Skip to content

Rescue: 3 s per notification subscription, only the channels it needs - #109

Merged
freethinkel merged 3 commits into
mainfrom
rescue-fast-fail
Nov 21, 2026
Merged

freethinkel merged 3 commits into
mainfrom
rescue-fast-fail

Conversation

@freethinkel

Copy link
Copy Markdown
Collaborator

Already deployed to fmc.tools from this branch; merging so main matches prod.

A Watch 3 Pro in a boot loop stalls every attempt on the first startNotifications, the first step the watch itself has to answer, until it restarts about 17 s later. Rescue now subscribes only to the command and data channels (plus the shell when it has to pair) and gives up on a subscription after 3 s, so failed attempts end quickly and more fit into the windows when the watch is up.

Test: tests/ble-within.test.ts. Not verified on a real watch.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CQSU8Nr9jJ3ALiRfiDuC6m

freethinkel and others added 3 commits November 21, 2026 09:31
…nes it doesn't need

On a Watch 3 Pro stuck in a boot loop, every attempt now stalls on the first
startNotifications — the first step the watch itself has to answer, since Windows
can serve the service list from its cache — until the watch restarts about 17 s
later. Rescue now listens only on the command and data channels (plus the shell
when it has to pair) and gives up on a subscription after 3 s, so a dead try ends
fast and leaves time for more tries in the windows the watch is up.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CQSU8Nr9jJ3ALiRfiDuC6m
…il answered

Found in the Watch 3 Pro firmware and confirmed in the cmf-watch-utils emulator:
- the watch ignores a factory reset until a time write (0x8004) has marked the
  link live, and a clock more than 2 min off takes the time only on the second
  write — so the reset now goes out after the time, twice, in the reference
  client's format;
- the watch drops every frame for its first 200 ms on a link, so the first
  authName was lost whenever rescue connected quickly; rescue now re-sends it
  every 100 ms until the watch answers.

Before this, the rescue page's reset could not have worked on any watch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CQSU8Nr9jJ3ALiRfiDuC6m
…or timings

The cmf-watch-utils emulator (WP3 firmware) puts numbers on both: a live watch
answers a subscription in 8-11 ms, while one past its crash keeps advertising and
accepting a link but answers nothing until it reboots, so waiting 3 s only delays
the next try. And authName re-sent every 20 ms lands 5-6 ms after the 200 ms gate
instead of 38-61 ms, with exactly one handled.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CQSU8Nr9jJ3ALiRfiDuC6m
@freethinkel
freethinkel merged commit ba43783 into main Nov 21, 2026
1 check failed
@freethinkel
freethinkel deleted the rescue-fast-fail branch November 21, 2026 08:45
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant