Per-app override for forced NVIDIA GLX/EGL vendor (default/hypr/nvidia.lua)

[CJHarmath]

default/hypr/nvidia.lua sets __GLX_VENDOR_LIBRARY_NAME=nvidia (and NVD_BACKEND) session-wide for any Pascal/Maxwell-class NVIDIA GPU (the no-GSP branch, nvidia-580xx). That's the right default for almost everything, but it forces every GL/EGL app in the session onto NVIDIA's driver with no per-app opt-out.

Hit this concretely with Citrix Workspace App (wfica, AUR icaclient): its EGL-on-Wayland probe intermittently segfaults inside NVIDIA's proprietary libnvidia-egl-wayland.so.1 (a thread race in wl_proxy_create_wrapper, ~40% of launches on this machine, not deterministic). Full crash analysis, logs, and timeline: https://gist.github.com/CJHarmath/53ee5cdc6c02e4467ec051cfac9b1bb9

This isn't an Omarchy bug — the forced env var is working as designed, and the crash is entirely inside proprietary Citrix + NVIDIA code. The known community workaround is a wrapper script around wfica that strips Wayland env vars for that one process (forcing XWayland) so it never touches NVIDIA's EGL-Wayland path, while leaving the rest of the session on NVIDIA as normal: https://gist.github.com/loudambiance/bd4c1d8ddba74c52633f9a8497a3051e

Suggestion: document (or provide a supported hook for) a per-app GPU/EGL vendor override, so apps that are incompatible with NVIDIA's EGL-Wayland implementation don't require hand-patching the installed binary. Something in the spirit of default/hypr/nvidia.lua's detection, but exposed as an app-launch wrapper pattern users can opt individual .desktop entries into, would cover this and any future case in the same category (related to the hybrid-GPU routing issues already tracked in #8215).