prestonbrown/helixscreen v1.0.0 on GitHub

v1.0.0

Upgrading from 0.99? Nothing to do differently. The version jump is cosmetic:
every 0.99 install is on config_version 23 already, so this is an ordinary forward
step for the updater and your settings, layout and printer list carry over untouched.

1.0. Eleven months and 203 builds after the first commit, HelixScreen is a
complete touchscreen UI rather than a promising one: 80+ printers it configures itself
for, a dashboard you arrange yourself, bed mesh and input shaper rendered on the
machine, multi-material across seven filament systems, and nine languages. It does that
in about 15MB of RAM on a printer's own board, which is the whole point - the screen you
already own is enough.

Calling it 1.0 is a statement about the line, not about this build in isolation. The
0.99 series ran long on purpose, and stable now means the update channel most installs
track points here.

The last stretch of work went to three places. Networking: on printers whose firmware
runs its own network daemon, the Wi-Fi and Ethernet page drives that daemon instead of
sitting beside it. The AD5X's filament system, drawn the way the machine is actually
built, now understanding the IFS module split out of Z-Mod. And sound on the FlashForge
machines, where the AD5M's buzzer plays music instead of static and the AD5X's silent
buzzer was found and fixed. Around those: USB mice on 64-bit machines, touch calibration
on rotated displays, previews going monochrome on routed multi-tool prints, a quieter
pairing prompt, home-panel crashes chased from a field report, long-untranslated section
titles, and a memory bug in confirmation dialogs that the nightly sanitizer run caught
before any field report.

Added

Sound

Music on the AD5M's buzzer - the built-in buzzer never really worked: the channel it
needs is not enabled after boot, and the speaker cannot reproduce recorded audio, so
attempts came out as static. Melodies now play as clean square-wave notes, PC-speaker
style - following the lead line, shifted into the range the buzzer is actually loud in,
and ending in silence instead of holding the last note. The buzzer channel is enabled
automatically at startup, so sound survives every boot, and the playback engine is built
to never compete with the print host for the processor.

Networking

Wi-Fi and Ethernet through the printer's own network daemon - on printers whose
firmware runs its own network daemon (Forge-X 1.4.2 on the AD5M today), the network page
now talks to that daemon directly instead of working around it. Joining a network,
scanning, and the Ethernet row all go through it, so the screen and the firmware no
longer fight over the radio; the address appears the moment you connect; Ethernet stops
reporting "no cable" on machines where the firmware deliberately parks the port to allow
only one transport at a time; and a wrong password reports an authentication failure
right away instead of failing as a generic timeout 45 seconds later.
Losing the daemon mid-session falls back to what the kernel still knows rather than
blanking the page.

Filament

The standalone IFS module on the AD5X - the filament system now ships as a module
in its own right in Z-Mod, reporting its state through its own status objects. HelixScreen now recognizes those directly: which lanes hold material and
what is on them, which lane is loaded and its colour, read from the module instead of
polled through the older Z-Mod-era macros - which stand down on their own once the
module is present.
QIDI's stock load and unload macros - QIDI ships M604/M603 rather than the names
HelixScreen looked for, so load and unload did nothing on a stock QIDI machine.
Cancelling a print from the runout dialog now asks first - the cancel button on the
filament-runout dialog ended the print immediately, with no confirmation.
Lanes of unknown type report their real fill (#1367) - a lane whose type could not
be determined reported empty rather than the level it actually had.
Updates
Choose Stable or Beta on any install (#1236) - the update channel was a hidden
developer setting; it is now a normal choice in Settings. Stable tracks the 1.0 line,
Beta tracks what becomes 1.1. Picking a channel that offers an older build than the one
you are running asks before it downgrades, and never notifies you unprompted.

Pairing

Denied pairing requests stop coming back (#1376) - denying a slicer or phone app's
request used to buy silence only until its next reconnect, which re-prompted
immediately, forever. A denied client is now quiet for a minute, remembered per device:
denying one phone does not hush a different one, and approving a client clears its
record.

Fixed

Performance

Idle CPU on two-core machines (#1440) - the screen kept a processor core busy while
showing nothing new: a filename too long for its space scrolled continuously, the heater
icon pulsed, and the update queue woke a thousand times a second rather than once per
display refresh. Turning animations off in Settings now genuinely stops the motion, the
queue drains on the refresh period, and a filename that is not scrolling is clipped
instead of animated. This matters most on a two-core board such as the K1 Max, where the
print host needs the other core.
Permanent memory warnings on small machines - the thresholds that decide when
HelixScreen sheds caches were scaled to installed RAM only on the roomiest machines, so
a K1-class box sat at a warning level it could never clear: its warn threshold landed on
the app's own steady-state footprint, with the clear threshold below it. One field
session warned every five minutes for over an hour with 130 MB free, and each response
freed nothing while one of them discarded a cached view the same machine had
deliberately kept. Every band now scales with the RAM actually present.

