fix(verify): put the browser acceptance records back when the run is over

packages/webui/test/integration/browser-use-real-input.test.ts

@@ -23,7 +23,7 @@ import { afterAll, beforeAll, describe, expect, it } from "vitest";
 
 import { WebuiIsolatedChromiumProvider } from "../../src/runtime/browser/isolated-chromium/provider.js";
 import type { WebuiBrowserTargetId } from "../../src/shared/contracts/browser.js";
-import { beginAcceptanceSuite, observeAcceptance } from "../browser/acceptance-record.js";
+import { beginAcceptanceSuite, observeAcceptance, restoreAcceptanceRecords } from "../browser/acceptance-record.js";
 
 /** The record file this suite writes. */
 const SUITE = "isolated-chromium";
@@ -105,6 +105,9 @@ describe.skipIf(!available)("the browser input path, against a real page", () =>
   }, 60_000);
 
   afterAll(async () => {
+    // This suite rewrites the record file, and the file is tracked, so on CI
+    // the run puts it back the way the checkout had it.
+    restoreAcceptanceRecords();
     await provider?.close();
     if (dataDir) await rm(dataDir, { recursive: true, force: true, maxRetries: 5, retryDelay: 100 });
   });

scripts/lib/acceptance-records.mjs

@@ -0,0 +1,112 @@
+// The browser acceptance records, and putting them back.
+//
+// A record file is tracked: the plan's `complete/` directory is where the
+// accepted evidence lives, and the file in the checkout is the record of the
+// run that was accepted. A run rewrites it — every record carries the time and
+// the machine it ran on — so the working tree is dirty once the suite is over.
+//
+// That dirt is not a source change, and the release packaging refuses to run
+// with a dirty tree ("Packaging requires a clean committed working tree"), so
+// a CI run has to put the records back before it is over.
+//
+// The restore is git-based, and it restores the committed content of every
+// record file. The suites that write a record live in their own processes and
+// each of them restores what it wrote from the copy it kept in memory
+// (`restoreAcceptanceRecords` in the acceptance-record module); this one does
+// not depend on which process wrote what, so a suite that forgot to restore
+// its own cannot leave the tree dirty for the packaging step.
+//
+// Locally the records stay: they are what the run just produced, and the
+// published matrix is written from them.
+
+import { spawnSync } from "node:child_process";
+import { realpathSync } from "node:fs";
+import path from "node:path";
+
+// The tracked acceptance record files, wherever the plans put them. A plan that
+// adds a suite gets its record file covered without this file knowing the
+// suite's name.
+const RECORD_PATHSPEC = ":(glob).plan/*/complete/*.acceptance.json";
+
+/**
+ * Whether this is a CI run.
+ *
+ * `CI` is the variable the suites already test, and `GITHUB_ACTIONS` is the
+ * one the runner always sets; either is enough to say the run's own records
+ * are throwaway and the committed ones are the evidence.
+ */
+function isCi() {
+  return process.env.CI === "true" || process.env.GITHUB_ACTIONS === "true";
+}
+
+/**
+ * The path as the filesystem sees it.
+ *
+ * Git answers with the real path, and a checkout reached through a symbolic
+ * link — a temporary directory on macOS, for one — is not the same string as
+ * the one the caller passed. Comparing the two as written would decide that
+ * the checkout is somewhere else, and nothing would be restored.
+ */
+function realPath(target) {
+  try {
+    return realpathSync.native(target);
+  } catch {
+    return path.resolve(target);
+  }
+}
+
+/**
+ * Whether `root` is the top of a git checkout.
+ *
+ * An exported source archive has no repository at all, and there is nothing to
+ * restore there. A directory that merely happens to sit inside some other
+ * repository is worse than that: git would answer for the wrong tree, so the
+ * top level has to be this tree before anything is restored.
+ */
+function isCheckoutRoot(root) {
+  const top = spawnSync("git", ["rev-parse", "--show-toplevel"], {
+    cwd: root,
+    encoding: "utf8",
+  });
+  if (top.status !== 0) return false;
+  return realPath(top.stdout.trim()) === realPath(root);
+}
+
+/**
+ * The acceptance record files this checkout tracks.
+ *
+ * Read from the index rather than from a list of suite names, so a plan that
+ * adds a suite is covered without this file being told about it.
+ */
+export function acceptanceRecordFiles(root) {
+  if (!isCheckoutRoot(root)) return [];
+  const listed = spawnSync("git", ["ls-files", "-z", "--", RECORD_PATHSPEC], {
+    cwd: root,
+    encoding: "utf8",
+  });
+  if (listed.status !== 0) return [];
+  return listed.stdout.split("\0").filter((file) => file !== "");
+}
+
+/**
+ * Puts the acceptance records back the way the checkout had them.
+ *
+ * On CI only: locally the records are what the run just produced and they are
+ * kept. Returns the files it restored.
+ */
+export function restoreAcceptanceRecords(root) {
+  if (!isCi()) return [];
+  const files = acceptanceRecordFiles(root);
+  if (files.length === 0) return [];
+  const restored = spawnSync(
+    "git",
+    ["restore", "--source=HEAD", "--worktree", "--", ...files],
+    { cwd: root, encoding: "utf8" },
+  );
+  if (restored.status !== 0) {
+    throw new Error(
+      `Could not restore the acceptance records: ${restored.stderr.trim()}`,
+    );
+  }
+  return files;
+}

