- The map screen's own fill is only painted where the map area doesn't
cover it (the strip it uncovers as it drifts in); the area and the
opaque status bar hide the rest. ~2 ms a frame.
- With every visible cell holding a tile, the map area drops its fill and
grid, which lie under the tiles anyway, and still reports that it
covers what's below (coversOwnArea, shared with the frozen popup
image). The Home minimap drops them the same way. ~5 ms a frame.
Full map frame ~50 -> 46-48 ms on the L2.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Kept for the optimisation pass before a release:
- perfDrawProfile(): a full redraw timed per object (its own drawing),
totals by kind and the costliest listed -- on settings pages, Contacts,
Home, the map and under a frozen popup; the screen's entrance frames too.
- Per screen: build, layout, first draw and redraw apart.
- A popup over Home: a full frame dimmed live and frozen.
- The reset reason and a core dump summary at start.
- -D PERF_WRAP_TEXT with -Wl,--wrap= of lv_text_get_size_attributes,
lv_layout_apply, lv_obj_get_style_prop_internal and lv_obj_send_event:
text measured, flex runs, style reads and events per screen.
- -D PERF_KEEP_USB leaves the USB popup up.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Once a popup's dim is up, the screen below is drawn once into an image
(LV_USE_SNAPSHOT), darkened as the dim layer darkens it, and shown opaque
over the screen with the dim layer cleared. LVGL then starts drawing from
that image; an lv_image with no background never reports that it covers,
so it answers the cover check itself. A full frame under a popup: 45.7 ->
32.6 ms; freezing costs ~29 ms once, as the popup settles.
One popup at a time (a second one dims live), a strip keeps the status
bar dimmed under top-layer popups, the image goes with its popup, and the
popup dims live again if its screen goes first. POP_BOTTOM (over a live
map) isn't frozen.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
LVGL lays out children before their parent, and a new box's size is only
known after a pass, so every level sized in % of the one above cost a whole
extra layout pass over the screen. innerW() works the width out from the
styles instead; fillWidth() sets it on infoCard, infoLine (every row),
noteLabel and scrollList, and the screen body is px-wide.
Display & power page: 101 -> 76 flex runs, layout 68 -> 54 ms, first
frame 161 -> 146 ms; the other settings pages and Radio / Bot / Quick
alike. Four passes remain: a row's text column grows beside its control,
so its width is only known after the row's own layout.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Measured on the L2 (steady full frames): Settings page 37.7 -> 31.4 ms,
Display 41.5 -> 37.9, Contacts 34.6 -> 32.9, Home 37.3 -> 32.4.
- styleSurface() leaves boxes in the page colour unpainted; styleOpaque()
for the screen and the status bar, which things slide under.
- group() cards clip their corners only while a row is held
(rowPressable(), shared by the tappable rows).
- rowSlider(): the track an opaque tint (theme::mix(), also behind
ACCENT_DIM) instead of 20 % accent blended every frame.
- Map buttons, pills, scale, nav bar and the Home minimap labels opaque.
- Screens drift in without the fading cover: its full-screen blend was in
every entrance frame (page 39 -> 29 ms).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PNGdec (bitbank2, 1.1.6) decodes a tile a row at a time straight into
RGB565, without lodepng's 256 KB RGBA buffer; the trail overlay goes through
it too, blended by hand for palette-with-alpha and RGBA. It lives in its own
file (ui-lvgl/TilePng.cpp) because its zlib headers put DONE/EXTRA in the
global namespace. lodepng stays the fallback and the sim's decoder.
Same 20 tiles: decode 78 -> 50 ms, trails 42 -> 17.5 ms, a whole tile
170 -> 105 ms; the map with trails is pixel-identical.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The list fill becomes a small job queue (fill/then/fillStart, fillTick with a
25 ms budget per pass). Settings pages build section by section in
schemaRows, with the trailing extras (melodies, prune, quick messages) queued
after them so the order stays; Radio and Repeater build by group, Admin's
repeater pick like the contact list, Diag's Live tab by section. Rebuilds
that restore the scroll flush the queue first. Screens went from 220-360 ms
to 80-175 ms (build + first frame).
Diag's crash line read the core dump in flash every second (~44 ms each);
it is read once now. The perf run times every page and dismisses the USB
popup, which had covered all earlier measurements.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A widget costs ~1 ms on the L2, so Nodes with 48 rows held the screen for
~600 ms and a full contact list for over a second. fillStart() builds the
first screenful at once and fillTick() adds the rest ~25 ms a pass.
