-      const local = localName(child.name);
-      if (local === "Tags" || local === "Includes" || local === "Defines") continue;
-      if (local === "include") {
-        await this.handleXiInclude(child, parsed, stream, depth);
-        continue;
     }
-      const idAttr = child.attrs.find((a) => a.name === "id");
+  }
-      if (idAttr) {
+
+  /**
+   * Applies a document's compact index records: top-level assets, defines,
+   * <Includes>, nested <xi:include> and root-level <xi:include> targets.
+   * Works identically for fully parsed XML and shallow-scanned art assets.
+   */
+  private async applyRecords(
+    parsed: ParsedFile,
+    stream: StreamInfo,
+    depth: number,
+    viaInstance: boolean,
+  ): Promise<void> {
+    const records = parsed.records;
+    if (!records) return;
+    const file = parsed.file.path;
+    const origin = this.originOf(file);
+
+    for (const asset of records.assets) {
       this.addAsset({
-          type: local,
+        type: asset.type,
-          id: idAttr.value,
+        id: asset.id,
-          file: parsed.file.path,
+        file,
-          line: lineOf(parsed, idAttr.valueStart),
+        line: asset.line,
-          origin: this.originOf(parsed.file.path),
+        origin,
         stream: stream.name,
-          viaInstance: mode === "instance",
+        viaInstance,
       });
     }
-    }
 
-    for (const child of root.children) {
+    for (const define of records.defines) {
-      if (localName(child.name) !== "Defines") continue;
-      for (const define of child.children) {
-        if (localName(define.name) !== "Define") continue;
-        const name = define.attrs.find((a) => a.name === "name")?.value;
-        const value = define.attrs.find((a) => a.name === "value")?.value;
-        if (!name) continue;
       const entry: DefineDef = {
-          name,
+        name: define.name,
-          value: value ?? "",
+        value: define.value,
-          file: parsed.file.path,
+        file,
-          line: lineOf(parsed, define.start),
+        line: define.line,
-          origin: this.originOf(parsed.file.path),
+        origin,
       };
-        const arr = this.defines.get(name.toLowerCase());
+      const arr = this.defines.get(define.name.toLowerCase());
       if (arr) arr.push(entry);
-        else this.defines.set(name.toLowerCase(), [entry]);
+      else this.defines.set(define.name.toLowerCase(), [entry]);
-      }
     }
 
-    const includesElem = root.children.find((c) => localName(c.name) === "Includes");
+    for (const inc of records.includes) {
-    if (includesElem) {
+      const resolved = this.resolveCached(inc.source, dirname(file));
-      for (const inc of includesElem.children) {
-        if (localName(inc.name) !== "Include") continue;
-        const type = inc.attrs.find((a) => a.name === "type")?.value;
-        const source = inc.attrs.find((a) => a.name === "source")?.value;
-        if (!source) continue;
-        const resolved = resolveSource(source, dirname(parsed.file.path), this.searchPaths);
       if (!resolved.path) {
         this.diagnostics.push({
-            file: parsed.file.path,
+          file,
-            line: lineOf(parsed, inc.start),
+          line: inc.line,
-            message: `Include target not found: ${source}`,
+          message: `Include target not found: ${inc.source}`,
           severity: "warning",
           code: "include-not-found",
         });
         continue;
       }
-        if (type === "all" || type === "instance") {
+      if (inc.type === "all" || inc.type === "instance") {
-          await this.walk(resolved.path, type === "all" ? "all" : "instance", stream, depth + 1);
+        await this.walk(
-        } else if (type === "reference") {
+          resolved.path,
-          const manifestPath = manifestPathForReference(source, this.opts.builtmodsDirs);
+          inc.type === "all" ? "all" : "instance",
+          stream,
+          depth + 1,
+        );
+      } else if (inc.type === "reference") {
+        const manifestPath = this.manifestPathCached(inc.source);
         if (manifestPath) {
           const loaded = await this.loadManifest(manifestPath, stream.name);
-            if (!loaded && isXmlPath(resolved.path)) {
+          if (!loaded) await this.walk(resolved.path, "instance", stream, depth + 1);
-              // The manifest could not be parsed (missing/invalid): fall back
+        } else {
-              // to the placeholder XML so its content is still available.
           await this.walk(resolved.path, "instance", stream, depth + 1);
         }
-          } else if (isXmlPath(resolved.path)) {
-            // reference to a real XML file: treat its assets as available
-            await this.walk(resolved.path, "instance", stream, depth + 1);
-          }
-        }
       }
     }
 
-    // Nested <xi:include> anywhere in the tree (not just under the root):
+    for (const xi of records.nestedXiIncludes) {
-    // the target content is inlined into the parent element. We make the
+      const resolved = this.resolveCached(xi.href, dirname(file));
-    // target file available and surface missing targets instead of ignoring
-    // them silently.
-    for (const el of parsed.parse.elements) {
-      if (localName(el.name) !== "include") continue;
-      if (el.parent === root) continue; // already handled in the loop above
-      const href = el.attrs.find((a) => a.name === "href")?.value;
-      if (!href) continue;
-      const resolved = resolveSource(href, dirname(parsed.file.path), this.searchPaths);
       if (!resolved.path) {
         this.diagnostics.push({
-          file: parsed.file.path,
+          file,
-          line: lineOf(parsed, el.start),
+          line: xi.line,
-          message: `xi:include target not found: ${href}`,
+          message: `xi:include target not found: ${xi.href}`,
           severity: "warning",
           code: "include-not-found",
         });
         continue;
       }
       stream.files.add(normKey(resolved.path));
-      if (isXmlPath(resolved.path)) {
       await this.walk(resolved.path, "all", stream, depth + 1);
     }
+
+    for (const xi of records.rootXiIncludes) {
+      await this.handleRootXiInclude(xi, file, stream, depth);
     }
   }
 
-  private async handleXiInclude(
+  /**
-    xi: XmlElement,
+   * Root-level <xi:include> with an xpointer selects a named container's
-    parent: ParsedFile,
+   * children in the target document, so the target's DOM is needed. These
+   * targets are rare, so they are parsed on demand (the tree is also cached).
+   */
+  private async handleRootXiInclude(
+    xi: IndexRecordXi,
+    parentFile: string,
     stream: StreamInfo,
     depth: number,
   ): Promise<void> {
-    const href = xi.attrs.find((a) => a.name === "href")?.value;
+    const resolved = this.resolveCached(xi.href, dirname(parentFile));
-    if (!href) return;
-    const resolved = resolveSource(href, dirname(parent.file.path), this.searchPaths);
     if (!resolved.path) {
       this.diagnostics.push({
-        file: parent.file.path,
+        file: parentFile,
-        line: lineOf(parent, xi.start),
+        line: xi.line,
-        message: `xi:include target not found: ${href}`,
+        message: `xi:include target not found: ${xi.href}`,
         severity: "warning",
         code: "include-not-found",
       });
       return;
     }
-    if (!isXmlPath(resolved.path)) return;
-    const target = await this.readDocument(resolved.path);
-    if (!target?.parse?.root) return;
 
-    const xpointer = xi.attrs.find((a) => a.name === "xpointer")?.value ?? "";
+    const target = await this.readDom(resolved.path);
+    if (target?.parse?.root) {
+      const xpointer = xi.xpointer ?? "";
       let candidates: XmlElement[];
       if (xpointer) {
         const container = findXPointerContainer(target.parse, xpointer);
@@ -452,8 +632,9 @@ export class ModIndexer {
           });
         }
       }
-    stream.files.add(normKey(target.file.path));
+    }
-    await this.walk(target.file.path, "all", stream, depth + 1);
+    stream.files.add(normKey(resolved.path));
+    await this.walk(resolved.path, "all", stream, depth + 1);
   }
 
   // ── Manifest loading ──────────────────────────────────────────────
@@ -551,9 +732,32 @@ function lineOf(parsed: ParsedFile, offset: number): number {
   return parsed.lineMap.positionAt(offset).line + 1;
 }
 
-function isXmlPath(path: string): boolean {
+/**
-  const ext = extname(path).toLowerCase();
+ * Content sniffing for include targets with unknown extensions (e.g. art
-  return XML_EXTENSIONS.has(ext);
+ * formats beyond .w3x): a small header that starts with "<" after an
+ * optional UTF-8 BOM / whitespace and contains no NUL bytes is treated as
+ * XML text; anything else is a binary asset (registered, never parsed).
+ */
+async function looksLikeXml(path: string): Promise<boolean> {
+  try {
+    const fh = await open(path, "r");
+    try {
+      const buf = Buffer.alloc(SNIFF_BYTES);
+      const { bytesRead } = await fh.read(buf, 0, SNIFF_BYTES, 0);
+      const head = buf.subarray(0, bytesRead);
+      if (head.includes(0)) return false;
+      let i = 0;
+      if (head[0] === 0xef && head[1] === 0xbb && head[2] === 0xbf) i = 3;
+      while (i < head.length && (head[i] === 0x20 || head[i] === 0x09 || head[i] === 0x0a || head[i] === 0x0d)) {
+        i++;
+      }
+      return i < head.length && head[i] === 0x3c; // "<"
+    } finally {
+      await fh.close();
+    }
+  } catch {
+    return false;
+  }
 }
 
