The SH1106 path always pushed the full 1 KB buffer over I2C every endFrame,
even on static screens (clock, home) that don't change between updates. Hash
the GFX buffer (FNV-1a, no external dep — the CRC32 lib is only wired into
e-ink builds) and skip display.display() when it matches the last frame
pushed. _force_redraw forces a flush on the first frame and after
turnOn()/clear() so a post-wake or post-clear frame can't be wrongly skipped.
Mirrors the existing e-ink CRC-skip. Builds green: WioTrackerL1_companion_solo_dual.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
GPS support was implemented incorrectly for the ThinkNode M3. The reset
pin was being driven when it should be left floating for this unit. The
enable pin also wasn't being picked up by `MicroNMEALocationProvider`
because of a mismatch in constant naming conventions. I did a general
cleanup of the GPS and ThinkNode M3 bring-up code so that constant names
line up and "*_ACTIVE" constants are used consistently vs hardcoding
`HIGH`/`LOW`. After making these changes, serial data immediately starts
streaming in from the NMEA on boot and GPS detection just works.
LED handling was also not quite right for the ThinkNode M3. The LoRa TX
LED was being driven high to turn it on when it should actually be
driven low. I changed the code to use the `LED_STATE_ON` constant and
also added a little code to the shutdown path to properly make sure that
all LEDs are turned off.
Tested and confirmed working on real hardware.
Resolves both #1864 and #2879.
hasSeen() was simultaneously a predicate and a mutator — it inserted the
packet hash on every miss, making five call sites that only wanted to mark
a packet as sent call it with the return value discarded.
Split into:
- wasSeen() — pure predicate, no side effects
- markSeen() — explicit insert
All query sites now call markSeen() immediately after wasSeen() returns
false, preserving identical runtime behaviour. The five mark-only send
sites (sendFlood, sendDirect, sendZeroHop x2) now call markSeen directly.
Also fixes three bridge sites (BridgeBase, ESPNowBridge, RS232Bridge)
that had the same query+implicit-insert pattern.
Tests: add test/test_mesh_tables/ covering wasSeen purity, markSeen,
dup stats, and clear. Update SHA256 mock to produce deterministic output
(previously finalize() was a no-op). Add Packet.cpp to native build filter.
A single button press could surface as two CLICKs on the e-ink build, most
visibly as start+stop on the stopwatch. Two contact-bounce paths fed the
IRQ edge-capture machinery a phantom press/release pair:
- a bounce edge accepted just after a clean release was replayed as a real
press — ISR_DEBOUNCE_MS (5 ms) was too short for the joystick switch; raised
to 25 ms so the settling burst is swallowed.
- the live-pin self-heal reconciled prev against a single raw digitalRead,
which can sample a bouncing contact mid-flap and synthesise a transition.
It now acts only once the divergence has been stable for ISR_DEBOUNCE_MS, so
a momentary read can't inject a click; a genuinely lost edge still heals
(~25 ms later) so the button can't stick.
Both thresholds stay far below any human tap cadence (>100 ms), so rapid
multi-tap navigation still registers every tap.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The joystick directions and Back were never begin()'d, so on the e-ink build
they never claimed a GPIOTE channel — edges landing during a blocking panel
refresh were lost and the IRQ-capture work didn't reach them. begin() them.
Even with edges captured, rapid taps of one direction replayed into a single
check() and collapsed into a double/triple-click event the navigation handler
ignores, and the loop only ever dispatched one key per render. So:
- MomentaryButton: for multiclick=false buttons emit one CLICK per completed
release (one per check() call) instead of collapsing — each tap stays a
discrete key. multiclick=true buttons (double/triple) are unchanged.
- UITask: add a key FIFO; drain each direction fully into it, apply the whole
queued burst, then redraw once. N taps captured during a refresh become N
navigation steps at the cost of a single refresh. Also fixes losing a key
when two buttons fire in the same loop iteration.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Added build flag "SX126X_USE_REGULATOR_LDO" (set to 1 to enable) to use LDO for the SX1262 radio instead of DC-DC when the DCDC pin isn't wired up (e.g. t-echo lite)
- Added additional debug mode diagnostic messages during SX1262 init to better highlight faults
Replace the ad-hoc enter()/markClean() entry methods (which lived outside
the UIScreen interface and were invoked via casts from each gotoX) with a
virtual onShow() lifecycle hook called centrally by setCurrScreen().
