fix(install-e2e): boot the app-update legs against a REAL configured provider

The app-update legs died on the onboarding overlay - a fullscreen div that intercepts every click (the Settings click timed out under it). The old plan seeded a fake provider key, which lies: the overlay vanishes but the app is broken. Instead the driver now runs the desktop E2E suite's own mock inference server (tests-js/scripts/mock-server.ts, zero deps, bare-node type stripping >=22.18) and configures it into HERMES_HOME byte-for-byte like the dev:mock flow: config.yaml provider + MOCK_API_KEY env. The app boots genuinely configured - no overlay, real chat surface. e2e-assets/mock-provider.{sh,mjs} own start/stop (pid + url files; the wrapper lives until SIGTERM - gating on stdin-close made the server die instantly, a background process's stdin is already EOF) and the config write. Wired into the posix script driver's hermes-desktop-app-update arm and the macos driver's update phase (both app-update methods). The Playwright flow keeps its defense-in- depth: the real escape hatch ('I'll choose a provider later') and the verified 'Open settings' selector. Probed locally: models + streamed/non-streamed completions answer, server stops cleanly on kill.
2026-08-12 10:36:09 -04:00
parent fd02bd8621
commit 7721b26a9a
5 changed files with 134 additions and 7 deletions
@@ -120,15 +120,24 @@ async function main() {
  }
  const deadline = Date.now() + Number(values['timeout-ms']);

-  // Dismiss the onboarding overlay when present (fresh HERMES_HOME).
-  const skip = window.getByRole('button', { name: /skip|get started|continue/i }).first();
-  if (await skip.isVisible({ timeout: 5_000 }).catch(() => false)) {
-    await skip.click().catch(() => {});
+  // Dismiss the onboarding overlay when present. Two layers of defense:
+  // the drivers seed a provider key so the app's runtime check reports
+  // configured=true and the overlay never mounts; if it shows anyway
+  // (fresh HERMES_HOME, slow readiness check), click the real escape
+  // hatch - "I'll choose a provider later" (i18n en: chooseLater). The
+  // overlay is a fullscreen div that intercepts ALL clicks, so this must
+  // resolve before any Settings navigation.
+  const later = window
+    .getByRole('button', { name: /choose a provider later|skip/i })
+    .first();
+  if (await later.isVisible({ timeout: 10_000 }).catch(() => false)) {
+    await later.click().catch(() => {});
+    await later.waitFor({ state: 'hidden', timeout: 15_000 }).catch(() => {});
  }

