Make sleep tracking work, and suppress only auto-reconnect rather than discovery

[kmatzen]

Third of four stacked PRs. Base is fix/packet-reassembly (#88), not main.

This one changes user-visible behaviour, so the reasoning matters more than the diff.

Sleep tracking never worked at all

setDeviceSleeping wrote through an optional chain into discoveredDevices and connectedDevices. Both are only ever populated by addDiscoveredDevice — which nothing calls. Both dictionaries are permanently empty, so the setter was a silent no-op and isDeviceSleeping always returned false.

Net effect today: the app immediately re-discovers and reconnects a camera you just put to sleep.

after setDeviceSleeping(true), isDeviceSleeping = false
>>> SILENT NO-OP: sleep never recorded

Why just recording the flag is not the fix

Suppressing discovery for a camera believed asleep makes it permanently unreachable: it is withheld from discoveredGoPros both on disconnect and on every advertisement, and connectToGoPro requires it to be there. The user has no way back.

I modelled the alternatives in TLA+ (specs/SleepState.tla, included in the next PR). Every advertisement-based design fails one of two properties:

Design	Does not undo own sleep	Never permanently invisible
current — flag never set	✗	✓
always ignore while flagged	✓	✗
ignore inside a timed window	✗	✓
clear only after observed silence	✓	✗
suppress auto-reconnect only	✓	✓

The timed-window version is the one I wrote first and believed was correct. TLC found the counterexample in seven steps: the camera is still advertising when the window closes, so the app re-discovers it mid-shutdown.

The ambiguity is fundamental — a camera still shutting down and a camera that just woke up emit identical advertisements. No rule over advertisements plus elapsed time can separate them, so every such design must choose which way to be wrong.

The change

The old guard conflated two concerns — do not auto-reconnect a camera the user slept and do not show it — and only the first was required. Discovery is never suppressed now; the flag gates scheduleReconnectIfNeeded and the straggler retries. No assumption about shutdown duration anywhere.

One property I deliberately did not assert: never auto-reconnecting during shutdown. TLC shows it conflicts with honouring manual override — if the user taps connect mid-shutdown, that clears the flag, and a later auto-reconnect is then correct.

Testing

135 tests pass, including seven new ones. Mutation-verified: reverting sleep storage to the dictionary-backed version fails the two tests that assert the flag actually persists.

Three of the seven still pass under that mutation, because they assert absence of sleep state — trivially true when the getter always returns false. They guard against regressions in the other direction; they are not evidence the fix works.

Review notes

Not verified against hardware. The behaviour change is worth exercising with a real camera: sleep a camera, confirm the app stops reconnecting it, confirm it stays visible in the list, and confirm tapping it reconnects.

🤖 Generated with Claude Code

https://claude.ai/code/session_01AqfipPBp3eErDgZboooqxP