Enable auto-indexing for nested field paths (#728)

* fix: enable auto-indexing for nested field paths

This fix allows auto-indexes to be created for nested field paths
(e.g., `profile.score`, `metadata.stats.views`), not just top-level
fields. This resolves performance issues where queries with `eq()`,
`gt()`, etc. on nested fields were forced to do full table scans
instead of using indexes.

Changes:
- Remove the `fieldPath.length !== 1` restriction in `extractIndexableExpressions()`
- Update `ensureIndexForField()` to properly traverse nested paths when creating index accessors
- Add comprehensive tests for nested path auto-indexing with 1, 2, and 3-level nesting
- Verify that nested path indexes are properly used by the query optimizer

Fixes #727

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

* fix: use colon-prefixed naming for auto-indexes to avoid conflicts

Change auto-index naming from 'auto_field_path' to 'auto:field.path'
to prevent ambiguity between nested paths and fields with underscores.

Examples:
- user.profile -> auto:user.profile
- user_profile -> auto:user_profile
(no conflict!)

Co-authored-by: Sam Willis <sam.willis@gmail.com>

* chore: add changeset for nested auto-index fix

* style: format changeset with prettier

---------

Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Sam Willis <sam.willis@gmail.com>

Author: 3 people

.changeset/enable-nested-auto-index.md

@@ -0,0 +1,50 @@
+---
+"@tanstack/db": patch
+---
+
+Enable auto-indexing for nested field paths
+
+Previously, auto-indexes were only created for top-level fields. Queries filtering on nested fields like `vehicleDispatch.date` or `profile.score` were forced to perform full table scans, causing significant performance issues.
+
+Now, auto-indexes are automatically created for nested field paths of any depth when using `eq()`, `gt()`, `gte()`, `lt()`, `lte()`, or `in()` operations.
+
+**Performance Impact:**
+
+Before this fix, filtering on nested fields resulted in expensive full scans:
+
+- Query time: ~353ms for 39 executions (from issue #727)
+- "graph run" and "d2ts join" operations dominated execution time
+
+After this fix, nested field queries use indexes:
+
+- Query time: Sub-millisecond (typical indexed lookup)
+- Proper index utilization verified through query optimizer
+
+**Example:**
+
+```typescript
+const collection = createCollection({
+  getKey: (item) => item.id,
+  autoIndex: "eager", // default
+  // ... sync config
+})
+
+// These now automatically create and use indexes:
+collection.subscribeChanges((items) => console.log(items), {
+  whereExpression: eq(row.vehicleDispatch?.date, "2024-01-01"),
+})
+
+collection.subscribeChanges((items) => console.log(items), {
+  whereExpression: gt(row.profile?.stats.rating, 4.5),
+})
+```
+
+**Index Naming:**
+
+Auto-indexes for nested paths use the format `auto:field.path` to avoid naming conflicts:
+
+- `auto:status` for top-level field `status`
+- `auto:profile.score` for nested field `profile.score`
+- `auto:metadata.stats.views` for deeply nested field `metadata.stats.views`
+
+Fixes #727

packages/db/src/indexes/auto-index.ts

@@ -44,14 +44,25 @@ export function ensureIndexForField<
 
   // Create a new index for this field using the collection's createIndex method
   try {
-    collection.createIndex((row) => (row as any)[fieldName], {
-      name: `auto_${fieldName}`,
-      indexType: BTreeIndex,
-      options: compareFn ? { compareFn, compareOptions } : {},
-    })
+    // Use the proxy-based approach to create the proper accessor for nested paths
+    collection.createIndex(
+      (row) => {
+        // Navigate through the field path
+        let current: any = row
+        for (const part of fieldPath) {
+          current = current[part]
+        }
+        return current
+      },
+      {
+        name: `auto:${fieldPath.join(`.`)}`,
+        indexType: BTreeIndex,
+        options: compareFn ? { compareFn, compareOptions } : {},
+      }
+    )
   } catch (error) {
     console.warn(
-      `${collection.id ? `[${collection.id}] ` : ``}Failed to create auto-index for field "${fieldName}":`,
+      `${collection.id ? `[${collection.id}] ` : ``}Failed to create auto-index for field path "${fieldPath.join(`.`)}":`,
       error
     )
   }
@@ -108,20 +119,22 @@ function extractIndexableExpressions(
       return
     }
 
-    // Check if the first argument is a property reference (single field)
+    // Check if the first argument is a property reference
     if (func.args.length < 1 || func.args[0].type !== `ref`) {
       return
     }
 
     const fieldRef = func.args[0]
     const fieldPath = fieldRef.path
 
-    // Skip if it's not a simple field (e.g., nested properties or array access)
-    if (fieldPath.length !== 1) {
+    // Skip if the path is empty
+    if (fieldPath.length === 0) {
       return
     }
 
