Skip to content

build-rootfs: [omarchy] stable aarch64 aquamarine 0.15.1 (so=14) breaks hyprtoolkit from ALARM extra (needs so=15) #433

Open

Description

@matthewjaybarr-png

tools/build-rootfs currently fails for every device on the default OMARCHY_CHANNEL=stable. Found while building the enchilada image for #432; nothing about it is device-specific.

What fails

==> Building omamobile-rootfs-enchilada-2026.11.04-aarch64 for enchilada (kernel 7.1.0-rc1-sdm845, [omamobile] channel local)
error: not in any repository: omarchy

omarchy is in the repository. It's its dependency chain that can't be resolved. pacman -Sp omarchy in the build image:

:: unable to satisfy dependency 'libaquamarine.so=15-64' required by hyprtoolkit
:: unable to satisfy dependency 'hyprtoolkit' required by xdg-desktop-portal-hyprland
:: unable to satisfy dependency 'xdg-desktop-portal-hyprland' required by omarchy
:: unable to satisfy dependency 'hyprtoolkit' required by hyprland-guiutils
:: unable to satisfy dependency 'hyprland-guiutils' required by hyprland
:: unable to satisfy dependency 'hyprland' required by omarchy

Why

[omarchy] comes before [extra] in pacman.conf, so its aquamarine hides ALARM's:

repo aquamarine provides
pkgs.omarchy.org/stable/aarch64 0.15.1-1.3 libaquamarine.so=14-64
pkgs.omarchy.org/edge/aarch64 0.16.0-1.1 libaquamarine.so=15-64
ALARM extra 0.16.0-1 libaquamarine.so=15-64

ALARM extra's hyprtoolkit 0.6.0-3 depends on libaquamarine.so=15-64, which stable's aquamarine doesn't provide, and stable has no hyprtoolkit of its own. So [omarchy] stable aarch64 is one aquamarine rebuild behind ALARM. (The stable db's Last-Modified: Thu, 05 Nov 2026 00:44:09 GMT.)

Workaround (verified at the pacman level)

OMARCHY_CHANNEL=edge tools/build-rootfs …. Edge carries the same omarchy 4.5.0-2, plus aquamarine 0.16.0, and pacman -Sp omarchy resolves there. I haven't finished a full image build on edge yet. Note that edge then also becomes the phone's [omarchy] channel (via rootfs/pacman.conf).

Two small things on our side, whatever happens upstream of us

  1. rootfs/build-inside:44: available() treats any pacman -Sp failure as "not in any repository", which pointed me the wrong way. Printing pacman's own unable to satisfy dependency lines (or wording it "cannot be installed") would have saved the hunt. Happy to send a PR if you want one.
  2. Whoever publishes [omarchy]'s aarch64 stable repo may want to know. I don't know where that's tracked, so I'm reporting it here.

Host: Omarchy (x86_64), Docker 29.9.2 + qemu-user binfmt, repo at d1fb7b6.

Activity

  1. matthewjaybarr-png commented on Nov 5, 2026

    @matthewjaybarr-png
    ContributorAuthor

    Follow-up: switching to OMARCHY_CHANNEL=edge gets past the aquamarine problem but hits a second skew one step later, in the [omamobile] pacstrap stage:

    warning: downgrading package hyprland (0.57.1-2 => 0.57.0-3.1)
    error: failed to prepare transaction (could not satisfy dependencies)
    :: installing hyprland (0.57.0-3.1) breaks dependency 'hyprland=0.57.1' required by omarchy-hyprland-titlebars
    
    • edge [omarchy] aarch64: omarchy-hyprland-titlebars 0.3.0-4 depends on hyprland=0.57.1 (stable's 0.2.2-2 pins hyprland=0.56.2)
    • published [omamobile]: hyprland 0.57.0-3.1, which build-inside installs over edge's 0.57.1-2

    So right now neither channel builds a rootfs as-is: stable on aquamarine/hyprtoolkit, edge on titlebars/hyprland. Locally I'm working around it by adding omarchy-hyprland-titlebars to rootfs/packages.omarchy-skip (nothing in [omarchy] or our packages depends on it). The real fix is probably bumping packages/hyprland to 0.57.1, or whichever version stable [omarchy] lands on once aarch64 catches up.

  2. derekross commented on Nov 5, 2026

    @derekross
    Owner

    Thanks, this was a great report, and the follow-up saved me a step. I checked it against today's databases and it holds, with one addition.

