fix(resolution): filter stale-target edges so watch sync survives FK violations (#455)

CHANGELOG.md

@@ -10,6 +10,7 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 ## [Unreleased]
 
 ### Fixed
+- **Watch sync no longer aborts with `FOREIGN KEY constraint failed`.** PR #62 plugged this FK violation at the extraction-layer `insertEdges` site (empty-named nodes whose containment edges had no target), but the same violation kept reappearing on v0.9.5 during the daemon's *watch sync* — not on initial index — once an agent's daemon had been running long enough to accumulate edits. The resolution-layer `insertEdges` (and the callback-synthesizer pass) wasn't guarded the same way: a per-resolver name cache or a framework resolver's WeakMap-keyed lookup map could hand back a `Node` whose row had been removed by a recent file rewrite, and the FK check then aborted the entire resolution batch, leaving the user's daemon log filling with `Watch sync failed { error: 'FOREIGN KEY constraint failed' }`. The resolution layer now applies the same defense-in-depth filter the extraction layer already had — one cache-aware `getNodesByIds` per pass drops any edge whose source or target is no longer in the `nodes` table, so the rest of the resolved batch still lands. Surfaces as a fresh `codegraph init`/`index` cycle now surviving its first watch-sync cycle without the FK error, and the daemon recovering naturally instead of compounding into further failures. Closes #455.
 - **Hermes Agent: `codegraph install --target hermes` no longer corrupts `~/.hermes/config.yaml`.** Hermes serializes its config with PyYAML's default block style, which writes list items at the *same* indent as the parent mapping key (`cli:` and `- hermes-cli` both at column 2). The previous line-based YAML patcher mistook that first `  - hermes-cli` for the next sibling key, truncated the `cli:` block, and then spliced `- mcp-codegraph` at indent 4 *before* the existing items — leaving subsequent entries (`- browser`, `- clarify`, …) and even other platforms (`telegram:`, `discord:`) appearing at the `platform_toolsets:` level, which is no longer parseable YAML. The installer now recognizes the same-indent list style, finds the real end of the block at the next sibling key, and appends `- mcp-codegraph` at whatever indent the existing items already use. Re-installing on an already-corrupted file (or a 4-space-nested config that worked before) still produces a clean, parseable result. Closes #456.
 - **NestJS: `RouterModule.register([...])` route prefixes now propagate to controller routes.** Previously a controller declared inside a module wired through NestJS's `RouterModule` (a common pattern for modular apps with nested route prefixes) was indexed with its raw `@Controller(...) + @Get(...)` path — so `UsersController` under `RouterModule.register([{ path: 'admin', module: AdminModule, children: [{ path: 'users', module: UsersModule }] }])` showed up as `GET /` instead of `GET /admin/users`. The new cross-file pass walks every `*.module.{ts,js}` for `RouterModule.register/forRoot/forChild([...])` (recursive `children`) and `@Module({ controllers: [...] })`, then prepends the correct prefix to each affected route — including non-empty `@Controller` paths and method-level params (`/admin/users/:id`). The route node's `id` is preserved across the update so existing route→handler edges stay intact, and the pass is idempotent so incremental sync recovers when `app.module.ts` itself is edited. Closes #459.
 