-    const fieldName = fieldPath[0]
+    // For nested paths, use the full path joined with underscores as the field name
+    // For simple paths, use the first (and only) element
+    const fieldName = fieldPath.join(`_`)
     results.push({ fieldName, fieldPath })
   }
 

packages/db/tests/collection-auto-index.test.ts

@@ -750,4 +750,148 @@ describe(`Collection Auto-Indexing`, () => {
 
     subscription.unsubscribe()
   })
+
+  it(`should create auto-indexes for nested field paths`, async () => {
+    interface NestedTestItem {
+      id: string
+      name: string
+      profile?: {
+        score: number
+        bio: string
+      }
+      metadata?: {
+        tags: Array<string>
+        stats: {
+          views: number
+          likes: number
+        }
+      }
+    }
+
+    const nestedTestData: Array<NestedTestItem> = [
+      {
+        id: `1`,
+        name: `Alice`,
+        profile: { score: 85, bio: `Developer` },
+        metadata: {
+          tags: [`tech`, `coding`],
+          stats: { views: 100, likes: 50 },
+        },
+      },
+      {
+        id: `2`,
+        name: `Bob`,
+        profile: { score: 92, bio: `Designer` },
+        metadata: {
+          tags: [`design`, `ui`],
+          stats: { views: 200, likes: 75 },
+        },
+      },
+      {
+        id: `3`,
+        name: `Charlie`,
+        profile: { score: 78, bio: `Manager` },
+        metadata: {
+          tags: [`management`, `leadership`],
+          stats: { views: 150, likes: 60 },
+        },
+      },
+    ]
+
+    const collection = createCollection<NestedTestItem, string>({
+      getKey: (item) => item.id,
+      autoIndex: `eager`,
+      startSync: true,
+      sync: {
+        sync: ({ begin, write, commit, markReady }) => {
+          begin()
+          for (const item of nestedTestData) {
+            write({
+              type: `insert`,
+              value: item,
+            })
+          }
+          commit()
+          markReady()
+        },
+      },
+    })
+
+    await collection.stateWhenReady()
+
+    // Should have no indexes initially
+    expect(collection.indexes.size).toBe(0)
+
+    // Test 1: Nested field one level deep (profile.score)
+    const changes1: Array<any> = []
+    const subscription1 = collection.subscribeChanges(
+      (items) => {
+        changes1.push(...items)
+      },
+      {
+        includeInitialState: true,
+        whereExpression: gt(new PropRef([`profile`, `score`]), 80),
+      }
+    )
+
+    // Should have created an auto-index for profile.score
+    const profileScoreIndex = Array.from(collection.indexes.values()).find(
+      (index) =>
+        index.expression.type === `ref` &&
+        (index.expression as any).path.length === 2 &&
+        (index.expression as any).path[0] === `profile` &&
+        (index.expression as any).path[1] === `score`
+    )
+    expect(profileScoreIndex).toBeDefined()
+
+    // Verify the filtered results are correct
+    expect(changes1.filter((c) => c.type === `insert`).length).toBe(2) // Alice (85) and Bob (92)
+
+    subscription1.unsubscribe()
+
+    // Test 2: Deeply nested field (metadata.stats.views)
+    const changes2: Array<any> = []
+    const subscription2 = collection.subscribeChanges(
+      (items) => {
+        changes2.push(...items)
+      },
+      {
+        includeInitialState: true,
+        whereExpression: eq(new PropRef([`metadata`, `stats`, `views`]), 200),
+      }
+    )
+
+    // Should have created an auto-index for metadata.stats.views
+    const viewsIndex = Array.from(collection.indexes.values()).find(
+      (index) =>
+        index.expression.type === `ref` &&
+        (index.expression as any).path.length === 3 &&
+        (index.expression as any).path[0] === `metadata` &&
+        (index.expression as any).path[1] === `stats` &&
+        (index.expression as any).path[2] === `views`
+    )
+    expect(viewsIndex).toBeDefined()
+
+    // Verify the filtered results are correct
+    expect(changes2.filter((c) => c.type === `insert`).length).toBe(1) // Only Bob has 200 views
+
+    subscription2.unsubscribe()
+
+    // Test 3: Index usage verification with tracker
+    withIndexTracking(collection, (tracker) => {
+      const result = collection.currentStateAsChanges({
+        where: gt(new PropRef([`profile`, `score`]), 80),
+      })!
+
+      expect(result.length).toBe(2) // Alice and Bob
+
+      // Verify it used the auto-created index
+      expectIndexUsage(tracker.stats, {
+        shouldUseIndex: true,
+        shouldUseFullScan: false,
+        indexCallCount: 1,
+        fullScanCallCount: 0,
+      })
+    })
+  })
 })