Bring in the 65 upstream commits since 0.11.1, with ^K lock and the self chip.
Both sides kept their features. React gives way to Lock in two places. On the
PicoCalc lane, Lock keeps the Shift half of F2 and React is ^E only. In the
hint, ^E react is the first atom to go when ^K lock has no room.
The long-message code now refills the compose line with MentionEditor, which
replaced LineEditor upstream.
Each section of the outbox now has column headers: TARGET and TRIGGER for
waiting entries, TARGET and OUTCOME for finished ones. The rows no longer
show the message. Enter on a waiting entry opens it to edit (the courier
holds the entry while the prompt is open, so the old text cannot go out
meanwhile), and Enter on a finished one shows it. Del deletes an entry
after a red confirm, and ^R sends a waiting entry now; on the PicoCalc
they ride F3 as Send, with Delete behind it. On a narrow screen a long
name gives cells to the trigger and breaks off on its crack.
The message field has the colours of the chat's compose line: a white
prompt mark and white text.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
On battery with the Cap on, 79% of the gauge reads failed, in groups of
up to 110 s. The poll read 3 times in 11 s and then waited 20 s, so it
missed the good periods, and after 7 failed polls the gauge went away.
- After a failed read, the next read comes after 5.5 s, not 20 s.
- The last good reading stays for 10 minutes instead of 7 polls.
- A change of the charging flag waits for a second read that agrees,
because a garbled read said Charging while the pack discharged.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Select a message in a channel or a DM and press ^E (Shift+F2 on the handhelds) to
react with an emoji. A reaction is a usual message on the wire: MeshTerm writes the
format of MeshCore One and reads it and the two formats of meshcore-open. The
transcript folds each reaction under the message that it names, with a count, or
the names of the nodes while that message is selected. A reaction does not raise
the unread badge, and the Messages list previews it as "reacted 👍".
A reaction names its target by the sender's timestamp. Thus MeshTerm now gives the
device the timestamp of each message it sends and stores exactly that second. A
retry of a DM keeps the timestamp and counts the try number, as the protocol
specifies. The emoji that ^E offers are the reaction_emoji preference. Rooms have
no reactions.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
The star of our node had the hue of our key, set from the self-info at run
time. It is now the white of "you" on each surface: the map, the routes,
the lists, the legends, and the chat. The run-time colour, the device-state
hook, and the style generation of the render cache are removed. A faded
star keeps its grey.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The star of our node was the yellow of the map. SELF_MARK now names a
theme style, you.mark, and the theme writes our key's hue into it when the
device-state cache reads the self-info. The context makes that connection,
so the services still do not import the UI. The map, the routes, the lists,
the legends, and the chat all draw it, and the render cache keys on a style
generation so no old bytes stay. A faded star keeps its grey.
In a chat, a sender or a mention with no key had the light blue keyless
chip. It is now quiet: the grey of plain text on that grey darkened by the
chip rule. Routes keep the keyless chip, arrow mode keeps node.unknown, and
the PicoCalc keeps the keyless pair.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Greying every name that two keys ever used was too strict: a reflash with an
erase makes a new key, and 34 names on this mesh have one, JP's own among
them. Smog went grey this morning because of two keys silent since September.
The newest key now gets the name, and the name is grey only when another key
used it within name_overlap_days (a new Display preference, "Shared names",
default 7) of it: two live claimants. A contact that was never heard never
wins. On this machine's real data, LOG.IC 3R goes to its active key 47368a7c,
Smog to fa331c83, and 8 of 883 names stay grey.
The stored names come with their times from the one scan node_names() already
ran (node_names_with_times), so opening a chat costs no second scan.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A name is a label any key can claim, and a channel message carries only its
sender's name. When two nodes use one name, the resolver used to let the last
contact win, which handed one node's hue, map, and page to a message that
nothing tied to it: a second "LOG.IC 3R" with a new key took the first one's
green and location. Such a name now resolves to None, like any unplaced name.
A contact's full key and the stored id of the same node still count as one.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A final pass over the docstrings and comments: "the own X of Y" is now
"Y's own X" (84 places), a navigation entry is a screen on the stack, our
node is our node, a paint is a paint, the quoted UX labels are in code
font, and a list of punctuation glyphs is in code font.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The same rewrite as for the core package. Three comments now agree with
the code: the activity bucket of the header is one minute
(ACTIVITY_BUCKET_S), the RX log holds packets, and a reference to a
module note that does not exist is gone.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Hold-to-quit was the Cardputer's alone. It is now MeshTerm's gesture
everywhere a key can be seen coming up, which a terminal never reports:
- The emulator's window, for either handheld, which reads its own keys
as the Cardputer's panel does.
- A terminal's Esc, where the keyboard can be asked whether it is still
down: GetAsyncKeyState on Windows, as the right-Ctrl rescue already
asks, and EVIOCGKEY on Linux wherever the keyboard's event device
opens (the PicoCalc, a uConsole, a user in `input`). It reads the key
state only, no events.
An Esc arriving while its key is still down is held back as a hold:
delivered as it comes up, a box after a second, gone at three. The
keyboard's repeats are part of the hold, a key typed during it goes in
behind the Esc, and on a VT stream the last repeat, which prompt_toolkit
can parse after the key is up, is dropped rather than going back a
screen behind a box the reader just let go of. On macOS and over SSH Esc
is unchanged.
SIGTERM is a quit for the interactive menu on every platform now, not
only under the Cardputer's launcher: a system shutting down gets the
same clean exit, terminal and console font restored.