 async function findCaseInsensitiveDir(dir: string): Promise<string | null> {

src/indexer/records.ts

@@ -0,0 +1,177 @@
+/**
+ * Compact per-file index records.
+ *
+ * A rebuild only needs each file's top-level assets, defines, includes and
+ * xi:include targets — not the DOM. Extracting these records at parse time
+ * and caching them across rebuilds lets Corona-scale rebuilds skip both the
+ * DOM and the file I/O for unchanged files, while keeping the DOM cache small
+ * for on-demand features (hover / navigation / outline).
+ *
+ * Pure TypeScript: no vscode dependency.
+ */
+
+import type { LineMap, XmlDocument } from "../language/xmlParser";
+import type { ShallowDocument } from "./shallowScan";
+
+export interface IndexRecordAsset {
+  /** Top-level element name, e.g. "W3DContainer". */
+  type: string;
+  id: string;
+  /** 1-based line of the id attribute value. */
+  line: number;
+}
+
+export interface IndexRecordDefine {
+  name: string;
+  value: string;
+  /** 1-based line of the <Define> element. */
+  line: number;
+}
+
+export interface IndexRecordInclude {
+  type: "all" | "instance" | "reference" | null;
+  source: string;
+  /** 1-based line of the <Include> element. */
+  line: number;
+}
+
+export interface IndexRecordXi {
+  href: string;
+  xpointer: string | null;
+  /** 1-based line of the <xi:include> element. */
+  line: number;
+}
+
+export interface IndexRecords {
+  assets: IndexRecordAsset[];
+  defines: IndexRecordDefine[];
+  includes: IndexRecordInclude[];
+  /** <xi:include> elements that are direct children of the root. */
+  rootXiIncludes: IndexRecordXi[];
+  /** <xi:include> elements nested anywhere else in the document. */
+  nestedXiIncludes: IndexRecordXi[];
+}
+
+const INCLUDE_TYPES = new Set(["all", "instance", "reference"]);
+
+function lineOf(lineMap: LineMap, offset: number): number {
+  return lineMap.positionAt(offset).line + 1;
+}
+
+function localName(tag: string): string {
+  const idx = tag.lastIndexOf(":");
+  return idx >= 0 ? tag.slice(idx + 1) : tag;
+}
+
+/**
+ * Extracts the index records of a fully parsed document. Mirrors the walk
+ * semantics of the indexer exactly: top-level assets (excluding
+ * Tags/Includes/Defines), $DEFINE constants, the top-level <Includes> block
+ * and root/nested <xi:include> elements.
+ */
+export function extractIndexRecords(parse: XmlDocument, lineMap: LineMap): IndexRecords {
+  const assets: IndexRecordAsset[] = [];
+  const defines: IndexRecordDefine[] = [];
+  const includes: IndexRecordInclude[] = [];
+  const rootXiIncludes: IndexRecordXi[] = [];
+  const nestedXiIncludes: IndexRecordXi[] = [];
+  const root = parse.root;
+  if (!root) return { assets, defines, includes, rootXiIncludes, nestedXiIncludes };
+
+  for (const child of root.children) {
+    const local = localName(child.name);
+    if (local === "Tags" || local === "Includes" || local === "Defines") continue;
+    if (local === "include") {
+      const href = child.attrs.find((a) => a.name === "href")?.value;
+      if (href) {
+        rootXiIncludes.push({
+          href,
+          xpointer: child.attrs.find((a) => a.name === "xpointer")?.value ?? null,
+          line: lineOf(lineMap, child.start),
+        });
+      }
+      continue;
+    }
+    const idAttr = child.attrs.find((a) => a.name === "id");
+    if (idAttr) {
+      assets.push({ type: local, id: idAttr.value, line: lineOf(lineMap, idAttr.valueStart) });
+    }
+  }
+
+  for (const child of root.children) {
+    if (localName(child.name) !== "Defines") continue;
+    for (const define of child.children) {
+      if (localName(define.name) !== "Define") continue;
+      const name = define.attrs.find((a) => a.name === "name")?.value;
+      if (!name) continue;
+      defines.push({
+        name,
+        value: define.attrs.find((a) => a.name === "value")?.value ?? "",
+        line: lineOf(lineMap, define.start),
+      });
+    }
+  }
+
+  const includesElem = root.children.find((c) => localName(c.name) === "Includes");
+  if (includesElem) {
+    for (const inc of includesElem.children) {
+      if (localName(inc.name) !== "Include") continue;
+      const source = inc.attrs.find((a) => a.name === "source")?.value;
+      if (!source) continue;
+      const type = inc.attrs.find((a) => a.name === "type")?.value;
+      includes.push({
+        type:
+          type && INCLUDE_TYPES.has(type)
+            ? (type as "all" | "instance" | "reference")
+            : null,
+        source,
+        line: lineOf(lineMap, inc.start),
+      });
+    }
+  }
+
+  for (const el of parse.elements) {
+    if (localName(el.name) !== "include") continue;
+    if (el.parent === root) continue; // already handled above
+    const href = el.attrs.find((a) => a.name === "href")?.value;
+    if (!href) continue;
+    nestedXiIncludes.push({
+      href,
+      xpointer: el.attrs.find((a) => a.name === "xpointer")?.value ?? null,
+      line: lineOf(lineMap, el.start),
+    });
+  }
+
+  return { assets, defines, includes, rootXiIncludes, nestedXiIncludes };
+}
+
+/** Converts a shallow scan (offsets) into index records (1-based lines). */
+export function recordsFromShallow(scan: ShallowDocument, lineMap: LineMap): IndexRecords {
+  return {
+    assets: scan.assets.map((a) => ({
+      type: a.name,
+      id: a.id,
+      line: lineOf(lineMap, a.idValueStart),
+    })),
+    defines: scan.defines.map((d) => ({
+      name: d.name,
+      value: d.value,
+      line: lineOf(lineMap, d.start),
+    })),
+    includes: scan.includes.map((i) => ({
+      type: i.type,
+      source: i.source,
+      line: lineOf(lineMap, i.start),
+    })),
+    rootXiIncludes: scan.rootXiIncludes.map((x) => ({
+      href: x.href,
+      xpointer: x.xpointer,
+      line: lineOf(lineMap, x.start),
+    })),
+    nestedXiIncludes: scan.nestedXiIncludes.map((x) => ({
+      href: x.href,
+      xpointer: x.xpointer,
+      line: lineOf(lineMap, x.start),
+    })),
+  };
+}

src/indexer/refs.ts

@@ -10,6 +10,37 @@ export interface ReferenceTarget {
   score: number;
 }
 
+/**
+ * True when an attribute is a "pipeline-local" reference that the global
+ * asset index cannot judge:
+ *
+ * - `id` attributes declare the element's own identity. When the attribute
+ *   has no refType (plain Poid / pipeline ids) or its refType is compatible
+ *   with the element's own type (e.g. ModuleData@id -> ModuleData), the
+ *   element itself is the definition site, so no global definition is
+ *   required. An `id` whose refType points at a *different* asset type
+ *   (e.g. RoadObject@id -> Road) is a real cross-asset reference and keeps
+ *   its reference semantics.
+ * - Poid-typed attributes (ModuleId, AutoResolveBody, SoundRef, ...) are
+ *   "pipeline object id" references that are resolved within the same
+ *   asset/subtree (modules, sub-objects, pivots, shader materials...),
+ *   never against the global asset index.
+ */
+export function isLocalReferenceAttribute(
+  typeName: string | null,
+  attrName: string,
+): boolean {
+  const attr = attributesOfType(typeName).find((a) => a.name === attrName);
+  if (!attr) return false;
+  const isId = attrName.toLowerCase() === "id";
+  if (isId) {
+    if (!attr.refType) return true;
+    if (typeName && isAssignableTo(typeName, attr.refType)) return true;
+    return false;
+  }
+  return attr.type === "Poid";
+}
+
 /**
  * True when an attribute is a typed reference: either the instance
  * inheritance attribute `inheritFrom`, or an XSD attribute whose simple type
@@ -27,7 +58,11 @@ export function isReferenceAttributeOfType(
 ): boolean {
   if (attrName.toLowerCase() === "inheritfrom") return true;
   const attr = attributesOfType(typeName).find((a) => a.name === attrName);
-  return attr != null && (attr.refType != null || attr.isRef);
+  if (attr == null || !(attr.refType != null || attr.isRef)) return false;
+  // Definitions (id) and pipeline-local references (Poid) are not references
+  // to global assets, so they never need a definition in the global index.
+  if (isLocalReferenceAttribute(typeName, attrName)) return false;
+  return true;
 }
 
 /**
@@ -73,6 +108,9 @@ export function resolveReferenceTargetsForType(
   } else {
     const attr = attributesOfType(typeName).find((a) => a.name === attrName);
     if (!attr || (!attr.refType && !attr.isRef)) return [];
+    // id definitions and Poid pipeline-local references are never resolved
+    // against the global asset index.
+    if (isLocalReferenceAttribute(typeName, attrName)) return [];
     refType = attr.refType;
   }
 

src/indexer/shallowScan.ts

@@ -0,0 +1,274 @@
+/**
+ * Shallow XML scanner for large art-asset documents (e.g. .w3x).
+ *
+ * Full XML parsing builds an element tree whose memory footprint is roughly
+ * 17x the source text (measured on Corona model files). Model exports like
+ * <W3DMesh> contain hundreds of thousands of tiny numeric elements
+ * (Vertices/V, Triangles/T, ...) that the extension never needs: the index
+ * only consumes top-level asset definitions (name + id), top-level
+ * <Includes>, nested <xi:include> targets and <Defines> constants.
+ *
+ * This scanner performs a single linear pass without building child nodes,
+ * so multi-megabyte model files can be indexed in linear time with ~zero
+ * retained memory.
+ *
+ * Pure TypeScript: no vscode dependency, reusable outside the extension.
+ */
+
+export interface ShallowAssetRecord {
+  /** Top-level element name, e.g. "W3DContainer". */
+  name: string;
+  /** Value of the id attribute. */
+  id: string;
+  /** Offset of the first id value character. */
+  idValueStart: number;
+  /** Offset one past the last id value character. */
+  idValueEnd: number;
+  /** Offset of the element's "<". */
+  start: number;
+  /** Offset one past the ">" of the start tag. */
+  startTagEnd: number;
+}
+
+export interface ShallowIncludeRecord {
+  /** "all" | "instance" | "reference", or null when absent/unknown. */
+  type: "all" | "instance" | "reference" | null;
+  source: string;
+  /** Offset of the <Include> element. */
+  start: number;
+}
+
+export interface ShallowXiIncludeRecord {
+  href: string;
+  xpointer: string | null;
+  /** Offset of the <xi:include> element. */
+  start: number;
+}
+
+export interface ShallowDefineRecord {
+  name: string;
+  value: string;
+  /** Offset of the <Define> element. */
+  start: number;
+}
+
+export interface ShallowScanError {
+  message: string;
+  offset: number;
+}
+
+export interface ShallowDocument {
+  assets: ShallowAssetRecord[];
+  includes: ShallowIncludeRecord[];
+  /** <xi:include> elements that are direct children of the root. */
+  rootXiIncludes: ShallowXiIncludeRecord[];
+  /** <xi:include> elements nested anywhere else in the document. */
+  nestedXiIncludes: ShallowXiIncludeRecord[];
+  defines: ShallowDefineRecord[];
+  errors: ShallowScanError[];
+}
+
+interface AttrHit {
+  value: string;
+  valueStart: number;
+  valueEnd: number;
+}
+
+/**
+ * Matches a complete XML tag while respecting quoted attribute values, so a
+ * ">" or "/" inside a value never terminates the tag early.
+ */
+const TAG_RE = /<(?:"[^"]*"|'[^']*'|[^'"<>])*>/g;
+
+export function scanXmlShallow(text: string): ShallowDocument {
+  const errors: ShallowScanError[] = [];
+  const assets: ShallowAssetRecord[] = [];
+  const includes: ShallowIncludeRecord[] = [];
+  const rootXiIncludes: ShallowXiIncludeRecord[] = [];
+  const nestedXiIncludes: ShallowXiIncludeRecord[] = [];
+  const defines: ShallowDefineRecord[] = [];
+
+  // Depth of the currently open element stack. The document root opens at
+  // depth 0 -> 1, so its direct children open when depth === 1.
+  let depth = 0;
+  let inIncludes = false;
+  let inDefines = false;
+  let i = 0;
+
+  while (i < text.length) {
+    const lt = text.indexOf("<", i);
+    if (lt < 0) break;
+
+    // Non-element constructs: comments, CDATA, DOCTYPE and processing
+    // instructions are skipped whole so their content never looks like tags.
+    if (text.startsWith("<!--", lt)) {
+      const close = text.indexOf("-->", lt + 4);
+      if (close < 0) {
+        errors.push({ message: "Unterminated comment", offset: lt });
+        break;
+      }
+      i = close + 3;
+      continue;
+    }
+    if (text.startsWith("<![CDATA[", lt)) {
+      const close = text.indexOf("]]>", lt + 9);
+      if (close < 0) {
+        errors.push({ message: "Unterminated CDATA section", offset: lt });
+        break;
+      }
+      i = close + 3;
+      continue;
+    }
+    if (text.startsWith("<!", lt)) {
+      const close = text.indexOf(">", lt + 2);
+      if (close < 0) {
+        errors.push({ message: "Unterminated DOCTYPE", offset: lt });
+        break;
+      }
+      i = close + 1;
+      continue;
+    }
+    if (text.startsWith("<?", lt)) {
+      const close = text.indexOf("?>", lt + 2);
+      if (close < 0) {
+        errors.push({ message: "Unterminated processing instruction", offset: lt });
+        break;
+      }
+      i = close + 2;
+      continue;
+    }
+
+    TAG_RE.lastIndex = lt;
+    const m = TAG_RE.exec(text);
+    if (!m) {
+      errors.push({ message: "Unterminated tag", offset: lt });
+      break;
+    }
+    const tag = m[0];
+    const gt = m.index + tag.length - 1;
+    const inner = tag.slice(1, -1);
+    const closing = inner.startsWith("/");
+    const selfClosing = !closing && /\/\s*$/.test(inner);
+    const body = closing ? inner.slice(1) : inner;
+    let nameEnd = 0;
+    while (nameEnd < body.length && !/[\s/>]/.test(body[nameEnd])) nameEnd++;
+    const name = body.slice(0, nameEnd);
+    const base = lt + 1;
+
+    if (!closing) {
+      // Top-level elements (direct children of the root).
+      if (depth === 1) {
+        inIncludes = name === "Includes";
+        inDefines = name === "Defines";
+        const idAttr = findAttr(inner, base, "id");
+        if (idAttr && idAttr.value) {
+          assets.push({
+            name,
+            id: idAttr.value,
+            idValueStart: idAttr.valueStart,
+            idValueEnd: idAttr.valueEnd,
+            start: lt,
+            startTagEnd: gt + 1,
+          });
+        }
+      }
+      // <Include> entries inside the top-level <Includes> block.
+      if (inIncludes && depth === 2 && name === "Include") {
+        const type = findAttr(inner, base, "type");
+        const source = findAttr(inner, base, "source");
+        if (source?.value) {
+          includes.push({
+            type:
+              type?.value === "all" || type?.value === "instance" || type?.value === "reference"
+                ? type.value
+                : null,
+            source: source.value,
+            start: lt,
+          });
+        }
+      }
+      // <Define> entries inside the top-level <Defines> block.
+      if (inDefines && depth === 2 && name === "Define") {
+        const nameAttr = findAttr(inner, base, "name");
+        const valueAttr = findAttr(inner, base, "value");
+        if (nameAttr?.value) {
+          defines.push({ name: nameAttr.value, value: valueAttr?.value ?? "", start: lt });
+        }
+      }
+      // Nested <xi:include> (or any *:include) anywhere in the document.
+      if (localName(name) === "include") {
+        const href = findAttr(inner, base, "href");
+        const xpointer = findAttr(inner, base, "xpointer");
+        if (href?.value) {
+          const rec: ShallowXiIncludeRecord = {
+            href: href.value,
+            xpointer: xpointer?.value ?? null,
+            start: lt,
+          };
+          if (depth === 1) rootXiIncludes.push(rec);
+          else nestedXiIncludes.push(rec);
+        }
+      }
+      if (!selfClosing) depth++;
+    } else {
+      depth--;
+    }
+    i = gt + 1;
+  }
+
+  return { assets, includes, rootXiIncludes, nestedXiIncludes, defines, errors };
+}
+
+/**
+ * Scans the attributes of a tag (its inner text, without "<" and ">") for
+ * the first occurrence of `want` and returns its value with absolute offsets
+ * (base = offset one past the "<").
+ */
+function findAttr(inner: string, base: number, want: string): AttrHit | null {
+  let i = 0;
+  while (i < inner.length) {
+    while (i < inner.length && /\s/.test(inner[i])) i++;
+    // Self-closing marker or end of tag: no more attributes.
+    if (i >= inner.length || inner[i] === "/" || inner[i] === ">") break;
+    const nameStart = i;
+    while (i < inner.length && !/[\s=/>]/.test(inner[i])) i++;
+    const name = inner.slice(nameStart, i);
+    while (i < inner.length && /\s/.test(inner[i])) i++;
+    if (inner[i] === "=") {
+      i++;
+      while (i < inner.length && /\s/.test(inner[i])) i++;
+      const q = inner[i];
+      if (q === '"' || q === "'") {
+        const vStart = i + 1;
+        const vEnd = inner.indexOf(q, vStart);
+        if (vEnd < 0) {
+          return name === want ? { value: "", valueStart: -1, valueEnd: -1 } : null;
+        }
+        if (name === want) {
+          return {
+            value: inner.slice(vStart, vEnd),
+            valueStart: base + vStart,
+            valueEnd: base + vEnd,
+          };
+        }
+        i = vEnd + 1;
+        continue;
+      }
+      // Unquoted value (tolerated).
+      const vs = i;
+      while (i < inner.length && !/[\s>]/.test(inner[i])) i++;
+      if (name === want) {
+        return { value: inner.slice(vs, i), valueStart: base + vs, valueEnd: base + i };
+      }
+      continue;
+    }
+    // Attribute without a value (rare but tolerated).
+    if (name === want) return { value: "", valueStart: -1, valueEnd: -1 };
+  }
+  return null;
+}
+
+function localName(name: string): string {
+  const idx = name.lastIndexOf(":");
+  return idx >= 0 ? name.slice(idx + 1) : name;
+}

