ci(web): Phase P6 — vite build + headless E2E gate on every trunk merge (#1445) (#1454)

.woodpecker/ci.yml

@@ -254,6 +254,23 @@ steps:
     depends_on:
       - typecheck
 
+  # Canonical verify:release stage `build` (#1445, P6): every PR proves the
+  # full workspace build — including the SPA `vite build` — before merge,
+  # instead of leaving build breakage to surface post-merge in publish.yml's
+  # verify step. Same canonical command the publish pipeline's build step runs.
+  build:
+    image: *node_image
+    commands:
+      - *enable_pnpm
+      - pnpm build
+    depends_on:
+      # after test, not typecheck: turbo gives `test` a ^build dependency, so
+      # running this step concurrently with test would put two independent
+      # turbo builds on the same shared-workspace dist/ and turbo cache with
+      # no cross-process locking — the same serialization invariant
+      # publish.yml documents for #1411.
+      - test
+
 services:
   ci-postgres:
     image: pgvector/pgvector:pg17

.woodpecker/publish.yml

@@ -407,6 +407,96 @@ steps:
       - build
       - verify
 
+  # #1445 (P6): headless Playwright E2E gate on every trunk merge. Boots the
+  # real gateway on the embedded PGlite path (no DATABASE_URL, no services)
+  # serving the built SPA bundle via WEB_DIST_DIR — the exact serving path the
+  # gateway image ships (docker/gateway.Dockerfile sets WEB_DIST_DIR to the
+  # baked bundle), which keeps #1407's parity guarantee: the image build steps
+  # below depend on this gate, so a bundle that fails E2E never publishes.
+  #
+  # Image pinned to the @playwright/test version in pnpm-lock.yaml so the
+  # image's bundled browsers match the workspace driver exactly (bump the two
+  # together). The step installs no workspace packages (corepack does fetch
+  # the pinned pnpm itself): it reuses the workspace node_modules
+  # from `install` and the dist outputs from `build` — the gateway's runtime
+  # dependency path is pure JS/WASM (PGlite is WASM, postgres-js is pure JS),
+  # so the alpine-installed modules run unchanged under this glibc image.
+  # depends_on publish-next-npm per the #1411 serialization invariant: this
+  # step reads the workspace and must never run inside the manifest-transform
+  # window.
+  e2e:
+    image: mcr.microsoft.com/playwright:v1.58.2-noble
+    environment:
+      GATEWAY_PORT: '14242'
+      PLAYWRIGHT_BASE_URL: http://localhost:14242
+      # The database is seeded by Playwright's globalSetup in this step, so
+      # login failures are real failures: without this flag the suite's
+      # skip-when-login-fails guards (a live-environment affordance) could
+      # skip every authenticated spec and go green while proving nothing.
+      E2E_REQUIRE_SEEDED_AUTH: '1'
+    commands:
+      - corepack enable
+      - |
+        # Throwaway signing secret for this step's ephemeral embedded database
+        # (the gateway refuses to boot without one). Generated per run so no
+        # usable literal lives in the tree.
+        export BETTER_AUTH_SECRET="$(head -c 32 /dev/urandom | base64)"
+        export WEB_DIST_DIR="$(pwd)/apps/web/dist"
+        if [ ! -f "$WEB_DIST_DIR/index.html" ]; then
+          echo "[e2e] FATAL: $WEB_DIST_DIR/index.html missing — did the build step run?" >&2
+          exit 1
+        fi
+        # Boot the gateway from the built dist, cwd- AND HOME-isolated: the
+        # local-tier PGlite database lives under $HOME/.config/mosaic/gateway/
+        # (database.module.ts), not under cwd, so HOME must point at the
+        # throwaway dir too or the run would share a database with anything
+        # else in the container's home.
+        GATEWAY_RUN_DIR="$(mktemp -d /tmp/e2e-gateway.XXXXXX)"
+        (cd "$GATEWAY_RUN_DIR" && export HOME="$GATEWAY_RUN_DIR" && exec node "$OLDPWD/apps/gateway/dist/main.js") > /tmp/gateway.log 2>&1 &
+        GATEWAY_PID=$!
+        ready=0
+        for i in $(seq 1 90); do
+          if node -e "fetch('http://localhost:' + process.env.GATEWAY_PORT + '/health', { signal: AbortSignal.timeout(2000) }).then((r) => process.exit(r.ok ? 0 : 1), () => process.exit(1))"; then
+            ready=1
+            break
+          fi
+          if ! kill -0 "$GATEWAY_PID" 2>/dev/null; then
+            echo "[e2e] FATAL: gateway process exited during startup" >&2
+            cat /tmp/gateway.log >&2
+            exit 1
+          fi
+          echo "[e2e] waiting for gateway ($i/90)..."
+          sleep 1
+        done
+        if [ "$ready" -ne 1 ]; then
+          echo "[e2e] FATAL: gateway did not become ready in 90s" >&2
+          cat /tmp/gateway.log >&2
+          exit 1
+        fi
+        echo "[e2e] gateway ready; running Playwright suite"
+        set +e
+        pnpm --filter @mosaicstack/web exec playwright test
+        E2E_EXIT=$?
+        set -e
+        kill "$GATEWAY_PID" 2>/dev/null || true
+        if [ "$E2E_EXIT" -ne 0 ]; then
+          echo "[e2e] FATAL: Playwright suite failed (exit $E2E_EXIT); gateway log follows" >&2
+          tail -100 /tmp/gateway.log >&2
+          echo "[e2e] browser-side traces/screenshots are under apps/web/test-results/ in the step workspace (not persisted past the pod)" >&2
+        fi
+        exit "$E2E_EXIT"
+    # Same filter as the image builds it gates: a merge that publishes no
+    # image (docs-only on main) pays no browser suite, and a skipped e2e does
+    # not block anything (skipped-dependency semantics, same as
+    # publish-next-npm on tag events).
+    when: *image_build_when
+    depends_on:
+      - build
+      - verify
+      # #1411: never read the workspace inside publish-next-npm's
+      # manifest-transform window.
+      - publish-next-npm
+
   # TODO: Uncomment when ready to publish to npmjs.org
   # publish-npmjs:
   #   image: *node_image
@@ -466,6 +556,8 @@ steps:
       # ERR_PNPM_OUTDATED_LOCKFILE despite a clean restore. This edge is the
       # serialization invariant; add it to every new workspace consumer.
       - publish-next-npm
+      # #1445 (P6): a bundle that fails the E2E gate never publishes an image.
+      - e2e
 
   build-appservice:
     image: gcr.io/kaniko-project/executor:debug
@@ -510,3 +602,5 @@ steps:
       # ERR_PNPM_OUTDATED_LOCKFILE despite a clean restore. This edge is the
       # serialization invariant; add it to every new workspace consumer.
       - publish-next-npm
+      # #1445 (P6): a bundle that fails the E2E gate never publishes an image.
+      - e2e