Dear Maintainer, I am seeing what appears to be a severe memory leak in xfwm4 during a long-running X11/Xfce session. The machine normally remains running for several days. On 2026-10-24, after returning to the graphical session, the desktop failed to display correctly. The result was a black screen with only the mouse pointer visible, accompanied by heavy disk activity. I switched to a virtual console and verified that the Xfce session was still running. In particular, xfce4-session, xfwm4, xfsettingsd, xfce4-panel and xfdesktop were all alive. The xfwm4 process, however, was consuming approximately 10 GB of resident memory. Values observed from /proc/<xfwm4-pid>/status were approximately: VmSize: 10,949,952 kB VmRSS: 10,201,776 to 10,288,032 kB RssAnon: 10,265,212 kB RssFile: 22,816 kB VmSwap: 19,004 to 28,324 kB Most of the memory was therefore anonymous resident memory. At the same time, atop did not show sustained CPU or disk saturation that would otherwise explain the unusable graphical session. Kernel logs covering the incident did not show an i915/DRM GPU hang, GPU reset, GPU fault, or similar graphics-driver failure. I recovered the graphical session by replacing xfwm4 without rebooting the machine: xfwm4 --replace Immediately after replacement, the new xfwm4 process was approximately: VmSize: 690128 kB VmRSS: 103684 kB RssAnon: 21484 kB RssFile: 81548 kB VmSwap: 0 kB The desktop became usable again immediately after replacing xfwm4. System configuration: Debian GNU/Linux 13 (trixie) X11 session Xfce xfwm4 4.20.0-1 Intel HD Graphics 4600 is the active desktop/OpenGL renderer NVIDIA GeForce GTX 850M is installed as a PRIME render-offload GPU Two-monitor configuration xfwm4 compositing was enabled when the leak occurred xrandr reports the Intel-driven modesetting provider as the display provider and NVIDIA-G0 as the secondary PRIME provider. OpenGL for the desktop is rendered using Mesa Intel HD Graphics 4600. As a diagnostic/workaround after the incident, I disabled xfwm4 compositing. I am currently monitoring the new xfwm4 process to see whether the memory growth recurs with compositing disabled. The machine has atop process accounting/logging enabled at one-minute intervals. I am attaching historical xfwm4 memory data, system statistics from the incident period, kernel information and the relevant Xfce/X11 configuration. Expected behavior: xfwm4 should maintain relatively stable memory usage during a long-running X11/Xfce session. Actual behavior: During a session lasting several days, xfwm4 grew to approximately 10 GB RSS, eventually leaving the graphical session with a black screen and mouse cursor. Replacing xfwm4 immediately reduced its RSS to approximately 100 MB and restored the desktop. At present I do not have a short deterministic reproducer. The issue appears after a long-running session. Attachments include: system and package information xfwm4 configuration XRandR/provider information OpenGL renderer information historical atop process-memory data for xfwm4 atop system statistics around the incident filtered kernel logs around the incident manually recorded process memory values from the leaking xfwm4 process Please let me know if additional diagnostics, debug symbols, heap profiling, or testing with a newer xfwm4 build would be useful. Regards, Adan Nada