meshcore-solo-site's new cross-visitor relay bridge (Phase 3) needs a way
to read a sim instance's live freq/bw/sf/cr so a WebSocket relay server
can fan traffic out only to other visitors tuned to the same params,
mirroring real LoRa channel isolation. Added sim_radio_get_params()
(examples/companion_radio/main.cpp), reading NodePrefs directly -- the
real live source of truth, since SimRadio::setParams() has always been a
complete no-op (accepts and discards its arguments). Out-params via
pointer, same buffer-pointer idiom sim_radio_poll_tx() already uses;
build_wasm.sh needed HEAPF32/HEAP32 added to EXPORTED_RUNTIME_METHODS so
JS can read them back.
While testing the relay bridge with two independent browser contexts, hit
a real bug: sim_instance_salt() (SimInstance.h) is a pure function of the
simInstanceTag STRING ('hero'/'B'/'R'), so it's only useful for telling
apart same-tab instances -- it's identical for two genuinely different
visitors who both boot a 'hero' instance. Combined with SimRNG::begin()'s
other seed ingredients -- time(NULL) (1-second resolution) and a WASM
heap pointer (no ASLR inside the sandbox, so it's fully deterministic
across independent boots of the same binary) -- two different visitors
landing on the same wall-clock second reliably generated byte-identical
Ed25519 identities (confirmed: two fresh browser contexts launched
together produced provably identical advert packets end to end).
Fixed by mixing in real crypto.getRandomValues()-sourced entropy from the
host page (new sim_instance_entropy(), read the same way simInstanceTag
already is) into both SimRNG::begin() and the twin seeding pattern in
SimRadio::getRngSeed(). Applies to every sim instance uniformly (hero/B/R
all get proper per-session entropy now), not just the new relay path --
strictly an improvement with no compatibility concern, since identity
generation only ever runs once per fresh/empty IDBFS.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
- Turn on ENV_INCLUDE_GPS/UI_SENSORS_PAGE for the sim build (both were
compiled out entirely, so the GPS and Sensors home-page carousel
entries didn't exist regardless of home_pages_mask) -- SimSensorManager
already fed a real JS-settable GPS fix and a temperature/battery
channel with nothing to display them.
- Add sim_test_advert_flood() to the repeater build too, mirroring
companion_radio's hook: MyMesh::updateAdvertTimer() otherwise leaves
the repeater's first self-advert 2 minutes out, so a host page
couldn't make it discoverable as a contact (needed for admin login)
right away.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
Real SSD1306/SH1106 OLEDs this sim mirrors are monochrome white-on-black,
not amber -- swap the canvas fillStyle/strokeStyle from #ffb000 to #fff
everywhere the sim display driver blits lit pixels.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
NodePrefs::auto_off_secs defaults to 15 -- real power-saving behaviour a
battery-powered board needs (screen off, optionally locked, 15s after
the last input). UITask::autoOffMillis() treats 0 as "never" and skips
that whole branch in UITask::loop() entirely, so there's no separate
flag to touch. A demo running on a visitor's screen has no battery to
save and no reason to go dark while they're reading it.
sim_test_disable_screen_timeout() is a new sim-only hook, same shape as
the other sim_test_* boot-time hooks: meshcore-solo-site calls it once
after boot.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
NodePrefs::tz_offset_hours defaults to 0 (UTC) -- correct for real
hardware, which has no other way to know the visitor's timezone besides
Settings > System > Timezone. SimRTCClock (see 718dbdbe) is already
correct live UTC (time(NULL)), but without a timezone applied, a demo
instance's Clock screen displayed correct-but-UTC time, which just reads
as "wrong" to a visitor who never opened Settings -- e.g. 20:40 shown
while it was really 22:40 CEST. This is the other half of the site's
"would be nice if the time was synced with the computer" ask; 718dbdbe
only fixed the stale-IDBFS-restore half.
sim_test_set_timezone_hours(int) is a new sim-only hook, same shape as
sim_test_show_all_home_pages(): meshcore-solo-site calls it once after
boot with the browser's own timezone offset. Whole hours only
(tz_offset_hours is an int8_t) -- same limit a real device's own
Settings field has.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
A Reset control on a host page (meshcore-solo-site's RESET button,
mesh.html's own Reset buttons) re-invokes the MODULARIZE factory for the
same simInstanceTag to simulate a device restart, since board.reboot()
just exit()s the wasm process (inert / freezes the tab under
-sEXIT_RUNTIME=0). Nothing ever tore down the OLD instance's own
emscripten_set_main_loop() callback, so it keeps ticking forever after
being "replaced" -- and keeps drawing onto the same simInstanceTag-keyed
<canvas> the new instance draws onto too. Two visible symptoms this
caused on meshcore-solo-site: flicker/reversion after resetting the same
device more than once (multiple orphaned instances competing for one
canvas), and -- together with the HomeScreen::poll() fix in 5fbfd7c5 --
would have still left a residual single-old-instance repaint race even
after that fix alone.
sim_stop_main_loop() wraps emscripten_cancel_main_loop() so a host page
can explicitly stop the OLD Module reference right before discarding it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
_shutdown_init (set true once KEY_ENTER is pressed on the Shutdown home
page) was never cleared, so this branch kept calling _task->shutdown()
on every single poll() tick indefinitely once triggered.
Invisible on real hardware: _board->powerOff() halts the MCU in the
non-restart path, so there's no next tick to matter. But
SimMainBoard::powerOff() is a deliberate no-op (no real hardware to
power off), so a sim instance keeps running after "shutdown" -- and each
repeated shutdown() call re-fires _display->turnOff(), which blacks out
its <canvas> (keyed by simInstanceTag) again on every frame.
This is what made meshcore-solo-site's RESET button unusable after a
device had been shut down: the still-running old instance kept
re-blacking the very canvas a freshly reset instance (same tag, same
canvas element) was trying to render its own boot splash onto -- visible
as a black screen, with the new instance's splash winning a single frame
every so often before being painted over again.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
DataStore::restoreRTCTime() runs on every boot and, if a prior save
exists, overwrites RTCClock with that stale timestamp. Correct on real
hardware (no other way to know the time before a GPS fix or a phone/CLI
sync), but SimRTCClock is already backed by the real host wall clock from
construction -- overwriting it with an old IDBFS-persisted save made a
returning meshcore-solo-site visitor's on-screen clock drift away from
their own real time instead of just showing it. Guarded behind #ifndef
SIM_PLATFORM.
sim_test_show_all_home_pages() is a new sim-only test hook, same shape as
the existing sim_test_* hooks in this file: real hardware ships with a
curated 5-page Home carousel (NodePrefs::HP_DEFAULT) so a first-time user
isn't handed 13 pages to joystick through, with the rest opt-in via
Settings > Home Pages. meshcore-solo-site calls this once after boot to
show the whole feature set instead, without touching the real-hardware
default.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
UITask::shutdown() busy-waits on buzzer.isPlaying() for up to 2.5s before
powering off. On real hardware that's a real (if crude) wait; in the
Emscripten sim it's a synchronous block on the browser's single JS/wasm
thread, which freezes the whole page for the duration -- reported as the
site "zacinanie się" (stuttering) whenever hibernate/shutdown triggers.
Guards it with #ifdef SIM_PLATFORM, mirroring the identical pattern
already used a few lines below for the low-battery pre-shutdown pause.
Verified empirically (not just by reading the diff): measured real
requestAnimationFrame throughput on meshcore-solo-site while triggering
hibernate for real (Home -> Shutdown page -> Enter). Before: 135 frames
in 3.5s (607ms max stall). After: 633 frames in the same window (110ms
max) -- confirmed with a real stash/rebuild before-after control.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018iubftDmKNWmkNnhJRz8UH
Code-review pass over the buzzer/sim commits turned up several real bugs,
plus two issues found afterward from manual browser testing:
Rendering (SimDisplayDriverCanvas, variants/sim/SimDisplayDriver.h + target.cpp):
- getTextWidth() measured UTF-8 BYTES (strlen()*6), not codepoints. Since
b067e95b stopped stripping accents, any accented string now measures
double its real width -- mis-centred titles, premature ellipsis/marquee,
badges pushed off-screen. Now uses the real MiscFixedRenderer measurement
(miscFixedTextWidth()), same as SH1106Display/SSD1306Display.
- Added the matching getCodepointWidth() override (O(1) single-glyph
advance), same pattern as SSD1306Display.
- isSingleFont() was left at the base class's `false`, though this backend
only ever renders MiscFixed -- UITask.cpp's status-bar indicator height
keys off this (`lh-2` vs `lh`), so the sim drew it 2px taller than a real
board.
- print() blitted the full 128x64 canvas on every call (dozens per frame,
60fps) -- now tracks a dirty bounding box and only clears/blits the
region actually touched.
Web Audio (buzzer bridge, index.html + mesh.html):
- No AudioContext.resume() -- a context created (or later suspended) in the
'suspended' state (Safari/Firefox, or any browser backgrounding the tab)
stayed silent forever. Now resumed on every gesture.
- linearRampToValueAtTime with no anchoring setValueAtTime interpolates
from the LAST scheduled event, not "now" -- so the anti-click ramps could
effectively snap instead of fading. Fixed with cancelScheduledValues +
setValueAtTime(current) before each ramp.
- mesh.html: a gesture only armed the clicked instance's audio. Click A,
send A->B, and B (the one actually meant to beep on receipt) stayed
silent. Now any gesture arms both A and B.
- RTTTL rests (freq=0, still "playing") now explicitly hold pitch and drop
gain instead of it happening to work by coincidence.
Misc: sim_test_get_num_contacts() was missing the g_sim_ready gate every
other sim_test_* hook has, so it could return a bogus negative count before
setup() finishes seeding num_contacts.
Splash screen missing "Solo <version>" bar: variants/sim never defined
FIRMWARE_SOLO_BUILD (every real Solo board does), so SplashScreen silently
skipped that whole line -- the sim looked like a plain non-Solo companion
build. Added -D FIRMWARE_SOLO_BUILD=1 to platformio.ini and build_wasm.sh.
Verified on a real canvas screenshot: "MESHCORE 1.17.1 / 19 Aug 2026 /
Solo v1.27".
Wasm-fetch error message: "failed to start: RuntimeError: Aborted(both
async and sync fetching of the wasm failed)" is Emscripten's own opaque
message for the single most common real cause -- the page opened via
file://...index.html instead of served over http(s) (fetch() on a local
file is blocked by CORS in both Chrome and Safari, confirmed by reproducing
the exact same error/stack via file://). Both harnesses now detect
location.protocol === 'file:' and show an actionable message with the
one-line fix instead of the raw stack trace.
Battery-set latency: SimMainBoard's battery value is an exact, instantaneous
JS-set integer (see sim_battery_set_mv()), but UITask's battery-check code
polls it every 8s and runs it through an EMA (alpha=0.2) meant to smooth a
REAL board's noisy ADC -- so a value typed into the demo UI could take tens
of seconds to visibly settle. SIM_PLATFORM now checks every 250ms and skips
the EMA (nothing to smooth), since the reading is already clean. Measured
on real canvas pixels: indicator update now lands within one screen-refresh
cycle instead of up to 8s+.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
genericBuzzer (src/helpers/ui/buzzer.h/.cpp) gets a third platform branch,
#elif defined(SIM_PLATFORM), alongside the existing NRF52 (direct PWM) and
NonBlockingRtttl paths -- purely additive, no changes to either real-hardware
branch. It reuses the NRF52 branch's already hardware-free RTTTL parser
(_parseHeader/_parseNext/_noteFreq, now shared via a widened guard) but
tracks (current frequency, note-end-time) instead of touching real PWM/timer
registers, advancing on plain millis() polling from loop() -- same
non-blocking shape UITask already drives every tick.
Wired into the sim build the same way every real board sets its buzzer pin
(-D PIN_BUZZER=<n> in build_flags/DEFINES; here it's a dummy sentinel since
there's no real pin, just something to activate the existing #ifdef
PIN_BUZZER guards in UITask.h/.cpp/SoundNotifier.h unchanged), plus two new
small UITask accessors (isBuzzerPlaying/buzzerFreqHz/buzzerVolume) and three
EMSCRIPTEN_KEEPALIVE exports so a host page can poll the buzzer's state.
Browser side: one Web Audio oscillator+gain per companion instance (index.html
single-instance; mesh.html per A/B, not R which is headless), created lazily
on the first real user gesture (AudioContext autoplay policy), polled every
20ms and mapped to the oscillator frequency/gain -- so every notification
sound, ringtone, alarm, and volume-blip that already worked on real hardware
now actually produces audio in the browser, unchanged at the call-site level.
Verified end-to-end with real RTTTL playback traces (not just "no errors"):
the startup jingle's exact note frequencies (C6/E6/G6) and a real DM-received
notification triggering the receiving instance's Web Audio gain node from 0
to its mapped volume and back, matching the actual "MsgRcv3" melody's notes.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
getContactByIdx() indexes the raw contacts[] table directly, whose first
MAX_ANON_CONTACTS (8) slots are reserved for anon requests -- getNumContacts()
already excludes them from its count, so real contacts start at index
MAX_ANON_CONTACTS, not 0 (NearbyScreen.h's own contact scan already applies
this offset; its comment documents why).
MessagesScreen.h's buildContactList() didn't, so its loop only ever read
the reserved anon slots (empty name, type 0) for any total at or under
MAX_ANON_CONTACTS -- e.g. a device with exactly one known contact would
show "SELECT CONTACT" / "No favourites" with an empty list forever,
regardless of the dm_show_all/fav_only setting or that contact's own
favourite flag. Reproduced live in the browser sim: a fresh companion_radio
instance with one real ADV_TYPE_CHAT contact showed nothing until this fix.
Also fixed _sorted[]'s stored index (was the bare loop counter, needed to
be the raw table index every other call site in the file already assumes)
and the same missing-offset bug in BotScreen.h's room-contact counter and
MessageHistory.h's pub_key-prefix contact lookup.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Selecting a variant from the Hold-Enter accent popup (e.g. 'o' -> 'o with
acute') correctly stored the real UTF-8 character in the text buffer, but
the live-typing preview line rendered it as the plain base letter -- the
accent was invisible even though the data was right.
Root cause: DisplayDriver::translateUTF8ToBlocks() transliterates extended
Latin characters down to ASCII for display drivers whose font can't render
them; real boards with the pixel-perfect MiscFixedFont override it to a
plain passthrough once _single_font is set (SH1106Display.cpp etc).
SimDisplayDriverCanvas reuses that same MiscFixedFont/MiscFixedRenderer
(confirmed the popup's own preview renders diacritics correctly, since it
prints its own variants directly rather than through this path) but never
added the matching override, so every string routed through
translateUTF8ToBlocks() -- not just the keyboard preview -- silently lost
its diacritics.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Rounds out the browser sim harness with the rest of the physical board's
interactions: a reset button (JS-driven, since board.reboot() is inert
under -sEXIT_RUNTIME=0), GPS input wired into a real LocationProvider via
new SimSensorManager, JS-settable battery/environment telemetry, an
admin/repeater-login test hook (sendRoomLogin against the default
"password"), and full physical-keyboard text entry (printable ASCII
passthrough into the existing KeyboardWidget, Tab->KEY_KB_ENTER submit).
Also fixes three real bugs found while exercising all of this in a real
browser:
- UITask.cpp's native-only stdin poll branch had no __EMSCRIPTEN__
exclusion, so it also compiled into the wasm build and called a real,
blocking window.prompt() on nearly every frame -- the actual cause of
the reported time/controls jumping. Now gated to native only.
- 'n'/'p' were mapped as Next/Prev keyboard shortcuts, colliding with
typing those literal letters. Removed the shortcuts; added explicit
Next/Prev buttons to mesh.html (previously relied solely on them).
- Buttons grabbed native browser keyboard focus on click, so a later
stray Enter/Space could silently re-trigger a previously-clicked button
(e.g. Reset). mousedown now calls preventDefault() on all buttons.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two more real-vs-sim mismatches, found after seeing the rendered UI:
1. The Clock screen's big time display (setTextSize(2)) rendered at size 1
-- SimDisplayDriverCanvas ignored setTextSize() entirely (a leftover
no-op from the old system-font renderer) and never overrode
getCharWidth()/getLineHeight(), so the big-digit layout math in
UITask.cpp's drawBig() came out wrong even once print() itself gained
real font support. Track _text_sz, scale both metrics by it (matching
SSD1306Display's own getCharWidth()==6*_text_sz pattern), and pass it
through to miscFixedPrint() in target.cpp instead of a hardcoded 1.
2. The Home carousel's "<PRESS_LABEL> to open" hint said "long press to
open" -- true only for touchscreen-only boards with no dedicated Enter
button (PRESS_LABEL's #if UI_HAS_JOYSTICK / #else split in
examples/companion_radio/ui-new/UITask.cpp). The sim's D-pad + OK key
behaves like a joystick board (a SHORT Enter press opens each page;
holding it separately reaches the real context menu via
handleLongPress()), so showing the touchscreen wording was both
inaccurate and different from what a real joystick board like Heltec V3
displays. Added SIM_PLATFORM to that #if alongside UI_HAS_JOYSTICK --
UI_HAS_JOYSTICK itself stays unset, since its other two gates
(begin()-ing/polling real joystick MomentaryButton objects) need
hardware the sim's target.cpp doesn't declare.
Verified in real Chromium: Clock screen shows "08:11:10" at real double
size above the normal-size date line; Home carousel now says "press Enter
to open". Full regression clean: 3 native envs, wasm companion_radio, the
2-instance+repeater mesh demo, and the long-press context-menu test.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Two prototype polish fixes ahead of the website embed work:
1. Holding Enter (the sim's main way into "the rest of the options") did
nothing -- the JS bridge's sim_enqueue_key() went straight to
enqueueKey(), bypassing UITask::handleLongPress() entirely, so
KEY_CONTEXT_MENU could never be reached. Add injectSimKeyLongPress()/
sim_enqueue_key_longpress(), which does route through the real
handleLongPress() (same code a real MomentaryButton(pin, 1000, ...)
reaches), and wire up press-and-hold (buttons + Enter/Space key) in both
web harnesses with the same 1000ms threshold real hardware uses.
2. SimDisplayDriverCanvas::print() drew text with the browser's own system
font (ctx.fillText, '8px monospace') instead of the real bitmap font a
MeshCore-Solo board renders with OLED_MISC_FIXED_FONT=1 (see
solo/heltec_v3/platformio.ini). Vendor the real Adafruit_GFX (unmodified,
from the same PlatformIO registry package a real board build pulls) into
variants/sim/thirdparty/gfx/, and render print() through the real,
shared src/helpers/ui/MiscFixedRenderer.h + MiscFixedFont.h -- byte-
identical glyphs to real hardware, not a look-alike.
Verified in a real Chromium (Playwright): short Enter -> "CLOCK TOOLS",
held Enter -> "CLOCK FIELDS" (gotoDashboardConfig(), proving
KEY_CONTEXT_MENU is really reached); font renders as hard square pixels,
inverse/selected-row text still punches correctly through a filled bar.
Full regression re-run clean: all 3 native envs, wasm companion_radio, and
the two-instance + repeater mesh demo (relay routing, DM delivery).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Mirrors the simple_repeater sim port exactly: headless (no DISPLAY_CLASS,
UITask.cpp excluded from the build), new sim_simple_room_server native env
plus build_wasm_room_server.sh, own SimFS root ./sim_data_room so its
identity storage can't collide with the companion or repeater instances on
the same page/cwd.
Verified beyond "it compiles": ran the native binary and confirmed a real
_main.id identity file gets persisted through the actual SimFS/IdentityStore
path, same as the other two sim targets.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Ports examples/simple_repeater to variants/sim/ (new sim_simple_repeater
native env + build_wasm_repeater.sh) and adds a JS "ether"
(variants/sim/web/mesh.html) that bridges two real companion_radio WASM
instances through a real simple_repeater instance in a strict A<->R<->B
topology (no direct A-B link), proving genuine relay routing rather than
a shortcut.
Also fixes multi-instance issues Phase 2's single-instance design never
surfaced: SimDisplayDriver's canvas context/id caching was keyed on a
single global instead of per-instance, and both wasm builds were missing
_malloc/_free/HEAPU8 runtime exports needed for the ether to poke bytes
into an instance's memory.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New board variant compiling the unmodified MyMesh/UITask/DataStore app
logic against real mesh::Radio/MainBoard/RTCClock/RNG interfaces, for
running the actual firmware outside embedded hardware:
- Native (plain g++, platform = native): ASCII-art display over stdout,
stdin-driven input, local-disk-backed DataStore/IdentityStore.
- Emscripten/WASM (variants/sim/build_wasm.sh, since PlatformIO's native
platform force-overrides any CC/CXX toolchain override back to system
clang++): canvas-backed display, IDBFS-backed persistence across page
reloads, JS-callable input via sim_enqueue_key(), emscripten_set_main_loop.