    Stable: [omarchy] stable aarch64 picked up aquamarine 0.15.1-1.3 (libaquamarine.so=14-64) with the 4.5.0 promotion. Our 2026.11.03 image, built from stable on 11-03, still got aquamarine 0.16.0 from extra. Stable's hyprland 0.56.2-7 needs libhyprgraphics.so=4-64, which no longer exists on aarch64, so stable's own Hyprland stack can't be installed there either. Edge has had 0.16.0-1.1 since 11-02. That's Omarchy's to fix, and we'll report it to their package repository. Installed phones aren't affected: pacman -Syu keeps the newer aquamarine they already have.

    Titlebars: omarchy-hyprland-titlebars pins an exact Hyprland version, and Omarchy 4.5.0 also installs it from a migration (1791403310). On an installed phone today that migration fails and stops omarchy update partway. Worse, once Omarchy's Hyprland and ours differ in the other direction, it would succeed by replacing our Hyprland and its DPMS fix. So we're doing two things: titlebars goes into rootfs/packages.omarchy-skip (#442; your workaround is the right fix, since a phone has no floating workspaces to put titlebars on), and omamobile-update will hold that migration, in the omamobile-omarchy update that ports our patch series to 4.5.0 (#434).

    We won't make edge the default: omamobile-update only updates against Omarchy's stable channel, so an edge image would refuse every update. Until Omarchy's stable aarch64 catches up, an edge build is fine for bring-up, but not for an image you hand to someone.

    Yes please on the build-inside PR. Please keep "no such package" separate from "its dependencies can't be met", and print pacman's reasons. The same check also files a package in Omarchy's list that can't be installed under "not built for aarch64" in the MANIFEST, so please fix that path too.

    Hyprland 0.57.1 is coming as well (our patches apply unchanged), as a normal phone-tested bump.

  3. nesmeck commented on Nov 5, 2026

    @nesmeck
    Contributor

    @derekross, one data point on "installed phones aren't affected": a phone installed before the 2026.11.03 image can still have the old aquamarine.

    My Pixel 7 was set up on 2026-10-27, and it has aquamarine 0.15.1-1 installed, which provides libaquamarine.so=14-64:

    • /var/lib/pacman/local/aquamarine-0.15.1-1/desc: %PROVIDES% libaquamarine.so=14-64
    • also installed: hyprland 0.56.2-5.1, hyprtoolkit 0.6.0-1, hyprgraphics 0.5.1-4

    So pacman -Syu has no newer aquamarine to keep. It would take [omarchy] stable's aquamarine 0.15.1-1.3 (still so=14), while the update brings hyprtoolkit 0.6.0-3 from ALARM extra and hyprland 0.57.0-3.1 from [omamobile] development, and both need libaquamarine.so=15-64. I read that from today's databases (pkgs.omarchy.org/stable/aarch64, pkgs.omamobile.org/development/aarch64, ALARM extra). I haven't run the update, so I expect it to stop at dependency resolution, as it did for @Somnius on the 8 Pro. That part is a prediction, not something I ran.

    omamobile-update check (omamobile-omarchy 0.1.12 on this phone) reports ok with 86 updates, so it doesn't catch this. As far as I can tell it lists versions but doesn't resolve the transaction. If it ran something like pacman -Sup against its temporary databases, a phone like this one would get a clear refusal before the update starts.

    I'm holding this phone's update until #434 and Omarchy's stable aarch64 aquamarine land. I'm happy to run pacman -Sup on it against the temporary databases and post the exact output, if that's useful.

    🤖 Generated with Claude Code

  4. gitpash commented on Nov 6, 2026

    @gitpash

    Confirmed on a real phone (Pixel 7 Pro, stable, omarchy 4.5.0-3, hyprland 0.57.0-3.1): the update stops at the titlebars migration.

    Running migration (1791403310)
    Install titlebars for floating workspaces
    warning: cannot resolve "libhyprgraphics.so=4-64", a dependency of "hyprland"
    warning: cannot resolve "hyprland=0.56.2", a dependency of "omarchy-hyprland-titlebars"
    :: The following package cannot be upgraded due to unresolvable dependencies: omarchy-hyprland-titlebars
    :: Do you want to skip the above package for this upgrade? [y/N]
    error: failed to prepare transaction (could not satisfy dependencies)
    Something went wrong during the update!
    