__tests__/sync-fk-regression.test.ts

@@ -0,0 +1,232 @@
+/**
+ * Sync FK regression test (issue #455).
+ *
+ * #62 plugged FK violations at the extraction layer (empty-named nodes whose
+ * containment edges had no target). #455 reports the same `FOREIGN KEY constraint
+ * failed` reappearing on v0.9.5, but during *watch sync* on a Python-only project —
+ * a different trigger than the C/C++ header empty-name issue #62 covered.
+ *
+ * The reproducer below drives the same path the daemon takes: extract → resolve →
+ * insert edges. The resolution pass's `insertEdges` was not guarded the way the
+ * extraction-layer insert was after #62, so any edge with a stale source/target
+ * (e.g. a synthesized framework target whose node was deleted by a concurrent
+ * file rewrite) throws and aborts the sync, leaving the FK error the user sees.
+ *
+ * The test asserts: a sequence of file rewrites + sync()s never throws, and the
+ * graph stays internally consistent (every edge's source + target are real nodes).
+ */
+import { describe, it, expect, beforeAll, afterEach } from 'vitest';
+import * as fs from 'fs';
+import * as path from 'path';
+import * as os from 'os';
+import CodeGraph from '../src/index';
+import { initGrammars, loadAllGrammars } from '../src/extraction/grammars';
+
+beforeAll(async () => {
+  await initGrammars();
+  await loadAllGrammars();
+});
+
+describe('watch sync FK regression (#455)', () => {
+  let tmpDir: string | undefined;
+  let cg: CodeGraph | undefined;
+
+  afterEach(() => {
+    if (cg) {
+      cg.close();
+      cg = undefined;
+    }
+    if (tmpDir) {
+      fs.rmSync(tmpDir, { recursive: true, force: true });
+      tmpDir = undefined;
+    }
+  });
+
+  function assertGraphIntegrity(cg: CodeGraph): void {
+    // Every edge must reference real nodes. If FK was disabled or violated,
+    // dangling refs would show up here.
+    const db = (cg as unknown as { db: { getDb(): { prepare(sql: string): { get(): unknown } } } }).db;
+    const sqlite = db.getDb();
+    const dangling = sqlite
+      .prepare(
+        `SELECT count(*) as c FROM edges e
+         WHERE NOT EXISTS (SELECT 1 FROM nodes n WHERE n.id = e.source)
+            OR NOT EXISTS (SELECT 1 FROM nodes n WHERE n.id = e.target)`
+      )
+      .get() as { c: number };
+    expect(dangling.c).toBe(0);
+  }
+
+  it('survives repeated sync() cycles on a Django-style Python project without FK errors', async () => {
+    tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'cg-fk455-'));
+
+    // Mimic a small Django app: requirements + manage.py marker, models/views/urls
+    // in two app packages that cross-reference each other.
+    fs.writeFileSync(path.join(tmpDir, 'manage.py'), '# django marker\n');
+    fs.writeFileSync(path.join(tmpDir, 'requirements.txt'), 'django==4.2\n');
+
+    fs.mkdirSync(path.join(tmpDir, 'users'));
+    fs.writeFileSync(path.join(tmpDir, 'users/__init__.py'), '');
+    fs.writeFileSync(
+      path.join(tmpDir, 'users/models.py'),
+      'class User:\n' +
+        '    def __init__(self, name):\n' +
+        '        self.name = name\n'
+    );
+    fs.writeFileSync(
+      path.join(tmpDir, 'users/views.py'),
+      'from users.models import User\n' +
+        'class UserListView:\n' +
+        '    def get(self, request):\n' +
+        '        return User("a")\n'
+    );
+    fs.writeFileSync(
+      path.join(tmpDir, 'users/urls.py'),
+      'from django.urls import path\n' +
+        'from users.views import UserListView\n' +
+        'urlpatterns = [path("users/", UserListView.as_view(), name="user-list")]\n'
+    );
+
+    fs.mkdirSync(path.join(tmpDir, 'posts'));
+    fs.writeFileSync(path.join(tmpDir, 'posts/__init__.py'), '');
+    fs.writeFileSync(
+      path.join(tmpDir, 'posts/models.py'),
+      'from users.models import User\n' +
+        'class Post:\n' +
+        '    def __init__(self, author):\n' +
+        '        self.author = author\n'
+    );
+    fs.writeFileSync(
+      path.join(tmpDir, 'posts/views.py'),
+      'from posts.models import Post\n' +
+        'class PostListView:\n' +
+        '    def get(self, request):\n' +
+        '        return Post(None)\n'
+    );
+    fs.writeFileSync(
+      path.join(tmpDir, 'posts/urls.py'),
+      'from django.urls import path\n' +
+        'from posts.views import PostListView\n' +
+        'urlpatterns = [path("posts/", PostListView.as_view(), name="post-list")]\n'
+    );
+
+    cg = CodeGraph.initSync(tmpDir);
+    await cg.indexAll();
+    assertGraphIntegrity(cg);
+
+    // Drive the same path the daemon's file watcher drives: a series of file
+    // rewrites + sync()s. We shuffle line counts on each rewrite so node IDs
