fix(scripts): resolve token-benchmark.ts --perf imports to real nested module paths

runPerfBenchmarks and its callers (buildCodegraph, buildSessionOptions)
dynamically imported/spawned codegraph's own modules via flat paths
(src/builder.js, src/queries.js, src/native.js, src/parser.js,
src/cli.js) that never existed — src/ has no compiled .js siblings, and
the real modules live nested (domain/graph/builder.ts, domain/queries.ts,
infrastructure/native.ts, domain/parser.ts), mirroring dist/'s layout.

Fix the four --perf imports by reusing srcImport() from
scripts/lib/bench-config.ts — the same .ts-preferring resolution the
other benchmark scripts (benchmark.ts, query-benchmark.ts,
incremental-benchmark.ts, resolution-benchmark.ts) already rely on for
this exact problem.

The CLI-spawning cliPath had the same defect plus a second one: even
with a correct path, a spawned child `node` process doesn't inherit the
parent's --experimental-strip-types/--import <loader> flags needed to
run TypeScript source directly. Extract the fix into a new
scripts/lib/cli-invocation.ts (mirroring tests/integration/cli.test.ts's
NODE_TS_FLAGS pattern) so both call sites share one correct
implementation and it can be unit-tested directly.

Impact: 4 functions changed, 5 affected

Author: carlos-alm

scripts/lib/cli-invocation.ts

@@ -0,0 +1,26 @@
+/**
+ * Resolve `node` argv for spawning the codegraph CLI as a subprocess.
+ *
+ * Benchmark scripts (`token-benchmark.ts`) need to run the actual
+ * `codegraph` CLI (`build`, `mcp`) against a target directory, not just
+ * import its programmatic API. This repo ships TypeScript source with no
+ * compiled `dist/` required for local development, so `node <root>/src/cli.js`
+ * fails outright — only `cli.ts` exists (#1907). Even after fixing the
+ * extension, a spawned child process doesn't inherit the parent's
+ * `--experimental-strip-types` / `--import <loader>` flags the way
+ * statically- or dynamically-imported modules in the parent process do, so
+ * those flags must be passed explicitly. Mirrors the pattern already used by
+ * `tests/integration/cli.test.ts`'s `NODE_TS_FLAGS`.
+ */
+import path from 'node:path';
+
+/**
+ * @param root Absolute path to the codegraph repo root (the directory
+ *   containing `src/`, not `src/` itself).
+ * @returns argv (excluding `node` itself) to spawn `src/cli.ts` with the
+ *   flags required to run TypeScript source directly.
+ */
+export function resolveCliNodeArgs(root: string): string[] {
+	const loaderUrl = new URL('../ts-resolve-loader.ts', import.meta.url).href;
+	return ['--experimental-strip-types', '--import', loaderUrl, path.join(root, 'src', 'cli.ts')];
+}

scripts/token-benchmark.ts

@@ -27,14 +27,23 @@ import { parseArgs } from 'node:util';
 
 import { ISSUES, extractAgentOutput, validateResult } from './token-benchmark-issues.js';
 import { getBenchmarkVersion } from './bench-version.js';
+import { srcImport } from './lib/bench-config.js';
 import { median, round1, timeMedian } from './lib/bench-timing.js';
+import { resolveCliNodeArgs } from './lib/cli-invocation.js';
 import { extractUsageMetrics, tallyToolCalls } from './lib/session-metrics.js';
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 const root = path.resolve(__dirname, '..');
 const pkg = JSON.parse(fs.readFileSync(path.join(root, 'package.json'), 'utf8'));
 const benchVersion = getBenchmarkVersion(pkg.version, root);
 
+// Codegraph's own module tree lives nested under src/ (mirrors dist/'s
+// layout), not flat — see scripts/lib/bench-config.ts's srcImport() for the
+// same .ts-preferring resolution used by the other benchmark scripts.
+const SRC_DIR = path.join(root, 'src');
+
+const CLI_NODE_ARGS = resolveCliNodeArgs(root);
+
 // Redirect console.log to stderr so only JSON goes to stdout
 const origLog = console.log;
 console.log = (...args) => console.error(...args);
@@ -153,10 +162,9 @@ function checkoutCommit(nextjsDir, sha) {
 // ── Graph building ────────────────────────────────────────────────────────
 
 async function buildCodegraph(nextjsDir) {
-	const cliPath = path.join(root, 'src', 'cli.js');
 	console.error('Building codegraph graph for Next.js...');
 	const start = performance.now();
-	execFileSync('node', [cliPath, 'build', nextjsDir], {
+	execFileSync('node', [...CLI_NODE_ARGS, 'build', nextjsDir], {
 		cwd: nextjsDir,
 		stdio: 'pipe',
 		timeout: 600_000, // 10 min
@@ -189,12 +197,11 @@ function buildSessionOptions(mode, nextjsDir) {
 