    The update ends with exit 1 (nothing after this migration runs). The phone UI stays on "starting" because the sheet keeps tailing the events file after the run ended. Nothing was broken by it (hyprland and the shell keep running); it just never reaches the later migrations. Glad to know this one is held in the same update as 0.1.34, I'll test it on the 7 Pro when it's up.

  5. derekross commented on Nov 7, 2026

    @derekross
    Owner

    @nesmeck, you're right, and thank you for checking it against the databases. I ran it through pacman in your phone's repository order against today's databases: with aquamarine 0.15.1-1 installed, the update fails with "unable to satisfy dependency 'libaquamarine.so=15-64'" (hyprtoolkit from ALARM, our Hyprland 0.57). With 0.16.0 installed it resolves. So "installed phones aren't affected" was only true for phones installed from the 11-03 image on, and I'm sorry for saying otherwise. Omarchy's stable channel still has 0.15.1-1.3 today. I'll put aquamarine 0.16.0 into our own repository, the same way we now carry Omarchy 4.0.6, so phones like yours can update again without waiting for Omarchy. I'll post here when it's published.

    You're also right about omamobile-update check. It does ask pacman for the plan, but it ignores a failed plan and falls back to the plain list, so it said ok. It should refuse and show pacman's reason. I'll fix that in omamobile-omarchy (a small separate change).

    @gitpash, thanks for confirming the titlebars stop on a real phone. That migration isn't held yet. I've asked for the hold in the PRs that carry the 4.5 work (#469, #471), and phones stay on Omarchy 4.0.6 until it's in. Please leave your phone's update where it stopped rather than skipping the migration by hand (more on #434). The sheet staying on "starting" after the run ended sounds like a separate bug. Could you open an issue with the output of omamobile-update status --json and the last lines of $XDG_RUNTIME_DIR/omamobile-update/events from that run, if they're still there?

  6. derekross commented on Nov 7, 2026

    @derekross
    Owner

    Both fixes from my last comment are in:

  7. Somnius commented on Nov 9, 2026

    @Somnius
    Contributor

    @derekross a data point for this issue from my Pixel 8 Pro, which was installed on 2026-11-01 and was following the local channel (a local repo filled from my build machine), so it had hyprland 0.56.2-5.1 and aquamarine 0.15.1-1.

    Before the update omamobile-update check --json (omamobile-omarchy 0.1.31 on the phone) said ok: true, 78 updates, Omarchy 4.0.6 to 4.5.0-3, no boot packages, while pacman -Su --print failed exactly as @nesmeck predicted: unable to satisfy dependency 'libaquamarine.so=15-64' required by hyprtoolkit. So the phone's check still gave the false ok that #478 fixes.

    What worked: omamobile-update channel stable, which made the plan clean (90 updates, boot_packages: [], kernel: null, no Omarchy bump, as far as I can tell because [omamobile] stable carries 4.0.6 and comes first). I checked that plan against a temporary database first, then tapped Update on the phone's sheet and entered the PIN. omamobile-update status ended with exit 0, "Updated", nothing held, omarchy 4.0.6-1 before and after, and one restart reason: "Hyprland was updated". The result was hyprland 0.57.0-3.2, aquamarine 0.16.0-1, omamobile-omarchy 0.1.37, omamobile-shell 0.2.107 and omamobile-device-husky 0.1.8. I restarted the session with systemctl restart sddm instead of rebooting, and Hyprland 0.57.0 came up with the panel at 1344x2992@120. tools/phone-smoke later that evening: all required checks ok.

    One thing broke, and I don't know why: the on-screen keyboard didn't come back after the session restart. squeekboard was only running as an omamobile-osk.service that was not a unit file any more (systemd said "not found", failed since the restart) and that I had started by hand earlier, and nothing launched it at login. ~/.config/hypr/autostart.lua had no o.launch_on_start("squeekboard") line, even though omamobile-setup-user writes it once and its omamobile-osk done marker was set. Starting mobi.phosh.OSK.service brought the keyboard back, and I added the line to autostart.lua by hand; squeekboard then came up by itself at the next cold boot. I can't tell whether the update removed something or my hand setup was the odd one out. So a phone that has the marker but no autostart line, as mine did, would lose the keyboard on a session restart, that's all I know.

    I haven't tried the 7 Pro path, and none of this involved a kernel package.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions