test(server): regression tests for SPA fallback on client-side routes (fabro-q5 brief)

[zenprocess]

Summary

Regression tests for the SPA fallback that serves the app shell for client-side routes (/runs, /runs/<id>/stages, etc.). Brief: ao-company#306.

The server-side SPA fallback already lives in lib/apps/fabro-server/src/static_files.rs (the if accepts_html(headers) { load_asset_for_mode("index.html", …) } block at the end of serve_with_mode) and is wired into the router at lib/apps/fabro-server/src/server.rs:1878 (the else if web_enabled && matches!(req.method(), &Method::GET | &Method::HEAD) branch of the fallback service).

This PR adds the missing regression surface: tests that pin both halves of the contract, so a future revert of the fallback is caught.

Why

The operator's bookmarked /runs link currently works because the fallback is in place — but no test in the suite exercised it. Only a behavioral test catches a future revert. Four tests now do.

Acceptance

ulimit -n 4096 && cargo nextest run -p fabro-server spa_fallback_for_client_routes::

Expected: 4 passed, 0 failed.

Two-state proof (test fails when fallback is reverted)

Test run #1 — with the fallback REVERTED in lib/apps/fabro-server/src/static_files.rs (the if accepts_html(headers) { … } block commented out, no other code changed):

cargo test -p fabro-server --lib spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link

thread 'server::tests::spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link' (15384548) panicked at lib/apps/fabro-server/src/server/tests.rs:16808:9:
assertion `left == right` failed: GET /runs must return the SPA shell, not a 404 — deep links and bookmarks depend on this fallback
  left: 404
 right: 200

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 780 filtered out

Test run #2 — fallback restored, all four tests:

cargo test -p fabro-server --lib spa_fallback_for_client_routes

PASS [0.112s] spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link
PASS [0.122s] spa_fallback_for_client_routes::runs_route_404s_for_post_requests
PASS [0.131s] spa_fallback_for_client_routes::runs_route_404s_for_non_html_clients_to_avoid_silent_shells
PASS [0.141s] spa_fallback_for_client_routes::runs_detail_route_serves_spa_index_for_bookmarked_link

Summary [0.145s] 4 tests run: 4 passed, 777 skipped

This is the regression surface: a revert of the fallback produces a non-zero exit (base_rc != 0); with the fallback in place, exit 0 (head_rc = 0).

What the four tests pin

• runs_route_serves_spa_index_for_bookmarked_link — GET /runs with browser Accept: text/html,… returns 200 + the embedded SPA shell. Primary regression guard.
• runs_detail_route_serves_spa_index_for_bookmarked_link — deep route /runs/<id>/stages returns the same. The deep-link case is the operator's actual bookmark shape.
• runs_route_404s_for_non_html_clients_to_avoid_silent_shells — GET /runs with Accept: application/json returns 404. Prevents widening the fallback so a stray fetch() / typo'd API request silently receives a 25 KB HTML shell.
• runs_route_404s_for_post_requests — POST /runs returns 404. Only GET/HEAD reach the SPA fallback; this test fails if the method match is ever widened.

Diff scope

lib/apps/fabro-server/src/server/tests.rs | 106 ++++++++++++++++++++++++++++++
1 file changed, 106 insertions(+)

No production code change. The fallback was already correct; the brief's "hash-router" premise was wrong (the repo has used createBrowserRouter since 9b9ebdf), and the server already handled browser-router URLs.

Live path / unrelated notes

• The Cargo.lock change that builds up under this worktree is an unrelated fabro-referee workspace entry that regenerates on build. Not part of this PR; reverted locally before the branch was pushed.
• Brief's second change (gate POSTing /api/v1/runs/registrations, live since fabro server 38b26303 with zero prior callers) is being tracked separately on ao-company#306. End-to-end live path there is unproven today (sandbox-creation failures with firecracker snapshot-load 400); the brief instructed using a local mock stand-in and stating plainly that the live path is unproven — that change lives in the ao-company repo, not here.
• qa-pipeline reports no-tests on this repo and exits 0 — a manufactured green. Not cited. Real runner is cargo nextest run -p fabro-server per the brief.