+    // (file:kind:name:line) shift around, forcing real INSERT OR REPLACE +
+    // CASCADE behavior across files that cross-reference each other.
+    const targets = [
+      'users/views.py',
+      'posts/views.py',
+      'users/urls.py',
+      'posts/urls.py',
+      'users/models.py',
+    ];
+
+    for (let iter = 0; iter < 8; iter++) {
+      const file = targets[iter % targets.length]!;
+      const full = path.join(tmpDir, file);
+      const content = fs.readFileSync(full, 'utf8');
+      // Insert N blank lines at the top to shift every node's line number.
+      const padded = '\n'.repeat(iter + 1) + content;
+      // Use a future mtime so the size+mtime pre-filter in
+      // ExtractionOrchestrator.sync can't skip the file.
+      fs.writeFileSync(full, padded);
+      const now = Date.now() + (iter + 1) * 1_000;
+      fs.utimesSync(full, now / 1000, now / 1000);
+
+      // The fix should make this never throw; before the fix, FK errors fire
+      // during the resolution-layer insertEdges call inside sync().
+      await expect(cg.sync()).resolves.toBeDefined();
+      assertGraphIntegrity(cg);
+    }
+  });
+
+  it("drops resolution edges whose target node is no longer in the graph (the pathology #455 reports)", async () => {
+    // This narrower test reproduces the exact failure mode the user sees in
+    // their daemon log: the resolver hands `insertEdges` an edge whose target
+    // doesn't exist in `nodes`, and the FK constraint aborts the whole sync.
+    //
+    // We force the bug by populating the resolver's per-name cache with a
+    // stale node (whose id is *not* in the DB) and then asking it to resolve
+    // a reference to that name. Without the fix this throws
+    // `FOREIGN KEY constraint failed`; with it, the bad edge is filtered out
+    // and resolution returns normally.
+    tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'cg-fk455-stale-'));
+    fs.writeFileSync(path.join(tmpDir, 'a.py'), 'def caller():\n    Target()\n');
+    fs.writeFileSync(path.join(tmpDir, 'b.py'), 'class Target:\n    pass\n');
+
+    cg = CodeGraph.initSync(tmpDir);
+    await cg.indexAll();
+
+    // Reach in to the internals — the simplest way to forge the "stale node
+    // ID in the resolver's lookup path" condition the production bug arises
+    // from. The fix is what the test is verifying; touching internals here
+    // is a means to that end, not a contract we're asserting.
+    type Internals = {
+      queries: {
+        getNodesByName(name: string): Array<{ id: string; name: string }>;
+        getAllNodeNames(): string[];
+      };
+      resolver: {
+        warmCaches(): void;
+        resolveAndPersist(
+          refs: Array<{
+            fromNodeId: string;
+            referenceName: string;
+            referenceKind: string;
+            line: number;
+            column: number;
+            filePath: string;
+            language: string;
+          }>
+        ): { resolved: unknown[] };
+        // eslint-disable-next-line @typescript-eslint/no-explicit-any
+        nameCache: { set(key: string, value: any): void };
+      };
+    };
+    const internals = cg as unknown as Internals;
+    const queries = internals.queries;
+    const resolver = internals.resolver;
+
+    const caller = queries.getNodesByName('caller')[0];
+    const target = queries.getNodesByName('Target')[0];
+    expect(caller).toBeDefined();
+    expect(target).toBeDefined();
+
+    // Warm caches so warmCaches no-ops on the resolveAndPersist call below
+    // and our seeded nameCache entry isn't overwritten.
+    resolver.warmCaches();
+
+    // Forge a stale lookup result: a Node whose `id` doesn't exist in the
+    // `nodes` table. This is structurally what happens when a framework
+    // resolver's WeakMap cache hands back a Node that was deleted by a
+    // concurrent file rewrite — the user's #455 scenario.
+    const staleNode = { ...target!, id: 'class:stale.py:Target:1' };
+    resolver.nameCache.set('Target', [staleNode]);
+
+    // Ask the resolver to persist an edge that will resolve via the seeded
+    // (stale) cache entry. Without the FK filter this would throw
+    // `FOREIGN KEY constraint failed` and abort the whole batch.
+    expect(() =>
+      resolver.resolveAndPersist([
+        {
+          fromNodeId: caller!.id,
+          referenceName: 'Target',
+          referenceKind: 'calls',
+          line: 2,
+          column: 4,
+          filePath: 'a.py',
+          language: 'python',
+        },
+      ])
+    ).not.toThrow();
+
+    // The bad edge must not have been persisted either — FK enforcement is
+    // still on, and post-fix the dangling-edge count remains zero.
+    assertGraphIntegrity(cg);
+  });
+});