Real rweather/Crypto (AES128/SHA256/Ed25519) vendored unmodified and
proven working on both targets. variants/sim/web/index.html is a bare
verification harness, not the polished website embed.
hop_count was clamped to MAX_HIST_PATH_BYTES regardless of hash_size, but
path[] is only MAX_HIST_PATH_BYTES bytes total -- with hash_size>1 the old
clamp let i*hash_size run past the buffer in resolveHopName(). Not reachable
today since capturePath()/markChannelRelayed() already bound hop_count to
MAX_HIST_PATH_BYTES/hash_size on write, but the reader shouldn't rely on
writer discipline alone. Now clamps to the same MAX_HIST_PATH_BYTES/hash_size
bound.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
DeviceDiag::getHeapStats()/getStackFreeBytes() only had an implementation
for NRF52_PLATFORM; every other platform fell through to the 0/0 stub, so
Diagnostics > Live showed "N/A" for Heap free and Stack free on ESP32
boards (Heltec v3/v4) -- exactly the boards where that number is most
useful to see. Added an ESP32 branch using heap_caps_get_free_size()/
heap_caps_get_total_size() (MALLOC_CAP_8BIT, matching the general-purpose
heap `new`/malloc() actually draw from) and the same
uxTaskGetStackHighWaterMark() call nRF52 already uses, since Arduino-ESP32
runs on FreeRTOS too.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
NodePrefs.h's field declaration order used to just be historical append
order (on-disk format is defined solely by DataStore's explicit rd()/wr()
sequence, not struct layout), making the file hard to navigate. Reordered
fields into thematic groups (radio, repeater, bot, GPS/trail/location,
display/keyboard, etc.) with no on-disk/schema change; fixed two comments
that had gone stale (favourite_contacts/_kinds' [del→...] tags only named
one of the two handlers that actually clear them; dashboard_fields was
miscategorized under favourites). sizeof(NodePrefs) shifted twice as a
side effect of packing (2760→2752→2760) — verified via real builds on all
four canonical envs and re-checked against the serialization tripwire.
Also replaced test_companion_node_prefs.cpp's dead body (a disabled test
against a saveSerial/loadSerial API this struct never got) with real
coverage of the pure helper functions NodePrefs.h already carries -- band
bucketing, repeater-profile bounds, alarm-repeat round-trip, and every
option-lookup table, including their inconsistent out-of-range fallback
behaviour.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Four call sites (three bot reply paths, one app-originated-send mirror)
each hand-built the same "Me: <text>" + own_message=true framing that
MessagesScreen relies on to render an outgoing bubble -- one of them
(now fixed) had already drifted to the wrong prefix once. Centralizing
it in AbstractUITask::addOwnChannelMsg() means a caller can no longer
get the framing wrong.
tryBotReplyChannel() mirrored its own reply into the on-device history
prefixed with the node's own name instead of the "Me: " convention the
history view uses to tell an outgoing bubble from an incoming one, so
the bot's own reply rendered left-anchored like a message from someone
else labelled with the device's own name.
Cut meta-commentary asides, a restated default, and a stale changelog
footnote that didn't help someone using the feature -- plus tightened
a few release-notes bullets that repeated the same rationale two or
three times over.
Reboot fired on a single Enter with no way back, unlike Start OTA on
the same tab -- inconsistent given both take an unattended remote node
out of action for a while. Reuses the same beginConfirm() idiom.
Network mode, the dedicated profile and the flood filters were only
built into the row list once client_repeat was already ON, so setting
up a repeater meant enabling it blind first. Every other on/off tool
in ui-new (Bot, Live Share, Locator) already keeps its settings
visible while off; Repeater now matches.
The Hold-Enter popup was a single "Reset counters" item with no Cancel
row, so one Enter zeroed all stats immediately -- the same shape the
Trail/channel/preset resets had before beginConfirm() fixed them.
Reuse the existing botChannelSenderSplit() helper for the channel
[LOC]-share sender name instead of re-implementing the same "Name: "
split inline, and fix a stray lowercase "gps:" reply that didn't
match the rest of the !gps command's replies.
The animated marquee that lets a selected row scroll into view instead of
truncating behind an ellipsis uses a small "is this the same text as last
frame" buffer, _marquee_text[64], compared against the current frame's
temp_str[256] via strcmp(). For any text at or past that 64-byte cutoff,
temp_str simply has more bytes than _marquee_text can hold, so strcmp
always finds a difference at the truncation point -- is_new evaluates true
on every single call, which resets _marquee_skip_cp and _marquee_phase back
to the start every frame. Short labels and names (what this was written
for) stayed under the cutoff and animated fine; a longer message-body
preview -- letting a message be read from the list without opening
fullscreen, an emergent use of the same feature -- got stuck holding at its
first frame forever.
Bumped _marquee_text to 256 to match temp_str/window's own cap in the same
function, so the comparison is exact instead of silently truncated. Static
+192 bytes on the one global DisplayDriver instance; verified against RAM
usage on all four build targets (all comfortably under headroom).
This lives in src/helpers/ui/, shared with upstream, unlike the rest of
this session's ui-new/-only work -- kept as its own commit.
Found while re-reviewing this session's own commits: five screens each
hand-built the same 2-row Action/Cancel confirm popup, defaulting the
highlight to Cancel -- NearbyScreen's contact-delete, AdminScreen's
OTA-start, and the three just added (Trail's reset, Messages' channel
delete, RadioPresetPicker's preset delete). The plan that added those
three had already flagged this exact duplication without acting on it, so
it just tripled instead of getting fixed.
One PopupMenu::beginConfirm(title, action_label, cancel_label="Cancel")
replaces all five call sites, and makes "defaults to Cancel" a property of
the popup itself rather than something each new confirm has to remember.
Also drops two small redundancies spotted along the way: NearbyScreen's
_confirm.active = true, dead since begin() already sets it, and
RadioPresetPicker's deleting = false being set twice in a row (once inside
openConfirm(), once again by its only caller).
No behavior change; verified against the actual PopupMenu/menu-level state
machines in each of the five call sites before touching them.
Continuing the consistency review: sweep for the same three defect shapes
elsewhere in ui-new/ (own read pass plus two parallel research agents),
verified against source before acting.
Three destructive actions fired on a single Enter with no way back, unlike
contact-delete's existing confirm-defaulted-to-Cancel popup: Trail's "Reset
trail" (wipes the whole recorded route, no undo short of a prior manual
Save -- reuses Trail's own multi-level menu machinery, alongside its
GPS-off confirm), Messages' channel Delete, and RadioPresetPicker's saved-
preset delete (shared by Settings > Radio and Tools > Repeater, so one fix
covers both). All three now confirm the same way, defaulting to Cancel.
KeyboardWidget was the one place Hold-Enter still doubled as Cancel: Shift,
Backspace and a Latin letter's accent popup already have real, kept
meanings under a hold, but every other special-row cell (Space, OK/Done,
the {} placeholder) fell through to a bare CANCELLED, closing the keyboard
exactly like the real Cancel key. Now a no-op there too, matching the "only
Back closes it" rule already applied to popups and screens.
MessagesScreen defined the same two label arrays (Notif states, melody
slots) four times over, once per context-menu handler. Hoisted to one
pair of static class members -- constexpr wasn't enough to get the linker
to emit them on this toolchain, so they follow the same declare-in-class/
define-out-of-class shape NearbyScreen::FILTER_LABELS already uses.
Alert text: "Advert sent!"/"Advert failed.." and "Sent!" were the only
toasts anywhere with trailing punctuation; normalized to the plain style
every other confirmation uses. Unpinning from the Favourites Dial reported
the freed slot number from the Messages screens but not from Nodes or the
dial's own tile menu; now consistent everywhere pinning already was.
DiagnosticsScreen's Live/System/Font tab renderers hand-rolled the same
scroll-clamp/loop/indicator skeleton drawList() already bundles; switched
both to drawList (passing the screen's own _scroll as its `sel` too, since
neither tab has a row cursor -- makes drawList's internal clamp a no-op and
leaves clampScroll() as the only thing bounding it, unchanged). Pure
internal tidy, no behavior change.
Continuing the consistency pass: the four screens that show the
distance/bearing "navigate to a point" view (Nodes, Waypoints, Trail's
Track back, and navigating to a location shared in a message) had drifted
apart in three ways.
Only two of the four passed an EtaTracker to navview::draw(), so only
Nodes and Track back showed the closing-speed/ETA line -- navigating to a
waypoint or a shared location left it off for no reason. All four get one
now. They also left the view on three different key sets (Back alone,
Back+LEFT/RIGHT, Back+Enter); Back is now the only way out of any of them,
so a stray sideways nudge can't drop you out of a running track-back.
Messages' renderNav() also switched from reading node_prefs directly to
the shared useImperial() helper the other three already used.
Set as target -- the row Nodes and Waypoints both offer for a coordinate
-- was missing from the message-location Options menu; added alongside
Navigate and Save waypoint.
Nodes' own Set as target required a full 32-byte public key, which a
name-only live-track entry (someone sharing position on a channel who
isn't a saved contact -- the group-outing case this exists for) never
has. One flag was doing two jobs: "can be pinged" (needs the full key)
and "can be identified" (needs only the 6-byte prefix a person target
actually uses). Split into has_key/has_prefix; Set as target now only
needs a position, resolving to a person target (follows them) when a
prefix is available and a place target (pinned where they were) when it
isn't -- the same distinction Locator's own picker already draws.
Locator's target picker separately still listed the people pinned to the
Favourites Dial as its privileged top tier, which stopped making sense
once pinning and favouriting became separate concepts. It now leads with
favourites instead, matching every other list in the firmware.
Three interaction inconsistencies found while auditing the favourites work,
all of the same shape: the same gesture meaning different things depending
on which screen you were on.
Value rows in popup menus. Rows like "Notif: ON" or "Sort: Dist" show a
value the user steps through with LEFT/RIGHT, but Enter treated them as
ordinary menu picks and closed the popup, so changing two of them meant
reopening the menu in between. Trail's settings submenu was the lone
exception, working around it by rebuilding and re-selecting after each
Enter. PopupMenu now knows the difference: addValueItem() marks a row, and
Enter on it returns the new VALUE_NEXT instead of SELECTED, leaving the
menu open. Only Back closes a menu now. Applied to the Messages
contact/room/channel menus, Nodes, the Ringtone editor and Trail, which
drops its reopenSettingsAt() workaround. The LEFT/RIGHT cycling bodies
moved into one helper per menu, since Enter and RIGHT now share them.
Nodes' Fav row was the worst case: LEFT/RIGHT did nothing there at all, so
the only way to toggle a favourite was an Enter that dismissed the menu on
every flip. Its label moved to a member buffer (as the Pin row already had)
so it can be retitled in place.
Settings rows Auto-off, Low battery, GPS pwr and Battery ignored Enter,
though their options wrap exactly like the melody/keyboard/clock rows
beside them, where Enter has always stepped forward. They accept it now.
