Files
meshcore-analyzer/.github/workflows
efiten b2c8693cbc ci(e2e): run the axe gate's viewport/theme pairs concurrently (#2433)
Refs #2424. After #2432 the E2E shards are the longest path of a PR run:
9m34s, 11m00s and 10m51s on its run 44394972332.

## Where the E2E time goes

From the timestamped logs of that run: 1,655 s of suites over three
shards, plus about 1 min of setup per runner.

- **`test-a11y-axe-1668.js`: 359 s, 22% of the total, alone on shard
1.** It checks 104 cells (26 routes × 2 themes × 2 viewports) one after
another, and each cell waits a fixed 1.5 s for the SPA to render.
- **Next largest:** `test-e2e-playwright.js` 153 s,
`test-issue-1147-section-order` 89 s, `test-observer-detail-ownership`
88 s, and the slide-over flake gate about 100 s.
- **117 suites under 10 s each:** 409 s together, mostly the start of a
node process and a browser.

## What changes

1. **The axe gate runs its four (viewport, theme) pairs at the same
time.**
- Each pair already had its own browser context; the route loop moved
unchanged into `runCombo()`, and `main()` awaits the four with
`Promise.all`.
- Each pair buffers its log lines, which print afterwards in the old
order, viewport headers included.
- The 1.5 s render wait is unchanged, so axe sees the same page state;
the waits only overlap.
2. **Four suites move to shard 1, which gains about 240 s:**
- from shard 2: `test-issue-1147-section-order` (89 s) and
`test-health-status` (29 s);
- from shard 3: `test-observer-detail-ownership` (88 s) and
`test-issue-1281-location-row` (25 s).

None of them uses the shard-specific servers (`:13582` on shard 3,
`:8788`/`:13591` on shard 1). Expected, with the axe gate scaled by the
local ratio: about 479 / 468 / 461 s of tests per shard, down from 490 /
586 / 574.

## Checked locally

Windows, Chrome, the e2e fixture migrated, server on `:13581`:

- **Same output, three times faster:**

  | version | time | result |
  |---|---|---|
  | upstream/master | 282 s | 104 cells, net=0 |
  | this branch | 92 s | 104 cells, net=0 |

The per-cell lines (`✅/❌`, viewport headers, `SUMMARY`) are identical in
both runs (`diff` of the grep).
- **The failure path still works.** Against a port with no server
(`AXE_ROUTES_ONLY=/packets`), all four cells report `❌ net=1`, four
screenshots are written, and the exit code is 1.
- **Unit tests pass:** `test-test-inventory.js`, the a11y unit tests
that load this file (selftest, routes coverage), and
`test-issue-1956-release-routing.js`.

## Not checked

- **The CI time.** It comes from this PR's run. Four pairs at once on a
4-core runner may speed up less than on my laptop; I'll post the shard
times.
- **The moved suites, run locally.** Their code is unchanged and they
use the shared server on `:13581`. On shard 1 they run after the
split-frontend suite instead of their old neighbours.
- **The other suites themselves.** `test-e2e-playwright.js` holds 54
`waitForTimeout` calls. Replacing fixed waits with readiness checks
would change what each suite waits for, so that is not in this PR.
2026-11-18 23:11:35 +01:00
..