Public: anyone can read this; the crew can change it.
Migrated from github.com/Trevato/SO-ARM100
workbenchthis machine
STOP can lose to a command already in flight: the relay must drop what it holds, and the Pi must refuse commands older than its last STOP #434
Sign in to join this conversation.
new above
offline
Safety, P0. On deploy run 01m5fhgt94h97cs5n5s3v19nm0 (commit 037dab00, 2026-10-21 15:3xZ), 691 tests passed and this one failed:
stop.test.ts› "STOP through the relay, to a Pi that holds nothing itself" › (a) any interleaving of drag, slider, STOP, resume, disconnect and turn change: nothing reaches the Pi between a STOP and a resume plus a new commandFrame 11 was made before a STOP but reached the Pi after the STOP. A command already in flight in the relay beat the STOP under load (the check cell was busy). The seed is fixed, so the outcome depends on timing. It passed on 2bd01031 at 13:49. The page-only change in 037dab00 cannot touch this path.
On the real arm, a STOP is only a STOP if nothing the relay already accepted is sent after it.
Done when:
stress, or with--rerun-each 50in a busy cell). Add a case with a slow Pi link, where the write is delayed 50 ms.Also seen in the same log:
arm server 100.89.95.68:8101: reconnecting … unreachable. A test in the check reached for the real Pi's address. Tests must only ever talk to fakes on 127.0.0.1. Find which test reads the live boot config and make it hermetic.