Rows that ramp between fixed ends (Brightness, Volume, TX Pwr, Timezone,
SF/BW/CR) stay LEFT/RIGHT-only -- there is nothing to wrap to.
Hold Enter no longer doubles as Back. It quietly meant "go back" on Tools,
Locator, Live Share, Repeater, Bot, Auto-Advert, GPIO, Compass, the
Dashboard config and the Messages navigate view, while elsewhere the same
long press opens a context menu. It now only ever opens a menu, or does
nothing where there is none. Same for dismissing an open popup, which it
used to do. Checked that this strands nobody: every board that can reach
these screens has a real Back key (back_btn on joystick boards, Esc on
CardKB/TCA8418/T-Deck). Single-button boards produce no KEY_ENTER at all,
so they never leave the home pages in the first place.
Three names had grown around one idea. "Favourite" was a filter in
Settings, an invisible app-only flag on a contact, a device-settable bit on
a channel, and — on the Nodes screen — a menu row that actually pinned to
the Favourites dial. Nothing marked a favourite on screen, and the dial
only took chat contacts.
A favourite is now the starred flag (ContactInfo::flags bit 0 for contacts
and rooms, ch_fav_bitmask for channels), settable on the device everywhere
via a Fav: ON/OFF row, marked with a star on its row, and sorted to the top
of the list — in Messages, Tools > Nodes and the Locator target picker.
Settings > Contacts > "Favs top" turns the sorting off; it defaults on, and
is stored inverted so an upgraded prefs file reads back as on rather than
off.
MyMesh::setContactFavourite() writes the same bit the app sets and bumps
lastmod, so the two stay in sync. The DMs/Rooms = Fav list filters no
longer depend on having starred someone in the app first.
Pinning is now separate and explicit, and the dial takes contacts, room
servers and channels. Slots carry a kind (NodePrefs::favourite_kinds,
schema 0xC0DE0029) — a channel slot holds an index, so emptiness is decided
by the kind first, since channel 0's payload is all zeroes. Choosing what
to pin reuses the Messages screen's own Direct/Channels/Rooms browse
instead of a second picker built on the dial, which drops that picker, its
key/label tables and the now-unused getRecentDMContacts(). A filled tile
gained Unpin/Replace under Hold Enter.
Fixes a room server being pinnable and then unremovable: the picker's
recent-conversations tier didn't filter by contact type and room posts
share the DM history, opening one from the dial skipped the login
handshake, and Unpin only existed in the chat contact list that rooms never
appear in.
Also: PopupMenu::_visible was written and never read (render recomputes the
cap from display height), which is why menus with more items than the
"visible" argument always worked; Settings' all/fav values and the
dm_show_all comment said things that were not true.
Built for Heltec V3, Wio Tracker L1, Wio Tracker L1 e-ink and T-Echo Card.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- RepeaterScreen: drop the redundant "Rpt " prefix on the profile fields
(Preset/Freq/SF/BW/CR), matching Settings > Radio's own terminology --
the screen is already dedicated to the repeater's own profile, so the
prefix disambiguated nothing.
- Channel context menu: "Fav: yes/no" -> "Fav: ON/OFF", matching every
other toggle in the app.
- Settings (System tab): six labels left as raw concatenated identifiers
(AutoOff/AutoLock/TimeZone/LowBat/BattDisp/BzrVol) now read like their
space-separated neighbours (Auto pwr/Pwr save/DM sound/GPS pwr), and
BzrVol no longer clashes with "Buzzer" one row above it for the same
feature. Three value-label arrays also had one mismatched-case entry
fixed to match its siblings: Auto-off's "never" -> "OFF", Batt display's
"icon" -> "Icon", Sound's "built-in" -> "Built-in".
- ToolsScreen: re-enabled the mini-icons next to each tool, which had
been commented out ("don't fit visually"). Root cause: the screen's own
drawIcon() centred against lineStep() (line height + inter-row gap)
instead of getLineHeight() alone, 2px too generous for a near-full-
height icon -- now delegates to the already-correct miniIconDraw() used
elsewhere (e.g. the message-list ack checkmark). Also gave Admin and
GPIO their own icons (padlock, 3-pin header) instead of both sharing
System's cog.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The plain ✓ on your own channel post only said "at least one repeater
heard it" -- now it shows how many distinct repeaters echoed it back,
drawn as tiny 3x5 digit icons (icons.h) rather than the normal font,
since the slot next to the sender name is icon-sized, not text-row-
sized. DM delivery ticks are untouched (no repeater-count concept
there, so they keep the plain checkmark).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
An app-originated channel send (mirrored into the on-device history)
bumped that channel's unread badge whenever the device's own UI
wasn't already showing that exact channel -- unlike an on-device
compose, which sidesteps this by forcing itself into that channel's
view right before sending. Adds an explicit own_message flag through
addChannelMsg (MessageHistory -> AbstractUITask -> UITask ->
MessagesScreen) so an own post is never counted unread regardless of
what's on screen when it's sent.
Found the same bug in MyMeshBot.h's three auto-reply-into-channel call
sites (Remote Bot's own reply showing as unread on itself) and fixed
those with the same mechanism.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Extends the existing single-boolean channel relay-echo marker into a
full count + list of distinct confirming repeaters, since each
repeater retransmit already appends its own identity hash to the
packet's path and the echo-matching hash deliberately ignores that
mutable path -- so every distinct repeater's echo of one send now
matches the same tracking slot instead of only the first.
Symmetrically captures the hop path a received DM/channel message
actually took, so a new "Path"/"Relayed by" row in the existing
Hold-Enter Options popup can show the resolved sequence of repeaters
(by contact name, or a hex fallback for an unknown one).
Also fixes a real bug caught during testing: the popup row's own
label ("Path (N hops)"/"Relayed by (N)") was built into a stack-local
buffer handed to PopupMenu, which only stores the pointer -- it
rendered as garbage once the building function returned. Moved to a
persistent member buffer.
Bumps the dev-build fallback version and adds release notes/docs
for this plus the two other 1.27 features already on this branch
(BLE retry backoff, marquee-scroll for selected long text).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CMD_SEND_TXT_MSG and CMD_SEND_CHANNEL_TXT_MSG (the phone app's send path)
transmitted over the mesh but never touched the device's own MessagesScreen
history, unlike a message composed on-device (MessagesScreen::afterSend) --
so a DM/channel post sent from the app was invisible if that same
conversation was later opened on the device's own screen. Both handlers now
also call into the same history-store entry points incoming messages use.
Also wires up delivery-status parity with an on-device send, not just the
raw text:
- Channels: arms the existing "relayed into mesh" repeater-echo tracker
(trackRelaySend()/armChannelRelay()) on the new entry -- sendGroupMessage
already runs that tracker regardless of who originated the send, this
just attaches it to the right history entry. Required threading a ring
position back out through AbstractUITask::addChannelMsg (now returns int)
and a new armChannelRelay() passthrough.
- DMs: addDMMsg gained ack_tag/ack_deadline_ms/resends params (threaded
through MessageHistory -> MessagesScreen -> AbstractUITask/UITask) so an
app-sent DM gets the same pending -> \xe2\x9c\x93/\xe2\x9c\x97 status the on-device compose
path shows. resends stays 0 deliberately: the app owns its own retry
decision, so this only drives the on-screen status, never a second,
independent auto-resend from the device itself.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Selecting a row whose ellipsized text overflows now animates a "swing"
marquee: holds at the start, scrolls to reveal the full tail, holds
there, then scrolls back and repeats. Unselected/non-overflowing text
is unchanged (still a static "..."). E-ink gets slower, coarser steps
(fewer, cheaper partial refreshes) than OLED; unchanged frames are
already skipped by the display's CRC diff, so idle holds are free.
Wired into every screen with a selectable row: home favourites, DM/
channel lists and message bodies, Settings, popup menus, Bot, Admin,
Nearby, Waypoints, Locator, Live Share, and the alarm screen.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
At the edge of BLE range the phone repeatedly drops and reconnects, and
each queued outgoing frame's write can block the main loop up to 100ms
(Bluefruit's notify() waits on a free HVN packet slot). The fixed 250ms
retry throttle re-triggered that blocking wait constantly during a bad
link, making the device feel almost unresponsive. Retries now back off
exponentially (250ms -> 2s cap) on repeated failure and reset on the
next successful send.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
release-notes.md's v1.25 section already documents "Updated upstream
base to companion-v1.17.1", and that merge (68527e7b) is confirmed in
main's history -- but every MESHCORE_VERSION string, including
UITask.cpp's fallback default for boards that don't set it explicitly
(Heltec v3/v4, ThinkNode, Mesh Pocket, T-Echo), was still hardcoded to
the pre-bump "1.17". Bumped every occurrence to "1.17.1" to match what
actually shipped.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Scope/repeater-scope filtering, magnetic Hall-sensor screen lock,
solo build configs restructured into their own solo/<board>/ folder
(one _solo_dual build per board), ProMicro + CardKB support (PR #31,
tchellow), new Build Flags reference doc, book-order fullscreen message
paging, and a confirmed "Start OTA" admin action -- plus a full
pre-merge documentation audit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Full documentation audit before merging 1.26 into main turned up:
- README's Supported Devices table was missing GAT562 Mesh Watch13
(a CI-built, shipping firmware with zero end-user docs -- pre-existing
gap, not introduced by this branch) and ProMicro (added to the build
table by an earlier commit but never to the device list up top).
Building-from-source's env table was also missing Watch13.
- settings_screen.md's Keyboard table never documented the Ext. KB
(Full/Compact) setting, despite external_keyboard.md already
describing it in detail.
- nrf52_power_management.md's board-support table only listed 5 of the
13 boards that actually have a PowerMgtConfig implementation wired up
(grep-verified against every variants/*/*.cpp defining `power_config`)
-- added the other 8, and fixed one board's listed name (RAK 3401,
not "RAK WisMesh Tag") to match its actual getManufacturerName().
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
start ota was already sendable via Admin's Custom-command row (and
CLI-reachable directly), but had no dedicated menu entry. Adds a row to
the Actions tab that confirms first (Start/Cancel, defaulting to
Cancel) before sending -- unlike Reboot, OTA parks the remote in BLE
DFU mode for the duration of the update, disruptive enough to warrant
the extra step.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
PR #31 (ProMicro CardKB support) landed a few commits back but the
Contributors list was never updated.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- README's env table and external_keyboard.md's device-support table both
get a ProMicro row -- missed when the PR #31 merge landed since neither
was in scope for that commit.
- external_keyboard.md and build_flags.md now cover CARDKB_I2C alongside
ENV_PIN_SDA/ENV_PIN_SCL: the PR generalised CardKB bus selection to a
single macro after build_flags.md was first written, so the doc only had
the older (still valid, but no longer the only) form.