EVIOCGKEY checked on the Cardputer's arm64 kernel; the Lyra was offline.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The Cardputer Zero's launcher ends any app whose Esc is held: SIGTERM to
its process group at three seconds, SIGKILL three seconds after that
(cp0_esc_exit_policy.hpp in CardputerZero/launcher; M5's packaging guide
still describes an older Home-for-five-seconds rule). Its own hint never
reaches the panel while an app owns it, and MeshTerm took 3.4 s to exit,
so a held Esc ended in SIGKILL.
Now a tap of Esc is typed on its release, and a hold of one second
floats a "Hold Esc to quit" box whose bar fills over the two seconds
left. While it is up the app holds the device's transmit lock: nothing
new goes on the air, and a scoped channel send under way closes its
window first. Letting go gives the lock back and types nothing. Held to
the end, MeshTerm leaves the way Quit does, and the launcher's SIGTERM
joins that exit instead of interrupting it wherever the main thread
happened to be.
The exit itself fits: openhop-core's cleanup joined the radio's IRQ
thread, parked in an uninterruptible 30 s poll, for its full 2 s
timeout. The node drops that daemon thread from the books first. On the
device, under the real launcher, the process group now drains 1.3 s
after the SIGTERM, with no SIGKILL.
The emulator's window does the same on a held Esc, debounced for an X
server's release-press repeats.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The Cardputer Zero's Cap carries a GPS on the header's UART, which the
device screen listed as "~ n/a" (pyserial's name for a port with no USB
identity): a companion to connect to, and with nothing remembered, the
lone port a command-line connect would pick. A wiring now names its
board's GPS (`gps_port`, `gps_baud`; the Cap's is /dev/serial0 at
115200), and wherever that radio is listed, its GPS port is not: the
device screen, `meshterm devices`, and autodetect.
The radio node reads it the way MeshCore companion firmware reads a
board's GPS: `gps` (off until switched on, remembered in prefs.json)
and `gps_interval` (0, every fix, once a second), reported and set as
custom variables, which is where Device config already shows them. A
valid RMC or GGA fix becomes the node's position, the one its
self-info reports and, with Share location on, its adverts carry. The
last fix is saved as the node shuts down. Standard library only, as
the node's contract asks; NMEA checksums are checked.
On MeshTerm's side, our own self-info is now refreshed behind the
answer after a minute, like the contacts, so screens opened after the
node moves show where it is; and the connect-time settings offer no
longer flags a position the running GPS moved.
Checked on the Cardputer: the ttyS0 row is gone from `meshterm
devices`; `config custom gps 1` opens the port, is saved, and reads
back on; the real-terminal test passes there.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
With the Cap powered, I2C bus 1 doesn't only fail reads: some come
back garbled (a capacity of 56414, clamped to a confident 100), and
the kernel's bq27xxx driver caches whatever it read, failure or
garbage, for five seconds. Retrying 0.1 s apart only re-read the
cache, so every retry missed.
Now an unreadable or out-of-range percent is asked again 5.5 s later,
past the cache; a step of more than 5 points between polls waits for
the next poll to confirm it, which catches garbage that lands inside
0-100; and the sysfs reads leave the event loop, since a failing
transfer on that bus takes its time. Measured on the device with the
Cap on: a steady 96%, where it showed nothing or 100 before.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Tried on a real Cardputer Zero for the first time. It draws on the
panel, takes the keyboard, and opens from M5's app launcher:
scripts/cardputer-zero/launcher-entry.sh puts a source checkout there
(and `remove` takes it off) until there is a store package.
Three things the hardware showed:
- The splash opened with the highlight on /dev/ttyS0, the Cap's GPS
UART, because serial is listed first as found. Below the companions
we've used, anything that looks like one now comes before a port
nothing vouches for, across transports.
- The title bar's battery is the handheld's own BQ27220, as the
PicoCalc's is. The host pack is now found by what it is (a Battery
supply that isn't a peripheral's), not by the PicoCalc's name.
- While the Cap is powered, I2C bus 1 turns noisy: a quarter to three
quarters of the gauge's reads fail with an I/O error (one in ten of
the clock's), none with the header off. A failed read is asked again,
a status that won't read keeps its last verdict, and a good reading
stands through six failed polls instead of blinking out.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Only 24 of 144 direct messages and posts in the history were ever marked
delivered. Three faults, each fixed:
- The wait was the radio's own estimate times 1.2: about 5.7 s for a
flood, while a flooded round trip from a few hops out takes ten (a
room's answer from five hops landed at 10.1 s on hardware). It is now
the login's patience, the estimate stretched a quarter plus a round
trip sized to the route, about 15 s for a flood.
- An ack that landed after the wait was dropped. The radio pushes one
whenever it arrives, for any of its last eight sends, but nothing was
listening, so a delivered message stayed failed for good. The history
shows ACK and PATH packets heard seconds after messages recorded as
failed. The send now returns a Delivery carrying the ack code, and the
chat service keeps the last sixteen messages still waiting. A late ack
flips one to delivered, in history and in the message the chat screen
holds, and every code a retried message went out under counts.
- The Courier trusted the failure and sent the message again: a
duplicate for the recipient whenever the ack was only late. A late ack
now marks the entry delivered.
The ack listener is also armed before the send rather than after it,
so a neighbour's ack quicker than the send's own return is not missed.