src/indexer/types.ts

@@ -1,6 +1,8 @@
 import type { XmlDocument } from "../language/xmlParser";
 import type { ManifestInfo } from "./manifestParser";
 import type { LineMap } from "../language/xmlParser";
+import type { IndexRecords } from "./records";
+import type { DocumentCache, IncludeResolveCache, IndexRecordsCache } from "./caches";
 
 export type AssetOrigin = "project" | "sdk" | "manifest";
 
@@ -65,6 +67,20 @@ export interface IndexStats {
   sdkDir: string;
   indexedFiles: number;
   parsedFiles: number;
+  /** Art-asset documents indexed via shallow scan (no DOM tree). */
+  shallowScannedFiles: number;
+  /** Shallow scans served from the persistent cache (unchanged files). */
+  shallowCacheHits: number;
+  /** Parsed XML files served from the persistent records cache. */
+  recordsCacheHits: number;
+  /** Include/xi:include resolutions served from the resolve cache. */
+  resolveCacheHits: number;
+  /** Include/xi:include resolutions performed during this build. */
+  resolveCalls: number;
+  /** Time spent enumerating Include source candidates (ms). */
+  candidatesMs: number;
+  /** Time spent walking the include graph (ms). */
+  walkMs: number;
   assetCount: number;
   defineCount: number;
   manifestFiles: number;
@@ -102,6 +118,23 @@ export interface IndexOptions {
   additionalDataSearchPaths: string[];
   /** Directory walker used to enumerate files for source completion. */
   walker: FileWalker;
+  /** Optional parse-tree cache shared across rebuilds (owned by the workspace). */
+  documentCache?: DocumentCache;
+  /** Optional index-records cache shared across rebuilds (owned by the workspace). */
+  recordsCache?: IndexRecordsCache;
+  /** Optional include-resolution cache shared across rebuilds (owned by the workspace). */
+  resolveCache?: IncludeResolveCache;
+  /**
+   * When true, cached documents whose path is not in `changedFiles` are used
+   * without a per-file stat. The workspace enables this while a file watcher
+   * invalidates caches for changed paths; a forced reindex passes false.
+   */
+  trustUnchanged?: boolean;
+  /**
+   * Normalized paths (see `normKey`) known to have changed since the caches
+   * were populated. Only consulted when `trustUnchanged` is true.
+   */
+  changedFiles?: ReadonlySet<string>;
 }
 
 export interface FileWalker {
@@ -121,5 +154,10 @@ export interface ParseCache {
 export interface ParsedFile {
   file: IndexedFile;
   parse: XmlDocument | null;
+  /**
+   * Compact index records (assets/defines/includes/xi:include with lines).
+   * Present for every indexable XML document; null for binary files.
+   */
+  records: IndexRecords | null;
   lineMap: LineMap | null;
 }

src/language/context.ts

@@ -76,7 +76,15 @@ function analyzeStartTag(
 
   // Inside an attribute value?
   for (const attr of el.attrs) {
-    if (attr.hasValue && offset >= attr.quoteStart && offset <= attr.quoteEnd) {
+    // An unterminated value (quoteEnd < 0) happens while the user is typing
+    // the opening quote of a new attribute value; it must still be treated as
+    // an attribute-value context so enum/ref/define completions show up.
+    if (
+      attr.hasValue &&
+      attr.quoteStart >= 0 &&
+      offset >= attr.quoteStart &&
+      (attr.quoteEnd < 0 || offset <= attr.quoteEnd)
+    ) {
       const start = attr.valueStart;
       const prefix = offset > start ? text.slice(start, offset) : "";
       return {
@@ -128,3 +136,20 @@ function empty(kind: ContextKind): CompletionContext {
     existingAttrs: [],
   };
 }
+
+/**
+ * Splits the typed prefix of a whitespace-separated list value (xs:list, e.g.
+ * bit flags such as Surfaces="GROUND WATER") into the token being edited and
+ * the offset of that token inside the prefix.
+ *
+ * "GROUND WA" -> { token: "WA", start: 7 }
+ * "GROUND "   -> { token: "",  start: 7 }
+ * "WA"        -> { token: "WA", start: 0 }
+ */
+export function splitListValuePrefix(prefix: string): { token: string; start: number } {
+  let start = prefix.length;
+  while (start > 0 && !/\s/.test(prefix[start - 1])) {
+    start--;
+  }
+  return { token: prefix.slice(start), start };
+}

src/language/semanticTokens.ts

@@ -0,0 +1,63 @@
+import { LineMap, type XmlDocument } from "./xmlParser";
+
+/**
+ * Semantic token types used by the highlighting fallback. They are standard
+ * vscode token types, so every theme already has colors for them.
+ */
+export type SemanticTokenType = "type" | "property" | "string";
+
+export interface SemanticTokenRange {
+  line: number;
+  startChar: number;
+  length: number;
+  tokenType: SemanticTokenType;
+}
+
+/**
+ * Builds semantic token ranges from a tolerant parse tree.
+ *
+ * This is the highlighting fallback for malformed XML: while the TextMate
+ * grammar loses structure (e.g. an attribute value whose closing quote has
+ * not been typed yet turns the rest of the file into one string), semantic
+ * tokens keep element names, attribute names and attribute values colored.
+ * The ranges are sorted by position for the vscode encoder.
+ */
+export function buildSemanticTokenRanges(
+  doc: XmlDocument,
+  text: string,
+): SemanticTokenRange[] {
+  const lineMap = new LineMap(text);
+  const out: SemanticTokenRange[] = [];
+
+  const push = (offset: number, length: number, tokenType: SemanticTokenType) => {
+    if (length <= 0 || offset < 0 || offset + length > text.length) return;
+    const pos = lineMap.positionAt(offset);
+    out.push({
+      line: pos.line,
+      startChar: pos.character,
+      length,
+      tokenType,
+    });
+  };
+
+  for (const el of doc.elements) {
+    // Element name in the start tag.
+    push(el.start + 1, el.name.length, "type");
+    // Element name in the closing tag (when present).
+    if (el.closeTagStart >= 0) {
+      push(el.closeTagStart + 2, el.name.length, "type");
+    }
+    for (const attr of el.attrs) {
+      push(attr.nameStart, attr.name.length, "property");
+      if (!attr.hasValue) continue;
+      // Include the surrounding quotes when available; for an unterminated
+      // value quoteEnd is -1 and the token ends at the recovered value end.
+      const start = attr.quoteStart >= 0 ? attr.quoteStart : attr.valueStart;
+      const end = attr.quoteEnd >= 0 ? attr.quoteEnd : attr.valueEnd;
+      push(start, end - start, "string");
+    }
+  }
+
+  out.sort((a, b) => a.line - b.line || a.startChar - b.startChar);
+  return out;
+}

src/language/xmlParser.ts

@@ -66,6 +66,14 @@ export interface Position {
   character: number;
 }
 
+/**
+ * Removes a leading UTF-8 byte-order mark (U+FEFF) so source offsets match
+ * the text as editors expose it (VS Code strips the BOM from document text).
+ */
+export function stripBom(text: string): string {
+  return text.charCodeAt(0) === 0xfeff ? text.slice(1) : text;
+}
+
 /** Precomputes line start offsets for offset <-> position conversion. */
 export class LineMap {
   private lineStarts: number[] = [0];
@@ -336,7 +344,16 @@ export function parseXml(text: string): XmlDocument {
     const gt = findTagEnd(text, i + 1);
     if (gt < 0) {
       err("Unterminated start tag", i);
-      const content = text.slice(i + 1);
+      // Recovery while typing: an attribute value whose closing quote has not
+      // been typed yet makes the scanner run to EOF. End the malformed start
+      // tag at the first line break (or EOF) so the rest of the document is
+      // still parsed and completion/hover keep working for the elements after
+      // the broken tag. The missing quote/tag end is still reported above.
+      let recoverTo = i + 1;
+      while (recoverTo < text.length && text[recoverTo] !== "\n" && text[recoverTo] !== "\r") {
+        recoverTo++;
+      }
+      const content = text.slice(i + 1, recoverTo);
       const raw = parseTag(content, i + 1);
       if (raw.name) {
         const el = buildElement(raw, stack.length);
@@ -344,7 +361,8 @@ export function parseXml(text: string): XmlDocument {
         root = root ?? el;
         stack.push(el);
       }
-      break;
+      i = recoverTo + 1;
+      continue;
     }
     const content = text.slice(i + 1, gt);
     const raw = parseTag(content, i + 1);

src/model/schemaModel.ts

@@ -20,6 +20,8 @@ export interface AttributeInfo {
   /** True for reference-typed attributes whose simple type has no refType. */
   isRef: boolean;
   enumValues: string[];
+  /** True for xs:list types (whitespace-separated bit flags / lists). */
+  isList: boolean;
   allowsDefine: boolean;
   isBoolean: boolean;
   base: string | null;
@@ -39,6 +41,7 @@ export interface SimpleTypeInfo {
   refType: string | null;
   isRef: boolean;
   enumValues: string[];
+  isList: boolean;
   allowsDefine: boolean;
   doc: string;
 }
@@ -212,3 +215,26 @@ export const STRUCTURAL_ELEMENTS = [
   "Defines",
   "Define",
 ];
+
+/**
+ * Element names that live outside the RA3 XSD model's namespace. The model
+ * only covers `uri:ea.com:eala:asset`; other namespaces (so far the W3C
+ * XInclude `xi:`) must not be validated against it.
+ */
+const FOREIGN_ELEMENT_PREFIXES = ["xi:"];
+
+/** True when an element name belongs to the RA3 XSD model's namespace. */
+export function isXsdElementName(name: string): boolean {
+  const lower = name.toLowerCase();
+  return !FOREIGN_ELEMENT_PREFIXES.some((p) => lower.startsWith(p));
+}
+
+/**
+ * True when an attribute name can be validated against the RA3 XSD model.
+ * EA schema attributes are unprefixed; prefixed attributes (`xai:`,
+ * `xi:`, `xlink:`, `xml:`, `xsi:`, `xmlns:*`) are namespace machinery and
+ * are not defined by the XSD.
+ */
+export function isXsdAttributeName(name: string): boolean {
+  return !name.includes(":");
+}

src/workspace.ts

@@ -3,6 +3,7 @@ import { existsSync } from "node:fs";
 import { join, dirname } from "node:path";
 import { CachedDirectoryWalker } from "./indexer/fileScanner";
 import { ModIndexer } from "./indexer/indexer";
+import { DocumentCache, IncludeResolveCache, IndexRecordsCache } from "./indexer/caches";
 import type { ModIndex } from "./indexer/types";
 import { readSettings, type ExtensionSettings } from "./settings";
 
