Hyprland crashes (SIGSEGV) in Aquamarine backend after 3.8.3 update — aquamarine ABI skew

[rtome85]

System details

Omarchy: 3.8.3, GPU:AMD Radeon HawkPoint iGPU (1002:1900) — Mesa/RADV, Kernel: 7.1.3-arch1-2

What's wrong?

Summary

After the system update to version 3.8.3, Hyprland crashed mid-session with SIGSEGV inside the Aquamarine backend (Aquamarine::CBackend::dispatchIdle). The session dropped to the SDDM login screen; logging back in started a fresh, stable session (no reboot needed).

The crash report shows a library version skew: Hyprland was built against Aquamarine 0.12.0 but the system now has Aquamarine 0.12.1. This looks like a packaging-alignment issue — Hyprland 0.55.4-1 was not rebuilt against the newer aquamarine 0.12.1-1 that shipped in the same update.

Environment

• Omarchy: 3.8.3
• GPU: AMD Radeon HawkPoint iGPU (1002:1900) — Mesa/RADV
• Kernel: 7.1.3-arch1-2

Installed package versions:

hyprland     0.55.4-1
aquamarine   0.12.1-1     <-- Hyprland built against 0.12.0
hyprutils    0.13.1-1
mesa         1:26.1.4-1
vulkan-radeon 1:26.1.4-1
linux        7.1.3.arch1-2
libdrm       2.4.134-1

What changed (pacman.log, 2026-07-14 12:31)

This update was first run in the boot where the crash occurred:

hyprland          0.55.2-2  -> 0.55.4-1
aquamarine        0.12.0-1  -> 0.12.1-1
mesa              1:26.1.1-2 -> 1:26.1.4-1
vulkan-radeon     1:26.1.1-2 -> 1:26.1.4-1
linux             7.0.10.arch1-1 -> 7.1.3.arch1-2
libdrm            2.4.133-1 -> 2.4.134-1
wayland-protocols 1.48-1 -> 1.49-1
xorg-xwayland     24.1.11-2 -> 24.1.13-1

Crash signature

Hyprland received signal 11 (SEGV)
Version: v0.55.4 (a0136d8c04687bb36eb8a28eb9d1ff92aea99704)
Aquamarine: built against 0.12.0, system has 0.12.1

Stack trace (crashing thread):
  #0  CrashReporter::createAndSaveCrash
  #1  (Hyprland + 0x3bfade)               <- signal handler
  #2  libc  (signal delivery)
  #3  0x...5475b9b0  (n/a + 0x0)          <- jump to invalid/garbage function pointer
  #4  Hyprutils::Signal::CSignalBase::emitInternal   (libhyprutils.so.12)
  #5  Aquamarine::CBackend::dispatchIdle             (libaquamarine.so.11)
  #6  (Hyprland + 0x72901a)
  #7  wl_event_loop_dispatch   (libwayland-server.so.0)
  #8  wl_display_run           (libwayland-server.so.0)
  #9  CEventLoopManager::enterLoop
  #10 main

Other threads were parked in libgallium-26.1.4 (Mesa GPU worker threads).

The crash is a jump to an invalid function pointer while emitting a signal from Aquamarine::CBackend::dispatchIdle — consistent with an ABI mismatch between the Hyprland binary (compiled against aquamarine 0.12.0) and the running aquamarine 0.12.1 shared library.

Impact / frequency

• Occurred once, mid-session (~11 hours into the session), not a boot loop.
• Session recovered by re-logging in; machine did not reboot.

Suggested fix

Rebuild/repackage Hyprland against aquamarine 0.12.1 (or pin aligned versions) so the "built against" and "system has" versions match. If this is confirmed to be an upstream Hyprland/aquamarine bug rather than a packaging skew, please forward to https://github.com/hyprwm/Hyprland or https://github.com/hyprwm/aquamarine.

Attachments

• Full crash report + coredumpctl info backtrace: hyprland-crash-2026-07-15.txt
• Raw coredump available via coredumpctl (PID 1656) if needed.

hyprland-crash-2026-07-15.txt

[dhh]

Is there something better here than a backport @vaxerski?

[vaxerski]

this is not ABI, this is a rare UAF crash already fixed in 0.13.0. You'll have to wait for hyprland 0.56.0 though because API broke between hyprutils 0.13 and 0.14. 0.56.0 should come out tomorrow.