The debug log names how a message went out and the radio's own
round-trip time.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Joining a room moves out of Chat into a Rooms page beside Channels. It
lists every room server the radio knows, joined ones first, with each
room's access, unread count, last post and last-heard time. Each room
has a page: join, log in again, change password, open the board, forget
it, or delete its stored posts. Chat now lists only joined rooms, the way
it lists only channels on a slot, and only a joined room raises the
unread badge. chat join becomes rooms join, alongside rooms list and
rooms forget.
"No reply" told the reader nothing, because a room is just as silent for
a wrong password as for a login that never reached it. Testing on
hardware showed the commonest cause is neither: a login sent along the
route the radio had learned met silence, and the same login by flood was
accepted in eleven seconds. A silent login is now explained from what is
known: how it went out, when the room was last heard, and whether this
password has worked before. It offers the one next step that evidence
points at:
- Try by flood, after a learned route met silence. The radio forgets
the route (local, nothing transmitted), then sends the login once.
- Try again…, after a flood met silence. The prompt reopens with the
password filled in, so Enter resends it and typing replaces it.
- Add it & try again, when the radio doesn't hold the room as a contact.
Every retry is a button press, and none is ever sent on its own. A
password over 15 bytes is refused at the prompt, since MeshCore sends
only the first 15.
On the board, a failed or unfinished login used to leave the compose
line taking posts the room would drop. It now opens only while we are in
and may post, and otherwise says why and what to press. Opening a joined
room that has gone quiet logs in in the background. ^L runs the
explained flow.
Underneath, the login exchange listens for the radio's own confirmation,
matched to the login by its expected ack. It records whether the login
went by flood or along a route, or the radio's reason for not sending it
at all. This also corrects the debug log: "send unacknowledged" read the
library's wait for the *answer* as the send. On hardware the radio
confirms in under 0.1 s. Device.reset_route is new. The simulator learns
routes, can be given a stale one, and refloods.
Refs: #24
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A review of the room work turned up ways the remembered passwords could
go wrong, and a doubled login:
- A room that lets readers in lets any password in, so a login that only
earned read-only proved nothing about the password. It replaced the
room password anyway, so one mistyped ^L left a member read-only for
good, with no prompt to fix it. It now never replaces a password that
is already kept, and ^L asks for one while you are read-only.
- A blank login only asks "do you know me?". Accepted, it overwrote the
remembered room password, and as an admin it emptied the admin
password Repeater admin shares. It now replaces neither.
- A room's login replaces a known member's role with the one the
password earns, so an owner who held both passwords was demoted to
member each time the room password was tried first. The admin password
now goes first.
- Closing a room view and opening it again mid-login sent a second
login. The room service now shares a login already in flight, and a
room with one under way is not due another.
- A refusal forgot the room password even when it refused a different
one. ^R is no longer offered to a read-only member, whose resend the
room would drop. Enter on a picked post opens its paths when there is
nowhere to put a reply, and a paste is refused there too.
Refs: #24
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A room server is a message board on a radio: it keeps its members' last
posts and sends each member the ones they missed. MeshTerm could see rooms
and run your own, but not join one. Posts that reached the radio were filed
as direct messages from the room, with their authors thrown away.
Chat now has a Rooms section between Channels and Direct. Opening a room
shows its board at once, each post under the member who wrote it. A post
carries the first four bytes of its author's key, so it is named by any
contact or overheard advert with that key and coloured by the key itself.
An author nothing names stands as the hash, in the unknown-node grey. The
first visit asks for the room's password over the board. After that,
opening a room logs in again only once it has been quiet for
room_relogin_minutes (a new preference, 30 by default), and ^L (F6 on the
handhelds) logs in on demand. The title says what the room allowed:
member, admin or read-only. A read-only member gets a line saying the room
drops their posts, in place of a compose box. A post goes out once, never
soft-retried, since a duplicate would reach every member.
Underneath: messages carry an author (schema 18), a room's board leaves
out its command replies, and a re-sent post is recorded once. Room posts
raise the unread badge and a room's command replies don't. The device
gains room_login, sharing admin_login's exchange and reading the access
the room granted. Room passwords are remembered in rooms.json; an admin
login is kept in admin.json, shared with Repeater admin. The simulator
has a room, Lakeside BBS, to join with "hello".
Two fixes the research turned up. An admin command took the first direct
message from anyone as its reply, so a room post or a companion's message
landing mid-command was read as the node's answer. It now listens before
sending, and only for a reply from that node that is not a post. And
Repeater admin names a room server's guest password "Room password", the
password members join with.
Refs: #24
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Panning or zooming the map quickly on the PicoCalc could stop the whole app
for many seconds. A watchdog run on the device found two causes behind it,
plus the port walk fixed in the previous commit.
Memory. A decoded tile holds 1 to 4.6 MB of RAM on the device (65 bytes a
point), not the 0.9 MB the code assumed. The source kept up to 24 of them,
counted as tiles, and the map screen kept 8 or more of its own on top, so a
few pans and a zoom out filled a 104 MB device. Swap is on the SD card, so
the process then stopped while the kernel read its code back in: 16,000
major page faults and a 15 s freeze in one run. Now the source holds decoded
tiles to a byte budget, one eighth of the machine's RAM (12.8 MB on the
PicoCalc, at most 128 MB on a desktop), weighed by mvt.resident_bytes. That
estimate was fitted against tracemalloc on the device's 157 cached tiles: 1%
under to 9% over there, and 6% under to 21% over on a 64-bit desktop. The
screen keeps only the tiles in view, and takes any tile already in RAM
straight from the source on the paint (BasemapSource.resident), with no trip
through a thread. The prefetcher warms the disk cache (BasemapSource.warm)
instead of filling the RAM with guesses, and passes over any tile already
written down. Tile loads take turns, two at a time, and a tile the view has
left before its turn comes is dropped, not loaded.