@@ -15,12 +16,21 @@ export class ModWorkspace {
   settings: ExtensionSettings;
 
   private walker = new CachedDirectoryWalker();
+  // Caches owned here survive rebuilds: a fresh ModIndexer reuses them and
+  // only re-reads files whose stat changed (crucial for the ~2.6 GB of .w3x
+  // art assets in a project like Corona).
+  private documentCache = new DocumentCache();
+  private recordsCache = new IndexRecordsCache();
+  private resolveCache = new IncludeResolveCache();
+  private context: vscode.ExtensionContext;
+  private watchers: vscode.FileSystemWatcher[] = [];
   private statusBar: vscode.StatusBarItem;
   private rebuildTimer: ReturnType<typeof setTimeout> | null = null;
   private building = false;
   private dirty = false;
 
   constructor(context: vscode.ExtensionContext) {
+    this.context = context;
     this.settings = readSettings();
     this.statusBar = vscode.window.createStatusBarItem(
       vscode.StatusBarAlignment.Left,
@@ -51,11 +61,69 @@ export class ModWorkspace {
       this.statusBar.hide();
       return;
     }
+    this.startWatching();
     this.statusBar.text = "$(sync~spin) RA3 XML: indexing…";
     this.statusBar.show();
     await this.rebuild();
   }
 
+  /**
+   * Invalidates cached documents for a path (called by the file watcher and
+   * on document save), so the next rebuild re-reads it instead of trusting
+   * the cached copy.
+   */
+  invalidate(path: string): void {
+    if (!path) return;
+    this.documentCache.invalidate(path);
+    this.recordsCache.invalidate(path);
+  }
+
+  /**
+   * Called when files are created or deleted: include-resolution results
+   * (which encode file existence) are no longer trustworthy.
+   */
+  invalidateExistence(): void {
+    this.resolveCache.clear();
+  }
+
+  /**
+   * Watches the project, SDK and extra DATA roots for file changes and
+   * invalidates the corresponding cache entries. With caches invalidated
+   * precisely, rebuilds can trust every other cached file and skip per-file
+   * stats (huge win on mechanical drives; Corona rebuild dropped from ~38s
+   * to a few seconds).
+   */
+  private startWatching(): void {
+    if (!this.projectRoot) return;
+    const roots = new Set([
+      this.projectRoot,
+      this.settings.sdkPath,
+      ...this.settings.additionalDataSearchPaths,
+    ]);
+    for (const root of roots) {
+      if (!existsSync(root)) continue;
+      try {
+        const watcher = vscode.workspace.createFileSystemWatcher(
+          new vscode.RelativePattern(root, "**/*"),
+        );
+        watcher.onDidCreate((uri) => {
+          this.invalidate(uri.fsPath);
+          this.invalidateExistence();
+        });
+        watcher.onDidChange((uri) => this.invalidate(uri.fsPath));
+        watcher.onDidDelete((uri) => {
+          this.invalidate(uri.fsPath);
+          this.invalidateExistence();
+        });
+        this.watchers.push(watcher);
+        this.context.subscriptions.push(watcher);
+      } catch {
+        // The root may be temporarily unavailable (e.g. removable drive);
+        // indexing still works, just without watcher-based invalidation.
+      }
+    }
+  }
+
   scheduleRebuild(): void {
     if (!this.projectRoot) return;
     if (this.rebuildTimer) clearTimeout(this.rebuildTimer);
@@ -64,12 +132,13 @@ export class ModWorkspace {
     }, REBUILD_DEBOUNCE_MS);
   }
 
-  async rebuild(): Promise<void> {
+  async rebuild(force = false): Promise<void> {
     if (!this.projectRoot) return;
     if (this.building) {
       this.dirty = true;
       return;
     }
+    if (force) this.resolveCache.clear();
     this.building = true;
     this.settings = readSettings();
     try {
@@ -81,6 +150,12 @@ export class ModWorkspace {
         indexSageXml: this.settings.indexSageXml,
         additionalDataSearchPaths: this.settings.additionalDataSearchPaths,
         walker: this.walker,
+        documentCache: this.documentCache,
+        recordsCache: this.recordsCache,
+        resolveCache: this.resolveCache,
+        // Trust cache entries unless the user explicitly asked for a full
+        // verification (ra3modxml.reindex).
+        trustUnchanged: !force,
       });
       const started = Date.now();
       this.index = await indexer.build();
@@ -90,7 +165,7 @@ export class ModWorkspace {
       this.statusBar.text = `$(symbol-misc) RA3 XML: ${formatCount(s.assetCount)} assets`;
       this.statusBar.tooltip =
         `${s.projectDir}\n` +
-        `${s.indexedFiles} files indexed (${secs}s)\n` +
+        `${s.indexedFiles} files indexed (${s.parsedFiles} parsed, ${s.shallowScannedFiles} art assets shallow-scanned, ${secs}s)\n` +
         `${s.assetCount} assets (${s.manifestAssetCount} from ${s.manifestFiles} manifests)\n` +
         `${s.defineCount} defines, ${s.streams} streams, ${s.sourceCandidates} include candidates`;
     } catch (err) {
@@ -113,6 +188,7 @@ export class ModWorkspace {
     return {
       file: { path, stat: null },
       parse,
+      records: null,
       lineMap: new LineMap(text),
     };
   }

test/caches.test.mjs

@@ -0,0 +1,73 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import {
+  DocumentCache,
+  IncludeResolveCache,
+  IndexRecordsCache,
+} from "../out/indexer/caches.js";
+
+function parsed(path, elements) {
+  return {
+    file: { path, stat: { mtimeMs: 1, size: 1 } },
+    parse: { root: { name: "r" }, elements: new Array(elements), errors: [] },
+    records: null,
+    lineMap: null,
+  };
+}
+
+test("DocumentCache evicts the largest tree when over the element budget", () => {
+  const cache = new DocumentCache(64, 100);
+  cache.set(parsed("a.xml", 60));
+  cache.set(parsed("b.xml", 60));
+  assert.equal(cache.get("a.xml"), undefined, "largest tree evicted first");
+  assert.ok(cache.get("b.xml"));
+});
+
+test("DocumentCache evicts least recently used when over capacity", () => {
+  const cache = new DocumentCache(2, 1_000_000);
+  cache.set(parsed("a.xml", 10));
+  cache.set(parsed("b.xml", 10));
+  cache.set(parsed("c.xml", 10));
+  assert.equal(cache.get("a.xml"), undefined);
+  assert.ok(cache.get("b.xml"));
+  assert.ok(cache.get("c.xml"));
+});
+
+test("DocumentCache invalidate frees budget", () => {
+  const cache = new DocumentCache(64, 100);
+  cache.set(parsed("a.xml", 60));
+  cache.invalidate("a.xml");
+  cache.set(parsed("b.xml", 60));
+  assert.ok(cache.get("b.xml"), "invalidating a freed its budget share");
+  cache.set(parsed("c.xml", 60));
+  assert.equal(cache.get("a.xml"), undefined);
+  assert.ok(cache.get("c.xml"));
+});
+
+test("IndexRecordsCache stores and invalidates entries", () => {
+  const cache = new IndexRecordsCache();
+  const entry = {
+    stat: { mtimeMs: 1, size: 1 },
+    records: { assets: [], defines: [], includes: [], rootXiIncludes: [], nestedXiIncludes: [] },
+    kind: "full",
+  };
+  cache.set("a.xml", entry);
+  assert.equal(cache.get("a.xml"), entry);
+  cache.invalidate("a.xml");
+  assert.equal(cache.get("a.xml"), undefined);
+});
+
+test("IncludeResolveCache stores sources and manifest lookups", () => {
+  const cache = new IncludeResolveCache();
+  const key = "dir|DATA:static.xml";
+  const result = { path: "C:/sdk/static.xml", prefix: "DATA", raw: "DATA:static.xml" };
+  assert.equal(cache.get(key), undefined);
+  cache.set(key, result);
+  assert.equal(cache.get(key), result);
+  assert.equal(cache.getManifest("static.xml"), undefined);
+  cache.setManifest("static.xml", "C:/sdk/builtmods/static.manifest");
+  assert.equal(cache.getManifest("static.xml"), "C:/sdk/builtmods/static.manifest");
+  cache.clear();
+  assert.equal(cache.get(key), undefined);
+  assert.equal(cache.getManifest("static.xml"), undefined);
+});

