ci(pages): add smoke test step to verify build output is servable

[HuijungYoon]

Summary

• Add serve as a pages devDependency so CI can statically serve dist/ without downloading on every run.
• After the Build step in pages-ci.yml, smoke-test /, /docs/contributing, /docs/quickstart, and /features for HTTP 200 responses that include .bundle.js references (SPA mode via serve -s).
• Closes ci(pages): add smoke test step to verify build output is servable #725

Test plan

• Pages CI runs on this PR (pages/** path filter)
• Smoke test step passes (all four routes OK)
• Bundle size check still runs after smoke test
• Locally (Node 24.18.0): cd pages && npm install && npm run build, then run the smoke script from the workflow

[CLAassistant]

All committers have signed the CLA.

[github-actions]

🔍 OpenCodeReview found 2 issue(s) in this PR.

• ✅ Successfully posted inline: 2 comment(s)

[github-actions]

[maintainability · medium]
Reliability concern: fixed sleep 2 is fragile.

On a busy CI runner, 2 seconds may not be enough for the server to be ready. Consider using a polling loop with retries instead, which is more resilient to variable startup times:

for i in $(seq 1 20); do
  curl -sf http://127.0.0.1:3999/ > /dev/null && break
  sleep 0.5
done

[github-actions]

[bug · high]
Bug: set -e will kill the script before the $? check is reached.

GitHub Actions run: steps use bash -eo pipefail by default. When curl -sf fails (e.g., the server isn't ready or a path returns an error), the command substitution BODY=$(curl -sf ...) will propagate the non-zero exit code, and set -e will immediately terminate the script — before the if [ $? -ne 0 ] check ever runs. This means the first failing path aborts the entire loop, skipping remaining paths and the exit $FAILED line.

Restructure so that curl's failure is caught within an if condition (which is exempt from set -e):

Suggestion:

Suggested change

	BODY=$(curl -sf "http://127.0.0.1:3999${path}")
	if [ $? -ne 0 ]; then
	if ! BODY=$(curl -sf "http://127.0.0.1:3999${path}"); then

[lizhengfeng101]

@HuijungYoon Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

[HuijungYoon]

@lizhengfeng101 I've signed the CLA! I'm really looking forward to contributing more to this project. Thank you!

[lizhengfeng101]

@lizhengfeng101 I've signed the CLA! I'm really looking forward to contributing more to this project. Thank you!

ok

[lizhengfeng101]

Hey @HuijungYoon, nice addition — the smoke step is a sensible lightweight alternative to spinning up Playwright. Two things I'd love your take on before this goes in:

1. All four routes resolve to the same index.html under serve -s. Since this is a React Router SPA, serve -s rewrites every unmatched path to index.html, and dist/ only contains index.html / 404.html / the bundles. So /, /docs/contributing, /docs/quickstart, and /features all return the identical HTML, and the .bundle.js assertion is effectively the same check run four times. It'll catch a missing index.html or a broken bundle reference, but it won't catch a broken route component (that only surfaces once React Router renders client-side). Not a blocker, but the four-route list reads like it's testing more than it actually is — might be worth either a comment noting the limitation, or adding one assertion against a real static file (e.g. /404.html) so we're confirming the server is genuinely reading dist/.

2. Minor: the if [ $? -ne 0 ] branch is effectively dead under the Actions default shell. GitHub runs run: steps with bash --noprofile --norc -eo pipefail, so set -e is active. On a BODY=$(curl -sf ...) failure the step aborts before reaching the $? check, so the friendly "not reachable" message never prints. In practice this rarely bites (given point 1, routes won't individually 404 — only a fully-dead server would fail curl, and that still fails CI on the first / request), so I'd call it a cleanup rather than a must-fix. If you want the branch to actually work, moving curl into the tested context does it:

if ! BODY=$(curl -sf "http://127.0.0.1:3999${path}"); then
  echo "FAIL: ${path} — not reachable"; FAILED=1
elif ! echo "$BODY" | grep -q '\.bundle\.js'; then
  ...

Also +1 on the trap ... EXIT for cleanup — nice touch over the original issue snippet.

[HuijungYoon]

@lizhengfeng101

Thanks for the thoughtful review

I pushed a follow-up commit that addresses both items:

1. Added a comment clarifying the SPA limitation with serve -s (the route checks validate shell reachability and bundle references, not client-side route rendering correctness).
2. Updated the curl branch to if ! BODY=$(curl -sf ...) so the “not reachable” path works correctly under GitHub Actions’ bash -e behavior.
3. Added a static asset check for /404.html to confirm we’re actually serving files from dist/, not only exercising SPA fallback.

Really appreciate the careful read and the concrete suggestions.

[lizhengfeng101]

LGTM