CPU. fill_polygon asked every edge about every scanline, which at z7 means
coastlines with tens of thousands of edges: 6.8 s of a 7.6 s frame. Each
edge is now filed under only the rows it crosses, and draws the same dots
(checked against the old fill on 3,000 random polygons, and in a test).
_draw_tile also tests each ring's bounds before projecting it, and skips a
ring wholly outside the view. That changes no dot: 1,500 frames from the
device's tiles match rendering with no culling at all. The one frame that
differed from HEAD was HEAD's own building cull, which dropped a footprint
whose fill rounds onto the canvas edge.
Measured on the PicoCalc with the profile-picocalc map_burst tour (keys sent
at 10 a second, from a pinned z10 view, real radio and database). The new
code ran first, so the baseline had the warmer tile cache:
pans, catch-up after a burst HEAD 1.0-1.8 s now 0.4-0.6 s
longest silence, whole tour HEAD 11.6 s now 0.6 s
whole-process freezes HEAD 8.0, 3.1, 2.1, 11.6 s
now one, 1.6 s (idle pages swapped back in)
swap at its peak HEAD 25 MB now 7 MB
warm raster, z7 / z8 / z9 HEAD 7.6 / 1.9 / 1.8 s
now 0.57 / 0.29 / 0.81 s
warm raster, z10 / z13 HEAD 0.49 / 0.56 s now 0.26 / 0.26 s
Still open: z9 spends 0.23 s drawing lines; a prefetch decode of a heavy z7
tile is still 1.3 to 1.7 s on a worker thread; and pages swapped out while
the map sits idle can still cost a second or two to come back.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
"Host" already meant two things here: a TCP companion's network host,
and battery="host" for this machine itself. And with a PicoCalc preview
coming, the package is less a console host than an emulator of a
handheld's display: its panel, grid, font, colours and keys. So it is
meshterm.emulator, Platform.console_host is own_display, its desktop
window is window.py (EmulatorWindow) and its device front end
framebuffer.py, since "sim" and "device" now mean the radio simulator and
the device being emulated. MESHTERM_HOST_FONTS is MESHTERM_EMULATOR_FONTS.
The package's own description opens on what it is not: nothing runs the
handheld's processor or operating system; it mimics the display's
constraints so the window obeys the device's.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
M5Stack makes three Cardputers and only the Zero runs Linux; the PicoCalc
takes several cores and only the Luckfox Lyra runs MeshTerm. So the bare
names were ambiguous, and every reference to the two devices now carries
the variant: the --platform values and MESHTERM_PLATFORM, the Platform
constants (PICOCALC_LYRA, CARDPUTER_ZERO), their lane decks and every
screen's picocalc_lyra_lane / cardputer_zero_lane, the font inventories,
the lane chips' theme styles, the scripts/ folders, the two chrome test
files, and the docs and CLAUDE.md that name them.
There is no alias: --platform picocalc is refused with the three real
choices listed, rather than falling back to the desktop layout. The
PicoCalc is auto-detected from its device tree, so it needs nothing.
What names hardware or a user's data keeps its name: the PicoCalc
keyboard's input-device name, its battery's sysfs supply, and the serial
profile the Lyra setup script writes. .gitignore follows the firmware's
folder, and the changelog keeps the words each release used.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The stock image gives an app no text console, so MeshTerm draws its own: the TUI
runs in-process with prompt_toolkit's output feeding a small VT parser, the cells
are drawn in Terminus 6x12 with MeshTerm's marks over it, and keys go in through a
pipe in xterm's spelling. The front end is the only part that differs: framebuffer
and evdev on the device (written ahead of the hardware), a Tk window on the desktop.
Each handheld's glyph inventory is now named by Platform.font, which replaces
ascii_fold; the render-boundary fold derives its accents, emoji, and fallback from
that inventory and quantizes colour only where truecolor is off.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A lane's keyboard-bound half is now a LaneDeck named by Platform.lane_deck: which
keycodes drive the slots, where the chips sit, how they are drawn, and which of a
screen's lane definitions it reads. Screens define picocalc_lane; cardputer_lane
follows it until a screen overrides it, so the two can part ways one screen at a
time. The Cardputer deck is Fn+4..8, Shift with them as F16-F20, eight-cell chips
centred over their keys in the fn key's orange-red and the Shift blue.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
MeshCore firmware sets its clock only forward: CMD_SET_DEVICE_TIME with an earlier time is answered ERR_CODE_ILLEGAL_ARG, which read as 'the device rejected the request as malformed'. Since 2026-09-24 the companion's clock has been ahead of this computer's (itself within 0.7 s of NTP), so every connect's sync was refused. set_clock now reads the radio's clock first: within a minute ahead is in sync and nothing is written; further ahead raises ClockAheadError, which says by how much and that a reboot resets it. The device layer also maps the firmware's refusal to that error, and the simulator's clock now only moves forward, like the firmware's.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The firmware has no per-channel scope, so a channel send is now a window
on the companion's session scope: set it, send, restore it, under a
transmit lock every flood-capable send also takes (DMs, adverts, admin
commands, logins, regions/neighbour requests), so nothing interleaves.
Ack waits stay outside the lock. A scope the radio can't take stops the
message (FloodScopeError naming the firmware it needs) rather than sending
it unscoped or under the default; * checks the firmware version, since
before 1.16 the override would quietly fall back to the default scope.