scripts/verify.mjs

@@ -12,6 +12,8 @@ import path from "node:path";
 import { fileURLToPath } from "node:url";
 import { parseArgs } from "node:util";
 
+import { restoreAcceptanceRecords } from "./lib/acceptance-records.mjs";
+
 // Single definition of the full verification pipeline. GitHub CI runs `pnpm verify`
 // so a local run executes the same gates in the same order; previously the order
 // only existed inside the workflow file and could not be reproduced locally.
@@ -246,6 +248,22 @@ try {
   }
   report.status = failed ? "FAIL" : "PASS";
 } finally {
+  // The browser acceptance records are tracked files that a test run rewrites,
+  // and the release packaging runs in a separate process after this one and
+  // refuses to run with a dirty tree. Put them back before the run is over, so
+  // a verification run leaves the checkout the way it found it. A failure to
+  // restore is reported rather than thrown: the gate that failed is the
+  // interesting one, and the packaging step says so plainly if the tree is
+  // still dirty.
+  try {
+    const restored = restoreAcceptanceRecords(root);
+    if (restored.length > 0)
+      console.log(`acceptance records restored: ${restored.length} file(s)`);
+  } catch (error) {
+    console.error(
+      `could not restore the acceptance records: ${error instanceof Error ? error.message : String(error)}`,
+    );
+  }
   rmSync(temporary, { recursive: true, force: true });
   for (const gate of report.gates)
     if (gate.status === "NOT_RUN")

test/source-sync.test.mjs

@@ -23,6 +23,7 @@ import { copyMcodeToolsArtifact, downloadMcodeToolsArtifact, MCODE_TOOLS_ARTIFAC
 import { checkWindowsSourceLocation, runWindowsSourceLocationCheck } from '../scripts/check-windows-source-location.mjs';
 import { recordBuildIntegrity, verifyBuildIntegrity } from '../scripts/lib/webui-build-integrity.mjs';
 import { validateExceptionLedger, sourceFingerprint } from '../scripts/lib/webui-dependency-rules.mjs';
+import { acceptanceRecordFiles, restoreAcceptanceRecords } from '../scripts/lib/acceptance-records.mjs';
 
 test('permanent loader approval follows tokens rather than line numbers', () => {
   const loader = 'const entry = new URL("./embedded/cli.mjs", import.meta.url); await import(entry.href);';
@@ -657,9 +658,16 @@ function verificationFixture(t) {
   const directory = mkdtempSync(path.join(tmpdir(), 'verify-fixture-'));
   t.after(() => rmSync(directory, { recursive: true, force: true }));
   const root = path.join(directory, 'source');
-  mkdirSync(path.join(root, 'scripts'), { recursive: true });
+  mkdirSync(path.join(root, 'scripts/lib'), { recursive: true });
   const verifier = path.join(root, 'scripts/verify.mjs');
   copyFileSync(new URL('../scripts/verify.mjs', import.meta.url), verifier);
+  // The verifier imports the module that puts the acceptance records back, so
+  // the fixture carries it the same way it carries a stub for the export step.
+  // The fixture is not a checkout, so the restore finds nothing to do there.
+  copyFileSync(
+    new URL('../scripts/lib/acceptance-records.mjs', import.meta.url),
+    path.join(root, 'scripts/lib/acceptance-records.mjs'),
+  );
   const manager = path.join(directory, 'manager.cjs');
   writeFileSync(manager, `
     const fs = require('node:fs');
@@ -700,6 +708,55 @@ function verificationFixture(t) {
   };
 }
 