test/completion.test.mjs

@@ -0,0 +1,157 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { createRequire } from "node:module";
+
+// Minimal vscode shim so the compiled completion provider can run under plain
+// node. Only the APIs used by the completion call path are implemented.
+const CompletionItemKind = {
+  Field: 1,
+  Property: 2,
+  EnumMember: 3,
+  Value: 4,
+  Constant: 5,
+  File: 6,
+};
+
+class CompletionItem {
+  constructor(label, kind) {
+    this.label = label;
+    this.kind = kind;
+  }
+}
+
+class Position {
+  constructor(line, character) {
+    this.line = line;
+    this.character = character;
+  }
+}
+
+class Range {
+  constructor(start, end) {
+    this.start = start;
+    this.end = end;
+  }
+}
+
+class MarkdownString {
+  constructor(value) {
+    this.value = value ?? "";
+  }
+  appendMarkdown(text) {
+    this.value += text;
+    return this;
+  }
+  appendCodeblock(text) {
+    this.value += "\n```\n" + text + "\n```\n";
+    return this;
+  }
+}
+
+class SnippetString {
+  constructor(value) {
+    this.value = value;
+  }
+}
+
+const require = createRequire(import.meta.url);
+const Module = require("module");
+const origResolve = Module._resolveFilename;
+Module._resolveFilename = function (request, ...args) {
+  if (request === "vscode") return "vscode-stub";
+  return origResolve.call(this, request, ...args);
+};
+require.cache["vscode-stub"] = {
+  id: "vscode-stub",
+  filename: "vscode-stub",
+  loaded: true,
+  exports: {
+    CompletionItem,
+    CompletionItemKind,
+    Position,
+    Range,
+    MarkdownString,
+    SnippetString,
+  },
+};
+
+const { Ra3CompletionProvider } = require("../out/features/completion.js");
+
+function makeDocument(text) {
+  const lineStarts = [0];
+  for (let i = 0; i < text.length; i++) {
+    if (text.charCodeAt(i) === 10) lineStarts.push(i + 1);
+  }
+  return {
+    getText: () => text,
+    offsetAt: (pos) => lineStarts[pos.line] + pos.character,
+    positionAt: (offset) => {
+      let lo = 0;
+      let hi = lineStarts.length - 1;
+      while (lo < hi) {
+        const mid = (lo + hi + 1) >> 1;
+        if (lineStarts[mid] <= offset) lo = mid;
+        else hi = mid - 1;
+      }
+      return new Position(lo, offset - lineStarts[lo]);
+    },
+  };
+}
+
+// Enum completions do not consult the index, but provideCompletionItems only
+// routes value contexts when the workspace index is present.
+const provider = new Ra3CompletionProvider({ index: {} });
+const token = { isCancellationRequested: false };
+
+test("Surfaces enum completion works with an unclosed quote", async () => {
+  const text =
+    `<AssetDeclaration>\n  <LocomotorTemplate id="x" Surfaces="G>\n  <Other/>\n</LocomotorTemplate>\n</AssetDeclaration>`;
+  const line1 = text.split("\n")[1];
+  const line1Start = text.indexOf("\n") + 1;
+  const valueStart = text.indexOf('Surfaces="') + 'Surfaces="'.length;
+  const pos = new Position(1, line1.indexOf("G") + 1);
+
+  const items = await provider.provideCompletionItems(makeDocument(text), pos, token);
+  const labels = items.map((i) => i.label);
+  assert.ok(labels.includes("GROUND"));
+  assert.ok(!labels.includes("WATER"));
+  assert.ok(items.every((i) => i.kind === CompletionItemKind.EnumMember));
+
+  // Replacement range covers only the value (start..cursor).
+  const ground = items.find((i) => i.label === "GROUND");
+  assert.equal(ground.range.start.character, valueStart - line1Start);
+  assert.equal(ground.range.end.character, pos.character);
+});
+
+test("list values filter on the token after whitespace", async () => {
+  const text =
+    `<AssetDeclaration>\n  <LocomotorTemplate id="x" Surfaces="GROUND W>\n  <Other/>\n</LocomotorTemplate>\n</AssetDeclaration>`;
+  const line1 = text.split("\n")[1];
+  const line1Start = text.indexOf("\n") + 1;
+  const valueStart = text.indexOf('Surfaces="') + 'Surfaces="'.length;
+  const pos = new Position(1, line1.indexOf("W") + 1);
+
+  const items = await provider.provideCompletionItems(makeDocument(text), pos, token);
+  const labels = items.map((i) => i.label);
+  assert.ok(labels.includes("WATER"));
+  assert.ok(labels.includes("WALL_RAILING"));
+  assert.ok(!labels.includes("GROUND"));
+
+  // Replacement range covers only the second token, not "GROUND ".
+  const water = items.find((i) => i.label === "WATER");
+  assert.equal(water.range.start.character, valueStart + "GROUND ".length - line1Start);
+});
+
+test("empty unterminated value offers all enum values", async () => {
+  const text =
+    `<AssetDeclaration>\n  <LocomotorTemplate id="x" Surfaces=">\n  <Other/>\n</LocomotorTemplate>\n</AssetDeclaration>`;
+  const line1 = text.split("\n")[1];
+  const pos = new Position(1, line1.indexOf('Surfaces="') + 'Surfaces="'.length);
+
+  const items = await provider.provideCompletionItems(makeDocument(text), pos, token);
+  const labels = items.map((i) => i.label);
+  assert.equal(items.length, 11);
+  assert.ok(labels.includes("GROUND"));
+  assert.ok(labels.includes("WATER"));
+  assert.ok(labels.includes("CRUSHABLE_WALL"));
+});

test/context.test.mjs

@@ -0,0 +1,42 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { parseXml } from "../out/language/xmlParser.js";
+import { analyzeContext, splitListValuePrefix } from "../out/language/context.js";
+
+test("unterminated quote is still an attribute-value context", () => {
+  const text = `<Locomotor id="x" Surfaces="GROUND>\n  <Other/>\n</Locomotor>`;
+  const cursor = text.indexOf("GROUND") + 6;
+  const doc = parseXml(text);
+  const ctx = analyzeContext(doc, text, cursor);
+  assert.equal(ctx.kind, "attribute-value");
+  assert.equal(ctx.attr?.name, "Surfaces");
+  assert.equal(ctx.valuePrefix, "GROUND");
+  assert.equal(ctx.element?.name, "Locomotor");
+});
+
+test("empty unterminated value keeps attribute-value context", () => {
+  const text = `<Locomotor id="x" Surfaces="\n</Locomotor>`;
+  const cursor = text.indexOf('Surfaces="') + 'Surfaces="'.length;
+  const doc = parseXml(text);
+  const ctx = analyzeContext(doc, text, cursor);
+  assert.equal(ctx.kind, "attribute-value");
+  assert.equal(ctx.attr?.name, "Surfaces");
+  assert.equal(ctx.valuePrefix, "");
+});
+
+test("closed quote still resolves value context", () => {
+  const text = `<Locomotor id="x" Surfaces="GROUND">\n</Locomotor>`;
+  const cursor = text.indexOf("GROUND") + 6;
+  const doc = parseXml(text);
+  const ctx = analyzeContext(doc, text, cursor);
+  assert.equal(ctx.kind, "attribute-value");
+  assert.equal(ctx.attr?.name, "Surfaces");
+  assert.equal(ctx.valuePrefix, "GROUND");
+});
+
+test("splitListValuePrefix isolates the token being edited", () => {
+  assert.deepEqual(splitListValuePrefix("GROUND WA"), { token: "WA", start: 7 });
+  assert.deepEqual(splitListValuePrefix("GROUND "), { token: "", start: 7 });
+  assert.deepEqual(splitListValuePrefix("WA"), { token: "WA", start: 0 });
+  assert.deepEqual(splitListValuePrefix(""), { token: "", start: 0 });
+});

test/fixtures/minimod/Data/Includes/Models/Tank_Damaged.dds

@@ -0,0 +1,2 @@
+This is a fake DDS payload used to verify that binary assets included from
+an XML hub are registered as files but never parsed as XML.

test/fixtures/minimod/Data/Includes/Models/Tank_FP.w3d

@@ -0,0 +1,4 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<AssetDeclaration xmlns="uri:ea.com:eala:asset">
+	<W3DMesh id="Tank_FP" />
+</AssetDeclaration>

test/fixtures/minimod/Data/Includes/Models/Tank_SKN.w3x

@@ -0,0 +1,5 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<AssetDeclaration xmlns="uri:ea.com:eala:asset">
+	<W3DContainer id="Tank_SKN" Hierarchy="Tank_SKL" />
+	<W3DHierarchy id="Tank_SKL" />
+</AssetDeclaration>

test/fixtures/minimod/Data/Includes/VehicleArt.xml

@@ -0,0 +1,8 @@
+<?xml version="1.0" encoding="utf-8"?>
+<AssetDeclaration xmlns="uri:ea.com:eala:asset">
+	<Includes>
+		<Include type="all" source="Models/Tank_SKN.w3x" />
+		<Include type="all" source="Models/Tank_FP.w3d" />
+		<Include type="all" source="Models/Tank_Damaged.dds" />
+	</Includes>
+</AssetDeclaration>

test/fixtures/minimod/Data/Mod.xml

@@ -9,5 +9,6 @@
 		<Include type="all" source="Includes/Weapons.xml" />
 		<Include type="all" source="Includes/Shared.xml" />
 		<Include type="all" source="Includes/Refs.xml" />
+		<Include type="all" source="Includes/VehicleArt.xml" />
 	</Includes>
 </AssetDeclaration>

test/indexer.test.mjs

@@ -2,8 +2,16 @@ import { test } from "node:test";
 import assert from "node:assert/strict";
 import { fileURLToPath } from "node:url";
 import { dirname, join } from "node:path";
+import fs from "node:fs";
+import os from "node:os";
 import { ModIndexer } from "../out/indexer/indexer.js";
 import { CachedDirectoryWalker } from "../out/indexer/fileScanner.js";
+import {
+  DocumentCache,
+  IncludeResolveCache,
+  IndexRecordsCache,
+} from "../out/indexer/caches.js";
+import { resolveReferenceTargetsForType } from "../out/indexer/refs.js";
 
 const root = dirname(dirname(fileURLToPath(import.meta.url)));
 const project = join(root, "test", "fixtures", "minimod");
@@ -70,3 +78,182 @@ test("provides include source candidates", async () => {
   assert.ok(xml.some((c) => c.source === "Includes/Units.xml"));
   assert.ok(xml.some((c) => c.source === "DATA:static.xml"));
 });
+
+test("indexes art-asset XML (.w3x / sniffed .w3d) via shallow scan and skips binary", async () => {
+  const idx = await buildIndex();
+
+  // The .w3x hub chain: Mod.xml -> VehicleArt.xml -> Models/Tank_SKN.w3x.
+  const skn = idx.assetsById.get("tank_skn");
+  assert.ok(skn?.some((d) => d.type === "W3DContainer"), "W3DContainer from .w3x indexed");
+  assert.ok(
+    idx.assetsById.get("tank_skl")?.some((d) => d.type === "W3DHierarchy"),
+    "W3DHierarchy from .w3x indexed",
+  );
+  const container = skn.find((d) => d.type === "W3DContainer");
+  assert.match(container.file, /Tank_SKN\.w3x$/);
+  assert.ok(container.line > 0, "definition line recorded from shallow scan");
+
+  // Unknown extension with XML content is sniffed and indexed.
+  assert.ok(
+    idx.assetsById.get("tank_fp")?.some((d) => d.type === "W3DMesh"),
+    "unknown-extension XML (.w3d) sniffed and indexed",
+  );
+
+  // Binary content is registered as a file but never parsed.
+  assert.equal(idx.assetsById.get("tank_damaged"), undefined, "binary asset never indexed");
+  assert.ok(
+    idx.files.has(
+      join(project, "Data", "Includes", "Models", "Tank_Damaged.dds").toLowerCase(),
+    ),
+    "binary include target registered as a file",
+  );
+
+  // The reported Harbinger scenario: <Model Name="Tank_SKN"/> (refType
+  // BaseRenderAssetType) must resolve to the W3DContainer defined in the w3x.
+  const targets = resolveReferenceTargetsForType(
+    idx,
+    "ScriptedModelDrawModel",
+    "Name",
+    "Tank_SKN",
+  );
+  assert.equal(targets.length, 1);
+  assert.equal(targets[0].def.type, "W3DContainer");
+  assert.match(targets[0].def.file, /Tank_SKN\.w3x$/);
+});
+
+test("w3x files appear in Include source completion candidates", async () => {
+  const idx = await buildIndex();
+  assert.ok(
+    idx.sourceCandidates.some((c) => c.source === "Includes/Models/Tank_SKN.w3x"),
+    "project-relative w3x candidate",
+  );
+  assert.ok(
+    idx.sourceCandidates.some((c) => c.source === "DATA:Includes/Models/Tank_SKN.w3x"),
+    "DATA: w3x candidate",
+  );
+});
+
+test("shallow scans and full parses are cached across rebuilds", async () => {
+  const documentCache = new DocumentCache();
+  const recordsCache = new IndexRecordsCache();
+  const resolveCache = new IncludeResolveCache();
+  const makeIndexer = () =>
+    new ModIndexer({
+      projectDir: project,
+      sdkDir: sdk,
+      builtmodsDirs: [join(sdk, "builtmods")],
+      indexSageXml: true,
+      additionalDataSearchPaths: [],
+      walker: new CachedDirectoryWalker(),
+      documentCache,
+      recordsCache,
+      resolveCache,
+    });
+
+  const first = await makeIndexer().build();
+  const second = await makeIndexer().build();
+
+  assert.equal(first.stats.shallowScannedFiles, 2, "w3x + sniffed w3d scanned on first build");
+  assert.equal(second.stats.shallowScannedFiles, 0, "unchanged art assets are not re-scanned");
+  assert.equal(second.stats.shallowCacheHits, 2);
+  assert.ok(second.stats.recordsCacheHits > 0, "XML records served from cache");
+  assert.ok(second.stats.resolveCacheHits > 0, "include resolutions served from cache");
+  assert.equal(second.stats.resolveCalls, 0, "no include re-resolved on a trusted rebuild");
+  assert.equal(second.stats.assetCount, first.stats.assetCount);
+  assert.equal(second.stats.indexedFiles, first.stats.indexedFiles);
+});
+
+test("trusted rebuilds skip unchanged files; invalidation forces re-reads", async () => {
+  const documentCache = new DocumentCache();
+  const recordsCache = new IndexRecordsCache();
+  const resolveCache = new IncludeResolveCache();
+  const opts = () => ({
+    projectDir: project,
+    sdkDir: sdk,
+    builtmodsDirs: [join(sdk, "builtmods")],
+    indexSageXml: true,
+    additionalDataSearchPaths: [],
+    walker: new CachedDirectoryWalker(),
+    documentCache,
+    recordsCache,
+    resolveCache,
+  });
+
+  // Trusted rebuild: cached files are reused without any per-file stat/read.
+  const first = await new ModIndexer({ ...opts(), trustUnchanged: true }).build();
+  assert.equal(first.stats.shallowScannedFiles, 2);
+  const second = await new ModIndexer({ ...opts(), trustUnchanged: true }).build();
+  assert.equal(second.stats.shallowScannedFiles, 0);
+  assert.equal(second.stats.shallowCacheHits, 2);
+  assert.ok(second.stats.recordsCacheHits > 0);
+
+  // Simulate the file watcher / save handler: invalidate one art asset.
+  // The next trusted rebuild must re-scan exactly that file.
+  recordsCache.invalidate(join(project, "Data", "Includes", "Models", "Tank_SKN.w3x"));
+  const third = await new ModIndexer({ ...opts(), trustUnchanged: true }).build();
+  assert.equal(third.stats.shallowScannedFiles, 1);
+  assert.equal(third.stats.shallowCacheHits, 1);
+  assert.ok(
+    third.assetsById.get("tank_skn")?.some((d) => d.type === "W3DContainer"),
+    "re-scanned w3x asset present",
+  );
+
+  // Invalidate one XML file: exactly its records are re-extracted.
+  recordsCache.invalidate(join(project, "Data", "Includes", "Units.xml"));
+  const fourth = await new ModIndexer({ ...opts(), trustUnchanged: true }).build();
+  assert.equal(fourth.stats.recordsCacheHits, third.stats.recordsCacheHits - 1);
+  assert.ok(
+    fourth.assetsById.get("testtank")?.some((d) => d.type === "GameObject"),
+    "re-parsed XML asset present",
+  );
+
+  // A forced rebuild (ra3modxml.reindex) verifies stats but still reuses
+  // content caches for unchanged files.
+  const forced = await new ModIndexer({ ...opts(), trustUnchanged: false }).build();
+  assert.equal(forced.stats.shallowScannedFiles, 0);
+  assert.equal(forced.stats.shallowCacheHits, 2);
+});
+
+test("index stats include candidate/walk phase timings", async () => {
+  const idx = await buildIndex();
+  assert.equal(typeof idx.stats.candidatesMs, "number");
+  assert.equal(typeof idx.stats.walkMs, "number");
+});
+
+test("w3x with a UTF-8 BOM is indexed with correct offsets", async (t) => {
+  const tmp = fs.mkdtempSync(join(os.tmpdir(), "ra3modxml-bom-"));
+  t.after(() => fs.rmSync(tmp, { recursive: true, force: true }));
+  const projectDir = join(tmp, "project");
+  fs.mkdirSync(join(projectDir, "Data", "Models"), { recursive: true });
+  fs.writeFileSync(
+    join(projectDir, "Data", "Mod.xml"),
+    `<?xml version="1.0" encoding="utf-8"?>
+<AssetDeclaration>
+  <Includes>
+    <Include type="all" source="Models/Tank_BOM.w3x"/>
+  </Includes>
+</AssetDeclaration>`,
+    "utf8",
+  );
+  fs.writeFileSync(
+    join(projectDir, "Data", "Models", "Tank_BOM.w3x"),
+    "\uFEFF<?xml version=\"1.0\" encoding=\"UTF-8\"?>\n" +
+      "<AssetDeclaration>\n" +
+      "  <W3DContainer id=\"Tank_BOM\" />\n" +
+      "</AssetDeclaration>\n",
+    "utf8",
+  );
+
+  const idx = await new ModIndexer({
+    projectDir,
+    sdkDir: sdk,
+    builtmodsDirs: [join(sdk, "builtmods")],
+    indexSageXml: false,
+    additionalDataSearchPaths: [],
+    walker: new CachedDirectoryWalker(),
+  }).build();
+
+  const def = idx.assetsById.get("tank_bom")?.find((d) => d.type === "W3DContainer");
+  assert.ok(def, "BOM-prefixed w3x asset indexed");
+  assert.equal(def.line, 3, "id line is correct despite the BOM");
+});