-  // Settings -> About -> Update now. Selectors favor accessible names over
-  // DOM structure so renderer refactors don't break the leg.
-  await window.getByRole('button', { name: /settings/i }).first().click();
+  // Settings -> About -> Update now. The settings trigger is an icon
+  // button whose accessible name is "Open settings".
+  await window.getByRole('button', { name: /open settings|settings/i }).first().click();
  await window.getByRole('tab', { name: /about/i }).or(
    window.getByRole('button', { name: /about/i })).first().click();
  const updateNow = window.getByRole('button', { name: /update now/i }).first();
@@ -0,0 +1,31 @@
+// Driver-side wrapper around tests-js/scripts/mock-server.ts (the desktop
+// E2E suite's OpenAI-compatible mock): starts the server as a LIBRARY
+// (importing, not executing, so the dev-launcher block never runs) and
+// publishes its URL to a file for the shell driver to consume.
+//
+// Usage: node mock-provider.mjs <url-file>
+//   Writes "<url-file>" with the base URL (http://127.0.0.1:<port>) once
+//   the server is listening, then stays alive until stdin closes.
+
+// @ts-check
+import fs from 'node:fs';
+import process from 'node:process';
+import { fileURLToPath } from 'node:url';
+
+const mockUrl = fileURLToPath(new URL('../../../tests-js/scripts/mock-server.ts', import.meta.url));
+
+const { startMockServer } = await import(mockUrl);
+const mock = await startMockServer();
+
+const urlFile = process.argv[2];
+if (!urlFile) {
+  console.error('usage: node mock-provider.mjs <url-file>');
+  process.exit(1);
+}
+fs.writeFileSync(urlFile, mock.url);
+console.log(`[mock-provider] listening at ${mock.url}`);
+
+// Live until killed. NOT gated on stdin closing: a backgrounded process's
+// stdin is already at EOF, so an end-event exit would fire immediately
+// and the server would die right after writing its URL.
+await new Promise(() => {});
@@ -0,0 +1,71 @@
+#!/usr/bin/env bash
+# Start/stop the mock inference server and point the app at it as a REAL
+# configured provider.
+#
+# Why: the desktop app boots to the onboarding overlay when no provider is
+# configured - a fullscreen div that intercepts every click, which killed
+# the app-update legs. Seeding a fake key makes the overlay vanish but
+# leaves a lying app; this makes the app GENUINELY configured (config.yaml
+# provider + key env, exactly what tests-js/scripts/mock-server.ts's
+# dev:mock flow writes) with a real, chat-capable backend.
+#
+# Usage (sourced from a driver):
+#   mock_start <workroot>      start, write config into $HERMES_HOME
+#   mock_stop                  kill the background server
+#
+# Requires: ASSETS (e2e-assets dir), LOG_DIR, HERMES_HOME, ok/fail helpers.
+
+MOCK_PIDFILE=""
+MOCK_URLFILE=""
+
+mock_start() {
+  local workroot="${1:?mock_start needs a workroot}"
+  MOCK_PIDFILE="$workroot/mock.pid"
+  MOCK_URLFILE="$workroot/mock.url"
+  rm -f "$MOCK_PIDFILE" "$MOCK_URLFILE"
+
+  # Bare `node file.ts` type-stripping works on node >=22.18 (the images
+  # ship 22.22+; the installed managed node is >=26) - same contract as
+  # the repo's own `dev:mock` script.
+  node "$ASSETS/mock-provider.mjs" "$MOCK_URLFILE" > "$LOG_DIR/mock.log" 2>&1 &
+  echo $! > "$MOCK_PIDFILE"
+
+  local _i
+  for _i in 1 2 3 4 5 6 7 8 9 10; do
+    [ -s "$MOCK_URLFILE" ] && break
+    sleep 0.2
+  done
+  if [ ! -s "$MOCK_URLFILE" ]; then
+    log_group "mock server transcript" "$LOG_DIR/mock.log"
+    fail "mock inference server did not come up; transcript above"
+  fi
+  local url
+  url="$(cat "$MOCK_URLFILE")"
+  ok "mock inference server: $url"
+
+  # The provider config, byte-compatible with writeMockConfig() in
+  # tests-js/scripts/mock-server.ts.
+  cat > "$HERMES_HOME/config.yaml" <<EOF
+model:
+  default: mock-model
+  provider: mock
+providers:
+  mock:
+    api: $url/v1
+    name: Mock
+    api_mode: chat_completions
+    key_env: MOCK_API_KEY
+    models:
+      mock-model: {}
+    context_length: 4096
+EOF
+  printf 'MOCK_API_KEY=e2e-mock-key\n' >> "$HERMES_HOME/.env"
+  ok "provider 'mock' configured in $HERMES_HOME (api $url/v1)"
+}
+
+mock_stop() {
+  if [ -n "$MOCK_PIDFILE" ] && [ -f "$MOCK_PIDFILE" ]; then
+    kill "$(cat "$MOCK_PIDFILE")" 2>/dev/null || true
+    rm -f "$MOCK_PIDFILE"
+  fi
+}
@@ -357,6 +357,15 @@ case "$UPDATE_METHOD" in
    ASSETS="$REPO_ROOT/tests/install/e2e-assets"
    SPEC="$WORK_ROOT/launch-spec.json"

+    # A REAL configured provider: the mock inference server (the desktop E2E
+    # suite's own) is configured into HERMES_HOME exactly like the dev:mock
+    # flow does. The app then boots genuinely configured - no onboarding
+    # overlay (a fullscreen div that intercepts every click) - and the chat
+    # surface is real too.
+    source "$ASSETS/mock-provider.sh"
+    mock_start "$WORK_ROOT"
+    trap mock_stop EXIT
+
    step "capturing the hermes desktop launch spec (build runs for real)"
    rc=0
    (cd "$INSTALL_DIR" && \
@@ -270,6 +270,13 @@ phase_update() {
  ok "serve.git main = $HEAD_SHA"

  step "updating via $UPDATE_METHOD"
+  # The app must boot configured or the onboarding overlay (a fullscreen
+  # div) eats every click: configure the mock inference server exactly like
+  # the dev:mock flow does, so the app is genuinely configured.
+  # shellcheck source=../install/e2e-assets/mock-provider.sh
+  source "$ASSETS/mock-provider.sh"
+  mock_start "$WORK_ROOT"
+  trap mock_stop EXIT
  case "$UPDATE_METHOD" in
    open-app-update)
      # The installed app IS the user surface here (double-click the .app);