phase a/fix playwright webserver (#48)

* fix(e2e): wire playwright webServer to boot live code-iq backend

Adds a webServer block to src/main/frontend/playwright.config.ts so
npm run test:e2e boots a real code-iq serve against the
spring-petclinic fixture before any spec runs, removing the root cause
of the 0-pass / 575-fail Playwright audit result.

Design choices:
- /api/stats is the readiness probe, not /actuator/health (today
  returns 503 OUT_OF_SERVICE; tracked separately on
  phase-a/fix-graph-health).
- bash -c "exec java ..." so Playwright can signal the JVM cleanly on
  teardown, and so the *-cli.jar glob expands.
- Overridable via BASE_URL (skip webServer entirely), E2E_FIXTURE (pick
  a different pre-enriched repo), E2E_PORT (avoid port conflicts).
- baseURL unchanged for spec code — still http://localhost:8080 by
  default, overridable via BASE_URL.

Verified on this machine:
- webServer boots the backend; /api/stats returns 200 within 120s.
- Playwright connects to the backend.

Remaining system-level blocker (not a config issue):
chromium headless shell is missing 8 shared libraries on this host
(libatk-1.0, libXcomposite, libgbm, libatspi, ...). Fix once per host:

    sudo npx playwright install-deps chromium

Once that's run, npm run test:e2e exercises the full 131-test
chromium suite against a real backend. The original baseline
0p/575f number was the sum across 7 Playwright projects.
Documented in BASELINE.md.

* docs(baseline): capture real Playwright chromium pass rate (33/131)

With the webServer wiring in place + system deps installed via
`sudo npx playwright install-deps chromium`, ran the full chromium
project (19.2 min wall time). Results:

  33 passed / 94 failed / 4 skipped

Huge jump from the "0 passed / 575 failed" baseline — the 575 was the
sum across 7 Playwright projects (4 browsers × many responsive
breakpoints), and all of it was caused by not having a backend. With
a live backend against the spring-petclinic fixture, a third of the
chromium suite passes.

The remaining 94 failures are NOT environmental; they're genuine test
↔ UI divergence:

  - expect(locator).toBeVisible / element(s) not found — selectors
    targeting UI elements that don't render in the current build
    (component/CSS-class renames).
  - strict mode violations on `getByText(/0/)` — tests assume unique
    stat values; spring-petclinic's 691-node graph produces many
    zero-valued cards, so the match resolves to 8+ elements.
  - toHaveLength mismatches — expected N list items, got different count.
  - "Button has no aria-label" in accessibility.spec.ts — real a11y
    regressions, not a test bug.
  - 30s timeouts on locator.focus / locator.boundingBox — elements
    that never appear.

Treating the chromium 33/94/4 split as the honest Phase A baseline.
Test-suite maintenance (update selectors to match current UI, or
generate a deterministic fixture) is scoped out of Phase A.

Author: aksOps

docs/superpowers/baselines/2026-04-17/BASELINE.md

@@ -222,6 +222,19 @@ Ordered by severity. Each item cites the raw artifact it was derived from.
 