A failed restore is repaired by the next transmission.
Channel scopes are set from the channel page's Send scope picker (known
regions with their carriers, or a typed name) and from the CLI (channels
scope, chat send --scope). The chat's title carries the scope as a status
atom, ^R/Resend sends the newest scoped message again unscoped after an
amber confirm, and the scope each sent message went under is recorded
(messages.scope, schema v17) so the paths dialog can say what it went
under and when no known repeater carries that region. Device config now
teaches the region store the default scope it reads or sets.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The packet viewer's route row reads "flood · scope harbour" (or
"scoped · 3fa1" for a region nobody here has named, or "unscoped")
instead of the library's bare "tc flood"; a direct frame keeps its
route word and states no scope. The live feed gains a SCOPE lane where
the width holds it, 72 columns included (the subject lends it one
cell); the PicoCalc's row already scrolls at 53 and leaves the scope to
the viewer. The message paths dialog states a message's scope once, as
a title atom, read off its arrivals' frames. monitor streams a SCOPE
lane before NAME, and the scope object (or null) under --json.
The viewer and the feed take the region store's scope_of by injection,
falling back to reading the frame alone, and the viewer's card cache
keys on the scope so a region learned while it is open renames it.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A muted line under the Weekly advert row on the Preferences page, set in
the value lane: "next advert in 4d" while the week runs, "due — waiting
for a packet" and then "due — after 30 s of quiet" once it is due, "no
device connected", and "the week starts on Apply" while it is only staged
on. The line is a width-aware separator that reads the scheduler's new
status() on every paint, from state the loop already keeps, so it follows
the wait live without touching the file. It is drawn, even blank, for as
long as the preference is on, so the rows below never shift; every message
fits the PicoCalc's 53 columns, and a gallery specimen holds it there.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
A MeshCore companion never advertises by itself, so the hourly zero-hop and
daily flood adverts MeshTerm sent by default were the only automatic adverts
on the air, and more than a companion needs. The Background adverts rows leave
Device config and `config advert-cadence` goes with them.
In their place, two preferences under Sending: weekly_flood_advert (off by
default) and advert_quiet_s (5/15/30/60 s, default 30). Each device keeps its
own real-time week, running from the latest of the switch-on, its last flood
advert (automatic or by hand), and its first connection; a due advert survives
a restart. Before sending, each connection must hear a packet, then the air
must stay quiet (nothing heard, nothing of ours sent) for the spell plus a
random 0-5 s, drawn afresh at every break so two waiting radios part.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Import order, a docstring reworded rather than have "macOS" capitalised
into something that is not the product's name, two over-long test lines,
and _tls_context moved below the logger it uses so the module keeps its
shape: docstring, imports, constants, logger, then code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
On a Mac the map showed nodes floating over nothing: overlays come from the
local database and arrived fine, while every base tile silently failed.
A frozen build can have no trust store. The default TLS context asks OpenSSL for
one, and on macOS and Linux that is a filesystem path baked into the
interpreter's own build -- which need not exist on a machine that never
installed that Python, which is every machine a PyInstaller bundle lands on.
Each tile fetch ended in CERTIFICATE_VERIFY_FAILED, and since the fetcher cannot
tell a refusal from empty terrain, the map drew what it had. Windows is the
exception that hid this: its default context reads the OS certificate store, so
the binary there always worked, and so did every checkout on every platform.
So the context now keeps the default when that default actually loaded
certificates, and otherwise points it at the bundle the operating system ships.
Those are the trust decisions the machine's administrator has made and the OS
keeps current; carrying our own copy would freeze a snapshot of them into every
release, and certifi -- the obvious way to do that -- is MPL-2.0, which this
project's own dependency policy refuses. `tests/test_notices.py` said so, which
is what that gate is for.
Verification is not weakened anywhere. With no bundle to be found the fetch
still verifies and still fails, because a map tile is not worth teaching the app
to skip certificate checks, and an unverified fetcher would be one import away
from being reused where it matters.
Confirmed against the 0.3.6 arm64 build on a Mac: handing the same binary the
OS bundle turned one hard failure into a fetched TileJSON and 7.3 MB of tiles.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The Calculinux bring-up, both console-font builds, the GPL-2.0 text that belongs
beside the 6x8 one, and the xiao-radio firmware kit move under scripts/picocalc/;
the SPI bridge is scripts/uconsole/meshterm-spi-bridge. The folder's own README is
gone: its per-file index is now a section of docs/hardware.md, next to the table
that says which manual is yours. Every path in the manuals, NOTICE, CLAUDE.md, the
tests, the package docstrings and the scripts' own headers follows.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A new "Set clock on connect" preference (Device group, off by default) has the
interactive session write the host time to the radio once per connection — at
launch, on a deferred connect, and after a reconnect, since a link drop is exactly
when a board without a real-time clock may have rebooted and lost the time. The
set runs in a background task off the connect path and reports only in the log,
with the drift it corrected; a refused write costs a warning line, never the
session. Scripted CLI runs never start the service, so a one-shot command that
only reads the radio never writes its clock.
The device-side step is the one the Device config screen's Sync clock action
already ran inline in apply_ops; it is lifted into services/clock_sync.set_clock
and the action and the service both call it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
scripts/calculinux-console-font.sh is now calculinux-console-font-6x12.sh.
Every reference follows: the setup script's hand-off, the 6x8 script's
SHARED default and comments, the theme16 test that parses it, the module
docstrings, the two manuals, the scripts index and CLAUDE.md. The font files
it builds keep their names.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Apache-2.0 does not require per-file headers, but a file lifted out of the tree carried no marking at all. One comment line, first (or right after a shebang), on every .py under meshterm/, tests/, packaging/ and scripts/, plus the extensionless SPI bridge; tests/test_spdx.py keeps a new file from landing without one.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A Linux VT loads one font for the whole console, so asking for the 6x8 build
(53x40, fourteen rows more) changes what the shell is drawn in too. The new
console_font preference (6x12 default, 6x8 the alternative, offered on the
PicoCalc only, round-tripping through preferences.yaml on every platform)
therefore comes with services/consolefont.py: remember the font the shell had
with `setfont -O`, apply the one the preference names, and restore it on every
way out -- the menu's finally and an atexit hook, since ^Q and an unhandled
error leave by other doors. Verified on the device that setfont -O captures
the current font with its unicode table and that feeding it back reproduces it
exactly, with no sudo, because the VT is the process's own controlling
terminal. Only the interactive menu switches; a scripted run never touches the
console it was invited into. Changing the preference on its page applies live.
Shipping both fonts needed the licence untangled: the 6x8 base is the Linux
kernel's font_6x8 bitmap, GPL-2.0, so its build moves into
calculinux-console-font-6x8.sh under that licence with the GPLv2 text beside it
in scripts/LICENSE.GPL-2.0. The shared generator stays in the 6x12 script
between BEGIN/END markers and the 6x8 script executes that block, so the marks,
donors and aliases the font contract pins exist in one place. The produced PSF
is a data file the kernel console reads; MeshTerm never links or bundles it, and
NOTICE and the hardware manual say so. PrefSpec gains a platforms gate for the
page; the labels use an ASCII x because U+00D7 is not in the console font.
The 6x8 base has no Cyrillic; the manual says so and the owner accepts it.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A pre-release architecture and QA pass, applied as one change because the moves
share files and only the whole set was gate-tested (2480 passed, ruff clean).
Three new modules pull logic down out of layers that should not own it:
core/channel_probe.py (ChannelSlot, read_channel_slots, probe_channel_slots)
so services/device_state.py no longer imports a screen; services/markers.py
(gather_markers, with MapMarker a core/models dataclass) so two screens stop
importing a tool; core/tracing.py (trace_timeout, path_hash_flags) so
core/connection.py no longer imports services. The routing-prefix width is
DeviceState.routing_prefix_bytes(), replacing five private cross-module imports.
One definition per thing: is_path_hash, the NodeResolver/NameKeyResolver
aliases, TILE_RETRY_SECONDS, haversine_km, marks.SELF_MARK, MENTION; nine
private names that crossed module boundaries got public ones and menus.__all__
lists what other modules import. A preference's default is written once
(ALERT_CAP, DONE_CAP, DEFAULT_VIEW_FRACTION read the registry), the stranded
PREFERENCES doc comment sits on its subject, and three docstrings stop naming
questionary, a library dropped long ago.
The command line gains --version (eager, exit 0, documented, tested) and drops
the -p short form on trace, trace-path and tx-optimize: the root lifts -p as
--profile before a leaf ever sees it, so the leaf's could never work, and a
derived test now checks every leaf against every root short option.
tx-optimize with a one-hop --path is a usage error, exit 2, because nothing
transmitted is not a device failure. Log handlers are closed before they are
cleared (197 unclosed-file warnings gone), chat's fire-and-forget tasks keep a
reference, coroutines use get_running_loop(), the never-awaited MapScreen._load
in a test is awaited, and the CLI sweep passes --no-ble to devices so a test
run stops turning on the Bluetooth radio three times (the suite is 40 % faster).
The about page lists every dependency, and the release skill bumps the README's
download names, which still said 0.2.8.
apply_ops stays in tools/: it and its helpers build report blocks and open two
screens, so moving it to core/ would invert core over ui. The widgets CLAUDE.md
names by their private spellings (_format_age, _NODE_GLYPHS) keep them until
CLAUDE.md is edited alongside.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Pick location on map gathered its surrounding nodes before opening, and that
meant reading the contact list from the radio whenever the session hadn't
cached it. A companion refusing that read (ERR_CODE_BAD_STATE, which the log
shows all session) held the picker closed through every retry, twenty-odd
seconds, only to open with no contacts anyway. Our own position waited the
same way after Device config dropped its cached self-info.
The picker now opens on what is already in hand: cached contacts and
self-info (DeviceState.peek_contacts / peek_self_info) and our own history,
via gather_markers(wait=False). Whatever the cache lacked is fetched behind
the open map and joins it through MapScreen.set_markers when the radio
answers. The read is never cancelled, so a fetch another screen shares
survives the picker closing. The Map tool and the node page keep waiting,
unchanged.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The slot probe stops for two very different reasons and returned the
same short list either way. Off the end of the configured slots it is
the truth; a read that simply failed leaves whatever it had, and that
was cached for the session - an empty one reading exactly like a
device with no channels, a truncated one making the next free slot
look free with a channel sitting in it, which the add flows then wrote
straight over. The two endings raise different things, so the probe
now says whether it finished, an unfinished one is answered but not
kept, and every add confirms the slot is empty before anything is
written to it.
The chat service's slot map was also walking the radio itself, and
unbounded - it skipped empty slots all the way to the 64-slot cap, so
on firmware that never rejects an index every edit paid 64 round-trips
before the screen could redraw, on top of the manager's own re-read.
It now builds from the probe the manager was making anyway.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three things, all the same story - the manager was silent while it
worked and talkative once the work was already visible.