- Nodes: 6 rows first; the 2 s refresh rewrites built rows in place
- New message: contacts filled the same way
- Settings: a group at a time (all at once when coming back scrolled)
- Nodes 671 -> ~200 ms, Settings 376 -> ~190 ms longest UI stall
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- TileCache: request() hands z/x/y to a task on core 0, which renders into
a spare buffer; poll() swaps it into the LRU slot (stale results dropped
after drop()/invalidate()); the sim runs the job at once
- MapScreen / Home map place finished tiles as they arrive; the longest
UI stall with the map opening goes from ~340 to ~170 ms
- LVGL allocates from the PSRAM heap (lv_mem_psram.h), safe from both
tasks; LVGL's own pool is gone
- lv_conf: notes why LV_OS_FREERTOS with two draw units stays off
(measured slower: drawing is bound by PSRAM bandwidth)
- UI_PERF_TEST: the map (tile read/decode, pan, redraw), build time and
the longest loop pass per screen
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Trail on the L2: 32768 points every 5 m, no simplification, no point
spacing setting; copied to the card every minute and restored at power-on.
The line keeps up with the position dot (incremental sync, pending point).
Map: a tap's finger jitter no longer turns follow off; the GPS button
shows when the map follows. The trail is drawn as one polyline with
breaks and off-view chunks skipped (an lv_line queued a draw task per
point).
Fewer rebuilds: conversation bubbles patched (new messages, delivery
marks), home mini-map dots reused, Nearby rows rewritten in place,
history read in runs and each file opened once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
503/429 slow the pace down, transport errors retry after 2/8/30 s, failed
tiles get a second pass; the download only gives up after 10 minutes
without a single tile or on repeated 403s.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- CMD_GET_SCREENSHOT on colour screens: display type 2 (RGB565 LE),
32-bit payload size; ui-lvgl captureFrame() shared with the SD BMP
- ENABLE_SCREENSHOT in the L2 build; tools/screenshot.py decodes type 2
- docs and READMEs point to solo.marekzegarek.com/#pc-tools
- USB popup: "Keep card" instead of "Charge only" (only the card is lent;
charging and USB serial work either way)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- CPU at 80 MHz with the screen on after 2 s without touch or animation;
a touch restores the full clock before its frame is drawn
- setText(): periodic refreshes no longer redraw unchanged labels
(status bar, Home, lock, Diag, GPS, compass, clock, map, nav, nearby)
- map markers kept across the 3 s rebuild when unchanged
- app connected: the loop blocks until a BLE frame arrives
(serialFrameArrived / hasPendingFrames) instead of polling every 2 ms
- PSRAM: boot frame sprite and the HTTP response buffer released
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The WiFi and Bluetooth rows rebuilt the whole Settings page to update
their hint, cutting the switch's animation short; the hint is now
rewritten in place (rowHintSet). The "Bluetooth on" echo toast is gone.
- Map download: a 404 gets the same one retry as other errors
(OpenTopoMap sends the odd 404 for a tile not rendered yet), and a
download that ends with gaps says "Done, N missing - download again".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- build-solo-sim.yml clones LVGL at the version pinned in
solo/wio-tracker-l2/platformio.ini and builds the SIM_UI=lvgl sim into
wasm/lvgl/ of solo-sim-wasm.zip, for the site's device switch.
- build_wasm.sh: LVGL_DIR can be overridden (CI has no .pio libdeps).
- sim_test_set_node_name(): names a sim node, for site screenshots.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Short pages by feature (getting started, messages, contacts, navigation,
tools, settings, lock, hardware) instead of one page per OLED screen, with
Wio Tracker L2 differences as notes; facts checked against the code
(keyboard scripts vs. hold-for-accents, {batt}/{dist}, L2 limits and
settings layout). Developer guides and build flags under docs/solo/developer.
Old docs/solo_features and its ~50 screenshots removed; a hero banner and
three simulator screenshots added; links updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Working plans (docs/development, docs/design redesign notes) removed; the
two developer guides moved to docs/developer/. README cut to devices,
flashing, links and building (Wio Tracker L2 added); its feature list and
doc tables moved to docs/solo_features/README.md. RELEASE.md describes the
Solo release flow. Upstream's GitHub Pages workflow (docs.meshcore.io)
dropped, CLAUDE.md untracked, Buy Me a Coffee back in FUNDING.yml.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The phone app's offline queue took 256 x 177 B = 45 KB, which with the
on-device UI left the L1 ~5 KB of heap from boot (the sign buffer of #38
never fit). The UI keeps its own message history, and a full queue drops
the oldest channel messages first, so 128 is plenty: the L1 heap now has
~27 KB free. L1 OLED / e-ink, ProMicro, T-Echo Lite, both GAT562; the
upstream variants/ are unchanged (build_unflags in solo/).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The SD card (message history) could be lent to a computer from the lock
card before the PIN was entered. While locked with a PIN, usbPoll() no
longer reacts; a cable plugged in meanwhile is offered once unlocked.
