prestonbrown/helixscreen v0.99.114

on GitHub

v0.99.114

This is the 1.0 release candidate. It is the code intended to ship as 1.0.0, published as
an ordinary 0.99.x update so that everyone gets it rather than only people who opted into a
beta channel. If it holds up in the field, 1.0.0 follows with no further changes. If it does
not, the fix ships as 0.99.115 and no bad 1.0.0 ever exists.

This started as a sweep through every issue still open against 1.0 and turned into the largest
release the project has shipped. Two fixes stand out because both presented the same way, as
"the printer stopped responding", with nothing in the log to explain it. Waking the display
could park the UI thread on a lock and leave a lit screen ignoring every touch, in one case for
36 minutes. And a Klipper shutdown could be overwritten by a stale snapshot replayed at the end
of discovery, so the app showed Ready for two and a half minutes while the printer was down,
re-enabled navigation and kept sending commands to a dead printer.

Several other defects had gone unnoticed because the wrong thing was being measured: the
Lifetime Print Stats tile quietly reported only your most recent 500 jobs, accelerometers were
never discovered at all so Settings > Sensors was empty on every printer that has one, and two
printers were missing the database entry that applies their preset.

Added

• A new theme, Hazard. ANSI Z535 safety colours on powder-coated black: caution yellow,
  orange for warning, red for danger, with sharp corners and brass seams. Dark-only.
• HelixScreen now detects and reports a hung UI main loop, so a freeze leaves evidence behind
  instead of looking like a dead process. Detection only, nothing is killed. The threshold
  defaults to 60 seconds and is settable at runtime or with HELIX_HANG_THRESHOLD_SEC.
• When updates cannot be installed, the notice is now a row you can tap. It explains how to
  update from a terminal and shows a QR code to the documentation.
• The Android app is documented in the README and the install guide: which of the three APKs to
  take, why the .aab is not for users, sideloading, permissions, and the uninstall needed to
  move to a future Play Store build.
• helix-screen ctl can address unnamed widgets and set label text directly, which makes far
  more of the interface reachable for scripted testing and screenshots.

Fixed

Freezes and connection state

• A lit screen that ignored every touch. Waking the display forced a WebSocket reconnect on
  the UI thread, and that reconnect waits for in-flight callbacks with no time bound, so one
  undrained callback parked the main loop permanently. Seen on Centauri Carbon and AD5M and
  confirmed on two wedged devices. The reconnect is now Android-only, where it is actually
  needed, and the wait is bounded
  (#1245).
• A Klipper shutdown could be displayed as Ready. Klippy state arrives from two unordered
  sources, and the snapshot replayed at the end of discovery could put Ready back over a live
  shutdown. That re-enabled the nav buttons, auto-dismissed the recovery dialog moments after it
  correctly appeared, and let commands through to a dead printer. Live frames now win over
  replayed ones. Relatedly, an unrecognised printer state no longer falls back to Ready.
• Jogs refused as "Printer is busy" on an idle printer. Klipper reports the idle timeout as
  Printing while any gcode runs at all, so housekeeping loops such as bed fan timers, air filter
  timers and AFC's own prep made roughly 7% of jog attempts bounce. HelixScreen now waits for a
  full second of continuous activity before it treats the printer as busy.
• The change-Moonraker-address prompt landed on top of the setup wizard. On a fresh install
  it appeared about a minute in, over the wizard's own Connection step, giving two competing
  host-entry dialogs.

Creality CFS (#968, #1278)

• CFS operations now verify that filament actually moved. A load or unload the firmware
  reports as finished is checked against the toolhead sensor, and a mismatch raises a fault with
  its own recovery action rather than being reported as success.
• The part-cooling fan no longer blows on the nozzle during every K1 load, cut and flush, and
  two runout messages that previously paused the print with no dialog at all now show one. Both
  came from commands wrongly believed absent on K1; the K1 firmware image proves otherwise.
• CFS error recovery works on K1. It was sending a resume command that firmware does not
  register, so the box never resumed.
• Filament changes no longer pass a parameter the firmware does not accept, so the purge runs at
  your configured length instead of silently falling back to defaults.
• The filament code catalog is cached rather than re-parsed on every box update, which was a
  100 KB parse every time a spool moved.