test/records.test.mjs

@@ -0,0 +1,65 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { parseXml, LineMap } from "../out/language/xmlParser.js";
+import { extractIndexRecords, recordsFromShallow } from "../out/indexer/records.js";
+import { scanXmlShallow } from "../out/indexer/shallowScan.js";
+
+test("extractIndexRecords mirrors the walk semantics", () => {
+  const text = `<?xml version="1.0"?>
+<AssetDeclaration>
+  <Defines>
+    <Define name="HP" value="100"/>
+  </Defines>
+  <Includes>
+    <Include type="all" source="Units.xml"/>
+    <Include type="instance" source="Base.xml"/>
+  </Includes>
+  <GameObject id="Tank" CommandSet="TankCommandSet"/>
+  <WeaponTemplate id="TankGun"/>
+  <xi:include href="DATA:Extra.xml" xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:Extra/child::*)"/>
+  <GameObject id="Tank2">
+    <Draws>
+      <xi:include href="DATA:Nested.xml"/>
+    </Draws>
+  </GameObject>
+</AssetDeclaration>`;
+  const lineMap = new LineMap(text);
+  const records = extractIndexRecords(parseXml(text), lineMap);
+  assert.deepEqual(
+    records.assets.map((a) => [a.type, a.id, a.line]),
+    [
+      ["GameObject", "Tank", 10],
+      ["WeaponTemplate", "TankGun", 11],
+      ["GameObject", "Tank2", 13],
+    ],
+  );
+  assert.deepEqual(
+    records.defines.map((d) => [d.name, d.value]),
+    [["HP", "100"]],
+  );
+  assert.deepEqual(
+    records.includes.map((i) => [i.type, i.source]),
+    [
+      ["all", "Units.xml"],
+      ["instance", "Base.xml"],
+    ],
+  );
+  assert.deepEqual(
+    records.rootXiIncludes.map((x) => [x.href, x.line]),
+    [["DATA:Extra.xml", 12]],
+  );
+  assert.deepEqual(
+    records.nestedXiIncludes.map((x) => [x.href, x.line]),
+    [["DATA:Nested.xml", 15]],
+  );
+});
+
+test("recordsFromShallow converts offsets to 1-based lines", () => {
+  const text = `<AssetDeclaration>\n  <W3DContainer id="A"/>\n</AssetDeclaration>`;
+  const lineMap = new LineMap(text);
+  const records = recordsFromShallow(scanXmlShallow(text), lineMap);
+  assert.equal(records.assets.length, 1);
+  assert.equal(records.assets[0].type, "W3DContainer");
+  assert.equal(records.assets[0].id, "A");
+  assert.equal(records.assets[0].line, 2);
+});

test/refs.test.mjs

@@ -4,7 +4,15 @@ import { fileURLToPath } from "node:url";
 import { dirname, join } from "node:path";
 import { ModIndexer } from "../out/indexer/indexer.js";
 import { CachedDirectoryWalker } from "../out/indexer/fileScanner.js";
-import { isReferenceAttribute, resolveReferenceTargets } from "../out/indexer/refs.js";
+import {
+  isLocalReferenceAttribute,
+  isReferenceAttribute,
+  isReferenceAttributeOfType,
+  resolveReferenceTargets,
+  resolveReferenceTargetsForType,
+} from "../out/indexer/refs.js";
+import { parseXml } from "../out/language/xmlParser.js";
+import { resolveElementType } from "../out/language/typeContext.js";
 import * as model from "../out/model/schemaModel.js";
 
 const root = dirname(dirname(fileURLToPath(import.meta.url)));
@@ -63,10 +71,21 @@ test("isReferenceAttribute distinguishes references from enums/paths", () => {
   assert.equal(isReferenceAttribute("FireWeaponNugget", "WeaponName"), true);
 });
 
-test("untyped references (isRef without refType) resolve to any declared id", () => {
+test("attribute-level xas:refType is preserved in the model", () => {
-  // LocomotorSet/@Locomotor has type AssetReference: xas:isRef="true" with
+  // <xs:attribute name="Locomotor" type="AssetReference"
-  // no refType. It must be treated as a reference and match the id regardless
+  //                xas:refType="LocomotorTemplate" /> — the refType lives on
-  // of the target asset type.
+  // the attribute node, not on the AssetReference simple type.
+  assert.equal(
+    model.attributesOfElement("LocomotorSet").find((a) => a.name === "Locomotor")?.refType,
+    "LocomotorTemplate",
+  );
+  assert.equal(
+    model.attributesOfElement("ArmorSet").find((a) => a.name === "Armor")?.refType,
+    "ArmorTemplate",
+  );
+});
+
+test("typed Locomotor references resolve only to LocomotorTemplate defs", () => {
   const idx = {
     assetsById: new Map([
       [
@@ -79,6 +98,14 @@ test("untyped references (isRef without refType) resolve to any declared id", ()
             line: 5,
             origin: "project",
           },
+          {
+            // Same id under a different type must NOT satisfy the reference.
+            type: "GameObject",
+            id: "AlliedAntiVehicleVehicleTech1Locomotor",
+            file: "GameObject.xml",
+            line: 1,
+            origin: "project",
+          },
         ],
       ],
     ]),
@@ -100,3 +127,118 @@ test("untyped references (isRef without refType) resolve to any declared id", ()
   // A non-reference attribute still returns nothing.
   assert.equal(resolveReferenceTargets(idx, "LocomotorSet", "EditorName", "X").length, 0);
 });
+
+test("module ids declare the element itself and are never global references", () => {
+  // The exact reported scenario: <TruckDraw id="ModuleTag_Draw"/> inside a
+  // GameObject. ModuleData@id carries xas:refType="ModuleData", which is the
+  // element's own base type -> a definition site, not a reference.
+  const xml = `
+<AssetDeclaration>
+  <GameObject id="GuardianTank">
+    <Draws>
+      <TruckDraw id="ModuleTag_Draw" />
+    </Draws>
+  </GameObject>
+</AssetDeclaration>`;
+  const doc = parseXml(xml);
+  const truckDraw = doc.elements.find((e) => e.name === "TruckDraw");
+  assert.ok(truckDraw);
+  const elType = resolveElementType(truckDraw);
+  assert.equal(elType, "W3DTruckDrawModuleData");
+  assert.equal(isLocalReferenceAttribute(elType, "id"), true);
+  assert.equal(isReferenceAttributeOfType(elType, "id"), false);
+  const idx = { assetsById: new Map(), assets: new Map(), defines: new Map() };
+  assert.equal(
+    resolveReferenceTargetsForType(idx, elType, "id", "ModuleTag_Draw").length,
+    0,
+  );
+});
+
+test("Poid attributes reference pipeline-local objects, not global assets", () => {
+  // AttachModuleId / ModuleId / AutoResolveBody are Poid-typed references to
+  // modules inside the same GameObject; the global index cannot judge them.
+  assert.equal(isReferenceAttributeOfType("AttachNugget", "AttachModuleId"), false);
+  assert.equal(isLocalReferenceAttribute("AttachNugget", "AttachModuleId"), true);
+  const idx = { assetsById: new Map(), assets: new Map(), defines: new Map() };
+  assert.equal(
+    resolveReferenceTargetsForType(idx, "AttachNugget", "AttachModuleId", "ModuleTag_X").length,
+    0,
+  );
+});
+
+test("cross-type id references (RoadObject@id -> Road) stay real references", () => {
+  // RoadObject declares its id with xas:refType="Road" (an AssetReference,
+  // NOT the element's own type), so it must keep reference semantics.
+  assert.equal(
+    model.attributesOfType("RoadObject").find((a) => a.name === "id")?.refType,
+    "Road",
+  );
+  assert.equal(isReferenceAttributeOfType("RoadObject", "id"), true);
+  const idx = {
+    assetsById: new Map([
+      [
+        "road01",
+        [
+          {
+            type: "Road",
+            id: "Road01",
+            file: "Roads.xml",
+            line: 3,
+            origin: "project",
+          },
+        ],
+      ],
+    ]),
+    assets: new Map(),
+    projectDir: ".",
+    sdkDir: ".",
+    defines: new Map(),
+    files: new Map(),
+    streams: [],
+    manifests: new Map(),
+    sourceCandidates: [],
+    diagnostics: [],
+    stats: {},
+  };
+  const targets = resolveReferenceTargetsForType(idx, "RoadObject", "id", "Road01");
+  assert.equal(targets.length, 1);
+  assert.equal(targets[0].def.type, "Road");
+});
+
+test("top-level asset ids with plain pipeline ids are not references", () => {
+  // LocomotorSet@id / ArmorSet@id are Poid pipeline ids without a refType:
+  // they identify the element and never need a global definition.
+  assert.equal(isReferenceAttributeOfType("LocomotorSet", "id"), false);
+  assert.equal(isReferenceAttributeOfType("ArmorTemplateSet", "id"), false);
+  const idx = { assetsById: new Map(), assets: new Map(), defines: new Map() };
+  assert.equal(
+    resolveReferenceTargetsForType(idx, "LocomotorSet", "id", "AnyLocalId").length,
+    0,
+  );
+});
+
+test("xi:include elements are outside the XSD model and unvalidated", () => {
+  // The nested XInclude inside a GameObject draw list (HeadlightDraw2.xml)
+  // must resolve to no XSD type, so href/xpointer can never be flagged as
+  // unknown attributes.
+  const xml = `
+<AssetDeclaration>
+  <GameObject id="GuardianTank">
+    <Draws>
+      <xi:include
+        href="DATA:Includes/HeadlightDraw2.xml"
+        xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:HeadlightDraw2/child::*)" />
+    </Draws>
+  </GameObject>
+</AssetDeclaration>`;
+  const doc = parseXml(xml);
+  const inc = doc.elements.find((e) => e.name === "xi:include");
+  assert.ok(inc);
+  assert.equal(model.isXsdElementName(inc.name), false);
+  assert.equal(resolveElementType(inc), null);
+  // The XInclude attributes themselves are unprefixed (model-valid name
+  // shape); the element-level foreign check is what keeps them quiet.
+  for (const a of inc.attrs) {
+    assert.equal(model.isXsdAttributeName(a.name), true, a.name);
+  }
+});