The plain-text NVS PIN never shipped, so there's nothing to migrate.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Since bf1e6fca kbApplyCapsUtf8 wanted 4 bytes free before each codepoint;
the ABC grid passes a 3-byte cell buffer, so every cell came out empty (T9
and CardKB Compact draw no such cells, so it went unnoticed). A codepoint
is now copied when it fits with the NUL.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The contributor's feature -- a PIN asked before the screen unlocks, on
boot, and when the magnet cover opens, stored as a salted SHA-256 -- on
top of this branch, with:
- ui-core/ScreenLock.h: set / check / isSet and the tries limit, shared by
L1 and L2. isSet looks at every byte (the PR tested only the first, so one
PIN in 256 silently disabled the lock). 5 misses pause entry for 30 s.
- NodePrefs: the fields at the tail after quiet hours, sentinel 0xC0DE0031,
sizeof 2880; an older file's bytes there are cleared. A sentinel bump no
longer resets display_brightness_pct from a file that already has it.
- DataStore: no AGPL header (the project is MIT), no whitespace churn.
- Keyboard: the number pad is a widget flag (pin_mode) with a prompt in the
preview ("New PIN", "Again", "PIN"), not keyboard_type = 2 written into the
prefs and restored after.
- L1 Settings: "Lock PIN" right after the schema's "Lock screen" row, typed
twice, at least 4 characters, saved at once (a power-off right after
setting it kept no PIN). Home is told about a boot-time lock once it
exists (also fixes booting with the cover closed).
- L2: the screen PIN moves from plain text in NVS to the same hash in
NodePrefs; the old one is migrated on first boot and removed. The pad
unlocks as soon as the hash matches from 4 digits on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CMD_SIGN_START promised the app 8K before a malloc it never checked. On
the L1 the nRF52 heap has ~5K free from boot, so the malloc always failed
and every CMD_SIGN_DATA answered ERR_CODE_BAD_STATE. Now it allocates
first, halving down to 256 B, and reports the size it got; nothing at all
is an error on SIGN_START itself.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- L1's MessagesScreen sends through UiCore::sendDirectText / sendChannelText
(its own copy of send + file + ACK tracking gone).
- ui-core/Telemetry.h: the dashboard fields (numbers, names, formats) and
sensor readings as text, each frontend passing its Style; L2 Home, L1's
clock page, lock screen, sensors page and GPS altitude use it.
- L2: the side button held in the first 8 s is the CLI rescue, as on L1.
- L1's repeater radio goes through rptctl.
- L1 Settings are built from SettingsSchema: short labels for small
screens, SCHEMA_* placeholders expand to the schema's sections, one
generic row (switch ON/OFF, bars for brightness / volume). 20 hand-written
rows and their tables gone; L1 implements applyDisplayPrefs /
applySoundPrefs. settings::text / step / indexOf shared with L2.
- Schema: auto-off gains 5 s; the advert sound scope reads "Direct".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Dead code gone: anim::fadeHide, radioctl::presetCount, the L2 board's
setSdPower / setGrovePower / expanderOK (awRead, gnssReset now private).
- buttonBar() / barButton() replace 7 copies of the popup button row.
- confirmBody(): one red-button confirm for reboot / power off, reset
counters, admin commands, preset and scope delete.
- tapConfirmed(): the 11 "tap again" buttons share one mechanism, 3 s
everywhere, the label restored when left alone.
- noteLabel() for ~30 wrapped notes.
- localTm() in NodePrefs.h and one MONTHS table: lvgl, MsgExpand, the L1
clock and lock screen.
- The sim keeps UI settings in an in-memory nvs::, so one set of load /
save helpers serves both.
- Asleep, the loop wakes every 50 ms (the button / touch poll), not 20.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Settings > Sound > Quiet hours (L2: its own group; L1: Quiet hrs / from /
until): from quiet_from to quiet_to (local hours, wrapping past midnight)
the buzzer is muted and a message doesn't wake the screen; the clock alarm
still rings. A manual mute or unmute inside the window stands until the
window ends. Off with the clock unset.