Sound

The AD5X's buzzer was silent - speaker detection looked for a beeper pin that
printer does not have; its firmware drives the buzzer through an M300 macro instead, and
that whole class of machine went unheard while BEEP kept working. Such printers are now
recognized, and the speaker on/off override in Settings finally reaches the decision:
forcing it on installs the beeper even where no signal gives it away, forcing it off
really keeps it out.

Networking

Wi-Fi scanning stopped after a backend switch (#1405) - when the Wi-Fi backend was
swapped mid-scan, the outstanding scan could never complete and periodic scanning on the
network page silently died for the rest of the session. The switch now resolves the
scan, and scanning keeps running.
A Wi-Fi scan could hang forever (#1407) - losing the connection to wpa_supplicant in
the middle of a scan left that scan permanently outstanding: nothing further scanned
until restart. Every triggered scan is now guaranteed to complete, using whatever
results the daemon still holds.
Stale rows on the network page after a transport flip (#1398, #1399) - switching
between Wi-Fi and Ethernet while the page was open left the dead transport's address on
screen next to the new one until you tapped something. Both rows now refresh on every
change, and the Wi-Fi address reads correctly on machines that booted in Ethernet mode.
A dead network daemon no longer strands the radio - a boot that ships the firmware's
network daemon but does not actually run it selected that daemon anyway, failed to start
it, and left Wi-Fi down for the whole session while wpa_supplicant sat unused. The screen
now falls back to the supplicant when the daemon proves unreachable, and stays with the
daemon when it answers - so two clients never end up on one radio.
An action a backend cannot do is no longer reported as a failure - forgetting a
network, or toggling the radio, on a backend without that call raised a red "Failed to
forget WiFi network" and snapped the switch back. Those now say plainly that the
capability is missing, and a daemon-owned machine no longer silently rewrites a stored
radio-off back to on at startup.

