Skip to content

v0.3.1 — tablet mode folds both ways

Choose a tag to compare

@Brunux Brunux released this 22 Oct 18:51
· 294 commits to main since this release

v0.3.1 — tablet mode folds both ways

A maintenance release. The boot chain, the kernel and the compositor are the ones
v0.3.0 shipped; what changed is the session, and the largest change is
the one the 2026-10-22 card found: tablet mode could be entered by folding the
keyboard back and not left by folding it forward again.

What changed since v0.3.0

  • Folding the keyboard back exits tablet mode again. cros_ec_buttons reports
    the power and volume keys and the tablet switch on one libinput device, and
    wlroots builds both halves over the same handle - so input type:keyboard events disabled suspended the switch with the keys, and tablet:off never arrived.
    The mode now switches the typing keyboards off one identifier at a time
    (cros_ec and keyd's virtual keyboard) and leaves every identifier a switch
    also carries alone, which is why the volume keys keep working in tablet mode and
    the mode can always be left. check-system.sh FAILs on a type-wide disable and
    on a suspended switch, and the troubleshooting page has the folding back does
    nothing
    case with the way out. The on-screen keyboard is opaque now: with
    transparency the window underneath showed through the keys.
  • The desktop wears the mountain photograph again. v0.3.0 drew the mark as the
    wallpaper, and a wallpaper is a surface to live on rather than a logo. The mark
    stays as the launcher icon, the scalable SVG and the About card; the lock screen
    keeps the same photograph, which is what is up when nobody is looking.
  • Format a disk writes exFAT (mkfs.exfat -F -L LAZYOS) instead of FAT32,
    which refuses a file over 4 GB, and the candidate list no longer depends on the
    ChromeOS kernel partition alone: ui_disk_system refuses every disk that is
    built into the machine (the device-level removable attribute, TRAN=usb) or
    that the running system depends on (root, /boot, the partition named by
    kern_guid=, every active swap, each resolved with lsblk -s). Both Format and
    Auxiliary swap ask it again after the pick, because that step repartitions.
  • Alt+Search answers in either press order: keyd maps the Search key to F20
    under [alt] and sway binds that on the press, while a Search-then-Alt release
    falls through to the same window list. A search that cannot start now says why
    and waits for Enter instead of closing the window, which is what "the key does
    nothing" looked like.
  • The emoji picker only offers what the font draws. fonts-noto-color-emoji
    omits the text symbols inside the emoji blocks (chess pieces, arrows, alchemical
    signs, several ballot boxes), so 692 of 1988 rows inserted a tofu box;
    scripts/make-emojis.sh filters the table against the character map fontconfig
    resolves for emoji, and ~/.config/emojis.font pins the font so an upgrade is
    reported instead of leaving the table stale. 1296 rows remain.
  • The night light survives a suspend: the panel comes back with its own gamma
    table, so after-resume restarts the schedule and does nothing when the light
    was off. One helper now owns the command line and the mode, so the login line,
    the Settings switch and the resume hook cannot disagree. The idle screen's
    pkill is bracketed ([o]rg...) after the unbracketed one killed the shell
    that ran it before the restart.
  • The first boot opens one welcome page, once per installation rather than once
    per login: eight keys, the marker holds /etc/machine-id, and it is written only
    after the page closes, so a foot that failed tries again next login.
  • Tablet mode's off restores the rotation and both inputs with no state
    directory left
    , kills a leftover on-screen keyboard by name, and stops its
    poller first so nothing turns the screen back after it has been put upright.
  • The file browser remembers the last directory (one line in
    ~/.local/state/lazyos/files-last, --search excepted), and imv's keys are
    said once per session the first time a picture opens.
  • A match found with Alt+Ctrl+G opens at its line: bat | less +N answered
    Cannot seek to line number N because less cannot seek a pipe, so bat writes
    the page to a real file and less opens that file, which is removed when the
    window closes.
  • Notification cards carry a picture: the file browser's eight messages were
    the last notify-send -a APP calls without an icon, and check-system.sh warns
    when a new one appears.
  • Small repairs from the card's list: the About windows name Alt+Q (foot's
    Ctrl+Shift+W does not reach them on this keyboard), the root menu has a
    launcher entry (org.lazyos.menu.desktop), eject is installed so pcmanfm's
    Unmount works, the reminder prompts explain their formats and keep the title and
    note while asking, and zoxide is initialised after starship so cd - works.
  • The agents menu carries the session's glyphs on its rows (agent, accounts,
    install, default, terminal, remove, back) and a display/action table behind them,
    because rofi hands back the rendered row and markup cannot be parsed back.