AFC and BoxTurtle

• Buffer health is attributed to the right unit and stays there. On multi-unit rigs it landed
  on unit 0 for every unit, then went blank for minutes until the firmware happened to push a
  change. Found on a five-unit rig.
• A Snapmaker U1 with five units drew six toolheads for four extruders, all claiming T0. AFC
  publishes free-form config section names, and six places parsed them with a grammar that only
  fits Klipper's own. They now resolve names the way Klipper does.
• Spool names and vendors resolve from AFC's lane data using the key names AFC actually
  publishes.
• Spoolman users are no longer told to upgrade AFC to get filament names. Names already
  resolve through Spoolman; multi-colour spools are the part that needs the newer AFC, and the
  notice now says so.
• A spurious warning told users to add an extruder_name setting AFC refuses to start
  without. It fired because the config query lands just after the first status frames.
• AFC lanes follow the firmware's own current_map rather than guessing the lowest tool number,
  so multi-tool lanes drive the tool the firmware actually selected.
• Spool assignments are no longer rewritten on weight noise. A lane reporting weight as a float
  that drifts by hundredths caused 590 rewrites of the spool file, 590 database writes and 590
  filament-panel rebuilds in a single session.
• Config writes no longer trigger a full file-list reload. An AFC rig caused 113 complete
  directory round trips in one session while the user sat on the print status screen.

Tool and nozzle counts

• A single-hotend printer behind a 4-lane AMS no longer reports four tools. The nozzle label,
  the print-status tool badge, its chevron and the nozzle picker all counted lanes rather than
  hotends, and the picker went on to offer nozzles the printer does not have.
• Preheating "All" on such a printer sent the same target four times and the confirmation
  claimed it had heated four tools.
• A 4-port AD5X no longer advertises a 16-tool machine. It was reporting every addressable
  tool number the firmware register can hold rather than the ones actually mapped.

AMS and filament

• PLA+, PA6-CF, PETG-CF and every other punctuated material name were silently dropped
  while the save reported success. Our own filament database ships PLA+. Names containing a
  space, such as Silk PLA, never worked either.
• A fifth AMS unit produced parser warnings and a permanently dark temperature and humidity
  badge. The unit limit is raised to eight, and the environment overlay stops showing unit 0's
  humidity for every unit.
• Spools with no usable weight are drawn full rather than half full, which had read as half
  gone. Four places in the interface had drifted to different answers and now agree.
• A crash when opening the AMS context menu, and a deadlock risk on Snapmaker hardware, are both
  closed.

Screens, dialogs and widgets

• On small displays a modal's button row could fall off the bottom of the card, leaving a
  dialog that could not be dismissed. Every breakpoint's limit is now measured from the real
  modal on screen and re-verified from 480x272 up to 2560x1440
  (#1277).
• A home-screen widget that would not fit announced itself as "removed" on every launch, and
  a widget that was enabled but held no cell showed in the catalog as placed, greyed out and
  impossible to select or remove.
• Widget content no longer clips or wraps one letter per line on larger displays. The size
  bands were a single set of pixel values calibrated for small screens.
• Row 2 of Calibration & Tools no longer runs its labels together when it carries four
  conditional buttons.
• Toggling the printer switcher in the navbar takes effect immediately instead of waiting until
  you leave the overlay.
• A crash in grid edit mode when deleting a widget mid-resize is fixed.
• The print status header no longer shows an empty action button during a print.
• Page-scroll chevrons on the home screen stop landing on a widget's own content.
• Several dialogs that could grow past the screen are capped and scroll properly: AMS loading
  errors, the shutdown and recovery dialogs, and action prompts. The AFC fault diagram no longer
  sits inside the scrolling text area.

Z-offset

• The Z-Offset row showed 0.000 whenever an AD5M or AD5X was idle, and adjusting it while idle
  could throw away your real offset. ZMOD keeps the authoritative offset in its own storage and
  clears Klipper's live one when a print ends, so an idle nudge was applied to a phantom zero and
  that is what got saved. HelixScreen now reads the stored value, shows it while idle, and sends
  an absolute adjustment when the live offset is not the real base.
• A printer whose offset sits past 2mm is no longer yanked down to 2.000 on the first tap in
  the tune overlay, which drove the nozzle into the print. The limit now applies to how far one
  tuning session can travel, not to the offset itself
  (#1280).

Print history and status

• The Lifetime Print Stats tile now reports true lifetime totals from Moonraker instead of
  summing the most recent 500 jobs. If your totals disagreed with Mainsail, this was why
  (#1272).
• The idle "Reprint Last" tile no longer offers a file you deleted.
• A prompt button whose macro fails now reports Klipper's own error message instead of closing
  the dialog and leaving an empty screen.

Hardware, printers and presets

• Accelerometers are discovered again. Settings > Sensors listed nothing and showed a count of
  zero on every printer with an ADXL345 or LIS2DW
  (#1262).
• The Elegoo Centauri Carbon and Artillery M1 Pro now pick up their presets. Without the
  database link, neither printer received any of its preset settings unless it was installed
  from the factory image (#1260).
• Touchscreens that report only multi-touch axes are calibrated correctly. Every coordinate
  past the screen width had been saturating at the edge
  (#1259).
• An AD5X slot with a type but no colour no longer aborts the Change Type dialog with nothing
  left to reopen, and the phase display stops inventing a 230C target the printer never had.
• Screen rotation is now a setting in Display & Sound rather than something only reachable by
  editing config, and the rotation probe no longer misses every tap.
• A nozzle sitting just above its target no longer shows as heating while calling itself Ready.
  The temperature shown and the state described are now derived from the same value.
• Large g-code files are handled correctly on 32-bit builds, and the DRM backend keeps its
  64-bit buffer offset there, fixing display init on 32-bit Pi images.
• A printer preset no longer applies its panel's display and touch settings to a different
  machine's screen when HelixScreen is talking to that printer over the network.
• A fresh install no longer loses its shipped platform preset when an earlier printer_data
  directory outlives the install.