The acknowledgement goes. It reported changes the reader had just
watched land in the list in front of them, after the screen showing
them had closed. The count stays in summary, for the run log.
Every device action now reports while it runs, under the modal busy
card: write_channel draws its own (it is the one boundary every
mutation crosses), a reorder draws one card for the batch and retitles
it per move, and the re-read that follows a change draws one rather
than three. Modal is the point - the list underneath stays armed the
whole time it waits, so keys tapped during a slow write were landing
in its filter and coming back as a list showing nothing.
And "a write happened" moves off the change count onto the cache the
write drops. A write that landed and then raised - a QR view unwound
with ^W, a link lost mid-reorder - never reached changes += 1, so the
stale slot list and the chat service's slot map outlived a layout that
had already moved. write_channel now invalidates as it writes, which
the four CLI subcommands were never doing at all, and the manager
compares the epoch across a round.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`ruff format .` across the repo: 153 files, no behaviour change. This is the
second gate CONTRIBUTING.md promises, and until now it reported 154 files needing
reformat — so a first contributor ran the command the guide told them to run and
could not tell their own diff from the background.
Landed on its own, with nothing else in it, because a whole-repo reformat is only
safe to skip past in blame if it is provably nothing but layout. It was checked
that way rather than assumed: every changed file was formatted in memory and
compared by full `ast.dump()`, and 152 of 153 come out structurally identical
(the one exception was a docstring opening on a quote character, fixed in the
commit before this one). The string and comment token streams were diffed
separately, so a quote flip or an edited comment could not hide inside a
re-layout — five quote-style normalisations and sixteen implicit-concatenation
joins, all value-identical, and one commented-out line gaining the space after
its `#`. `meshterm specimen` is byte-identical on both platforms, and the suite
is 1888 green, gallery gate and `test_theme16` included.
What actually changed, in order of how much of the diff it is: wrapped calls and
signatures rejoin onto one line where they fit inside the hundred columns; the
magic trailing comma expands what it is there to expand; complex slices gain
their spaces (`[len(noise) :]`); a few redundant parens go. The tables that are
laid out to be read were fenced in the previous commit and are untouched here.
The aligned trailing-comment column is gone from about ninety lanes that were not
fenced — including the `heat.*` ladder in `MESH_THEME`, which cannot be fenced
where it sits. Every comment stays on the entry it describes, so nothing there
reads wrong; it just no longer lines up.
Add this commit to `.git-blame-ignore-revs`, which follows immediately.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`MeshTerm/0.1 (+https://github.com/; mesh node map)` went out on every
OpenFreeMap tile request and was wrong twice over: the URL is a bare host
nobody can visit, and `0.1` was a hand-written number that had already
drifted from the package's `0.1.0` and would drift again on the next bump.
OpenFreeMap issues no API key, so this header is the entire handle a tile
operator has on a client — the difference between being emailed about
traffic and being blocked for it.
The URL now points at the real repository, and the version is the package's
own `__version__`, which is what every other surface that shows a version
already reads (the About page, the splash wordmark, the menu mark, a trace
export). That is the single source of truth here, so reusing it is what
keeps the string honest.
Deliberately *not* `importlib.metadata.version("meshterm")`, despite that
being the usual advice: MeshTerm is run straight from a checkout at least as
often as from a wheel — this repo's own venv has no `meshterm` distribution
installed, nor does the PicoCalc deploy — so the metadata lookup would raise
`PackageNotFoundError` in exactly the setups it is developed and tested in,
and a fallback around it would be a second version literal to keep in step
with the first. Reading the package attribute has no failure mode at all,
which is also why the value stays a module constant rather than becoming a
function: it never changes, computing it cannot raise, and the tile fetch
path is no place to discover either.
Three tests in the basemap section of `test_map.py`: the string is exactly
the repo, the live version and the descriptive tail; it rides the header of
the request that actually leaves; and the module still imports with a real
version in it when the metadata lookup is made to fail outright, which is
the regression guard against someone reaching for `importlib.metadata` later
without a fallback.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`D205` across the 20 sites in `meshterm/` and `scripts/`, which leaves the
package and the scripts fully ruff-clean.
None of these were a summary that happened to wrap. They were a paragraph with
no summary at all, so the fix is additive: a plain sentence saying what the
thing *is* goes on the first line, and the prose that was there becomes the
description under it. "The transport of the current/selected connection:
``"serial"``, ``"ble"``, ``"tcp"``, or ``None``" now opens with "Which
transport the current or selected connection uses" and lists the four values
below; `_emit` leads with what it returns before the four paragraphs on why a
wide glyph forces a whole-row repaint.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`E501` across 69 lines, wrapped rather than exempted.
The prose ones — module docstrings in `menu`, `node_detail_screen`,
`pathline`, `pathgraph`, `walk_screen`, `trace_runner` and `test_about` — are
reflowed as paragraphs, so the block keeps its shape instead of gaining one
short orphan line. Nothing they say changed.
The code ones mostly gained a line break inside a call. Where a break would
have read worse than a name, the sub-expression got one instead: `reach` in
`topology`'s quality clamp, `plain_cells` in `pathline`, `numbering` in the
trace list, `listed` in `flash.py`'s refusal, and a small `shape`/`is_strip`
helper in two tests that were comparing or matching the same shape twice on one
line. Four docstring summaries lost a word they were not using.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`B905` over 56 sites, and the answer differs by idiom rather than by file.