This removes the "forgot to call enter() in a new navigator" footgun and
the unchecked cast smell: 12 navigators collapse to one-line
setCurrScreen(x) calls, and override enforces signature match. Two entries
that carry a parameter/variant keep an explicit typed call after
setCurrScreen(): RingtoneEditor::selectSlot(slot) and TrailScreen::showMapView().
Behaviour-preserving: only screens that previously had enter() get an
onShow() override; Splash/Home/QuickMsg/Diag keep no reset as before.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
begin() claimed the trampoline slot and set _isr_slot before calling
attachInterrupt(), so a button that lost the GPIOTE race (only 8 channels,
shared with the radio's DIO1) would take the ISR branch forever while its
interrupt never fired. attachInterrupt() returns 0 when channels are
exhausted — only commit the slot on success, else leave _isr_slot = -1 so
check() uses the polling path. Combined with the live-pin reconcile, an
exhausted button degrades cleanly to plain polling instead of misbehaving.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The interrupt path trusted the replayed edge stream alone (btn = prev) and
never sampled the live pin. Any lost edge — buffer overflow, a debounce-
dropped settling edge, or a missed GPIOTE event — left prev diverged from
the hardware until the next captured edge, i.e. a stuck button. After
draining the queue, reconcile against digitalRead(): a no-op when the edge
stream already matches, a one-call self-heal when it doesn't.
Also remove dead code in applyTransition() (a local `event` that was always
BUTTON_EVENT_NONE, making its cancel-on-click branch unreachable) carried
over verbatim from the original inline logic. No behaviour change.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
On nRF52 the buzzer advanced notes by polling millis() >= _note_end_ms in
loop(). A blocking display refresh (e-ink endFrame) starves loop(), so a
note boundary that falls inside a refresh is serviced late — the note
plays long, or the next is skipped. The keypress-time 300 ms render delay
only masked the first note.
Advance notes from a hardware TIMER1 compare interrupt instead, scheduled
for each note's exact duration, so timing is independent of render cadence.
TIMER0 is the SoftDevice's; TIMER1 is free (tone() uses PWM2, nrfx TIMER1
driver is disabled). loop() becomes a no-op on nRF52; the UITask keypress
render-delay workaround is removed.
_disarmNoteTimer() clears the latched NVIC pending IRQ (not just the event)
on stop()/_nrfBegin(), so a note-advance latched just before a stop can't
fire spuriously — which would skip a new melody's first note or blip after
an explicit stop. Event read-backs flush the write buffer per the nRF52
event anomaly.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
E-ink panel refreshes block the main loop for 500ms-3.6s (GxEPD2's
display.display() waits on the BUSY pin), and MomentaryButton::check()
only ever samples the live pin level once per loop iteration. A full
press+release that lands entirely inside a refresh is never sampled,
not just delayed - users on WioTracker L1 Eink were losing taps.
Add an opt-in ISR path (BUTTON_USE_INTERRUPTS) that latches each edge
with its own timestamp into a small ring buffer from attachInterrupt(),
independent of what the main loop is doing. check() replays buffered
edges through the existing transition/multi-click logic afterwards, so
click-duration and multi-click-window math still uses the edge's actual
capture time rather than "now".
Scoped to wio-tracker-l1-eink only for now: it's the one board this was
reported against and the only GxEPDDisplay-class variant that currently
builds clean on main. OLED boards refresh in ~20-30ms so polling already
catches every press; analog-button boards (rak4631/rak3401) can't use
attachInterrupt(CHANGE) at all.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Squash merge of feat/location-beacon-alerts-autopause (v1.21).
Features:
- Live Location Sharing — broadcast position as movement-gated [LOC]
messages to a channel or contact; live shares show as map pins with
distance/bearing in Nearby Nodes and a status-bar indicator. [LOC] is
parsed in DMs, channel messages and room messages; DM shares name the
sender.
