With Discord closed, every now-playing edge (two per track: the immediate
publish and the poll's) tried the IPC socket and the crate logged
"find_pipe: could not find pipe" each time, so a listening session read
as an error stream. A failed connect now sets a five-minute deadline,
logs one line at info, and later edges push nothing until it passes;
opting the integration in again clears the deadline. The crate's own
warning is demoted to error in the default log filter. Pure backoff rule
covered by a unit test.
The Discover preferences store also logged "unavailable" when its lazy
init ran before the session restored; that startup order is normal and
the handle is rebound on activation, so it is a debug line unless a real
user directory fails to open.
cc
Frontend-agnostic core in qbz-integrations::discord (DiscordRpc +
NowListening) — the Tauri backend minus the tauri::command/State
wrappers: lazy IPC connect, Listening activity (details/state/assets +
"Get QBZ" button + timestamps), and the opt-in Flatpak socket bridge.
qbz-slint discord_rpc controller: process-global client, init AFTER the
session is active (PR #477 persistence fix — both online and offline
entry, never at early boot), push on the track-change and play/pause
edges, clear on logout. Opt-in toggle in Settings > Integrations
(DiscordState, persisted in ui_prefs.discord_rpc_enabled) + a Flatpak
socket-access note. The opt-in is per-machine (ui_prefs) rather than
Tauri's per-user.
cc
vicrodh