 - **Playwright E2E: 0 passed / 575 failed.** 100% failure rate. Almost certainly environment/config rather than regressions — the audit script runs `npm run test:e2e` without starting the backend (`java -jar ... serve`), so any test that hits `/api/*` will fail. Needs a harness that spins up the server (or mocks it) before running Playwright, or a `webServer` entry in `playwright.config.ts`.
   - Raw: `raw/frontend/playwright.log`, `raw/frontend/playwright-summary.json`.
+  - **RESOLVED — config + system-deps + run (2026-04-17, branch `phase-a/fix-playwright-webserver`)**: added a `webServer` block to `src/main/frontend/playwright.config.ts` that boots `java -jar target/code-iq-*-cli.jar serve .seeds/spring-petclinic --port 8080` and gates readiness on `/api/stats` (HTTP 200). Installed system libs via `sudo npx playwright install-deps chromium`. Ran full chromium suite against the live backend (19.2 min wall time).
+
+  **New chromium baseline**: `33 passed / 94 failed / 4 skipped` out of **131 tests**. Huge improvement from "0 passed / 575 failed" (the 575 was the sum across 7 Playwright projects — chromium/firefox/edge/webkit + 3 responsive breakpoints). The 94 chromium failures are **not environmental** — they're real test ↔ UI divergence. Signature histogram of the 22 detailed failure blocks analyzed:
+
+  | Pattern | Interpretation |
+  |---|---|
+  | `expect(locator).toBeVisible() failed` / `element(s) not found` | Selectors targeting UI elements that no longer render (renamed / removed). |
+  | `strict mode violation: getByText(/0/) resolved to 8 elements` | Tests assumed unique stats values; spring-petclinic's 691-node graph produces many zero-valued stat cards. |
+  | `expect(received).toHaveLength(expected)` | List count mismatches (menu items, nav links, stats rows). |
+  | `Error: Button ... has no aria-label` | Real a11y regressions in `accessibility.spec.ts`. |
+  | `locator.focus: Test timeout 30000ms exceeded` | UI interactions hanging on elements that never appear. |
+
+  Test-suite maintenance / fixture generation is a follow-up (not a Phase A blocker). Candidate approaches: (1) rebuild the Ant-Design selectors to match current UI taxonomy; (2) pin a synthetic fixture with known, deterministic stats counts; (3) relax strict-mode selectors to `.first()` where tests genuinely target "any such element". 22 of the 94 failures have full error blocks captured in `raw/frontend/playwright.log`.
 
 ### High
 

src/main/frontend/playwright.config.ts

@@ -5,8 +5,14 @@ import { defineConfig, devices } from '@playwright/test';
  * Playwright E2E test configuration for Code IQ UI Redesign (Phase 7).
  *
  * Prerequisites:
- *   - `code-iq serve` running on localhost:8080 (Neo4j graph loaded)
- *   - Or use `webServer` below to start the Vite dev server pointing at a real backend
+ *   - A pre-built CLI jar under ../../../target/code-iq-*-cli.jar
+ *     (run `mvn -DskipTests package` from the repo root if missing).
+ *   - A pre-populated fixture at <repo-root>/.seeds/spring-petclinic
+ *     with .code-iq/cache + .code-iq/graph populated by
+ *     `./scripts/baseline/run-pipeline.sh spring-petclinic`. The webServer
+ *     block below boots `code-iq serve` against this fixture automatically.
+ *   - Set `BASE_URL` / `E2E_FIXTURE` to override defaults when running against
+ *     a different backend or fixture.
  *
  * Run all tests:       npm run test:e2e
  * Run headed:          npm run test:e2e:headed
@@ -22,8 +28,30 @@ export default defineConfig({
   workers: process.env.CI ? 1 : undefined,
   reporter: [['html', { outputFolder: 'playwright-report' }], ['line']],
 
+  // Boot the real code-iq backend against a pre-enriched fixture before any
+  // spec runs. Skipped if BASE_URL points elsewhere (developer has their own
+  // backend) or if a server is already listening on the target port locally.
+  webServer: process.env.BASE_URL ? undefined : {
+    command: [
+      'bash -c',
+      // Use shell so the target/*-cli.jar glob expands. `exec` hands the PID
+      // to Playwright so it can kill the process cleanly on test teardown.
+      `"exec java -jar ../../../target/code-iq-*-cli.jar serve `
+        + `${process.env.E2E_FIXTURE ?? '../../../.seeds/spring-petclinic'} `
+        + `--port ${process.env.E2E_PORT ?? '8080'}"`,
+    ].join(' '),
+    // /actuator/health is not a reliable readiness signal today (see
+    // BASELINE.md §GraphHealthIndicator). /api/stats returns 200 iff the
+    // server has finished starting and the graph has loaded.
+    url: `http://localhost:${process.env.E2E_PORT ?? '8080'}/api/stats`,
+    timeout: 120_000,
+    reuseExistingServer: !process.env.CI,
+    stdout: 'pipe',
+    stderr: 'pipe',
+  },
+
   use: {
-    baseURL: process.env.BASE_URL ?? 'http://localhost:8080',
+    baseURL: process.env.BASE_URL ?? `http://localhost:${process.env.E2E_PORT ?? '8080'}`,
     trace: 'on-first-retry',
     screenshot: 'only-on-failure',
     video: 'retain-on-failure',