Input

USB mice on 64-bit machines - the input probe misread the kernel's capability format
on 64-bit systems and concluded that no mouse had a left button, so every USB mouse was
rejected, and touch devices registering through the same path were lost with them. Mice
now work on 64-bit hosts.
Touch calibration on a rotated display (#1394) - running the touch-calibration
wizard while the display is rotated stored a mapping that sent every later tap through
the wrong transform, so touches landed far from your finger. The wizard now stores that
kind of range only on unrotated displays, and a bad range stored by the previous release
is ignored rather than applied.
Touch calibration on a rotated display, properly this time (#1394) - completing the
wizard on a rotated panel left every later touch a quarter turn out, because the affine
was solved against logical targets but applied before the display rotation, and its
clamp box was the transpose of the one it was solved against. The axis-swap check made
it worse by re-solving against transposed samples and installing a matrix nothing feeds.
Both are fixed, and the wizard now re-reads the screen size each time it is shown, so a
display rotated part-way through setup no longer lays the targets out at the wrong
ratios. Unrotated panels behave exactly as before.
Touchscreens that report their type in the other bit - the probe tested the wrong
bit for a direct-input device, so some panels were classified as pointing devices.
Page arrows on resistive panels - the home screen's page arrows never appeared on
resistive touchscreens, leaving the other pages unreachable.
Crashes
Random crashes on the home panel - the widgets that make up the home panel (nozzle
temperatures, the tool switcher, thermistors) kept pointers into a screen that had
already been torn down, so a pending update arriving afterwards wrote into freed memory
and took the app down with it - a Snapmaker U1 running four toolheads in the field among
the reporters, and to the user it looks like the screen randomly restarts. The print
status screen had the same hole while a print was preparing, and it is closed with them.
A memory bug in confirmation dialogs - a confirmation could keep a borrowed pointer
to its button caption after the screen that showed it was gone, and the next dialog read
released memory: the kind of fault that surfaces as a random crash or garbled button
text. It shipped in the previous release and was caught by the nightly sanitizer run,
before any field report.
A null event descriptor in the dispatch loop (#1470) - a null descriptor reaching
the event dispatcher took the app down; it is now guarded rather than dereferenced.
Z offset
Save Z Offset stacking with Helper-Script's save-zoffset (#1401) - with the Helper
save-zoffset macro in your config, Save Z Offset folded the offset into the probe while
the macro also saved it and re-applied it at boot: the offset grew on every save, and a
reporter's K1 climbed from 0.06 mm to 2.5 mm until the nozzle pressed into the bed.
HelixScreen now recognizes the macro and steps back, as it already did for Z-Mod - the
button is hidden and nothing is folded into the probe.
The firmware persistence enable is sent once, not on every launch - the command that
tells firmware to persist the offset was re-sent at each startup rather than when the
setting changed.
A macro wrapper is not proof the offset persists - a config merely wrapping
SET_GCODE_OFFSET was read as evidence the firmware saves the offset, so the screen
claimed a persistence it did not have.
A saved offset that would not clear - the pending delta stayed on screen after a
successful save, so the adjustment looked unapplied and inviting a second one.
Print screen
The last print's result dialog closes when the next print starts - "Print complete"
or "Print failed" used to stay on screen over a newly started job, and you had to
dismiss the previous print's obituary to watch the next one begin. It still stays while
the printer sits idle, so the result can be read.
The progress bar hides during pre-print - while the printer heats and probes before
a print, the progress bar is meant to step aside (keeping its space); the styling that
did it never applied, so the bar stayed fully visible through the whole preparation. It
behaves now.
The progress arc on the print screen (#1510) - the arc bound to a value that was
not the one being displayed, so it could disagree with the percentage beside it.
Speed and flow on narrow screens - the speed/flow row was shown on both narrow
breakpoints, where it had nowhere to go and crowded the rest of the panel.
A stutter when a print starts - the active-print thumbnail was written inside the
UI update batch, so a large image stalled the frame it landed on. It is written after
the batch now.
File detail and preview
Monochrome previews on routed multi-tool prints - a routing that sent every tool of a
multi-tool file to a single lane (seen at print start on an AD5X running the IFS module)
painted the whole model in that lane's colour, trampling the file's own palette. Such a
routing is now recognized as no colour answer at all, so the slicer's palette stands -
while lanes that deliberately hold matching spools as runout backups keep their say, and
a routing that empties takes its colours back off the screen instead of leaving them
stale for the rest of the file.
Colour lists from OrcaSlicer - a file whose colours arrive as one comma-joined list
was read as a single colour and painted the whole preview in the first tool's hue. The
list now parses into the per-tool palette it is.
Tool colours with the IFS module and a mapping plugin together - on a machine running
both the standalone IFS module and a tool-mapping plugin (lessWaste, bambufy), whichever
reported last owned the mapping, so the colours could flip back to raw lane numbers
whenever a module update landed after the plugin's. The plugin's mapping now wins,
whatever the arrival order.
The render takes over from the thumbnail cleanly on the file detail view - the swap
now happens the moment the first real frame lands on the canvas. Before, the thumbnail -
mostly transparent, in OrcaSlicer's case - stacked over the viewer for the whole build,
with the render and its Building label showing through at mismatched scales, and the
finished render stood about 18% shorter than the thumbnail it replaced, so the swap
visibly jumped. The render is now framed to match the thumbnail exactly, and a colour
change arriving mid-build no longer leaves the old image up.
The filaments card on the file detail view - the card had no surface of its own, so
its header and colour chips floated loose on the panel while every neighbouring card
sits raised, and the chips were one fixed height that clipped the lane number top and
bottom on smaller screens. The card renders like its siblings now, and the chips size
to the screen - checked at every breakpoint from 480x272 to 1280x720.
Multi-colour prints framed against the prime tower - OrcaSlicer spells its purge
tower differently from the slicers already recognized, so the tower counted as part of
the model: auto-fit framed the whole preview around it and shrank the actual part to a
thumbnail. The tower is now excluded from the fit and skipped by the renderer, which also
gives the rest of the model back the drawing budget the tower was consuming.
Lane colours stayed on a file after the routing cleared - when a tool routing emptied,
the request to take those colours back off the preview was rejected by the same guard
that protects a file's own palette, so stale lane colours stayed for the rest of the
file. Retraction is now its own path.

Fans

Blank fan icons in the Fans overlay - the mark beside each fan row, and the icon in
the no-fans state, were looked up as images against what is actually an icon glyph, so
every one of them drew a blank slot. They render now.
The fan widget's rows line up - each row of the fan stack was centred on its own, so
the icons sat at slightly different positions whenever the speeds read differently (a
fan at 0% next to one at 100%). Rows now share one left edge with the speeds
right-aligned to one edge, so nothing shifts as a fan ramps.

Translations

Ten section titles, in every language - titles reaching the screen through a less
common markup path were never picked up for translation, so section headers such as
Controllable Fans, Auto Fans, the sensor groups, New Spool and Device Section stayed
English on every non-English device. All nine languages now carry them, along with the
favourite-macro dialog's questions, which had sat as empty placeholders since they were
added.
The recovery dialog was translated twice - the text shown when the printer stops
unexpectedly was run through translation two times over: it rendered correctly only by
falling back, warned on every refresh, and could show the wrong string outright in a
locale whose translation is itself a lookup. It is translated once now.
The last untranslated strings for 1.0 - the 76 strings still sitting as English
placeholders are translated in all nine languages, and the word for a filament position
is now consistent: the generic interface says "slot" everywhere, in every language,
rather than alternating with the vendor-specific terms.