21 were the sliding window `zip(xs, xs[1:])` — ragged by exactly one, on
purpose, so `strict=` is the wrong question entirely. Those become
`itertools.pairwise`, which is what they always meant and needs no flag: the
walk's SNR stops, the heat ladder, the route chains in `topology`/`records`,
every polyline in `mapcanvas`, and the graph's edge sets in `pathgraph`.
The other 35 pair sequences that must line up, so they now say `strict=True`
and a mismatch raises instead of silently truncating the longer side. Running
them under that assertion is what sorted the two groups: the first pass put
`strict=True` everywhere and 208 tests failed, all of them a sliding window.
One goes the other way. `mvt._decode_feature` reads tags off a network tile,
where an odd-length list means the tile is malformed — and this decoder answers
that with a shrug everywhere else (`if not geom_ints: return None`, bounds-
checked key and value indices). It keeps `strict=False`, now with a comment
saying why, so a bad basemap tile still can't crash the map.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`ruff check --fix` over the whole tree, ahead of putting the linter in CI:
1977 mechanical fixes, overwhelmingly `Optional[X]` → `X | None` (1186 in the
package alone) and the quoted annotations that `from __future__ import
annotations` already made unnecessary, plus import sorting and unused imports.
Nothing here changes behaviour — the suite is 1885 green before and after.
Four findings were real rather than stylistic and are fixed by hand: a dead
`cells` range in `MapCanvas`, a dead slot dict in the heat-ladder test, a
`ProgressScreen` task handle nobody read, and a tab loop binding a `label` it
never used. The two `F821`s were quoted forward references to names imported
inside their function bodies — now declared under `TYPE_CHECKING`, which keeps
`test_performance`'s local import (it guards the startup import graph) while
letting the linter resolve the name.
412 findings remain and are deliberately untouched: they need a decision about
the rule, not an edit to the code.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Three kinds of "setting" were tangled into two files and one of them had no
screen at all. The radio's settings live on the radio and are edited on Device
config. Machine setup -- where the database is, which companion to talk to --
stays in config.toml, a text editor's job. Everything in between was app
behaviour: six keys sitting in config.toml with nothing to open them, and a
dozen more that were only module constants a reader could never reach.
Those are preferences now, and they have a page. core/preferences.py is a
PrefSpec registry in the shape of the device-config one -- group, type, bounds,
help, and a default -- and the page is built from the same parts as the device
editor, deliberately: one grouped lane-aligned list, current -> new staged in
warn, Apply/Back appearing only once something is staged, the shared discard
confirm on the way out. Two editors that stage values should not be two screens
to learn. What it adds is a Defaults section holding one row that stages every
changed preference back to its built-in value; it is drawn only while a reset
would do something, and it stages rather than writes, so it is as reviewable and
as Esc-discardable as any other change and the file is still only ever written
by the one action at the bottom. Its rows scroll sideways where the device
editor's don't: the explanation is the widest thing on the row and the only part
that runs past the edge, so <-> slide it under a pinned setting and value.
The file is preferences.yaml and holds only the values you disagree with the
code about -- grouped and commented, each entry above its help and its default.
That is what lets a later change to a default reach an install that never asked
to be exempt; delete a line and the code takes over again. A hand edit gone
wrong costs that line, never the session.
Every preference is wired to something that reads it. ctx.preferences is the
explicit path; preferences.current() is for the render layer, which has no
context to reach through, the same reason get_platform() exists.
The About pages come along into a This app section, the third scope after this
node and other nodes, with Preferences leading it -- the one row there that
changes MeshTerm rather than describing it. A section of one would have broken
the menu's own no-bucket-of-one rule, and the app's scope is the honest home for
both halves anyway.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
No, it isn't the ack — ACK is its own payload class and was never in the match.
Both symptoms come from the same place: a packet header's route type, which we
never recorded, decides what its `path` field *is*.
A FLOOD packet accumulates: every relay appends its hash, so the path is the
route it travelled to reach us. A DIRECT packet is the opposite — the path is a
routing instruction the sender wrote and the relays consume — so it says where
the packet is going, and an empty one means the route was used up, not that the
packet arrived in a single leap. Reading every path the first way is what drew a
message from five hops out as having appeared from nowhere.
The evidence is in your own log. The four-hop route on those sends is identical
across four consecutive messages, where a flood-accumulated path varies every
time; it ends on 40ef0f, which is the first hop of everything hairu sends *us*;
and a PATH frame carrying exactly those four hops crosses the log seconds before
the first message that uses it. That is a learned route to hairu being announced
and then used — not a route anything took to us.
So route_typename is stored (schema v15, a `route` column beside the crypto and
addressing ones added for the same reason), and an Arrival carries whether it
was routed. A routed frame with no path has route_known False: it is kept as a
copy that reached us — the count and the time are real — but it is drawn as
"routed · route not carried" instead of a confident zero hops, and it puts no
lane in the fan, because an empty lane there is a claim of adjacency the frame
never made. Frames from before the column read None and fall back to the old
behaviour rather than to a guess.
The round trip is a second bug and needed no new data. The path line closed on
our own ★ unconditionally, and for an outgoing message its head is our ★ too —
so every send drew ★ ▶ … ▶ ★. The graph did the same, DST_NODE having been
hardcoded to us. Both now take a destination: the far end for a message we sent,
ourselves for one we received. It is the rule the line's head already followed —
a path runs between the nodes it went between — applied at last to its tail. A
channel broadcast keeps us at both ends, because there the copies we log really
are our own message coming back.
collapse now keys on the route type as well as the hops, so a route travelled
and a route intended never fold into one row even when the hashes match.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