 	if (mode === 'codegraph') {
 		const dbPath = path.join(nextjsDir, '.codegraph', 'graph.db');
-		const cliPath = path.join(root, 'src', 'cli.js');
 		options.mcpServers = {
 			codegraph: {
 				type: 'stdio',
 				command: 'node',
-				args: [cliPath, 'mcp', '-d', dbPath],
+				args: [...CLI_NODE_ARGS, 'mcp', '-d', dbPath],
 			},
 		};
 	}
@@ -310,19 +317,12 @@ async function runQueryBenchmarks(hubName, dbPath, fnDepsData, fnImpactData) {
  * Reuses the same codegraph APIs as the existing benchmark scripts.
  */
 async function runPerfBenchmarks(nextjsDir) {
-	const { pathToFileURL } = await import('node:url');
-	const { buildGraph } = await import(
-		pathToFileURL(path.join(root, 'src', 'builder.js')).href
-	);
+	const { buildGraph } = await import(srcImport(SRC_DIR, 'domain/graph/builder.js'));
 	const { fnDepsData, fnImpactData, statsData } = await import(
-		pathToFileURL(path.join(root, 'src', 'queries.js')).href
-	);
-	const { isNativeAvailable } = await import(
-		pathToFileURL(path.join(root, 'src', 'native.js')).href
-	);
-	const { isWasmAvailable } = await import(
-		pathToFileURL(path.join(root, 'src', 'parser.js')).href
+		srcImport(SRC_DIR, 'domain/queries.js')
 	);
+	const { isNativeAvailable } = await import(srcImport(SRC_DIR, 'infrastructure/native.js'));
+	const { isWasmAvailable } = await import(srcImport(SRC_DIR, 'domain/parser.js'));
 
 	const dbPath = path.join(nextjsDir, '.codegraph', 'graph.db');
 

tests/unit/bench-config-src-import.test.ts

@@ -0,0 +1,44 @@
+/**
+ * Unit tests for scripts/lib/bench-config.ts's srcImport().
+ *
+ * Regression coverage for #1907: scripts/token-benchmark.ts's `--perf`
+ * dynamic imports used flat paths (e.g. `src/builder.js`, `src/queries.js`,
+ * `src/native.js`, `src/parser.js`) that never existed — the real modules
+ * live nested (`src/domain/graph/builder.ts`, `src/domain/queries.ts`,
+ * `src/infrastructure/native.ts`, `src/domain/parser.ts`), mirroring dist/'s
+ * layout. This locks in that the exact nested subpaths token-benchmark.ts
+ * now imports via srcImport() resolve to real, importable modules exporting
+ * the symbols runPerfBenchmarks() needs.
+ */
+
+import path from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { describe, expect, it } from 'vitest';
+import { srcImport } from '../../scripts/lib/bench-config.js';
+
+const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..', '..');
+const srcDir = path.join(repoRoot, 'src');
+
+describe('srcImport() against the module paths used by token-benchmark.ts --perf', () => {
+  it('resolves domain/graph/builder.js to a module exporting buildGraph', async () => {
+    const mod = await import(srcImport(srcDir, 'domain/graph/builder.js'));
+    expect(typeof mod.buildGraph).toBe('function');
+  });
+
+  it('resolves domain/queries.js to a module exporting fnDepsData, fnImpactData, statsData', async () => {
+    const mod = await import(srcImport(srcDir, 'domain/queries.js'));
+    expect(typeof mod.fnDepsData).toBe('function');
+    expect(typeof mod.fnImpactData).toBe('function');
+    expect(typeof mod.statsData).toBe('function');
+  });
+
+  it('resolves infrastructure/native.js to a module exporting isNativeAvailable', async () => {
+    const mod = await import(srcImport(srcDir, 'infrastructure/native.js'));
+    expect(typeof mod.isNativeAvailable).toBe('function');
+  });
+
+  it('resolves domain/parser.js to a module exporting isWasmAvailable', async () => {
+    const mod = await import(srcImport(srcDir, 'domain/parser.js'));
+    expect(typeof mod.isWasmAvailable).toBe('function');
+  });
+});

tests/unit/cli-invocation.test.ts

@@ -0,0 +1,46 @@
+/**
+ * Unit tests for scripts/lib/cli-invocation.ts
+ *
+ * Regression coverage for #1907: scripts/token-benchmark.ts spawned the
+ * codegraph CLI via `path.join(root, 'src', 'cli.js')` — a file that never
+ * existed (src/ has no compiled .js output; only src/cli.ts) — and, even
+ * after fixing the extension, a spawned child `node` process doesn't
+ * inherit the --experimental-strip-types / --import <loader> flags the
+ * parent script itself needs to run TypeScript source. resolveCliNodeArgs()
+ * centralizes the correct invocation so it can be exercised directly.
+ */
+
+import { execFileSync } from 'node:child_process';
+import fs from 'node:fs';
+import path from 'node:path';
+import { fileURLToPath } from 'node:url';
+import { describe, expect, it } from 'vitest';
+import { resolveCliNodeArgs } from '../../scripts/lib/cli-invocation.js';
+
+const repoRoot = path.resolve(path.dirname(fileURLToPath(import.meta.url)), '..', '..');
+
+describe('resolveCliNodeArgs', () => {
+  it('resolves the strip-types flag, the loader, and an existing cli.ts entry point', () => {
+    const args = resolveCliNodeArgs(repoRoot);
+
+    expect(args[0]).toBe('--experimental-strip-types');
+    expect(args[1]).toBe('--import');
+    expect(args[2]).toMatch(/^file:.*ts-resolve-loader\.ts$/);
+    expect(args[3]).toBe(path.join(repoRoot, 'src', 'cli.ts'));
+
+    // Both the loader and the CLI entry point must exist on disk — this is
+    // exactly the class of bug #1907 fixed (paths pointing at files that
+    // were never there).
+    expect(fs.existsSync(new URL(args[2]))).toBe(true);
+    expect(fs.existsSync(args[3])).toBe(true);
+  });
+
+  it('spawns the real CLI successfully via the resolved argv (mirrors tests/integration/cli.test.ts)', () => {
+    const args = resolveCliNodeArgs(repoRoot);
+    const out = execFileSync('node', [...args, '--version'], {
+      encoding: 'utf-8',
+      timeout: 30_000,
+    });
+    expect(out.trim()).toMatch(/^\d+\.\d+\.\d+/);
+  });
+});