src/resolution/callback-synthesizer.ts

@@ -812,6 +812,19 @@ export function synthesizeCallbackEdges(queries: QueryBuilder, ctx: ResolutionCo
     seen.add(key);
     merged.push(e);
   }
-  if (merged.length > 0) queries.insertEdges(merged);
-  return merged.length;
+  if (merged.length > 0) {
+    // Defense-in-depth (issues #42, #455): drop edges whose source/target
+    // no longer resolves to a real node. Some channel maps cache native
+    // Node refs across a resolver lifetime (WeakMap-keyed by context), so
+    // a file rewrite between map build and synthesis can leave stale IDs
+    // here. One FK violation aborts the whole batch — better to skip the
+    // dead edges and emit the rest than lose every synthesized edge.
+    const allIds = new Set<string>();
+    for (const e of merged) { allIds.add(e.source); allIds.add(e.target); }
+    const existing = queries.getNodesByIds([...allIds]);
+    const validEdges = merged.filter((e) => existing.has(e.source) && existing.has(e.target));
+    if (validEdges.length > 0) queries.insertEdges(validEdges);
+    return validEdges.length;
+  }
+  return 0;
 }

src/resolution/index.ts

@@ -607,6 +607,28 @@ export class ReferenceResolver {
     });
   }
 
+  /**
+   * Defense-in-depth: drop edges whose source or target is no longer in
+   * the nodes table. PR #62 (issue #42) applied this filter at the
+   * extraction-layer `insertEdges` site; #455 reports the same
+   * `FOREIGN KEY constraint failed` reappearing here at the
+   * resolution-layer site during watch sync, where a resolver lookup that
+   * crosses a framework-specific cache can hand us a target whose node
+   * was removed by a concurrent file rewrite. One batched, cache-aware
+   * `getNodesByIds` query is enough to skip those edges quietly instead
+   * of aborting the whole sync.
+   */
+  private filterEdgesByExistingNodes(edges: Edge[]): Edge[] {
+    if (edges.length === 0) return edges;
+    const allIds = new Set<string>();
+    for (const e of edges) {
+      allIds.add(e.source);
+      allIds.add(e.target);
+    }
+    const existing = this.queries.getNodesByIds([...allIds]);
+    return edges.filter((e) => existing.has(e.source) && existing.has(e.target));
+  }
+
   /**
    * Resolve and persist edges to database
    */
@@ -620,8 +642,9 @@ export class ReferenceResolver {
     const edges = this.createEdges(result.resolved);
 
     // Insert edges into database
-    if (edges.length > 0) {
-      this.queries.insertEdges(edges);
+    const validEdges = this.filterEdgesByExistingNodes(edges);
+    if (validEdges.length > 0) {
+      this.queries.insertEdges(validEdges);
     }
 
     // Clean up resolved refs from unresolved_refs table so metrics are accurate
@@ -668,8 +691,9 @@ export class ReferenceResolver {
 
       // Persist edges immediately
       const edges = this.createEdges(result.resolved);
-      if (edges.length > 0) {
-        this.queries.insertEdges(edges);
+      const validEdges = this.filterEdgesByExistingNodes(edges);
+      if (validEdges.length > 0) {
+        this.queries.insertEdges(validEdges);
       }
 
       // Clean up resolved refs so they don't appear in the next batch