Spoolman setup

• Spoolman setup no longer dead-ends on a stock Creality K2, where Moonraker's config lives
  outside the only writable file root.
• The AD5M's config is no longer rejected as unwritable when it is in fact writable through a
  linked path.
• Setup no longer claims Moonraker reads its config from somewhere else after any config edit
  since Moonraker last restarted, and a guessed config path must now match exactly before
  anything is written to it.

Updates and installation

• On a standalone-display install the entire Software Updates section vanished, and the fix
  could only reach users inside the update they were being kept from. HelixScreen now recognises
  the second update route the installer supports, and checking for updates is separated from
  installing them so a machine that cannot install can still tell you an update exists.
• The installer's own advice could not be followed. Nine messages told users to repair an
  install with a command that needs bash, absent on the BusyBox firmwares, and that would have
  performed a fresh install rather than an update.
• Upgrading from the CLI failed with a permission error even after the installer reported
  sudo was available, because it asked whether it could write into a directory rather than
  whether it could rename it.
• Unprivileged Raspberry Pi installs no longer land under a root-owned parent, which left
  only the destructive fallback path (#970).
• Three permanent PolKit warnings in Mainsail and Fluidd are gone on Centauri Carbon and K2
  Plus, and a Moonraker restart now works on firmware that uses neither systemd nor the init
  script we knew about.
• Unlocking beta features, selecting the Dev update channel, then locking beta again no longer
  leaves the app fetching from Dev with no way back to Stable.

Sound

• A printer with sounds turned off no longer holds the audio device open for the whole
  session, writing silence every period. On Android that also kept an AudioTrack open
  (#1253).
• The PWM buzzer backend no longer probes for a beeper on boards that have none, and the AD5M
  stops shipping 919 KB of music it cannot play. M300 still works everywhere.

Android

• The 1.0 release would have been refused as a downgrade on every existing Android install
  and rejected by the Play Store, because of how the version code was packed.

Translations

• All nine languages are now fully translated, up from 99.7% in the eight non-English ones.
  The runtime CJK fonts are rebuilt too: eight characters the new Japanese and Chinese text needs
  were in no font file, and a missing character draws as an empty box with no warning anywhere.
• Translations resolve all C escape sequences in keys, not just hex ones, so a further batch of
  text stops falling back to English.

Changed

• Memory use is down across the board, which matters most on the Centauri Carbon and other
  boards that swap at idle. About 1.2 MB of address space returned by capping allocator arenas,
  1 MB off the CC1 binary by dropping mocks and asserts from shipped builds, 345 KB of fonts that
  the per-platform tiers had never actually pruned, and 615 KB off the shipped layout files, of
  which roughly 310 KB stays resident for the whole session. The print history, the filament
  tables, the keypad and the tips database all stop holding memory they no longer need.
• Debug bundles now include the startup logs. Everything config-related happened before the
  logger was installed, so a bundle could show a "settings were corrupted" message with not one
  matching line anywhere in it.

Internal

• The sanitizer CI gates were reporting green while failing. A run with ten real
  AddressSanitizer reports was reported as passing, because the output was piped without
  pipefail and a success banner printed unconditionally. Fixed, along with the three standing
  findings the masked gate had been swallowing.
• The XML formatter's --check claimed files were clean that it had never managed to read,
  and its errors were being sent to /dev/null. Three layouts had been unreadable to it.
• Two use-after-frees closed (the AMS context menu's subjects outliving their storage, and a grid
  resize animation keyed on the wrong object), a latent out-of-bounds write removed, and a
  lock-order inversion between the Snapmaker backend and AMS state fixed.
• The Android release-signing gate was vacuous and would have passed on an artifact whose
  signature was never read. This is how v0.99.43 shipped a debug-signed bundle.
• The Play Store "What's New" text was being chopped mid-clause on every release, and its
  last two entries silently swapped. A release section can now carry an explicit block used
  verbatim, with a hard error if it exceeds the limit.
• Three Cloudflare Worker test suites had never run in CI. They do now, along with the analytics
  dashboard build.
• The Build workflow had been red for a day because the Moonraker plugin test dependency was
  never declared. Declared, with a gate so it cannot recur silently.
• Emergency-stop overlay observers now carry a lifetime token, closing a use-after-free that
  reproduces deterministically when the guard is removed.
• The install base is now estimated from CDN update polls rather than opt-in telemetry alone,
  which puts the real fleet at roughly double what had been quoted.