🤖 Generated with Claude Code

[zenprocess]

Evidence — reviewer could not fetch the diff blob

I am re-posting the acceptance evidence here so review does not have to wait for git push problems to clear. This is what re-running on a clean checkout should produce; the coach will independently re-run before accepting.

Acceptance command

ulimit -n 4096 && cargo nextest run -p fabro-server spa_fallback_for_client_routes::

Per the brief: cargo nextest run -p fabro-server (per-crate, not workspace). qa-pipeline reports no-tests on this repo and exits 0 — a manufactured green, not cited.

head_rc

PASS [0.112s] spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link
PASS [0.122s] spa_fallback_for_client_routes::runs_route_404s_for_post_requests
PASS [0.131s] spa_fallback_for_client_routes::runs_route_404s_for_non_html_clients_to_avoid_silent_shells
PASS [0.141s] spa_fallback_for_client_routes::runs_detail_route_serves_spa_index_for_bookmarked_link

Summary [0.145s] 4 tests run: 4 passed, 777 skipped

head_rc = 0.

base_rc — two-state proof that the test FAILS when the fallback is reverted

To produce this, I commented out the SPA fallback (if accepts_html(headers) { … } block at the end of serve_with_mode in lib/apps/fabro-server/src/static_files.rs), left all other code unchanged, and re-ran the relevant subset. The unmodified git tree is restored after the proof; this is captured for reproducibility.

State 1 — fallback REVERTED (Cargo.lock unchanged, server/tests.rs unchanged, only the if accepts_html(headers) { … } block commented out in static_files.rs):

cargo test -p fabro-server --lib spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link

thread 'server::tests::spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link' (15384548) panicked at lib/apps/fabro-server/src/server/tests.rs:16808:9:
assertion `left == right` failed: GET /runs must return the SPA shell, not a 404 — deep links and bookmarks depend on this fallback
  left: 404
 right: 200

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 780 filtered out

base_rc != 0 — the test fails when the fallback is removed.

State 2 — fallback RESTORED (this commit's source tree, as-pushed):

cargo test -p fabro-server --lib spa_fallback_for_client_routes

Summary [0.145s] 4 tests run: 4 passed, 777 skipped

head_rc = 0 — the test passes when the fallback is in place.

So the test produces base_rc != 0 under a hypothetical revert and head_rc = 0 on the committed tree. That is the regression surface the brief required.

How to reproduce independently

# Clean clone / fresh worktree, on this branch.
ulimit -n 4096 && cargo nextest run -p fabro-server spa_fallback_for_client_routes::
# expect: 4 passed, head_rc == 0

# To see the test catch a revert:
git checkout -b reproduce-tmp-fix/spa-fallback-runs
# edit lib/apps/fabro-server/src/static_files.rs and wrap the
# `if accepts_html(headers) { load_asset_for_mode("index.html", …) … }`
# block in /* … */ comments.
cargo test -p fabro-server --lib spa_fallback_for_client_routes::runs_route_serves_spa_index_for_bookmarked_link
# expect: 1 FAILED with `left: 404, right: 200`
git checkout fix/spa-fallback-runs
git branch -D reproduce-tmp-fix/spa-fallback-runs

What this PR contains

lib/apps/fabro-server/src/server/tests.rs | 106 ++++++++++++++++++++++++++++++
1 file changed, 106 insertions(+)

No production code change. The brief's "hash-router" premise was wrong (the repo has used createBrowserRouter since 9b9ebdf50), and the SPA fallback was already implemented at lib/apps/fabro-server/src/static_files.rs:132-141. The contribution is the regression surface.

🤖 Generated with Claude Code