- release-notes.md: added the ProMicro solo build and the new Build Flags
reference page to v1.26's What's New.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Merged after the fact, so it followed the old convention (solo env inside
variants/promicro/platformio.ini alongside the shared board configs). Moved
to match every other board post-restructuring -- extends still points at
Promicro, defined in variants/promicro/platformio.ini.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
pollHallSensor() acted on the raw pin reading immediately, unlike every other
physical-input path in this file (MomentaryButton, pollCardKB()'s own
last-raw edge check). A cheap mechanical reed switch -- one of the two
sensor types the docs explicitly recommend wiring here, alongside a
solid-state Hall IC -- can chatter for a few ms while the magnet crosses the
trigger distance, so a poll every loop() tick during that window could flip
_locked and fire _display->turnOff()/turnOn() repeatedly in that short span:
wasted work on any panel, and a real cost on e-ink where each is a slow
full-panel operation.
A raw reading now has to hold steady for HALL_DEBOUNCE_MS (25ms, same
threshold as MomentaryButton's ISR_DEBOUNCE_MS) before it replaces
_hall_magnet_present and triggers the lock/unlock actions.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Solo's `_solo_dual` env for each board lived inside variants/<board>/, mixed
in with the shared upstream board configs (repeater, room server, plain
companion) that every MeshCore fork carries. Split them out into their own
top-level solo/<board>/ folder -- each still `extends` the board base defined
in its original variants/<board>/platformio.ini, just no longer interleaved
with it file-wise.
- platformio.ini: extra_configs picks up solo/*/platformio.ini alongside
variants/*/platformio.ini.
- build-solo-firmwares.yml: env-discovery grep now scans solo/ instead of
variants/ (the release matrix is unchanged -- same 8 envs, same names).
- pr-build-check.yml: solo/** added to the trigger paths.
- GAT562 30S Mesh Kit: dropped the separate solo_ble env -- solo_dual is a
strict superset (BLE still works, plus USB), and every other board only
ever had one solo build to begin with.
- GAT562 Mesh Watch13: renamed solo_ble -> solo_dual (added DUAL_SERIAL=1) to
match. No comment anywhere recorded whether this board's USB data lines
are actually broken out to a connector -- DUAL_SERIAL compiles and works
over BLE regardless, so worst case the USB half goes unused.
- Removed a dead AUTO_SHUTDOWN_MILLIVOLTS from the three solo envs that set
it (Heltec V3/V4, T-Echo Lite+KeyShield): that macro is only ever read in
the old ui-tiny UITask, never ui-new, which every solo build (these
included) uses -- it did nothing on any of them. Left alone everywhere
else it's set (non-solo companion envs elsewhere use ui-tiny/ui-orig, or
the flag is legitimately read); out of scope here.
- Added docs/solo_features/build_flags.md: every optional -D flag a solo
build understands (GPIO, CardKB/joystick, Hall-sensor cover lock, buzzer/
vibration, GPS switch, display/battery tuning), verified against the code
rather than the existing per-board comments, with what's already baked
into every solo build kept separate from what's opt-in.
- README: doc index + Building from source section link to the new page;
path references to the moved solo files updated to solo/<board>/.
All 8 solo_dual envs rebuilt clean after the move; native test suite (40
cases) unaffected.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
No board in this repo has one built in, and no default pin is assumed
anywhere -- whoever wires a Hall-effect or reed sensor to a free GPIO sets
PIN_HALL_SENSOR (and HALL_ACTIVE_HIGH, for a sensor that pulls the pin high
rather than low on presence) as a build_flag on their own env. Entirely
opt-in and a no-op elsewhere, same pattern as PIN_GPIO1..4/ADC_MULTIPLIER/
CARDKB_ENABLE.
Level-triggered polling (like pollCardKB()) rather than an edge interrupt --
a magnet held near the sensor reads the same way every tick, so the new
pollHallSensor() only acts on the two transitions. Closing locks and blanks
the display with no wake grace (the cover is physically over the screen, so
there's nothing to show); opening unlocks and wakes it, with no key combo
either way. Both are independent of the Auto-lock setting, which is a
timeout, not a physical event.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
LEFT went to the newer message and RIGHT to the older one, which reads
backwards against the page metaphor the "<" / ">" markers set up. Swap it:
LEFT turns back to the older message, RIGHT forward to the newer one, and
the markers follow (they were keyed to the opposite flags).
PREV/NEXT are named in message order, not screen order -- MessagesScreen's
_hist_sel counts newest-first, so PREV is the older message -- so only the
key mapping and the two marker conditions change; the caller side is
untouched. Applies to both the DM and channel fullscreen views, which share
handleInput(). AdminScreen's reply view treats every non-NONE result as
"close", so it is unaffected.
Docs and release notes updated to match.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Follow-up review of f589b9b2 -- five defects in that commit's own changes.
- repeat_scope_only + repeat_extra_scopes were read/written in the MIDDLE of
the prefs stream, beside their repeat_* siblings. loadPrefsInt()'s rd() is a
plain sequential reader gated only on file.available(), with no per-field
versioning, so on any pre-existing file those 25 bytes were taken from the
fields that follow, shifting EVERY later field: repeater profile (incl. a
float freq), track_shared_loc, all of loc_share_*, trail, bot, GPIO modes.
Moved to the struct/file tail, sentinel bumped to 0xC0DE0027 with 0xC0DE0026
marked burned. sizeof stays 2752 (confirmed by build); the tripwire procedure
now spells out the append-only rule that "in struct order" left implicit.
- rebuildRepeatScopes() called getAutoKeyFor() with id 0 for every entry, but
that cache is keyed on the id alone and ignores the name on a hit -- so every
extra scope after the first silently got the first one's key, making the
comma-separated list do nothing. Distinct id per scope now.
- interference_threshold had no load clamp, so an upgrader read 0x23 (35) out
of the old file's sentinel tail instead of 0.
- CMD_SET_DEFAULT_FLOOD_SCOPE wrote default_scope_key without rebuilding the
relay filter, so setting or clearing the scope from the app left the repeater
filtering on the previous key until reboot. The on-device path already did.
- The keyboard preview derived the cursor's row a second time from byte
offsets, disagreeing with the cursor_line the scroll window already computes:
it pinned the cursor to the end of a full line (drawing '_' one character
past the display width) at every wrap boundary. Use cursor_line directly.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- Settings > Radio > Scope: type a community/region name on-device (derives
the shared key the same "#name" -> SHA256 way as DEFAULT_FLOOD_SCOPE_NAME),
previously only settable from a connected app.
- Tools > Repeater > Scope only + Extra scopes: only relay flood traffic
matching the device's own scope or a comma-separated list of additional
scopes, without changing what scope the device's own messages send under.
No-op while unconfigured.
- getCADEnabled()/getInterferenceThreshold() were hardcoded off on
companion_radio; CAD now auto-enables whenever RX power-save (duty-cycle)
is active, since the noise floor isn't kept fresh during duty-cycle sleep.
- Message truncation to fit the send frame could split a multi-byte UTF-8
character in half; now stops at the last complete character.
- The default "Public" channel was unconditionally re-added at every boot
before the saved channel list was loaded, so deleting it never stuck.
Only seeded now on a genuinely fresh device (no channel file yet).
- Tools > Nodes read contacts from the wrong starting offset, landing on
internally-reserved bookkeeping slots instead of real contacts -- showed
as blank "Unknown" rows and silently dropped that many real contacts off
the end of the list.
- resetContacts() only cleared the first few reserved slots, not the whole
contact table, contrary to its own comment; only reachable today via
private-key import, fixed to match stated intent regardless.
- Keyboard's multi-line text preview could render the cursor on an empty
line below short typed text instead of right after it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Brings in upstream companion-v1.17.0 (CAD, MCU temp telemetry, misc
fixes) alongside our RX duty-cycle power-save; see merge-upstream-v1.17
history for the full conflict-resolution rationale.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Adopts hardware Channel Activity Detection (wired into
RadioLibWrapper::isChannelActive() alongside our RSSI-threshold check
and RX duty-cycle power-save), MCU temperature telemetry, LR2021
standby workaround, DISPLAY_SCALE/FLIP overrides, NRF52Board
shutdownPeripherals() refactor, and misc upstream fixes.
Declines upstream's ConfigSerializer-based NodePrefs rewrite,
MultiSerialInterface/interface_manager, and UIColor palette system —
each would have broken large parts of the Solo-specific feature set
(NodePrefs fields, per-variant single serial_interface, enum-based
DisplayDriver::Color). Flagged as candidate follow-up migrations, not
permanent no's.
Also fixes several pre-existing bugs surfaced while chasing silent
merge breaks (stale newMsg() override signature in ui-tiny/ui-orig,
dead UIEventType::newContactMessage case, missing ContactsIterator
init), bumps FIRMWARE_VERSION/MESHCORE_VERSION to 1.17, and fixes a
missing <cstdlib> include that broke the native ConfigSerializer unit
tests.
Verified via 13+ pio run builds across ESP32/nRF52, all 3 companion UI
variants, and 7 display drivers, plus the full native unit test suite
(33/33 passing).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
New M5Stack Cardputer ADV variant (ESP32-S3, ST7789 TFT, built-in TCA8418
QWERTY keyboard, PI4IOE5V6408 LoRa-cap IO-expander autodetect), and a
KeyShield accessory variant for the existing LilyGO T-Echo Lite (external
TCA8418 T9 keypad + AW21009 backlight driver). Both keyboards share one
ENV_USE_TCA8418 polling block in UITask.cpp::loop(), coexisting with the
unrelated CardKB support (different chip/address/flag).
Fixes carried in from the contributed T-Echo Lite code: swapped GPS RX/TX
pins, TX-LED hooks, TCXO voltage, missing GxEPD2_122_T61 panel include.
Fixed during integration: I2C bus was probed for an RTC before Wire.begin()
configured its pins on Cardputer ADV (silent RTC autodetect failure).
Added dedicated *_solo_dual release envs for both boards (auto-picked up by
the solo-firmware release workflow). Gave the T-Echo Lite KeyShield solo
build -Os/-Ofast-unflag like every other nRF52 solo build (was missing,
cut flash usage from 90.7% to 61.4%).