NodePrefs gains quiet_hours / quiet_from / quiet_to (appended to the file,
sentinel 0xC0DE0030; sizeof 2832 on nRF52 and ESP32), off by default,
22:00-07:00. hourInWindow() / localHour() are shared with the bot's quiet
hours. soundctl::tick() replaces autoTick() (Auto and quiet hours).
Also fixes the L2 side-button mute: it went by the mode, not by whether
the buzzer was muted, so with Auto + an app connected (and now quiet hours)
the first press muted instead of unmuting.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Noise tab held the L2's interference hunt: parts of the board off one
by one, a slow cycle for a second radio, a spike hunt and three sweeps
(~90 s, L2 only). It is now one Measure: the noise floor on the mesh
frequency (median and 10th percentile of 5 s) and a +-1.1 MHz sweep round
it, ~12 s, with a short legend for reading it. Works wherever the variant
defines the radio.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Map tools: TRAIL, LIVE SHARE and ARRIVAL ALERT each with their own options
page (that section of Map options; back returns to Map tools), the arrival
alert switch where its state is shown, OFFLINE MAPS (Map areas, Live
tiles; the separate Download an area is gone -- a running or unfinished
download shows at the top of Map areas) and LAYERS.
Map areas: picking one frames it on the map above a small sheet (no Show
button, the map above isn't dimmed); closing the sheet, however it's
closed, puts the map back where it was and drops the outline -- it used
to stay on the map. New popup kind POP_BOTTOM.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The UI runs on Arduino's loop task, whose stock 8 KB stack had 340 B left
at its deepest point; a screenshot's file write went past it and reset the
device. The loop stack is now 16 KB (measured peak ~7.8 KB) and the
screenshot's line buffer is static. Diagnostics names the row
"UI stack free (lowest)".
With a cable in at boot, the USB popup waited for nothing and showed over
the splash; it now comes once the splash has faded.
Adds docs/development/plan-next.md, the plan up to the merge into dev.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Hold an app on Home (or Settings > Home apps) to arrange them: every app
is shown, hidden ones dimmed with a +; a drag drops the app into the slot
under the finger, held at a screen edge it goes to that page; a tap hides
or shows it. Settings stays shown. The order is kept in NVS (letters per
app, lower case = hidden), saved on Done or on leaving Home.
Roadmap: note the speaker-knock experiment (DAC ramp didn't help).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- USB through TinyUSB (ARDUINO_USB_MODE=0) instead of USB-Serial/JTAG: the
serial port stays, and plugged into a computer (enumerated, the bus not
suspended -- on battery an unplug only shows as a suspend) the device asks
"Charge only / USB drive". As a drive the firmware lets go of the card and
the computer gets its sectors; ejected or unplugged, the device restarts.
- Incoming messages: a banner slides down from under the status bar (icon,
sender or channel, two lines of the text from history); a tap opens the
conversation (not while locked), a swipe up dismisses it, not shown for
the conversation already open. Toasts moved up there too, under the banner
when both are up; the ones only echoing a switch or a choice are gone.
- Storage > Format card: a popup saying what goes, then Format tapped twice;
one FAT32 volume over the whole card (readable or not), then a restart.
- A deliberate restart shows a full "Restarting..." screen with the reason
instead of the screen just freezing.
- The screenshot flash is a thin frame round the edges: a full-screen fade
redrew all 320x240 each step and tore.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Two views, swapped with left / right: a polar sky plot (filled = used in the
fix, hollow = tracked, a pixel = no signal) beside fix, used / in view, HDOP,
PDOP and time to first fix; and a signal bar per satellite (used first, a
dotted 30 dB-Hz line) under per-constellation counts. The GPS stays awake
while it's open. Data from GpsSky (-D GPS_SKYVIEW, now on the L1 solo build).
The simulator's made-up sky moves from the L2's GpsScreen into
helpers/sensors/GpsSkySim.h, so both sims show it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Settings > WiFi keeps up to 8 networks (NVS "s0".."s7" / "p0".."p7"; the
single one older builds saved moves over). A tap loads one to change its
password, Forget asks with a second tap, a scan marks the known ones.
- Map downloads, live tiles and updates join the strongest saved network in
range (a short scan first when more than one is saved), else the newest
one -- which also covers a hidden network.
- Save joins the network once: "password OK", turned away (forgotten again,
the form keeps it to fix) or not in range (kept, checked later).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
JakubandClaude Opus 5.5
Huw DuddyandGitHub
ripplebizandGitHub
Liam CottleandGitHub
taco