wezterm crashes (NVIDIA EGL segfault), taking down all terminal windows

Problem

WezTerm crashes and takes all terminal windows down with it (they share a single wezterm-gui process). Recurring and accelerating: 6 crashes in ~2 weeks (Jun 17, 23, 28, Jul 1 x3).

Root cause

Segfault inside the NVIDIA proprietary driver, called from wezterm's OpenGL/EGL front-end. Every crash is preceded by MakeCurrent: BAD_ALLOC.

Core dump backtrace:

#0  libnvidia-eglcore.so.595.71.05 + 0xb6be88   <- crash is inside the NVIDIA driver
#1  libnvidia-eglcore.so.595.71.05 + 0xb7f0bd
#2  libnvidia-eglcore.so.595.71.05 + 0x849967
#3  wezterm-gui + 0x235a0d6                      <- wezterm OpenGL front-end calling in

Log immediately before the segfault:

window::egl > make_current failed ... MakeCurrent: BAD_ALLOC
systemd-coredump: Process (wezterm-gui) dumped core / Segmentation fault

BAD_ALLOC on eglMakeCurrent is a known NVIDIA EGL failure mode, typically triggered by GPU context loss / resource churn.

Environment

• GPU: NVIDIA GeForce RTX 3060
• Driver: 595.71.05
• Display: X11 (gdm-x-session)
• wezterm: 20260607-013257-093bf6bf (self-built from source)

Why all windows die together

All windows spawn into one shared wezterm-gui process ("Spawned your command via the existing GUI instance"). One segfault kills every window.

Fix

• Switch front_end off the crashing OpenGL/EGL path to WebGpu (Vulkan) in ~/.config/wezterm/wezterm.lua
• If WebGpu still crashes, fall back to front_end = "Software"
• Root-cause hardening: update NVIDIA driver from 595.71.05 (the actual bug is in the driver, not wezterm)

Monitoring

Watch for recurrence with:

coredumpctl list wezterm-gui

Any new SIGSEGV row means the new path is still hitting a driver bug.

Start with ~/.config/wezterm/wezterm.lua and review the configured front_end alongside the crash details in the issue. Run coredumpctl list wezterm-gui to check whether crashes recur after the configuration change. Done means the shared wezterm-gui process no longer produces SIGSEGV entries, or the remaining failure is documented as an NVIDIA driver issue.