Only ';' needs encoding, so the title and body fit in one TXT string.
Values that aren't valid percent-encoding are used as-is instead of
hiding the banner.
The "get a node" banner's link, and optionally its title, body and button
text, now come from the _banner.meshcoreopen.org TXT record, looked up
over Cloudflare DNS-over-HTTPS and cached for the record's TTL (1 min to
6 h). The record is `v=1;on=1;url=<https link>` with URL-encoded values;
`on=0`, a missing record or any lookup failure hides the banner.
The builder panel now labels its actions, lets the user pick round trip
or as-entered with an explanation, and only sends when "Run path trace"
is pressed. Hops, mode and the last result are kept in UiViewStateService
for the session, so going back no longer clears them and the last result
can be reopened without sending again.
Notification bodies now use the chat's parsers to show a localized
summary for shared locations (m: markers), channel images, GIFs and
reactions instead of the raw payload. The marker parser moves out of the
map screen into helpers/marker_message.dart so the service can use it.
- Detect mentions on the original text, as @[name] or a whole-word @name
- Save and restore the full mode map in review mode
- Open the mode picker from the chat bell and show the mode in its tooltip
- Share one mode picker sheet; fix the stale sheet context
- Persist modes by name and skip unknown entries
- Keep writing muted_channels so older builds keep the user's mutes
- Remove the muteChannel/unmuteChannel/isChannelMuted shims
- Update notifications docs to match the behaviour
- Respect the notifications-enabled setting
- Notify only the lowest newly crossed threshold
- Ignore implausible readings (boards without a battery sensor report 0 mV)
- Wait for the device identity so the saved chemistry is used
- Require two consecutive low readings before alerting
- Re-arm and clear alerts on any link reset, not just explicit disconnect
- Cancel stale notifications on re-arm and link reset
- Scope notification ids per device and name the device in the body
- Use the connector's shared NotificationService
- Add ChannelNotificationMode enum with all, mentionsOnly, muted options
- Replace mutedChannels Set with channelNotificationModes Map for per-channel settings
- Add mention detection for @username in channel messages
- Update channels list and chat UI with notification mode selector
- Add localization strings for new notification modes
- Automatic migration from old mutedChannels to new system
- Update user documentation
- Preserve backward compatibility with existing mute/unmute API
- Uploads carry a consent generation: nothing is sent if sharing is
turned off while a batch is being prepared, and a late response never
touches a queue that was cleared and refilled
- Parse only the signed 32 advert app-data bytes, as the firmware does,
so appended unsigned bytes can't add the hidden marker or change the
name
- Split online map requests that cross the antimeridian into two valid
boxes and merge the results
- README: link the network map next to the website
- Opt-in upload of header-only packet records and repeater/room server
adverts to map.meshcoreopen.org about once a minute; payloads, channel
and DM hashes and transport codes are stripped on the phone, locations
snapped to a ~2 km grid, install ID rotates daily
- One-time opt-in dialog on first radio connection; toggle and status in
App settings
- Local / Online / Both switch on the map screen with online repeater
hexagons, links, add-to-contacts and manage
- Node purchase banner on the connect screen (affiliate link, disclosed)
- Privacy policy and network map documentation
- Translations for all locales
Channel messages, settings, region, order and images were persisted by
radio slot index, so a slot reused for a different channel inherited the
old channel's history and settings (#359).
- Channel.storageId: first 8 bytes of SHA-256("meshcore-channel:" + psk).
- All per-channel stores key by channel ID; in-memory state stays by slot.
- The connector keeps a slot -> channel ID map, diffs it against the saved
map on every channel sync and reloads slots whose channel changed.
- Queued channel frames wait until this session's map is confirmed.
- One-shot per-radio migration from slot keys, using the cached channel
list the legacy data was written under; rerun-safe if interrupted.
- Deleting a channel clears its data; a channel removed elsewhere keeps its
history and gets it back when re-added. A PSK edit keeps its place in
the custom order.
- Slots sharing a PSK share history and settings.
- Image entries record their channel ID; a slot change resets that slot's
in-progress reassembly.
- Extract reusable widgets (chat composer/send pipeline, message bubble
frame, reaction chips, path map scaffold, radio settings dialog,
translation settings section, map POI clusters, nav tiles, etc.) and
slim down the screens that used them
- Map: hide/remove markers with restore, confirm before setting node
location, POI clustering, semantics labels
- Map cache: cancellable downloads and a bulk tile limit
- Repeater: unsaved-changes guard, CLI search/copy/share transcript,
framed-command debug preview, disconnected banner
- Confirm before clearing chats and debug logs
- Keep cached unread totals in sync on per-contact/channel updates
- Move remaining hardcoded strings into l10n and translate the 98 new
keys into all 17 non-English locales
Outgoing image entries were excluded from the maxEntries cap, and age
eviction only marks them evicted, so frequent senders accumulated entries
and sidecars without bound. Outgoing entries with nothing left on disk (or
failed) are now eligible, oldest first.