+test('a verification run puts the browser acceptance records back the way the checkout had them', t => {
+  const directory = mkdtempSync(path.join(tmpdir(), 'acceptance-records-'));
+  t.after(() => rmSync(directory, { recursive: true, force: true }));
+  const git = (...args) => execFileSync('git', ['-C', directory, ...args], { encoding: 'utf8', stdio: ['ignore', 'pipe', 'pipe'] });
+  const commit = (...args) => git('-c', 'user.name=fixture', '-c', 'user.email=fixture@local', ...args);
+  git('init');
+  // The repository's own attributes, so the checkout writes the records the
+  // way the repository writes them. Without them a Windows machine would
+  // check the file out with CRLF, and git would keep calling it modified
+  // after it had been restored.
+  copyFileSync(new URL('../.gitattributes', import.meta.url), path.join(directory, '.gitattributes'));
+  commit('add', '--', '.gitattributes');
+  commit('commit', '-m', 'the attributes');
+  commit('commit', '--allow-empty', '-m', 'base');
+  const record = '.plan/browser-use/complete/isolated-chromium.acceptance.json';
+  mkdirSync(path.join(directory, path.dirname(record)), { recursive: true });
+  writeFileSync(path.join(directory, record), '[]\n');
+  commit('add', '--', record);
+  commit('commit', '-m', 'the accepted record');
+  const dirty = () => git('status', '--porcelain', '--untracked-files=no').trim();
+
+  assert.deepEqual(acceptanceRecordFiles(directory), [record]);
+
+  // A run rewrites the record: every record carries the time and the machine
+  // it ran on, so the working tree is dirty once the suite is over.
+  writeFileSync(path.join(directory, record), '[{"at":"2026-01-01T00:00:00.000Z"}]\n');
+  assert.equal(dirty(), `M ${record}`);
+
+  const previous = { ci: process.env.CI, actions: process.env.GITHUB_ACTIONS };
+  delete process.env.CI;
+  delete process.env.GITHUB_ACTIONS;
+  try {
+    // Locally the records stay: they are what the run just produced, and the
+    // published matrix is written from them.
+    assert.deepEqual(restoreAcceptanceRecords(directory), []);
+    assert.equal(dirty(), `M ${record}`);
+
+    // On CI they go back, so the release packaging finds a clean tree.
+    process.env.CI = 'true';
+    assert.deepEqual(restoreAcceptanceRecords(directory), [record]);
+    assert.equal(dirty(), '');
+  } finally {
+    for (const [name, value] of [['CI', previous.ci], ['GITHUB_ACTIONS', previous.actions]]) {
+      if (value === undefined) delete process.env[name];
+      else process.env[name] = value;
+    }
+  }
+});
+
 test('platform verification omits only the compiler gates and invalid profiles fail closed', t => {
   const f = verificationFixture(t);
   const full = f.run(['--list']);
@@ -991,6 +1048,12 @@ test('Windows contract profile selects focused gates', () => {
   try {
     const fixture = path.join(root, 'verify.mjs');
     copyFileSync(new URL('../scripts/verify.mjs', import.meta.url), fixture);
+    // The verifier imports the module that puts the acceptance records back.
+    mkdirSync(path.join(root, 'lib'), { recursive: true });
+    copyFileSync(
+      new URL('../scripts/lib/acceptance-records.mjs', import.meta.url),
+      path.join(root, 'lib/acceptance-records.mjs'),
+    );
     const preload = path.join(root, 'platform.cjs');
     writeFileSync(preload, "Object.defineProperty(process, 'platform', { value: 'win32' });\n");
     const result = spawnSync(process.execPath, ['--require', preload, fixture, '--profile', 'windows', '--list'], {