- Locator (geofence) — arm a geofence around a saved waypoint or a person
(live [LOC] or last-known position), alert on arrive/leave or near/far,
with an optional homing beeper (gated to arrive/both modes). Arm from
Tools > Locator, Nearby Nodes, or Waypoints; target picker lists
favourites first, clearable via a "None" entry. Active target is drawn
as a flag on the map.
- One active target shared across Locator / Navigate / Map via a single
resolver (resolvePersonPos / activeTargetPos) that prefers a live
[LOC] share over the last-advertised GPS fix.
- Follow live contacts — Navigate to a live-sharing contact follows them
as they move and adds an ETA line; quick-share your own position from
the Map.
- Map & status-bar upgrades — home mini-map gets a north marker, scale
tick, and a connected trail line (was disconnected dots); status line
shows tracked-node count, an arrow + distance to the active Locator/Nav
target (falling back to the nearest live-tracked contact); GPS fix icon
in the top status bar, shown only on GPS boards with GPS enabled.
- Trail auto-pause — recording freezes on stops (banking elapsed time,
breaking the map line across the idle gap) and resumes on movement
without ending the session.
- Streaming trail simplification — GPS points are simplified in-stream
via a fixed-corridor (Reumann-Witkam) pass tuned for fidelity: straight
runs collapse to their endpoints, curves stay bounded to the Min-dist
tolerance, so the 512-point buffer covers a far longer route than a
flat point budget would suggest.
- Collapsible Tools (Location / Comms / System sections, fold-in-place
like Settings) and page-indicator icons on the home carousel.
- Waypoint coordinate editor — add a waypoint by scroll-editing lat/lon
digit by digit.
Fixes:
- Critical: low-heap hang and contact loss on RAM-tight builds. Halved
message-history scrollback rings (recovering ~14 KB free heap) and
made contacts/channels/prefs persistence atomic (temp file + rename),
so an interrupted save can no longer corrupt or wipe the store.
- Serial.write() bounded so a stalled USB host can't hang the device.
- Nearby Nodes: live [LOC] senders respect the type filter, sort by
shared position, and the list refreshes so live shares bubble up.
- Map: live contacts are labelled before waypoints.
- GPS status icon hidden when GPS is off in Settings.
- Splash screen no longer truncates a pre-release tag's own dash (e.g.
v1.21-rc1) when stripping the build's commit-hash suffix.
- Null-guarded the Locator target picker; clamped loc-share channel
index on load.
Under the hood:
- -Os size optimisation on the e-ink and GAT562 30S solo envs.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
While working on some sensor code implementations, I ran into
some hard crashes that root-caused to the 4KB loop task stack
being exhausted.
Looking for various optimization schemes resulted in this relatively
low-lift fix. RAK3401 and RAK4631 both support hardware crypto.
Rather than loading in one of the two software crypto libs,
we can just use the onboard hardware. This is faster, should
consume less power, and in testing used a scant up to 700
bytes in the run loop vs 2.5-3KB per advert.
This change **only** affects advert verification processing, which
currently consumes a significant chunk of the 4KB run loop.
I figured such a change should likely be implemented in phases.
After soaking, this hardware crypto verification
process could be implemented across the entire MeshCore
cryptographic function on RAK nodes. Also possible other nodes
have available hardware crypto, however, I have not checked,
so future improvements may also exist there.
Tested on:
- RAK3401 RAK 1W
- RAK4631 19001
The SH1106 OLED on the LilyGo T-Beam Supreme V3 never lights up because:
1. The primary I2C bus (Wire) that the OLED, BME280 and magnetometer share
is never initialized. The PMU/RTC use Wire1 (via XPowersLib); nothing
brings up Wire on PIN_BOARD_SDA/PIN_BOARD_SCL, so the display driver
talks on the wrong default pins.
2. The OLED is at I2C address 0x3D on this board (0x3C is taken by the
QMC6310 magnetometer), but SH1106Display::begin() only tried 0x3C.
Initialize Wire for TBEAM_SUPREME_SX1262, and have SH1106Display::begin()
prefer 0x3D (only ever used by the OLED) before falling back to 0x3C.
Backwards-compatible for boards with the OLED at 0x3C.