Rename _ChannelChatRow.timestamp to receivedAt: it already returned the
local arrival time used for sorting and day separators.
Review fixes:
- Keep stored message history beyond the 200-message window; stop
messages leaking between radios; reset link/session state on disconnect
- Accept firmware's 176-byte RX frames over USB/TCP
- Hold APP_START during contact sync; contact-sync watchdog; replay DMs
from not-yet-known senders; tombstone late command replies; finish
channel sync on ERR_NOT_FOUND; fail region-scoped sends cleanly
- DM retry watchdog when RESP_CODE_SENT never arrives; late repeater CLI
replies; ACK hash collisions; UTF-8-safe name truncation
- Image pipeline: delete-during-decode, lifecycle cancels, decoder memory
release, entry cap, dead worker recovery, atomic model writes
- Storage: deterministic reaction hash across web/native, Smaz
double-decode, per-entry parsing, per-radio path history and regions,
review-mode data isolation
- UI: safe channel slot selection, byte-length name limits, double-send
guard, mounted checks, localized hardcoded strings
- Skip persisted-store reload on repeat SELF_INFO
Features:
- Radio default flood scope (firmware ver 11+) and explicit unscoped
sends (ver 12+)
- Scan for nearby nodes (#570)
- Optional distinct first-hop repeater counting for channel resend (#589)
- Optional region next to hop count (#580)
Issues:
- Message text follows system font size (#569)
- Repeater/room password field accepts 20 chars (#576)
- TCP auto-reconnect and keepalive (#571)
- Order feeds by local receive time; flag future timestamps (#584)
- Coding rate help text (#568), French "Favoris" (#575)
- iOS/macOS app badge now clears
Translations for all new strings in every locale.
- Message notifications use Android MessagingStyle with Reply, Mark as read,
and Mute channel actions, so they work in Android Auto and on watches
- Replies go through the normal send path (length limits, SMAZ, Cyr2Lat,
retries), split into up to 3 messages, and report failures
- Declare Android Auto notification support and register the action receiver
- Add a bell icon to channel chat to mute/unmute; unify the channel mute key
so unnamed channels mute correctly
- Update README and docs for Android Auto support
dispose() no longer relies on FlSerial.closePort() to free the I/O buffers:
it checks the buffers' first byte instead of their address, so it usually
leaked both. Close via fl_close and always free by address.
A native close still running in a helper isolate is invisible to a fresh
isolate after hot restart, so connect() could fl_free() a slot mid-close.
Track the in-flight close with a per-process temp marker and wait for it
(bounded to 45s) before resetting the slot table.
- Updated LoRa airtime calculations in `lora_airtime_test.dart` to reflect new reference values.
- Introduced new tests for LDRO behavior and preamble symbol calculations.
- Added `contact_path_width_test.dart` to validate path hash width handling in contacts.
- Enhanced `model_changes_test.dart` to ensure legacy data migration works correctly.
- Created `remote_node_stats_test.dart` to decode and validate remote node statistics.
- Implemented `map_marker_parse_test.dart` for parsing marker text with coordinate validation.
- Developed `map_tile_cache_key_test.dart` to ensure API keys are excluded from cache keys.
- Added `message_retry_failure_test.dart` to test message retry service behavior on failures.
- Extended `path_history_service_test.dart` to preserve learned state across path attempts.
- Enhanced `received_image_store_test.dart` to handle concurrent writes and review fixes.
- Implemented `repeater_command_service_test.dart` to validate command response handling.
- Created `store_tolerant_parsing_test.dart` to ensure stores skip malformed records gracefully.
- Cleared untranslated JSON file for localization updates.
Sixth review round:
- disconnect(manual: true) joining an in-flight automatic teardown now
still marks it manual (and cancels the reconnect timer / stops the
background service), so a user-initiated disconnect can no longer be
followed by an auto-reconnect scheduled by the teardown it joined.
- _shouldAutoReconnect is false once _disposed, so a teardown that
outlives disposal cannot arm a timer that calls connect() on a disposed
connector.
- connect()/connectUsb()/connectTcp() and UsbSerialService.connect()
re-check the connection state after awaiting the in-flight teardown.
The await is a suspension point, so two callers could previously pass
the initial guard, resume together and both run the open sequence.
Fifth review round:
- MeshCoreConnector.dispose() sets _disposed, and _setState(),
notifyListeners() and markNotifyDirty() no-op afterwards. A teardown
started before disposal can still be running (the USB native close waits
out the kernel's closing_wait off the UI isolate), and its tail would
otherwise call notifyListeners() on a disposed ChangeNotifier.
- _handleSerialData() now only drops callbacks in the teardown states
(disconnecting/disconnected) instead of everything that is not
`connected`, so data arriving in the `connecting` window is still
ingested and buffered by the frame decoder.
