Run pumlv e2e tests

The suite lives in internal/frontend/tests/e2e/ and is driven by make e2e, which:

1. Runs make build — go generate ./... (which builds the embedded SPA via pnpm run build) then go build -trimpath -o pumlv ..
2. Runs cd internal/frontend && pnpm test:e2e (i.e. playwright test). Playwright's webServer spec spawns the freshly-built ./pumlv on 127.0.0.1:8766 against ../../examples and tears it down after the run.

The configured project is chromium only, and it launches Playwright's bundled Chromium (no branded channel), so the browser comes from playwright install chromium rather than a system Google Chrome.

Run it

make e2e

That is the whole happy path. Run from the repo root.

Recovering from failures

• Executable doesn't exist at .../chrome-linux/chrome — the bundled Chromium isn't installed. Run cd internal/frontend && pnpm exec playwright install chromium chromium-headless-shell. In sandboxes the browsers are usually preinstalled under /opt/pw-browsers (with PLAYWRIGHT_BROWSERS_PATH=/opt/pw-browsers already set), so Playwright finds them without a download. Do not edit playwright.config.ts to bake in a sandbox-local path.

• Server didn't come up within 30 s — the binary is missing or stale. Confirm ./pumlv --no-open --port 8766 ./examples boots by hand. If make build fails earlier with pattern all:dist: no matching files found from internal/static/embed.go, the frontend bundle wasn't produced — re-run make build from a clean tree (the go:generate directive runs pnpm install && pnpm run build).

• Port already in use — kill any stray pumlv from a previous run (pgrep -af pumlv) before retrying. playwright.config.ts has reuseExistingServer: !process.env.CI, so a stale local server won't error out by itself but will mask the binary you just built.

• A spec failed — Playwright writes internal/frontend/playwright-report/index.html and internal/frontend/test-results/<spec>/trace.zip per failure. Inspect those first; don't guess at fixes from the terminal output alone:

  cd internal/frontend && pnpm exec playwright show-trace test-results/<dir>/trace.zip
  

  Fix the underlying regression. Don't test.skip a failure to make CI green — either fix it or report it.

Iterating on one spec

While debugging, re-run only the spec you're working on:

cd internal/frontend && pnpm exec playwright test tests/e2e/<name>.spec.ts

Once it passes, run make e2e once more from the repo root to confirm the whole suite still passes before reporting back.

Don't

• Don't run pnpm test:e2e directly without first running make build — the webServer.command points at ../../pumlv, which may not exist or may be stale.
• Don't add tests to internal/frontend/tests/e2e/ that depend on files outside examples/ — that directory is what the e2e Playwright config serves.