fix(neo4j): write content_hash, so the incremental push actually diffs

src/build/neo4j/project.ts

@@ -233,10 +233,10 @@ function projectField(b: RowBuilder, f: TSField, owner: NodeRef, fileKey: string
 // ----------------------------------------------------------------------------------------------
 
 function moduleProps(mod: TSModule, fileKey: string): Props {
-  // The wire strips the internal fields (module_name, content_hash) — the graph name is the
-  // file key, exactly as the historical projection of the stripped tree produced.
+  // `name` is the file key. `content_hash` is what the incremental push diffs against on the next
+  // run (bolt.ts) -- schema.ts has always declared it on :TSModule, but nothing wrote it (#118).
   return prune({
-    id: mod.id, kind: "module", name: fileKey,
+    id: mod.id, kind: "module", name: fileKey, content_hash: mod.content_hash ?? null,
     is_tsx: mod.is_tsx, is_declaration_file: mod.is_declaration_file,
     ...span(mod), _module: fileKey,
   });

src/schema/emit.ts

@@ -58,7 +58,10 @@ function stripInternal(root: TSApplication): void {
     for (const nt of Object.values((t["types"] as Record<string, Record<string, unknown>>) ?? {})) stripType(nt);
   };
   for (const mod of Object.values(root.symbol_table) as unknown as Record<string, unknown>[]) {
-    delete mod["content_hash"];
+    // `content_hash` STAYS on the wire (#118). The Neo4j projection is built from this envelope
+    // and writes it onto :TSModule, where the incremental push reads it back to find changed
+    // modules -- stripping it here made that diff compare against NULL forever, so every push was
+    // a full re-upsert. codeanalyzer-python keeps it for the same reason (schema/py_schema.py).
     delete mod["last_modified"];
     delete mod["file_size"];
     for (const fn of Object.values((mod["functions"] as Record<string, Record<string, unknown>>) ?? {})) stripCallable(fn);

test/neo4j-schema.test.ts

@@ -122,6 +122,22 @@ describe("neo4j schema conformance", () => {
 // The JSON envelope advertises `analyzer{name,version}` (#29); the Neo4j :Application node is the
 // co-primary projection of the same envelope and must not diverge on analyzer identity.
 
+// #118: the incremental push diffs each module's stored `content_hash` to find what changed. The
+// field was stripped from the wire and never projected, so the diff compared against NULL forever
+// and every push was a full re-upsert. schema.ts declared the property, bolt.ts read it, and
+// project.ts never wrote it -- three parts of the system disagreeing. The old bolt test seeded the
+// value by hand, so it exercised the diff against data the projection could not produce.
+describe("module content_hash reaches the graph (#118)", () => {
+  test("every projected :TSModule carries a non-null content_hash", () => {
+    const modules = rows.nodes.filter((n) => n.labels.includes("TSModule"));
+    expect(modules.length).toBeGreaterThan(0);
+    for (const m of modules) {
+      expect(m.props["content_hash"], `no content_hash on ${m.value}`).toBeDefined();
+      expect(typeof m.props["content_hash"]).toBe("string");
+    }
+  });
+});
+
 describe(":Application node carries analyzer identity (issue #43)", () => {
   test("version matches package.json (the same source the JSON envelope's analyzer.version uses)", () => {
     const appNode = rows.nodes.find((n) => n.labels.includes("Application"));