test/schemaModel.test.mjs

@@ -17,12 +17,84 @@ test("GameObject has expected attributes", () => {
   assert.ok(attrs.some((a) => a.name === "inheritFrom"));
 });
 
+test("attribute-level xas:refType is captured (module ids, map objects)", () => {
+  // ModuleData@id is declared as <xs:attribute name="id" type="Poid"
+  // xas:refType="ModuleData" />; the refType must reach every module subtype.
+  const moduleId = model.attributesOfType("ModuleData").find((a) => a.name === "id");
+  assert.equal(moduleId?.refType, "ModuleData");
+  assert.equal(
+    model.attributesOfType("W3DTruckDrawModuleData").find((a) => a.name === "id")?.refType,
+    "ModuleData",
+  );
+  // MapObject@id declares itself; ThingTemplate references a GameObject.
+  assert.equal(
+    model.attributesOfType("MapObject").find((a) => a.name === "id")?.refType,
+    "MapObject",
+  );
+  assert.equal(
+    model.attributesOfType("MapObject").find((a) => a.name === "ThingTemplate")?.refType,
+    "GameObject",
+  );
+  // RoadObject@id references a different global asset type (Road).
+  assert.equal(
+    model.attributesOfType("RoadObject").find((a) => a.name === "id")?.refType,
+    "Road",
+  );
+  // Poid pipeline-local references keep their typed targets.
+  assert.equal(
+    model.attributesOfType("AttachNugget").find((a) => a.name === "AttachModuleId")?.refType,
+    "ModuleData",
+  );
+});
+
 test("Include has reference/instance/all enum", () => {
   const attrs = model.attributesOfElement("Include");
   const type = attrs.find((a) => a.name === "type");
   assert.deepEqual(type?.enumValues, ["reference", "instance", "all"]);
 });
 
+test("xs:list types expose their item enum and isList flag", () => {
+  const expected = [
+    "GROUND", "WATER", "CLIFF", "AIR", "RUBBLE", "OBSTACLE", "IMPASSABLE",
+    "DEEP_WATER", "WALL_RAILING", "CRUSHABLE_OBSTACLE", "CRUSHABLE_WALL",
+  ];
+  const flags = model.typeInfo("LocomotorSurfaceBitFlags");
+  assert.equal(flags.isList, true);
+  assert.deepEqual(flags.enumValues, expected);
+  const surfaces = model
+    .attributesOfType("LocomotorTemplate")
+    .find((a) => a.name === "Surfaces");
+  assert.equal(surfaces.isList, true);
+  assert.deepEqual(surfaces.enumValues, expected);
+});
+
+test("numeric list types do not invent enums", () => {
+  const list = model.typeInfo("SageUnsignedIntList");
+  assert.equal(list.isList, true);
+  assert.deepEqual(list.enumValues, []);
+  // Plain restriction enums keep isList=false.
+  const plain = model.typeInfo("Surface");
+  assert.equal(plain.isList, false);
+  assert.ok(plain.enumValues.length > 0);
+});
+
+test("foreign namespaces are excluded from XSD validation", () => {
+  // XInclude elements (xi:include / xi:fallback) come from the W3C XInclude
+  // namespace, not from the EA asset XSD.
+  assert.equal(model.isXsdElementName("xi:include"), false);
+  assert.equal(model.isXsdElementName("xi:fallback"), false);
+  assert.equal(model.isXsdElementName("GameObject"), true);
+  assert.equal(model.isXsdElementName("AssetDeclaration"), true);
+  // EA schema attributes are unprefixed; prefixed attributes (xai:, xi:,
+  // xlink:, xml:, xsi:, xmlns:*) are namespace machinery.
+  assert.equal(model.isXsdAttributeName("href"), true);
+  assert.equal(model.isXsdAttributeName("xpointer"), true);
+  assert.equal(model.isXsdAttributeName("xai:joinAction"), false);
+  assert.equal(model.isXsdAttributeName("xlink:href"), false);
+  assert.equal(model.isXsdAttributeName("xml:space"), false);
+  assert.equal(model.isXsdAttributeName("xmlns:xi"), false);
+});
+
 test("type assignability follows inheritance", () => {
   assert.ok(model.isAssignableTo("GameObject", "BaseInheritableAsset"));
   assert.ok(model.isAssignableTo("GameObject", "GameObject"));

test/semanticTokens.test.mjs

@@ -0,0 +1,145 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { createRequire } from "node:module";
+
+// Minimal vscode shim so the compiled semantic tokens provider can run under
+// plain node. Only the APIs used by the provider call path are implemented.
+class Range {
+  constructor(startLine, startCharacter, endLine, endCharacter) {
+    this.start = { line: startLine, character: startCharacter };
+    this.end = { line: endLine, character: endCharacter };
+  }
+}
+
+class SemanticTokensLegend {
+  constructor(tokenTypes) {
+    this.tokenTypes = tokenTypes;
+  }
+}
+
+class SemanticTokens {
+  constructor(data) {
+    this.data = data;
+  }
+}
+
+class SemanticTokensBuilder {
+  constructor(legend) {
+    this.legend = legend;
+    this.pushed = [];
+  }
+  push(range, tokenType) {
+    this.pushed.push({ range, tokenType });
+  }
+  build() {
+    return new SemanticTokens(new Uint32Array(this.pushed.length * 5));
+  }
+}
+
+const require = createRequire(import.meta.url);
+const Module = require("module");
+const origResolve = Module._resolveFilename;
+Module._resolveFilename = function (request, ...args) {
+  if (request === "vscode") return "vscode-stub";
+  return origResolve.call(this, request, ...args);
+};
+require.cache["vscode-stub"] = {
+  id: "vscode-stub",
+  filename: "vscode-stub",
+  loaded: true,
+  exports: {
+    Range,
+    SemanticTokens,
+    SemanticTokensBuilder,
+    SemanticTokensLegend,
+  },
+};
+
+const { parseXml } = require("../out/language/xmlParser.js");
+const { buildSemanticTokenRanges } = require("../out/language/semanticTokens.js");
+const { Ra3SemanticTokensProvider } = require("../out/features/semanticTokens.js");
+
+function lineStartsOf(text) {
+  const starts = [0];
+  for (let i = 0; i < text.length; i++) {
+    if (text.charCodeAt(i) === 10) starts.push(i + 1);
+  }
+  return starts;
+}
+
+function sliceAt(text, token) {
+  const starts = lineStartsOf(text);
+  const offset = starts[token.line] + token.startChar;
+  return text.slice(offset, offset + token.length);
+}
+
+function makeDocument(text) {
+  const lineStarts = lineStartsOf(text);
+  return {
+    getText: () => text,
+    offsetAt: (pos) => lineStarts[pos.line] + pos.character,
+    positionAt: (offset) => {
+      let lo = 0;
+      let hi = lineStarts.length - 1;
+      while (lo < hi) {
+        const mid = (lo + hi + 1) >> 1;
+        if (lineStarts[mid] <= offset) lo = mid;
+        else hi = mid - 1;
+      }
+      return { line: lo, character: offset - lineStarts[lo] };
+    },
+  };
+}
+
+const MALFORMED =
+  `<AssetDeclaration>\n  <LocomotorTemplate id="x" Surfaces="GROUND>\n  <Other/>\n</LocomotorTemplate>\n</AssetDeclaration>`;
+
+test("fallback tokens cover names, attributes and values", () => {
+  const doc = parseXml(MALFORMED);
+  assert.ok(doc.errors.length > 0);
+  const tokens = buildSemanticTokenRanges(doc, MALFORMED);
+
+  // Element names (opening and closing tags).
+  assert.ok(tokens.some((t) => t.tokenType === "type" && sliceAt(MALFORMED, t) === "AssetDeclaration"));
+  assert.ok(tokens.some((t) => t.tokenType === "type" && sliceAt(MALFORMED, t) === "LocomotorTemplate"));
+  assert.ok(tokens.some((t) => t.tokenType === "type" && sliceAt(MALFORMED, t) === "Other"));
+  assert.ok(
+    tokens.some(
+      (t) => t.tokenType === "type" && sliceAt(MALFORMED, t) === "LocomotorTemplate" && t.line === 3,
+    ),
+  );
+
+  // Attribute names.
+  assert.ok(tokens.some((t) => t.tokenType === "property" && sliceAt(MALFORMED, t) === "Surfaces"));
+  assert.ok(tokens.some((t) => t.tokenType === "property" && sliceAt(MALFORMED, t) === "id"));
+
+  // Attribute values: closed quotes include both quotes; the unterminated
+  // Surfaces value keeps the opening quote and everything up to the recovered
+  // line end (the `>` is still inside the unclosed string).
+  const strings = tokens.filter((t) => t.tokenType === "string").map((t) => sliceAt(MALFORMED, t));
+  assert.ok(strings.includes('"x"'));
+  assert.ok(strings.includes('"GROUND>'));
+
+  // Tokens are sorted by position for the vscode encoder.
+  for (let i = 1; i < tokens.length; i++) {
+    const a = tokens[i - 1];
+    const b = tokens[i];
+    assert.ok(a.line < b.line || (a.line === b.line && a.startChar <= b.startChar));
+  }
+});
+
+test("well-formed XML gets no semantic fallback tokens", async () => {
+  const text = `<AssetDeclaration>\n  <LocomotorTemplate id="x" Surfaces="GROUND"/>\n</AssetDeclaration>`;
+  const provider = new Ra3SemanticTokensProvider();
+  const result = await provider.provideDocumentSemanticTokens(makeDocument(text), {});
+  assert.equal(result.data.length, 0);
+});
+
+test("malformed XML gets semantic fallback tokens", async () => {
+  const provider = new Ra3SemanticTokensProvider();
+  const result = await provider.provideDocumentSemanticTokens(
+    makeDocument(MALFORMED),
+    {},
+  );
+  assert.ok(result.data.length > 0);
+});