Validation

The tablet failure was diagnosed from the upstream sources rather than guessed:
wlroots creates one wlr_input_device per capability over a single libinput
handle, sway's input ... events is
libinput_device_config_send_events_set_mode(), and libinput's
evdev_device_suspend() closes that handle when the mode is disabled - which
is why the switch went quiet with the keyboard. The script was then exercised
against a stub of swaymsg, jq and wvkbd-mobintl with the identifiers this
unit reports (0:0:cros_ec, 0:0:cros_ec_buttons, keyd's virtual keyboard and
the headset jack): on disables the two typing keyboards and touches neither
cros_ec_buttons nor the jack, off restores them, and repeated runs leave
nothing behind. The new check-system.sh checks were run against both the fixed
script and the old one.

The hardware run that signs this release off is
docs/test-checklist.md § Tablet mode on this card: fold
back out of the mode, the volume keys, the Lock key, the on-screen keyboard and
the rotation. The rest of the batch was verified on the build host and against the
card's report in reports/.

Assets

asset what it is
google-asurada.img.xz 6 GiB disk image: the three ChromeOS GPT partitions (kernel, /boot, root), ready for dd
google-asurada-rootfs.tar.zst the same root filesystem as a tarball (with /boot inside), for writing your own card with scripts/write-sd.sh
google-asurada.vmlinuz.kpart the signed FIT kernel partition of that image; it embeds the image's root=UUID, so it belongs to this image and no other
SHA256SUMS SHA-256 of the three assets
sha256sum -c SHA256SUMS
xz -dc google-asurada.img.xz | sudo dd of=/dev/mmcblk1 bs=4M conv=fsync

On the Chromebook: developer mode, crossystem dev_boot_usb=1 dev_boot_signed_only=0, insert the
card and pick external disk in the firmware menu. It comes up to a greetd prompt — user
brunux, password debian, passwordless sudo. Change the password before using it for
anything real.
scripts/install-emmc.sh installs it to internal storage.

What it does not carry

The artifact is sanitized and asserted clean by scripts/lib-dist.sh:
no Wi-Fi profiles (a released image knows no network and asks for one), no SSH host keys and no
machine-id (both generated on first boot, so every installation has its own identity), and no
personal files beyond the session configuration this repository ships.

Known limitations

  • The image is 6 GiB, and that is a fix in itself. The v0.3.0 artifact was 4 GiB and its root
    partition came out with 32 MB free (df said 100%): enough to boot and not enough to write a log,
    a screenshot or a browser cache. IMAGE_SIZE_MIB=6144 leaves about 2 GiB, and a larger card
    still keeps the space after p3 unused; write the tarball with scripts/write-sd.sh to size
    p3 to the whole card.
  • The tarball's /etc/fstab is the unconfigured one the build leaves behind: write-sd.sh
    generates the real one.
  • Artifacts are not byte-reproducible yet (FIT timestamp, filesystem UUIDs).
  • The battery gauge's chemistry and time-to-empty properties are upstream failures; waybar reads
    the capacity directly, which works.
  • There is no hardware video decoder, and Vulkan is not usable (PanVK answers Mali-G57 not
    supported
    ); the session renders through EGL and Panfrost, and video is CPU work.

Upstream

Vendored from pmaports commit 0e458a2417d6f5ed61c7b70621d5632e30136091 (2026-10-15), mainline
Linux 7.2.7, swayfx 0.6 on scenefx 0.5.