Ported the shared misc-fixed 6x9 font (full Latin/Greek/Cyrillic, opt-in via
OLED_MISC_FIXED_FONT) to ST7789Display for the Cardputer's on-screen
keyboard. ST7789Spi isn't Adafruit_GFX-based like the other single-font
drivers, and this panel's logical->physical scale is non-integer, so glyphs
are re-packed to XBM and blitted through the existing drawXbm(), which
already does correct fractional-scale boundary math, rather than
duplicating that logic.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Settings screen doc gets the new "GPS pwr" row and a watchdog note on
"Pwr save"; Diagnostics doc gets the new "RXPS wd s/h" row and updates
what "Reset counters" actually clears; release-notes.md gets Unreleased
entries for all three additions from the prior commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Three power-saving additions, prompted by comparing this fork's existing
RX duty-cycle support against IoTThinks/EasySkyMesh:
- RX duty-cycle watchdog: the SX126x's hardware RX<->sleep sequencer runs
with no MCU polling, so a desync (a known failure mode) previously had
nothing watching for it. A new watchdog samples the BUSY pin every tick;
no transition for too long triggers a soft re-arm, then a full chip
reset (with cached radio params reapplied, since std_init() resets to
compiled firmware defaults) if that doesn't clear it. Soft/hard recovery
counts surface on Tools > Diagnostics > Live as "RXPS wd s/h".
- Noise-floor recalibration during power-save: sampling was previously
skipped entirely while duty-cycling, freezing int.thresh interference
detection at whatever the floor was when power-save turned on. Now
borrows a brief continuous-RX window once a minute to take a fresh
reading before re-arming duty-cycle.
- GPS duty-cycling (Settings > System > "GPS pwr"): cycles GPS off between
fixes instead of running it continuously. Each wake waits for a fix
(capped at 60s) before sleeping again for the configured interval.
Repurposes the long-dead NodePrefs::gps_interval byte rather than adding
a new persisted field. A "is anything live using GPS right now" hold in
UITask keeps GPS continuously on whenever trail recording, live-share,
an armed Locator, or the Compass/Nearby-navigate view actually need a
live fix, so none of those features degrade. Locator crossing-state is
reset on each wake so a still-settling first fix can't read as a false
geofence crossing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
README had drifted from the firmware in several places:
- Supported Devices listed only the three nRF52 boards; Heltec V3/V4 were
missing entirely, as was any mention that ESP32-S3 flashes differently.
Flashing is now split per MCU, with the merged-vs-app-only .bin trap
spelled out — that one costs an afternoon to diagnose from a dark screen.
- It advertised a Lemon/Default font switch that was retired in v1.23; there
is one unified misc-fixed 6x9 font now and no font setting at all.
- tools/README claimed ENABLE_SCREENSHOT had to be added by hand, directly
contradicting README's "no special build flags required". The envs have
carried the flag for a while; rewrote the file to cover all four tools.
- "S key for screenshot" described screenshot.py's menu, not the device —
screenshots are driven entirely from the host via CMD_GET_SCREENSHOT.
Structurally, general notes (factory reset on migration, BLE-over-USB
priority) sat at the tail of the ESP32 subsection and read as ESP32-specific;
they're now placed where they apply. Heltec wiring moved out of Supported
Devices into its own section so the device table stays scannable, and the
Solo Tools heading is no longer a link (it was generating a garbage anchor).
New docs/solo_features/external_keyboard.md covers CardKB and the wired
joystick: shortcut table, Full vs Compact, and the pin assignment. The pins
are marked as verified on real V4 hardware only — V3 inherits them from
Heltec's documented pin-compatibility and hasn't been checked on a board.
FEATURES.md (roadmap + code audit, developer-only) moves to
docs/development/roadmap.md; nothing referenced it by path.
Adds Building from source / Releasing / Repository layout, since the release
flow was only discoverable by reading the workflow.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The solo release job only attached *.uf2 and *.zip, which covered every solo
board back when they were all nRF52. The Heltec V3/V4 .bin images were built
and then silently dropped, so tagging a release produced nothing downloadable
for those two boards.
Releases now carry a single -merged.bin per ESP32 board (bootloader +
partition table + app at 0x0). build.sh also writes an app-only .bin, but
that one needs offset 0x10000 and a bootloader already on the chip, so it's
deleted before upload rather than shipped next to the merged image — flashing
it at 0x0 looks exactly like a bricked device.
Also adds -D ENABLE_SCREENSHOT to both Heltec solo envs. Every other solo env
already had it, so Solo Tools couldn't capture these two boards' displays and
the README's "no special build flags required" was untrue for them. Both envs
rebuilt clean (V3: RAM 58.8%, Flash 43.2%).
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
V3 and V4 are pin-compatible per Heltec's docs, and V3's own PIN_GPS_RX/TX/EN
defines are dead code here (ENV_INCLUDE_GPS is never set to 1 in this file),
so there's no reason to keep the blind-guess pins from before. Reuses the
exact CardKB (SDA 3/SCL 4) and joystick (UP 23/DOWN 6/LEFT 47/RIGHT 48/BACK 33)
assignment already confirmed working on real V4 hardware.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Port the Wio Tracker L1 solo firmware (full on-device UI, dual BLE/USB
companion transport) to Heltec V3 and V4 OLED boards. Neither board has
a joystick or CardKB on-board, so each new env wires up both as optional
peripherals with default pins from what the board leaves free, gated
behind the existing UI_HAS_JOYSTICK/ENV_PIN_SDA+SCL flags.
DUAL_SERIAL was nRF52-only; added an ESP32 helpers/esp32/DualSerialInterface.h
counterpart so the flag isn't silently ignored on these boards. On V4's
native USB CDC, isClientConnected() also honours (bool)Serial (real DTR),
same as the nRF52 version; V3 has no native CDC so it stays BLE-only there.
Screen (SDA 17/SCL 18) and CardKB (SDA 3/SCL 4) confirmed working on real
V4 hardware.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The misc-fixed font (full Latin/Greek/Cyrillic) had only ever reached
SH1106Display and the e-ink driver. GAT562 30S -- like 24 other variants
-- uses SSD1306Display, which stayed on the built-in 5x7 font, so every
keyboard alphabet beyond ASCII and every accented contact name was
transliterated (Łódź -> Lodz) or drawn as a filled block.
Extract the renderer into MiscFixedRenderer.h rather than copying it:
Adafruit_SH110X and Adafruit_SSD1306 both derive from Adafruit_GFX and
glyph drawing is pure pixel plotting, so one implementation serves both
drivers (-79 lines from SH1106Display.cpp, same flash usage as before).
The header is .cpp-only by contract -- the font tables are static const,
so including it from a driver header would land a copy in every
translation unit.
On SSD1306 the path is opt-in behind OLED_MISC_FIXED_FONT, set only in
the two GAT562 30S solo envs: the font costs ~14 KB of flash and the
other 24 variants' repeater/companion builds have no keyboard to type
those alphabets on. Without the flag that driver is unchanged, verified
by building GAT562_30S_Mesh_Kit_repeater.
Drive-by consistency fix from the extraction: a newline in print() now
advances the cursor by yAdvance * text_size instead of a single yAdvance,
which was half a row short at setTextSize(2). The e-ink driver always
scaled it. No effect at size 1, which is everywhere the UI prints '\n'.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The screen faded as the device was used: text went grey a few updates
after it was drawn, while whatever had just changed stayed crisp. Only a
full refresh brought it back, and the next few updates ate it again.
A partial update is differential -- the controller drives only the pixels
that differ between its "current" (0x24) and "previous" (0x26) RAM and
leaves the rest to hold their own charge, which this panel doesn't do
well. Priming the previous-image RAM with the inverse of the incoming
frame makes every pixel a difference, so all of them get driven to their
target, in both directions.
The inverse matters: priming with flat white makes only white->black a
difference, so ink is re-driven but never erased and every screen ever
shown accumulates as a ghost. Confirmed on hardware, both ways round.
Costs one extra full-screen RAM write (a few ms of SPI). The refresh
itself is unchanged -- the waveform clocks the whole panel regardless of
how many pixels it drives -- so nothing got slower. Clearing ghosts is
still the periodic full refresh's job (Settings > Full rfsh), which can
now stay off.
The helper needs GxEPD2_BW's private _buffer/_page_height, so it goes in
the patched copy of the header, which every e-ink build now includes
unconditionally rather than only screenshot-enabled ones. All e-ink
variants pin GxEPD2 1.6.2, the version that copy tracks.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
A message containing a line break drew two words on top of each other in
the fullscreen reader. wrapLines() treated '\n' as an ordinary character:
it measured it via getCodepointWidth() -- which reports a full 6px cell
for it, since 0x0A sits below the font's first glyph -- and copied it into
the wrapped line. Both display drivers' print() then acts on '\n' by
resetting the cursor to x=0 and stepping down one row, so the tail of that
line was drawn straight over the following one.
wrapLines() now ends the line at '\n'/'\r' (CRLF counts as one break) and
consumes the byte rather than emitting it, preserving blank lines the
sender typed while still skipping degenerate empty wrap segments so the
loop can't stall. This covers the fullscreen view and the history list's
portrait bubbles, which share the function.
drawTextEllipsized() folds newlines into spaces for the same reason: it
draws one line clipped to max_width, and the compact one-line message
previews in the landscape list feed it raw message bodies. A space keeps
the words apart and measures the same, so the ellipsis maths is unchanged;
for names and labels it's a no-op.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The text preview was the last part of the keyboard still working in bytes
rather than codepoints. cpl is how many characters physically fit on a
line, so dividing byte offsets by it counted every 2-byte Cyrillic/Greek/
accented character as two: lines held half the text they had room for, and
a break could land inside a codepoint. Both display drivers are
permanently single-font, so translateUTF8ToBlocks() passes UTF-8 straight
through -- the truncated sequence reached print() and drew as garbage on
both sides of the break. Line boundaries now walk the buffer with the same
kbUtf8*() helpers insertion/backspace/T9 already use, and the per-line
buffers are sized for a full line of 2-byte characters.
Caps-lock also gets an underline on the shift key: it sets caps too, so
the highlight alone made a one-shot Shift and a held lock indistinguishable
despite capitalising one letter vs. every following one.
Drops UITask::applyFont() -- setSingleFont() is a no-op on both drivers
since they were pinned to misc-fixed, so it did nothing, and use_lemon_font
has had no Settings row for a while. The pref itself stays: it's part of
the on-disk layout. Retires the matching stale rationale on scriptHint().
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Compact mode (Settings > Keyboard's "Ext. KB" row) is meant to guarantee
operation with no joystick at all, but it was still half-tied to the
on-screen grid it hides:
- arrows now move the text cursor directly instead of a grid selection
nobody can see, and Tab opens the placeholder picker directly instead of
the row/col-dependent Hold-Enter dispatch
- plain Enter submits the field (there's no grid cell to have deliberately
landed on), same as Fn+Enter
- Fn+letter's accent popup no longer gates on the grid's script/T9
settings -- CardKB always types plain Latin regardless of them, so the
gate only made the gesture silently stop working
- the whole status line is gone: nothing it showed (script, T9-vs-ABC,
caps) is actionable from an external keyboard. The freed height goes to
message-preview lines, floored at the smallest grid's footprint so
cursor mode's own hint block still fits
- the accent popup gets a fixed slot instead of anchoring on a `row` that
is never deliberately navigated to in this mode
Also fixes a text-corrupting invariant break: moveCursorDirect() and
openPlaceholders() move the cursor without finalizing a pending T9
multi-tap cycle, so a later tap on the same cell within the timeout
overwrote an unrelated character. Every other cursor-moving path already
cleared it.