test/shallowScan.test.mjs

@@ -0,0 +1,122 @@
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { scanXmlShallow } from "../out/indexer/shallowScan.js";
+
+test("extracts top-level assets, includes, defines and xi:include", () => {
+  const text = `<?xml version="1.0"?>
+<AssetDeclaration xmlns="uri:ea.com:eala:asset">
+  <Defines>
+    <Define name="MODEL_SCALE" value="1.0"/>
+  </Defines>
+  <Includes>
+    <Include type="all" source="Art/Model_SKN.w3x"/>
+    <Include type="reference" source="DATA:static.xml"/>
+  </Includes>
+  <W3DContainer id="Model_SKN" Hierarchy="Model_SKL">
+    <SubObject SubObjectID="GUN">
+      <RenderObject><Mesh>Model_SKN.GUN</Mesh></RenderObject>
+    </SubObject>
+  </W3DContainer>
+  <W3DMesh id="Model_SKN.GUN"/>
+  <xi:include href="DATA:Includes/Extra.xml" xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:Extra/child::*)"/>
+</AssetDeclaration>`;
+  const doc = scanXmlShallow(text);
+  assert.equal(doc.errors.length, 0);
+  assert.deepEqual(
+    doc.assets.map((a) => `${a.name}:${a.id}`),
+    ["W3DContainer:Model_SKN", "W3DMesh:Model_SKN.GUN"],
+  );
+  assert.equal(doc.includes.length, 2);
+  assert.equal(doc.includes[0].type, "all");
+  assert.equal(doc.includes[0].source, "Art/Model_SKN.w3x");
+  assert.equal(doc.includes[1].type, "reference");
+  assert.equal(doc.defines.length, 1);
+  assert.equal(doc.defines[0].name, "MODEL_SCALE");
+  assert.equal(doc.defines[0].value, "1.0");
+  assert.equal(doc.rootXiIncludes.length, 1);
+  assert.equal(doc.rootXiIncludes[0].href, "DATA:Includes/Extra.xml");
+  assert.match(doc.rootXiIncludes[0].xpointer, /xpointer\(/);
+  assert.equal(doc.nestedXiIncludes.length, 0);
+
+  // Recorded offsets must slice the original text back out.
+  const container = doc.assets[0];
+  assert.equal(text.slice(container.idValueStart, container.idValueEnd), "Model_SKN");
+  assert.equal(text.slice(container.start, container.startTagEnd).startsWith("<W3DContainer"), true);
+  const mesh = doc.assets[1];
+  assert.equal(text.slice(mesh.idValueStart, mesh.idValueEnd), "Model_SKN.GUN");
+});
+
+test("handles > and / inside quoted values, self-closing tags and comments", () => {
+  const text = `<?xml version="1.0"?>
+<!-- a > comment with < inside -->
+<AssetDeclaration xmlns="uri:ea.com:eala:asset">
+  <W3DContainer id="Weird" Description="a > b < c" />
+  <W3DMesh id="X" VertexData="a / b"/>
+</AssetDeclaration>`;
+  const doc = scanXmlShallow(text);
+  assert.equal(doc.errors.length, 0);
+  assert.deepEqual(
+    doc.assets.map((a) => a.id),
+    ["Weird", "X"],
+  );
+});
+
+test("skips CDATA and processing instructions", () => {
+  const text = `<AssetDeclaration>
+  <!-- <W3DMesh id="FAKE"/> -->
+  <![CDATA[ <W3DMesh id="FAKE2"/> ]]>
+  <?xml-stylesheet href="x"?>
+  <W3DContainer id="REAL"/>
+</AssetDeclaration>`;
+  const doc = scanXmlShallow(text);
+  assert.equal(doc.errors.length, 0);
+  assert.deepEqual(
+    doc.assets.map((a) => a.id),
+    ["REAL"],
+  );
+});
+
+test("reports unterminated tags and keeps partial results", () => {
+  const doc = scanXmlShallow('<AssetDeclaration><W3DMesh id="A"/><W3DMesh id="B"');
+  assert.ok(doc.errors.length >= 1);
+  assert.deepEqual(
+    doc.assets.map((a) => a.id),
+    ["A"],
+  );
+});
+
+test("nested module payload does not create asset records", () => {
+  // Only top-level elements with an id are assets; the hundreds of thousands
+  // of numeric V/T elements inside a mesh are not.
+  const doc = scanXmlShallow(`<AssetDeclaration>
+  <W3DMesh id="MESH">
+    <Vertices><V X="1" Y="2" Z="3"/><V X="4" Y="5" Z="6"/></Vertices>
+    <Triangles><T A="0" B="1" C="2"/></Triangles>
+  </W3DMesh>
+</AssetDeclaration>`);
+  assert.deepEqual(
+    doc.assets.map((a) => a.id),
+    ["MESH"],
+  );
+});
+
+test("distinguishes root-level and nested xi:include", () => {
+  const doc = scanXmlShallow(`<AssetDeclaration>
+  <xi:include href="DATA:Top.xml"/>
+  <W3DMesh id="MESH">
+    <SubObject>
+      <RenderObject>
+        <xi:include href="DATA:Nested.xml" xpointer="xmlns(n=uri:ea.com:eala:asset) xpointer(/n:X/child::*)"/>
+      </RenderObject>
+    </SubObject>
+  </W3DMesh>
+</AssetDeclaration>`);
+  assert.deepEqual(
+    doc.rootXiIncludes.map((x) => x.href),
+    ["DATA:Top.xml"],
+  );
+  assert.deepEqual(
+    doc.nestedXiIncludes.map((x) => x.href),
+    ["DATA:Nested.xml"],
+  );
+});

test/xmlParser.test.mjs

@@ -1,6 +1,11 @@
 import { test } from "node:test";
 import assert from "node:assert/strict";
-import { parseXml, findElementAt } from "../out/language/xmlParser.js";
+import { parseXml, findElementAt, stripBom } from "../out/language/xmlParser.js";
+
+test("stripBom removes a leading UTF-8 byte-order mark", () => {
+  assert.equal(stripBom("\uFEFF<A/>"), "<A/>");
+  assert.equal(stripBom("<A/>"), "<A/>");
+});
 
 test("parses elements, attributes and positions", () => {
   const text = `<AssetDeclaration xmlns="uri:ea.com:eala:asset">\n\t<GameObject id="X" KindOf="A B"/>\n</AssetDeclaration>`;
@@ -45,3 +50,29 @@ test("tolerates partial input while typing", () => {
   assert.ok(doc.errors.length >= 1); // unclosed
   assert.equal(doc.elements.length, 2);
 });
+
+test("recovers from an unterminated attribute value at end of line", () => {
+  // Typing an attribute value quote without its closing quote makes the tag
+  // malformed; the parser must stop the broken start tag at the line break so
+  // the rest of the document (and completion for it) keeps working.
+  const text = `<AssetDeclaration>\n  <A x="1" y="abc\n  <B/>\n  </A>\n</AssetDeclaration>`;
+  const doc = parseXml(text);
+  assert.ok(doc.errors.some((e) => e.message === "Unterminated start tag"));
+  assert.deepEqual(doc.elements.map((e) => e.name), ["AssetDeclaration", "A", "B"]);
+  const a = doc.elements.find((e) => e.name === "A");
+  const y = a.attrs.find((at) => at.name === "y");
+  assert.equal(y.quoteEnd, -1);
+  assert.equal(text.slice(y.valueStart, y.valueEnd), "abc");
+  const b = doc.elements.find((e) => e.name === "B");
+  assert.ok(b.start > a.start);
+});
+
+test("reports an unterminated attribute value at EOF", () => {
+  const text = `<A x="abc`;
+  const doc = parseXml(text);
+  assert.ok(doc.errors.some((e) => /Unterminated start tag/.test(e.message)));
+  assert.equal(doc.elements.length, 1);
+  const a = doc.elements[0];
+  assert.equal(a.attrs[0].value, "abc");
+  assert.equal(a.attrs[0].quoteEnd, -1);
+});

tools/xsd-to-model.mjs

@@ -168,6 +168,7 @@ function resolveTypeDescriptor(typeName, memo = new Map()) {
       name,
       refType: null,
       enumValues: [],
+      isList: false,
       allowsDefine: false,
       isBoolean: name === "boolean",
       base: null,
@@ -180,6 +181,37 @@ function resolveTypeDescriptor(typeName, memo = new Map()) {
   const st = simpleTypes.get(name);
   if (st) {
     memo.set(name, null); // guard against cycles
+    const listNode = st?.list;
+    if (listNode) {
+      // xs:list types (bit flags such as LocomotorSurfaceBitFlags, AssetIdList,
+      // PercentageList, ...) serialize as whitespace-separated items. The
+      // usable enum/ref semantics come from the item type, so inherit them.
+      let itemDesc = null;
+      let inlineEnum = [];
+      let inlineAllowsDefine = false;
+      const itemTypeName = normalizeTypeName(listNode?.["@_itemType"]);
+      if (itemTypeName) {
+        itemDesc = resolveTypeDescriptor(itemTypeName, memo);
+      } else if (listNode.simpleType) {
+        inlineEnum = enumOfSimpleType(listNode.simpleType);
+        const pats = patternOfSimpleType(listNode.simpleType);
+        inlineAllowsDefine = pats ? pats.some(patternAllowsDefine) : false;
+      }
+      const desc = {
+        kind: "simple",
+        name,
+        refType: itemDesc?.refType ?? null,
+        enumValues: itemDesc?.enumValues?.length ? itemDesc.enumValues : inlineEnum,
+        isList: true,
+        allowsDefine: (itemDesc?.allowsDefine ?? false) || inlineAllowsDefine,
+        isRef: itemDesc?.isRef ?? false,
+        isBoolean: false,
+        base: itemDesc?.base ?? null,
+        doc: docOf(st),
+      };
+      memo.set(name, desc);
+      return desc;
+    }
     const base = normalizeTypeName(st?.restriction?.["@_base"]);
     const baseDesc = base ? resolveTypeDescriptor(base, memo) : null;
     const enumValues = enumOfSimpleType(st);
@@ -192,6 +224,7 @@ function resolveTypeDescriptor(typeName, memo = new Map()) {
       name,
       refType: normalizeTypeName(refType) || baseDesc?.refType || null,
       enumValues: enumValues.length ? enumValues : baseDesc?.enumValues ?? [],
+      isList: false,
       allowsDefine:
         (patterns ? patterns.some(patternAllowsDefine) : false) ||
         baseDesc?.allowsDefine ||
@@ -215,6 +248,7 @@ function resolveTypeDescriptor(typeName, memo = new Map()) {
       name,
       refType: null,
       enumValues: [],
+      isList: false,
       allowsDefine: false,
       isRef: false,
       isBoolean: false,
@@ -230,6 +264,7 @@ function resolveTypeDescriptor(typeName, memo = new Map()) {
     name,
     refType: null,
     enumValues: [],
+    isList: false,
     allowsDefine: false,
     isRef: false,
     isBoolean: false,
@@ -287,6 +322,7 @@ function collectAttributes(node, out = []) {
         name: `@attr:${name}`,
         refType: attr.simpleType?.["@_refType"] ?? null,
         enumValues: inlineEnum,
+        isList: false,
         allowsDefine: pats ? pats.some(patternAllowsDefine) : false,
         isBoolean: false,
         base: normalizeTypeName(attr.simpleType?.restriction?.["@_base"]),
@@ -300,8 +336,15 @@ function collectAttributes(node, out = []) {
       doc: docOf(attr),
       kind: desc?.kind ?? "unknown",
       type: desc?.name ?? null,
-      refType: normalizeTypeName(desc?.refType) ?? null,
+      // xas:refType may be declared on the attribute itself (e.g.
+      // <xs:attribute name="id" type="Poid" xas:refType="ModuleData" />) or
+      // on its simple type. The attribute-level declaration wins.
+      refType:
+        normalizeTypeName(attr["@_refType"]) ??
+        normalizeTypeName(desc?.refType) ??
+        null,
       enumValues: desc?.enumValues ?? inlineEnum ?? [],
+      isList: desc?.isList ?? false,
       allowsDefine: desc?.allowsDefine ?? false,
       isRef: desc?.isRef ?? false,
       isBoolean: desc?.isBoolean ?? false,
@@ -390,6 +433,7 @@ for (const [name, node] of simpleTypes) {
     refType: desc?.refType ?? node?.["@_refType"] ?? null,
     isRef: node?.["@_isRef"] === "true" || desc?.refType != null,
     enumValues: desc?.enumValues ?? [],
+    isList: desc?.isList ?? false,
     allowsDefine: desc?.allowsDefine ?? false,
     doc: docOf(node),
   };