v0.99.58
- Silent Moonraker request loss during connect/reconnect (#909) — libhv's
WebSocketClient::sendonly checks that the channel exists, not that the WS protocol is pastCONNECTING/WS_UPGRADING/RECONNECTING. A send issued in those windows wrote WS frame bytes onto a stream Moonraker treats as still-handshaking, so the bytes were silently dropped and the request sat inMoonrakerRequestTracker::pending_requests_until the 60s timeout — or forever, if the calling panel had already cleared its in-flight flag. Surfaced as the K2 Plus startup race wherePrintSelectPanel::refresh_filesissuedget_directorybeforeonopenand the panel stuck on its refresh spinner for hours with no timeout warning.MoonrakerClient::ready_to_send()now gates all foursend_jsonrpcoverloads onconnection_state_ == CONNECTED; the 5-arg overload fires the caller'serror_cbsynchronously withCONNECTION_LOSTso panels see immediate failure instead of silence. Adds debug-visibility instrumentation to the tracker's timeout sweep so future regressions surface as a stuck queue in bundles. - Background Moonraker/HTTP callbacks now marshal LVGL work to the main thread (#933) — several success/error callbacks in
PrintStatusPanel(gcode-for-viewing download, reprint error path) andCrashReportModal::send_with_bundlecaptured a lifetime token and checkedtok.expired()but ran the body inline on the WebSocket / HttpExecutor worker thread. Bodies that touched LVGL widgets (safe_delete,lv_qrcode_create, modalhide,ui_set_button_enabled) raced the render loop and produced the L081-cluster heap corruption surfacing as SIGSEGV inget_prop_core/layout_update_corethe next frame. All affected callbacks now wrap their LVGL touches intok.defer(...)so the work runs on the main thread. Telemetry: WKC5J9SK (ad5x v0.99.56), B4SG5M79 (k1 v0.99.53).