Fn+Tab is dropped as a separate shortcut -- plain Tab already covered
every case it did. Fn+Enter no longer reads as a dead key in cursor mode.
Direct typing moves into insertTyped(), one translation point documenting
what a future relabelled-keycap layout (Cyrillic/Greek) would need.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
- DM/room unread badges could claim messages the ring no longer held
(same class as the channel fix in 6470afaf, not covered by it):
getDMUnread()/getDMUnreadTotal() now clamp to dmHistCountForContact(),
and a new reconcileDMUnread() (called once per loop()) frees any
_dm_unread_table slot whose ring occupancy has dropped to 0, so a
17th sender isn't starved by stale entries. onContactRemoved() now
also clears _dm_unread_table -- the one per-contact table it was
missing.
- Shift didn't capitalise ł/ń/ź/ż (+ĺ/ľ/ň/ž): the Latin Extended-A
case-pairing rule assumed a single parity for the whole block, but it
flips around the unpaired codepoints ĸ/ʼn/Ÿ. Fixed with four
sub-ranges, verified exhaustively over U+0100-U+017F.
- Triple-click could still toggle the buzzer while locked on
PIN_USER_BTN/PIN_USER_BTN_ANA boards (joystick path already guarded
this).
- millis() wraparound: 4 absolute comparisons in UITask.cpp (battery
poll, auto-off, lock-wake, backlight) converted to the existing
(int32_t)(millis()-deadline)>=0 idiom; MyMeshBot.h's DM-throttle
eviction now picks the oldest slot by elapsed time instead of raw
t_ms, which picked the wrong slot right after a rollover.
- Long-press bypassed checkDisplayOn() on all 5 call sites -- neither
woke the display nor extended auto-off, and could deliver
KEY_CONTEXT_MENU to the invisible screen. Moved the gate inside
handleLongPress() itself instead of patching each site.
- CardKB's backspace/printable-insert branches didn't reset t9_cell,
so typing right after a T9 cycle tap could get silently overwritten
by a same-cell re-tap within the T9 timeout.
- buildContactList()'s counts[MAX_CONTACTS] was a 1400 B int array on
the 4 KB loop() stack; values are bounded by DM_HIST_MAX (32), so
now uint8_t.
- ACK table treated ack==0 as a wildcard: isAckPending(0) matched any
free slot, and processAck() with an all-zero ACK matched the first
free slot and returned its stale contact pointer. Both now skip/reject
ack==0, and the matched slot's contact pointer is cleared alongside
its ack hash.
- ensurePageOrderInit() could write one byte past page_order[13] when
migrating a saved order with all 13 slots full and CLOCK last --
guarded on insert_at < PAGE_ORDER_LEN.
Two findings from the same review were resolved as no-op decisions,
not code changes: !buzz over DM ignoring quiet hours is intentional
(the pull exemption is meant to cover the buzzer), and the offline
queue's full-queue drop-newest behaviour is upstream code, left alone.
Build-verified green on WioTrackerL1_companion_solo_dual (RAM 71.1%,
Flash 66.6%) and WioTrackerL1Eink_companion_solo_dual (RAM 73.0%,
Flash 67.9%).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The per-channel unread counter was independent of the ring's actual
contents, so the two drifted apart:
- Opening a channel whose entries had been evicted left the badge
claiming messages the list could no longer show. The viewing-session
bookkeeping computes the count from an _unread_at_entry snapshot, and
with an empty list _hist_visible is 0, so entering only knocked the
count down by one instead of clearing it (badge "7", empty list, then
"6").
- Eviction from a full ring decremented the counter for any dropped
entry, including already-read ones, undercounting the newer unread
messages the counter actually refers to.
chUnread()/getTotalChannelUnread() now clamp to the channel's ring
occupancy, so the badge can never promise more than the history holds
whatever the raw counter says, and eviction only decrements when the
entry being dropped was itself unread.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Post-review cleanups, no behaviour change:
- botScanCommands() parsed the command name and its two args with three
near-identical read-token loops; extracted a single readToken() lambda.
- Fn+Esc lock branch turned the display on twice (the unlock arm repeated
what the branch head already did); dropped the redundant call.
- setGpioMode()'s comment said "Cycle" (cycling lives in GpioScreen); now
describes what it actually does — set a specific mode + persist.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- Bot Actions (!buzz/!gps/!advert/!gpio1-4) ran their side effect
immediately during botScanCommands(), before quiet-hours/cooldown/
per-contact throttle were checked -- those gates only suppressed the
reply text, not the actual buzz/GPS toggle/advert/pin write. botCommandReply()
now only records what was requested; applyPendingBotActions() runs the
deferred effects once a wrapper's throttle checks pass and the ack sent,
mirroring the existing _locfix_requested pattern. resetPendingBotActions()
clears everything on every throttled/aborted path.
- CardKB's Fn+<letter> accent-popup shortcut bypassed the locked-input gate
(it called into KeyboardWidget directly instead of through the
enqueueKey()/dequeue path every other key uses, so it wasn't discarded
while _locked). Now checks _locked itself.
- Since a locked device now correctly ignores CardKB entirely, Fn+Esc
(single press) is added as CardKB's own lock/unlock gesture -- otherwise
a CardKB-only setup had no way to unlock. Esc rather than the adjacent
Fn+Backspace, to avoid an accidental press.
- botScanCommands() now parses up to two arguments per command instead of
one. Used by "!gps fix [seconds]" to override the default 90s timeout
(clamped 15-300s) for a poor sky view where 90s isn't always enough to
reach isLocFixReady()'s HDOP/satellite bar.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Satellite count alone is a poor proxy for fix quality -- few satellites
in good geometry can beat many in poor geometry. LocationProvider now
exposes getHDOP() (default -1 = unsupported); MicroNMEA implements it.
isLocFixReady() prefers HDOP <= 2.0 when available, falling back to the
old >=8 satellite threshold for providers that don't report it (e.g.
RAK12500/u-blox).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Same fix as hotfix/admin-login-timeout (96b44460). AdminScreen's only
guard was "_phase == LOGIN" (true for any node sat at the login
screen), not that the reply actually named _target. Now also checks
pub_key against _target.id.pub_key.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AdminScreen::onRoomLoginResult()'s only guard was "_phase == LOGIN" --
true for *any* node currently sat at the login screen, not specifically
_target. Combined with UITask::onRoomLoginResult()'s current-screen
dispatch (not requester-based), a slow reply for an earlier login
attempt (this screen's own previous target, or even MessagesScreen's)
arriving while the user has since opened Admin on a different,
password-less node -- still parked at the blank LOGIN keyboard, so
_phase == LOGIN here too -- was accepted as that new node's own login
result, flipping _admin_ok/_phase to COMMAND without ever actually
authenticating with it.
Root-caused by cancelUiPendingLogin() (previous commit): that fix
covers the "gave up, then it resolved late" path, but not "a reply for
a genuinely different pubkey arrives while merely _phase == LOGIN".
Checking pub_key against _target.id.pub_key closes that regardless of
which path let the reply through.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Same fix as hotfix/admin-login-timeout (5a5ebe9f). UITask::onRoomLoginResult()
dispatches by whichever screen is currently shown, not by who sent the
request, so a reply arriving after AdminScreen gave up (Cancel or the
timeout fix) could land on MessagesScreen instead and persist its own
unrelated _login_pw as the "confirmed" password for that pubkey.
MyMesh::cancelUiPendingLogin(pub_key) stops tracking the request on
give-up so a late reply matches nothing instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
UITask::onRoomLoginResult() dispatches a login reply to whichever
screen is *currently* shown (curr == admin_screen ? AdminScreen :
MessagesScreen), not to whoever actually sent the request. Neither
giving up path (manual Cancel, or the timeout added in 23f43cac) told
MyMesh to stop tracking the request, so a reply that still arrived
after the user had navigated away landed on whatever screen they'd
moved to instead -- most likely MessagesScreen, which then persisted
its own unrelated _login_pw as the "confirmed" password for that
pubkey, silently corrupting the saved password even on a genuine
success.
Adds MyMesh::cancelUiPendingLogin(pub_key), pubkey-guarded so it's a
no-op if a newer request has since overwritten ui_pending_login, called
from both of AdminScreen's give-up paths. A late reply now simply
matches nothing and is dropped.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Same fix as hotfix/admin-login-timeout (05609019). Tools > Admin >
System > "Admin password" changes the remote's admin credential but
never updated this device's saved copy, so the next login retried the
password just replaced -- likely the actual trigger behind the
"stuck on Logging in..." report. Parses CommonCLI's "password now: <v>"
success echo and saves that as the new on-device password.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Tools > Admin > System > "Admin password" (set-only, sends "password
<new>") changes the remote node's own admin credential, but nothing
updated this device's saved copy of it -- so the very next login
attempt to that node retried the password just replaced, landing
straight in the "stuck on Logging in..." case fixed in the previous
commit. Likely the actual trigger behind that report.
CommonCLI::handleCommand() always echoes a successful password change
back as "password now: <value>" (truncation and all), so parsing that
reply gives the exact value now required to log back in, rather than
trusting what we sent (which the remote may have truncated further).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Turns GPS on (if it wasn't already), waits for a stabilised fix
(isValid() + >=8 satellites, then averages 10s of readings), sends the
position, and restores GPS to whatever state it was in before -- up to
a 90s timeout, after which it reports a partial fix (if it got any
samples) or plain failure.
Replies in two parts since a fix takes seconds-to-minutes, unlike every
other bot command here: an immediate "acquiring fix..." ack (through
the existing synchronous command path), then the actual position as a
separate follow-up message once ready, delivered to whichever
destination (DM/room/channel) the request came from. Only one fix can
be in flight at a time -- a second request while one is pending gets an
immediate "already pending" instead of silently replacing it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AdminScreen's LOGIN phase had no timeout, unlike its COMMAND phase
(_cmd_deadline_ms). If a login reply never arrived -- most commonly a
saved password gone stale after the remote node's password changed,
silently dropped instead of nacked -- the screen stayed stuck with only
a manual Cancel to escape.
sendRoomLogin() now returns the same est_timeout sendAdminCommand()
already exposes; AdminScreen uses it to arm a deadline (poll(),
mirroring the COMMAND-phase pattern) that forgets the stale password
and returns to the picker on expiry, same as an explicit login
rejection already does.