Spoolman

Search in the spool picker - every keystroke rebuilt the whole spool list on the
spot; with a few hundred spools in Spoolman the first tap stalled the screen for a
noticeable beat on a desktop and for seconds on a K2, and the keyboard froze with it.
Typing now re-renders once, after a short pause; clearing the search applies
immediately; a search with no matches says so with its own card instead of flashing a
load-error retry; and the loading spinner sits centred.

Printers

A Qidi Q2 identified as a Plus 4 (#1431) - detection named a printer it could not
actually separate from a close rival, because two models scored the same on a
range-wide macro. It now declines to name a machine unless the winner leads every
candidate naming a different one, and the Qidi entries no longer claim a shared macro
identifies a single model.
A K1 no longer advertises hardware it does not have - the K1 preset seeded
hardware entries the machine does not ship, which then showed up as controls for
equipment that is not there.
Bed mesh calibration runs the sequence the printer ships - the mesh calibration ran
a generic sequence rather than the one the printer's own configuration defines.
Filament
A second, wrong error after a failed CFS load - when the homing move before a load or
unload failed, the real error arrived with a spurious "Unknown g-code state" stacked on
top of it. The cleanup step ran even though the operation it was meant to undo had never
started, and Klipper rejected it. Only the actual failure is reported now.
The runout dialog appearing when you pull filament out yourself - on a printer with a
plain filament sensor and no filament system, taking filament out while the printer was
idle raised the "Filament Runout" dialog, reporting a fault for something you had just
done with your own hands. Nothing is being extruded while the printer sits idle, so
filament leaving the sensor in that state is someone standing at the machine; the screen
now just shows the filament as empty and stays out of the way. A sensor already reading
empty when the screen arrives still offers to load, which is the case worth asking about.

Home screen

The home screen rearranging itself when Klipper drops out - a layout worked out while
the printer was down got written back to disk as though you had arranged it that way, so
widgets you had placed yourself reverted to an automatic arrangement after a power cycle
or a firmware restart. Your positions are now only saved once the printer is actually
reporting ready, so a fault no longer rewrites the layout underneath you.

Setup

The printer-mismatch button says what it opens - the button offered when the detected
printer disagrees with the saved one was labelled for something other than the screen it
actually opens.

Installing and updating

The Snapmaker U1 ships with its crash supervisor again - the U1 was the only
platform whose release never packaged helix-watchdog, so a crash left the screen
dead with nothing to restart it. That matters most on this machine, because the
installer also stands the stock touchscreen UI down.
A stock UI that reports itself missing - the installer now warns when a device known
to ship a vendor UI reports none, rather than continuing quietly and leaving two screens
competing for the display.
QIDI's stock screen across firmware generations - the installer stopped only the
screen process one firmware generation used; it now stops it on every generation.

Files and USB

Every mounted USB drive is scanned (#1373) - only the first mounted drive was
searched for printable files, so a second stick appeared empty.

Job queue

A Print-now tap while the printer is busy (#1373) - the tap did nothing and said
nothing; it now answers.

Settings

The log level shown is the one the sinks run at - the settings screen displayed the
ring buffer's floor rather than the level actually being written, so it could claim a
verbosity that was not in effect.

Telemetry

The send window no longer depends on the clock's direction (#1476) - the periodic
send gate was decided from a wall-clock comparison, so a clock stepping backwards (NTP
settling at boot is the usual cause) could hold sends off indefinitely. It is decided
from elapsed time now, and the periodic producers are sized to fit inside one window.

Changed

The temporary Firmware Restart tile is gone from the home screen - it was added to the
grid whenever Klipper was down, but placed automatically after every widget with a saved
position, so on a full home screen it was silently dropped and never appeared at all. The
recovery dialog that opens on a Klipper shutdown carries the restart action, and Firmware
Restart is still in the widget catalog if you want a permanent button for it.
The AD5X filament path drawing - the picture showed a machine the AD5X is not: a
merge unit mid-machine, or a selector with a single output line. The IFS is really four
lanes into a camshaft selector sitting under the spools, with the combiner hub on the
toolhead itself - that is what the canvas draws now, one continuous tube from the hub
down to the nozzle, and no bypass stub, because the machine has none. The toolhead
style picker in Settings also offers all eight styles instead of five.

Known Issues

Changing the UI scale rearranges the home screen - the widgets you placed on the
home screen are laid out again for the new scale, and your previous arrangement is not
kept. Setting the scale back to what it was does not restore it either. A saved widget
position is a coordinate in grid tracks, and the UI scale changes the grid those tracks
are counted against, so the positions are re-seated for the new grid rather than parked
and restored. If you have a home screen arranged the way you want it, pick the UI scale
first and arrange the widgets afterwards. (#1484)