Files
buzz/mobile/lib/shared/push
Wes b77ba86262 perf(mobile): cut test wall time and run mobile CI lanes in parallel (#8347)
## Problem

Mobile pushes and PRs are slow.

- **Pre-push takes ~133s on a mobile-only change.** About 115s of that
is `flutter test`, and the finish time is set by a few files that wait
on real clocks.
- `push_presentation_export_test` (75s) and
`push_presentation_recovery_test` (19s) sleep through the full 7.75s
retry backoff.
- `channel_messages_provider_test` (70s) sleeps through recount and
ownership retry backoff on real time.
- **The required Mobile CI job takes ~15–16 min.** It runs tests
(390–430s on a 4-vCPU runner, 2 test files at a time) and then the
Android debug APK build (~400s, no Gradle cache) back to back.
- **Mobile Swift takes 18–21 min.** The unsigned release build (~4.5
min) waits behind the simulator tests.

Pre-commit is ~3s and isn't part of the problem.

## Change

**Tests** (each one still asserts what it did before)
- **Push exports.** Production retries wait through
`debugPushPresentationRetryWait`, which defaults to `Future.delayed`.
The export and recovery tests skip the real waits and assert the exact
250ms → 4s backoff schedule instead. Export fixtures sign each event
once per file.
- **`channel_messages_provider_test`.** Every test runs on `fakeAsync`.
A file-local `test` wrapper keeps the 90 call sites unchanged. Timers
fire in their real order, and a test that waits on work outside fake
time fails instead of hanging. Raising the production retry bound
(`attempt < 2` → `3`) still fails the 5 exhausted-retry tests.
- **`signed qualified caller`.** This test mounts the page 48 times and
opens the profile sheet twice per mount. It now disables the sheet
animation; it asserts which pubkey and labels show, not motion.

**CI**
- **Mobile runs as five parallel matrix lanes:** checks (format,
analyze, unconfigured push test, recipe test), three `flutter test
--total-shards=3` shards, and the Android build. The required `Mobile`
gate still reads the combined result.
- **Android restores a Gradle cache,** which is saved only on pushes to
`main`.
- **Mobile Swift builds the unsigned release** in a parallel lane beside
the native build, tests and simulator checks.

## Validation

Measured locally on an 18-core Mac with warm caches:

| | Before | After |
|---|---|---|
| `flutter test` (full suite) | 115s | 74s |
| Summed per-test time | 411s | 307s |
| `push_presentation_export_test` | 75s | 38s |
| `channel_messages_provider_test` | 70s | 22s |
| Pre-push, mobile-only push | 133s | 95s |

- The CI lane timings come from this PR's own run.
- `actionlint` passes, and so do
`scripts/test-ci-required-context-isolation.sh` and
`scripts/ci-required-gates.test.mjs`.
- The pre-push hook (format, analyze, full `flutter test`) passed on
push.

## Not in this PR

- `channel_detail_page_test` (~55s) is now the local long pole.
Splitting it into shard files would let it run in parallel, but it is
one 16k-line library with `part` files, so that is a separate refactor.
- Caching Hermit and pub on macOS would cut ~100s of setup from each
Mobile Swift lane.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Signed-off-by: Wes <wesbillman@users.noreply.github.com>
Co-authored-by: Carl <32a2e2c9d428ee08902cab75d956da2c1d235a22d4766b0dd4138bf6e2e5db1d@buzz.block.builderlab.xyz>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-14 20:57:47 -06:00
..