Closes#2609
Address the open items from the FEATURES.md audit backlog:
- saveChannels() now skips unused slots (all-zero secret) instead of
writing all 40 every time, so /channels2 holds only configured channels
(was always ~2.7 KB) and wears the flash less. loadChannels() already
compacted empties on read, so the loaded result is unchanged.
- findChannelIdx() returns -1 for an all-zero secret, so a corrupted/empty
channel can't match an unused all-zero slot and misroute messages.
- TrailStore gains a static_assert that CAPACITY fits the uint16_t save
header count, failing the build instead of silently truncating.
Also re-classify two audit items verified to be non-issues in current
code: the bot strstr "199-char" truncation (BOT_SCRATCH=200 >= MAX_TEXT_LEN
=160, unreachable) and the MSG_PICK reply title (rlen clamped to 20, so
title[24] never overflows).
Verified: WioTrackerL1_companion_solo_dual builds clean.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Settings > Radio gains a repeater toggle, 16 community-suggested radio
presets plus manual Freq/SF/BW/CR tuning (digit-by-digit Freq editor),
4 persisted user preset slots, and the packet pool bumped 16->32 so
queued retransmits stop starving incoming-packet allocation while
relaying.
Four opt-in politeness knobs (skip-advert, max-hops, yield, min-SNR),
all flood-only and off by default, plus overhear suppression that
cancels a queued retransmit if a peer relays the same packet first.
Adaptive Power Control is suppressed while relaying (pins TX power to
the ceiling) and duty-cycle RX is forced off, since a repeater needs
to hear and relay at consistent power.
A dedicated Tools > Repeater screen consolidates the toggle, the
politeness knobs, and live forwarding stats, plus an optional "Custom"
radio profile — a dedicated frequency/SF/BW/CR for relaying, separate
from the companion's own network, band-matched to the companion
frequency by default and used everywhere a radio change can happen
(boot, on-device toggle, app-driven CMD_SET_RADIO_PARAMS, Settings
radio edits) so the device never silently falls back to the wrong
params mid-relay.
Diagnostics gains the actually-forwarded packet count, real heap-free
via mallinfo(), a reset-counters popup, and hardened loop-detect
bounds. Also: a frequency-floor fix and a float-equality preset-match
fix in the repeater profile logic, deduped BW-table/valCol/profile-
seeding helpers shared between Settings and the Repeater screen, and
a build.sh fix tolerating control characters in `pio project config`'s
JSON dump.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Shows packet counts by category (RX/TX), radio noise floor/RSSI/SNR,
packet-pool free count and outbound queue length, uptime, and heap/stack
headroom on a single scrollable screen.
Adds generic per-payload-type RX/TX counters to Dispatcher (mesh layer),
plus pool-free/queue-length getters, and a new DeviceDiag helper for
nRF52 heap (linker-symbol + sbrk) and stack (FreeRTOS high-water-mark)
stats — the only platform needed for the 3 in-scope boards (Wio Tracker
L1 OLED/Eink, GAT562 30S), all nRF52840.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Content drawn under the title separator touched the line on every standard
screen, because listStart() == headerH() (the row right after the separator).
Graphical screens worked around it with a hand-rolled hdr+2.
Bake a 2px breathing gap into listStart() so every list gains it at once, and
switch the screens that hand-rolled the offset (Compass, Nav, Nearby detail) to
listStart() so the content top is computed the same way everywhere.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Fixes from a full ui-new audit (OLED + e-ink both build green):
- remove dead DisplayDriver::drawScrollArrows (all lists use drawScrollIndicator)
- SettingsScreen selection bar -> lineStep()-1, matching the drawList screens
- HomeScreen: don't force the 1s blink refresh on the CLOCK page (status icons
are hidden there anyway)
- FullscreenMsgView: clamp KEY_DOWN scroll to a cached _max_scroll instead of
over-incrementing and leaning on the next render to clamp
- DataStore: clamp use_lemon_font (>1 -> 0) on load, like the other enum fields
- move the ~1.5KB wrap scratch (trans/lines) off the render stack into shared
file-scope statics in FullscreenMsgView (single-threaded UI; the fullscreen
view and history list never lay out in the same frame)
Cleanup:
- drop dead members RingtoneEditorScreen::DUR_VALS and HomeScreen::sensors_scroll
- delete redundant manual scroll-clamps in handleInput across the drawList
screens (drawList already reclamps each render); remove the now-unused _visible
from QuickMsgScreen/NearbyScreen/BotScreen and BotScreen::scrollToSel()
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- drawList(): scrollable list skeleton (visible window, keep-sel-in-view,
scroll-column reserve, per-row callback, indicator). Migrate Tools, Bot,
Nearby, Waypoints list and the three QuickMsg lists onto it; centralises the
scroll-clamp/reserve/indicator that each had open-coded (and inconsistently).
