mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-05 19:28:25 +00:00
Claude
3c65e601df
The convergence-based disband landed in `commons` and `quartz` but three things it depends on only exist per front end, and the app had none of them. **Nothing would have carried the pass to settlement.** Terminalization now happens when the convergence pass SETTLES, and the settler is started by inbound traffic that detected a fork — but a disband opens its pass from a local outbound Commit with no fork, so no carrier ever started. On Android the group would have sat in `Recovering` behind its own `Disbanding` gate forever: nothing sendable, never ending. The CLI never showed it because the harness drives settlement explicitly. `disbandGroup` and the regeneration path now start the carrier themselves. **The gate was not durable anywhere real.** Gate storage is defaulted on `MarmotPublishObligationStore` so existing stores keep compiling, and neither `AndroidPublishObligationStore` nor `FilePublishObligationStore` overrode it — so the previous commit's durability claim held only for the in-memory store the test used. Both now persist gates: one file per group beside the obligations. Android writes them unencrypted, unlike an obligation, because the value is one enum name and the filename is a group id the device already stores in the clear — no key material, no message content. `FileStoresGateTest` pins the round trip on the real file store, including that a gate write is not mistaken for an obligation on reload. **The UI announced an ending that may not have happened.** The toast said "Group disbanded" as soon as the call returned, which used to be true because an unacknowledged publish threw. It no longer throws — the request stays durable and pending — so `disbandMarmotGroup` now returns whether the group is terminal, and the screen says "Ending the group" when it is not. Leaving the screen is right either way: the group takes no further outbound work. Verified: quartz 4836, commons 1886, cli 53 green, and the full MDK interop harness is 29/29 at the new 0.9.21 pin with all of this built in — including test 05, whose earlier failure was the loopback relay dropping a websocket before OK rather than anything in 0.9.21. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016kCuA6tc4JQzHPCDd39GHq