test(e2e): slow lane: a target boots a real simulator and emulator, and off shuts both down

[V3RON]

Closes #372

Status

Implement: done (slow-lane tests only) Review: round 10, 0 open Mutate: n/a (no src change) Hardware: passed on 8d62bbc (3 of 3, slow lane, 2026-10-10) Gate: pending

Done when

• On a Mac with an iOS runtime and an Android system image: scripts/slow-e2e.sh runs e2e/slow-warm-pool.test.ts green, and its log names the two device.ready events before the first lease.
• On a machine without one of the platforms, the test skips with a reason naming the missing platform.

Review

Review passed at 8d62bbc after round 10 with no blocking finding open. No finding was overruled.

Assumptions

• Assumption: where the spec names the docs/internal/ARCHITECTURE.md test-lanes paragraph — that file has no test-lanes paragraph; the lane docs live in e2e/README.md, so its known-gaps line about real emulators now names the new file there.
• Assumption: the new file joins scripts/slow-e2e.sh by being passed as its argument (the script runs whatever files it is given; it lists none).
• Assumption: the spec's order (wait for both ready, then restart with the pool off) — the off test now starts with the pool off, because a shut-down extra device of the kind made the next boot ambiguous on hardware; the ready path is proven by the other two tests, and this is the conservative way to make "each boots on the next lease" assertable by device id.

Written by an agent.

[V3RON]

Hardware postponed

The review passed at 8d62bbc, with no blocking findings after round 10. The slow-lane check still has to run.

• Run 1 (f6e06de): tests 1 and 2 passed. Test 3 failed because it expected the second lease to get the device it had released. The planner boots the first shut-down device that fits, so test 3 was reworked: it now starts with the pool off, so the released device is the only one of its kind.
• Run 2 (f1088de): tests 1 and 2 passed. Test 3 hit its 30-minute timeout, and the run did not show which step hung. Test 3 now logs every step and gives each one its own time limit. The lane also uses --reporter=default, so the device.ready lines reach the log.
• Run 3 (8d62bbc): stopped by the maintainer about 20 minutes in, and moved to the next day. It had logged both device.ready events before any lease.

Next: run e2e/slow-warm-pool.test.ts through the slow lane on 8d62bbc. If test 3 fails, the log names the step that hung. This machine has both platforms, so the skip-when-missing case can't be shown here.

An emulator (qemu, pid 53897) was still running after the stop. It may come from run 3, but that is not proven, so it was left alone.

Written by an agent.

[V3RON]

Hardware: passed

Ran scripts/slow-e2e.sh e2e/slow-warm-pool.test.ts on 8d62bbc, on a Mac with an iOS runtime and an Android system image. This is the run the earlier "Hardware postponed" comment was waiting for.

• Result: 3 of 3 tests passed. Duration 957 s. Exit code 0.
• Test 1: one ready simulator and one ready emulator before any lease. Both device.ready events (boot 38 s and 170 s) are in the log before the first lease. 241 s.
• Test 2: the first lease of each targeted model carries no booting progress, and each device is ready again after release. 447 s.
• Test 3 (warmPool.enabled: false): the released simulator and emulator both end shutdown, and each boots on the next lease. Every step finished inside its time limit. 269 s.

Not shown: the skip-when-a-platform-is-missing case, because this machine has both platforms.

Issue #372 and its parent #359 are closed. This PR is still a draft; its Done when boxes and gate are for the maintainer.

Written by an agent.