- drawCenteredHeader(): centred title + separator, replacing the duplicated
two-liner across ~10 screens (incl. QuickMsg, which used width()/2 spacing and
was missed before); unifies the separator at headerH()-sepH().
- keyIsPrev()/keyIsNext() in UIScreen.h for the LEFT||PREV / RIGHT||NEXT idiom;
applied to 6 screens, behaviour-preserving. Cancel/context-menu stay
per-screen (they mean different things on different screens).
- rotLabel() dedups the 0/90/180/270 array in Settings.
- Unify selection-row height to lineStep()-1 (Tools/Bot matched the rest).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Pull repeated UI idioms into DisplayDriver and remove duplicated logic:
- DisplayDriver: add drawScrollArrows() and drawInvertedHeader(); replaces 11
copy-pasted scroll-arrow blocks across 5 screens and 3 inverted title-bar
blocks (NearbyScreen x2, NavView). Standardises the detail-view separator.
- useImperial(): single source on UITask; NearbyScreen/TrailScreen delegate
instead of each re-reading NodePrefs.units_imperial.
- NearbyScreen: extract saveSelectedWaypoint() (two identical blocks -> one);
drop the redundant local label buffer (WaypointStore truncates).
- UITask::shutdown(): power GPS off through LocationProvider::stop() instead of
poking PIN_GPS_EN directly — centralises enable/reset-pin + active-level logic.
Net -58 lines. Verified: GAT562 (SSD1306) and WioTrackerL1Eink (GxEPD) solo
builds compile clean.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- UITask::shutdown(): power GPS off using !PIN_GPS_EN_ACTIVE instead of hardcoded
LOW, matching MicroNMEALocationProvider::stop() (still LOW on current active-high
devices, but correct for a future active-low GPS).
- buzzer _noteFreq(): clamp octave to 8 instead of 7 so the parser's accepted range
(4-8) is fully consumed; previously an octave-8 digit was clamped away and could
leak into the stream as the next note's duration.
- Waypoint::readFrom(): check read() return for version/reserved/count header bytes
so a truncated file is rejected instead of using garbage count.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Two independent, default-off toggles under Settings › Radio.
Pwr save: hardware RX duty-cycle (SX126x SetRxDutyCycle via
startReceiveDutyCycleAuto). The chip cycles RX↔sleep and wakes on a preamble —
no MCU state machine; recvRaw reads the packet exactly as in continuous RX.
Falls back to continuous RX on non-SX126x. (Replaces an earlier software-CAD
state machine that fought the hardware: polling a warm-sleeping chip gave a
phantom-busy channel that stalled TX ~4 s and dropped ACKs in the scan gaps.)
Auto pwr: Adaptive Power Control. tx_power_dbm becomes a ceiling; actual TX
power tracks the reverse-link SNR margin (measured above the per-SF demod floor,
EWMA-smoothed, proportional step with a deadband). Feedback comes from direct /
room-server ACKs and, for channels (no ACK), from hearing a repeater rebroadcast
our own flood; a lost confirmation ramps power back up so channel sends can't get
stranded below what the repeaters can hear.
Prefs schema 0xC0DE0009 (rx_powersave, tx_apc). Radio page / name bar show the
live TX power; noise floor reads n/a while duty-cycling.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
JakubandClaude Opus 4.8
Kevin Le
Quency-DandGitHub
ripplebizandGitHub
Naomi Smith
Liam CottleandGitHub
Christoph Koehler
MarekZegare4andClaude Opus 4.8
entr0p1
taco
Nick Dunklee
UltimateCodeWarriorand=UltimateCodeWarrior
Ryan Gregg
Wessel Nieboer
Wessel NieboerandWessel Nieboer
Marco