- start binder before navigating from auth, mark isBinding synchronously
- overlap plex.tv resource refresh with optimistic cached-metadata connect,
reconcile tokens/membership in background once it lands
- cached endpoint probe gets a head start on the race instead of serially
blocking it
- defer phase-1 HTTPS upgrade off the splash critical path
- per-server connect progress stream for incremental splash checkmarks
- startup timing instrumentation (bind/fetch/race/connect elapsedMs)
- Skip redundant notifyListeners in HiddenLibrariesProvider init when
set is empty (empty→empty is a no-op for consumers)
- Track last-seen hidden keys in DiscoverScreen to avoid reloading
content when the listener fires without an actual change
- Track previous online server set in MultiServerProvider; only re-check
DVR availability when a new server comes online
- Remove dead content-loading pipeline from libraries_screen (sorts,
pagination, filters, items) that was never rendered
- Move /media/providers fetch into PlexClient.create() init, replacing
lazy EPG cache and providing libraries including shared items
Add per-platform cooldown on resume health probes (10s mobile, 2min
desktop). Debounce connectivity listener by 2s. Coalesce concurrent
checkServerHealth and reconnectOfflineServers into single in-flight
futures.
Connection probe requests in testConnectionWithLatency were sent without
X-Plex-Client-Identifier, X-Plex-Product, or X-Plex-Device-Name headers.
Plex treated each anonymous probe as a new unknown device and fired a
"New Device" notification for every server tested (one per shared/owned
server), while displaying "Flutter" as the device name from the HTTP
user-agent.
Pass clientIdentifier through findBestWorkingConnection and all connection
test helpers so every probe request identifies itself as "Plezy" with the
persistent client UUID. This prevents spurious notifications and ensures
the device shows the correct app name in Plex's device list.
https://claude.ai/code/session_01V5VraujkNmk5GGPLyZ33fN
edde746