Fourth review round:
- UsbSerialService.connect() now waits for _activeDisconnect before it
claims the `connecting` status, covering the Android USB-host path too;
previously the wait sat inside the desktop branch, so an Android
reconnect could still be overwritten by the old teardown.
- MeshCoreConnector.disconnect() is now a re-entrant-safe wrapper around
_disconnectInternal() that publishes the teardown as _activeDisconnect,
and connect()/connectUsb()/connectTcp() join it before starting. The
transport-level guard alone could not prevent the connector's own
teardown tail from setting the state back to `disconnected` after a new
connection was up. Repeat disconnect() calls now join the in-flight
teardown instead of returning early while it is still running.
flutter analyze clean; flutter test 570 passing, 2 pre-existing failures
in message_url_image_helper_test.dart (identical on a stashed tree);
flutter build linux --debug succeeds.
Third review round:
- disconnect() is now a re-entrant-safe wrapper around _disconnectInternal()
whose future is held in _activeDisconnect for the entire teardown, not just
the native close. connect() and dispose() wait on that instead of
_pendingNativeClose, so a reconnect can no longer open a new port and then
have the tail of the old teardown mark it disconnected.
- dispose() closes on any live handle (flh >= 0) instead of only when
isOpen() reports `open`: flserial keeps a valid slot and a running
SerialThread in the I/O error state, which would otherwise have been
orphaned. A setDTR() failure no longer skips the close.
- _releaseSerialBuffers() takes alreadyFreed: closePort() frees both buffers
but leaves the pointer fields dangling, so after a successful synchronous
close the fields are only cleared, never freed a second time.
Builds: flutter build linux --debug succeeds (after patching the unrelated
glibc >= 2.42 / tinycthread once_flag clash in the flserial checkout).
Second review round:
- connect() awaits _pendingNativeClose before touching flserial's global
slot table. fl_close() (helper isolate) and fl_free()/fl_init() mutate
the same flserial_tab, so reconnecting during the kernel's closing_wait
could double-close the old slot or kill the freshly opened one.
- dispose() no longer re-enters disconnect() when a close is in flight;
the running disconnect owns the ordering (native close, then
subscription cancel), so dispose only chains _closeFrameController onto
the pending close.
- dispose() frees the I/O buffers through the new _releaseSerialBuffers()
on every path, not just after a successful closePort(); with _serial
nulled, an error/closed-state dispose used to leak both 8 KiB buffers.
- _handleSerialData() ignores callbacks once the status leaves `connected`:
a callback queued before flh was invalidated made readList() throw and
published a spurious transport error mid-teardown.
Review follow-up:
- dispose() now takes ownership of the FlSerial (nulls _serial) and clears
serialReadBuff/serialWriteBuff after the synchronous closePort(); flserial
frees those buffers but leaves the pointer fields dangling, so a later
disconnect() reaching _closePortOffUiIsolate() with the same object would
have freed them a second time. _closePortOffUiIsolate() additionally
returns early when flh is already -1.
- dispose() skips its synchronous close when a close is already in flight:
fl_close() sets breakThread and joins the SerialThread before it reaches
the blocking ::close(), so the thread — and its NativeCallable — is
already gone while the helper isolate waits out closing_wait. The
in-flight future is tracked in _pendingNativeClose.
- If Isolate.run itself fails the port is now closed inline rather than
leaking a native slot and SerialThread with no reachable handle.
close(2) on a tty whose output the peer never drained blocks in the
kernel for the line discipline's closing_wait (measured 30.4s on Linux
against a CDC-ACM device that never reads its bulk OUT endpoint, e.g. a
MeshCore board flashed with a BLE-only companion build). FlSerial's
closePort() is a synchronous FFI call, so the whole app froze for that
window on every failed USB connect and on disconnect.
fl_close() only touches process-global native state, so run it through
Isolate.run() and replicate the Dart-side cleanup closePort() does
(stream close, calloc'd I/O buffers, flh reset). Log when the close
exceeds 1s so the "device is not draining" case is visible in the debug
log.
Verified against the wedged device: inline close blocked the calling
isolate for 30476ms with 0 timer ticks; the isolate close took 30302ms
with 302 ticks, i.e. the main isolate stayed responsive throughout.
Companion firmware (MeshCore commit 4a869163, 2025-12-30) replaces the
CLI frame timestamp with the companion node's RTC for TXT_TYPE_CLI_DATA,
so "clock sync" sets the repeater clock to the node RTC. Nodes with a
wrong RTC push repeater clocks years into the future; firmware cannot
set the clock backwards without clkreboot.
- RepeaterCommandService: translate "clock sync" to "time <phone epoch>"
before sending, matching official meshcore-cli behavior
- repeater login dialog: send "time <epoch>" for auto clock sync after
login (this path bypasses RepeaterCommandService)
- add unit tests for the command translation
Ref: https://github.com/zjs81/meshcore-open/issues/464