- Fix NULL block_data deref in stream_send_chunk_cb
- Test B1: create file in <db_dir>/media/, commit to media_files
- Store media_base in ui_state for streaming handler path resolution
- 8/9 tests pass (streaming works but timing issue with SUPER_REPL flood)
- md_handle_block_req: real chunk streaming via etcp_router_on_send_ready
- stream_ctx state machine: backpressure waiter → send chunk → repeat
- content_hash + block_sig on stream completion
- Fallback to author when supernode has no block info
- media_download: author_node_id for peer fallback
- test_media_delivery_full: 8/9 pass (SUPER_HELLO, HAVE_BLOCK, replication, admission, file create)
- Known issue: streaming test hung (BGP route for n1→n2 not established in time)
VoiceBubble is now a pure renderer — no local state, no LaunchedEffect, no timer.
All state from ViewModel.playbackState StateFlow:
- playing/paused/elapsedMs/progress/filePath — one atomic snapshot
- isActive = playbackState.filePath == message.filePath — only active clip shows cursor
- onProgress callback from AudioRecorderManager feeds real elapsed into ViewModel
- ViewModel timer (80ms) updates progress between onProgress callbacks
Removed: VoiceBubble's playing/paused/seekVersion/PlayState/onDone/onPaused.
Removed: AudioRecorderManager's isPaused/elapsedMs/pausedElapsedMs.
stopped flag was set before join() but cleared after, causing race:
if thread was still blocked in decodeRead when stopped was reset,
onComplete fired anyway. Monotonic playGeneration ++ never decreases,
thread captures myGen at start, compares after completion —
correctly skipped if generation changed (another stop/seek started).
PlayState data class bundles playId, elapsedMs, progress into one atomic snapshot.
LaunchedEffect keyed on playId — when it changes (seek/resume), the timer reads
elapsedMs and progress from the SAME snapshot, eliminating the race between
seekVersion++ and elapsedMs=seekMs.
When playing=true and seek happens, LaunchedEffect(playing) key didn't change,
so the elapsed timer kept using the old start time. Added seekVersion counter
as second key — increments on waveform tap and play/resume click, forcing
the timer coroutine to restart with the new elapsedMs position.
When stopPlayback/pausePlayback kills the playback thread, the thread's
onComplete callback was still firing, resetting UI state (playing=false,
elapsed=0). Added @Volatile 'stopped' flag — onComplete only fires on
natural completion, not on external stop/pause.
- Canvas pointerInput moved to wrapping Box (state updates trigger recomposition)
- showCursor = playing || paused (cursor and white bars visible on pause)
- elapsed/dur stays visible on pause too
- ViewModel: _currentPlayingPath tracks active track. Tap same → stop.
Tap different → stop old, play new. onComplete clears currentPlayingPath.
- AudioRecorderManager: stopPlayback() already called at start (existing)
- VoiceBubble: simplified click, always sends to ViewModel for state management
- inputbar.cpp: CSV waveform → base64 log-scale (same as voice_recorder.c)
- messagelist.cpp: sendVoiceMessage removes displayName|dur_sec| prefix, uses wfStr directly
setVoiceMessageRoles: decodes new format wf={base64};dur={ms};|... from data field
- messagedelegate.cpp: waveform values now pre-normalized 0..1, no log recalculation needed
- VoiceEncoder: no changes needed (was already compatible)
Both platforms now store identical format in chat_msg_submit.data:
wf=ABCDef...100chars;dur=NNNN;|file_size|block_size|num_blocks|sigs_hex
- voice_recorder: waveform_levels[100] accumulated during feed, normalized to 100 values in stop()
stored as 'wf=...;dur=...;' in chat_msg_submit data
- bridge: filePath in get_messages_json from local_attrs.fp, g_db_path moved global
- Message model: waveform: List<Float>, voiceDurationMs: Int, filePath: String
- VoiceBubble: Canvas waveform (100 bars, log scale), play button, duration
- ChatViewModel.playVoiceMessage → AudioRecorderManager.playVoiceFile
g_cc.db_path is '/data/.../files/chats.db' (file path), but media blocks
are written to '/data/.../files/media/<channel>/'. chat_msg was concatenating
'chats.db/media/...' — now strips the database filename, using only the
directory portion for the media path.
- voice_recorder: proper mkdir with errno/EEXIST check, errno logged on fopen failure
- MessageBubble: pointerInput wrapped in while(true) so PTT works after first gesture
- PTT Box: same composable identity in idle/recording states (just changes bg color)
Gesture coroutine survives recomposition → onPttReleased() fires on finger lift.
When locked, Box is replaced by cancel/send IconButtons.
- Min voice duration: 500→1000ms (C and ViewModel)
- audio_compressor.h/c moved to lib/ (alongside opus_codec, mem, debug_config)
- lib/Makefile.am: added to libuasync_a_SOURCES
- utun_sources.cmake: removed from _cfg_src (GLOB from lib/ picks it up)
- tests/Makefile.am: test_audio_compressor added to check_PROGRAMS
- test includes simplified (debug_config.h, mem.h — lib/ in include path)
- voice_recorder_init(db_path) in utun_bridge_init (was never called, g_rec stayed NULL)
- voice_recorder_deinit() in utun_bridge_destroy
- ViewModel: set _isRecording only after startRecording() succeeds
- voice_recorder: compressor config API (max_gain, lookback, lookahead, rise_rate, target_level)
- bridge/JNI: compressor_config export
- AudioRecorderManager: startMicTest/stopMicTest for level preview
- SettingsScreen: TabRow with Settings + Audio tabs
- Audio tab: mic test with live level bar, Opus preset selector, 5 compressor sliders matching desktop
evgeny