Same fix as hotfix/admin-login-timeout (23f43cac), split out of this
branch's other in-progress work.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
AdminScreen's LOGIN phase had no timeout, unlike its COMMAND phase
(_cmd_deadline_ms). If a login reply never arrived -- most commonly a
saved password gone stale after the remote node's password changed,
silently dropped instead of nacked -- the screen stayed stuck with only
a manual Cancel to escape.
sendRoomLogin() now returns the same est_timeout sendAdminCommand()
already exposes; AdminScreen uses it to arm a deadline (poll(),
mirroring the COMMAND-phase pattern) that forgets the stale password
and returns to the picker on expiry, same as an explicit login
rejection already does.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Add Settings > Keyboard "Ext. KB" row (boards with a CardKB-capable I2C
bus only): switching it to Compact hides the letter grid and special-row
icons in favour of a one-line status (script/page, caps) plus a Fn-shortcut
reminder, since an external-keyboard typist never looks at the on-screen
grid. Accent/placeholder popups still render as before. Off by default.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
CardKB is level-triggered (repeats the held byte every poll) and its Enter
key collided with the on-screen keyboard grid's own commit action, causing
duplicate characters and accidental message sends. Debounce polling and use
the CardKB v1.1 Fn modifier (confirmed working on real hardware) instead of
tracking navigation state: plain Enter now behaves like the physical centre
button, Fn+Enter submits, Fn+Tab opens the Hold-Enter equivalent, and
Fn+<letter> opens that letter's accent popup directly.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Auto-detected at boot on Wire1/Grove (addr 0x5F) -- no setting to flip, and a
no-op on boards without that bus or with nothing attached. This UI's key
codes (KEY_LEFT/UP/DOWN/RIGHT/ENTER/CANCEL) already match CardKB's own byte
protocol, so most input needs zero translation and flows through the same
key queue as physical buttons.
Two bytes get remapped in UITask::pollCardKB():
- Enter, only when the on-screen keyboard's plain grid state is active (no
placeholder/accent popup, not in cursor-mode), becomes a new KEY_KB_ENTER
sentinel meaning "submit the field" -- reusing plain KEY_ENTER there would
insert a stray character, since a CardKB typist's row/col never reflect an
intentional grid selection. Everywhere else Enter is untouched, so
selecting a placeholder or committing an accent still works normally.
- Tab (otherwise unused) becomes KEY_CONTEXT_MENU, standing in for the
"Hold-Enter" long-press gesture CardKB has no way to produce -- without it,
~30 context menus across the UI (message reply/navigate, Bot/Admin/
Repeater, ...) would be unreachable from the keyboard alone.
KeyboardWidget gains a direct-typing path: printable ASCII inserts straight
at the cursor bypassing the grid, Backspace deletes, KEY_KB_ENTER submits.
Build-verified: WioTrackerL1_companion_solo_dual and
WioTrackerL1Eink_companion_solo_dual both compile and link clean; also
smoke-tested Heltec_mesh_solar_companion_radio_ble (no ENV_PIN_SDA/SCL) to
confirm zero regression on boards without the feature.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
The bot outgrew "auto-reply" once it gained Actions (!buzz/!gps/!advert/
!gpio1-4) that control device hardware remotely, not just answer messages.
Renames the Tools screen entry and all doc cross-references; already-shipped
release notes (v1.23) are left as-is to match what actually shipped under
that name.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
- Auto-Reply Bot gains Actions (!buzz/!gps/!advert) behind a new per-target
toggle nested under Commands (bot_actions_dm/ch/room); off by default.
- Bot Trigger fields accept comma-separated multiple phrases, matching any
one fires the reply.
- New user-assignable GPIO feature (Wio Tracker L1): !gpio1..!gpio4 bot
commands plus a Tools > GPIO screen. Each pin cycles Off/Input/Output;
GPIO1/GPIO2 (P0.02/P0.29, the nRF52840's AIN0/AIN5) also offer a read-only
Analog mode via direct SAADC access. GPIO3/GPIO4 (P0.09/P0.10) are the
chip's NFC1/NFC2 pins, repurposed as plain GPIO via a one-time UICR
NFCPINS bit-clear in initVariant() (adapted from Adafruit's own
nfc_to_gpio example) -- confirmed working on real hardware.
- Fix: DM/room reply-prefix ("@[nick] ") stripping happened at the wrong
layer, hiding the "To:" header on DM replies and leaking the raw prefix
into room messages' list view; a related mismatch had the history
scrollbar's sizing pass wrap room messages with the sender name still
attached, disagreeing with the actual rendered text.
Build-verified: WioTrackerL1_companion_solo_dual and
WioTrackerL1Eink_companion_solo_dual both compile and link clean
(sizeof(NodePrefs) confirmed 2720 via real build, not guessed).
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
"+ Add channel" jumped straight into a generic Name+Secret form, so the
"hashtag channel" convention documented in docs/companion_protocol.md
(secret = first 16 bytes of sha256("#topic")) was only reachable by
already knowing to type a literal "#topic" into the passphrase field --
nothing in the UI surfaced it. The phone app instead shows an explicit
channel-type picker; this adds the same on-device.
"+ Add channel" now asks Public / Hashtag / Private first:
- Public commits immediately with the well-known default channel's name
and secret (8b3387e9c5cdea6ac9e5edbaa115cd72, confirmed to match
MyMesh.cpp's PUBLIC_GROUP_PSK and the docs' published key) -- useful
to restore it if deleted.
- Hashtag shows a single Topic field; Save synthesizes name="#topic" and
derives the secret via the existing SHA-256 passphrase path -- same
underlying mechanism Private already had, just discoverable without
knowing the "#" convention.
- Private is today's manual Name+Secret form, unchanged.
Editing an existing channel skips the picker (no ambiguity to resolve
there). Extracted hexToSecret() out of deriveSecret()'s hex-mode branch
so Public's fixed key parses through the same code instead of a second
hand-rolled loop. No other file needed changes -- openAdd()/openEdit()/
active()/render()/handleInput() keep their existing signatures.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
UITask::newMsg() (fired for every incoming DM/channel message) triggers a
3s "Msg: <sender>" overlay drawn on top of whatever screen is current --
including the shared KeyboardWidget when it's occupying the full screen
for text entry (message compose, room/repeater password, channel name,
device name, admin custom command, ...). A message arriving mid-typing
blanked out the letter grid for the full 3s with no way to see what was
being typed.
KeyboardWidget now tracks whether it was actually rendered this frame
(_visible, set at the top of render(), cleared by the new beginFrame()).
UITask's render loop calls _kb.beginFrame() before curr->render() and
skips the alert overlay when the keyboard turned out to be what got
drawn -- covers every screen that shares _kb, not just message compose.
The alert itself is unaffected (still fires, still expires after 3s) --
it just doesn't draw over the keyboard.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Renamed the vestigial Lemon/default font-switch naming (setLemonFont/
isLemonFont/drawLemonChar/lemonXAdvance/_use_lemon -> setSingleFont/
isSingleFont/drawGlyph/glyphXAdvance/_single_font) across DisplayDriver.h
and both concrete drivers -- both have been permanently single-font for
several commits, so the old names invited a future reader to think a
real switch still existed. Pure identifier rename, no logic changed.
Also extracted the byte-identical "SenderName: " prefix-splitting in
MyMeshBot.h's tryBotReplyChannel()/tryBotChannelCommand() into a shared
botChannelSenderSplit(), mirroring the existing botRoomSenderName().
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
drawLemonChar()'s "y" is the top of the current text row on SH1106 (real
glyphs render at y + 7 + yo + row), unlike GxEPDDisplay's version, where
y IS the baseline. The undefined-glyph fallback box copied GxEPD's
y - 7*sz formula verbatim during the font-unification pass, sending it
7px above the row's top edge -- into the previous line's space. Drawing
it at plain y (already baseline minus the font's 7px ascent) lands it in
the same relative position GxEPD's version occupies, within its own row.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Page 0 was hardcoded to Latin -- Cyrillic/Greek could only ever be the
second, cycled-to page. Settings > Keyboard's Alphabet row splits into
Main (which script the keyboard opens on by default) and Additional
(the second one reached via #@/abc), so a Cyrillic/Greek typist can make
their own script the default instead of always landing on Latin first.
Setting Additional equal to Main collapses back to a 2-page cycle (that
script + Symbols), same rule the old Latin-hardcoded design already used
implicitly.
KeyboardWidget.h: cellStr()/t9GroupStr() now dispatch through
scriptCellStr()/scriptT9GroupStr(), treating Latin as an ordinary peer of
Cyrillic/Greek instead of a special case; scriptHint() replaces
altAlphabetHint() so the #@/abc key's "next page" hint is correct
regardless of which script that lands on; the accent popup's gating
checks the current page's actual script instead of assuming page 0 is
always Latin. Removed the now-dead pageIsAltAlphabet().
NodePrefs gains keyboard_main_alphabet (schema sentinel 0xC0DE001F ->
0xC0DE0020, same append-at-tail/clamp-on-load pattern as every prior
schema growth this file uses). Verified via a real build that the new
field lands in existing tail padding -- sizeof(NodePrefs) is unchanged
at 2712.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Replaces the 8 separate Latin-diacritic alt-alphabet pages (Polish, Czech,
Slovak, German, French, Spanish, Portuguese, Nordic) with one popup: Hold
Enter on a Latin letter that has accented variants (a c d e i l n o r s
t u y z) opens a floating horizontal strip of that letter's accents,
anchored over its own row so the grid stays visible underneath (LEFT/
RIGHT picks, Enter inserts via a new shared insertGlyph() helper, Cancel
dismisses). Holding a letter with no variants is a no-op. Cyrillic/Greek
remain full alt-alphabet pages; NodePrefs::keyboard_alt_alphabet shrinks
from 11 to 3 values accordingly -- an old saved Polish..Nordic value just
clamps to Latin via DataStore.cpp's existing range check, no migration
code needed.
Freeing Hold-Enter on letter cells required moving cursor-mode's own
trigger: UP from the top letter row now enters it instead of wrapping to
the special row. To keep that wrap reachable, cursor mode's UP/DOWN
(Home/End) continue the wrap once already at that boundary -- UP again
lands on the special row, DOWN again back on the letter grid -- reusing
the same proportional column mapping the old direct wrap used.
Diagnostics' font-coverage sample swaps its 8 per-language lines for one
line sampling the new accent table. Docs (message_screen, settings_screen,
solo_ui_framework) and release-notes updated to match.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
JakubandClaude Sonnet 5