zed crashing on launch because NoMachine version 8.19.2 breaks vulkan

[cmstrickland]

Just dropping this in here in case anyone else runs into it. Zed stopped launching for me entirely this morning, after a reboot on my Debian laptop (which I don't tend to reboot that often)

All troubleshooting and triaging of zed pointed towards a crash in vulkan, and the logs also seemed to indicate that it was falling back to software rendering maybe .. 2025-11-12T15:14:40+00:00 INFO [zed::zed] Using GPU: GpuSpecs { is_software_emulated: true, device_name: "llvmpipe (LLVM 19.1.7, 256 bits)", driver_name: "llvmpipe", driver_info: "Mesa 25.0.7-2 (LLVM 19.1.7)" }

Digging more into that I found that vulkan was fully broken - vkgears et al also dump core with a panic. The mysterious part was how anything could have changed in the graphics drivers - I run debian stable, I hadn't been doing any reconfiguration. But clearly something had changed my driver stack since last boot. I tried reverting to the previous kernel, still no help.

Eventually I spotted an odd library reference in the vkgears crash backtrace - libnxegl.so , and libnx pointed me at nomachine , which i have installed on my laptop to use it's remote client. This tickled a memory of previous problems I'd seen from this app, and I found this in my login environment

echo $LD_PRELOAD 
/usr/NX/lib/libnxegl.so

unset LD_PRELOAD , and vkgears and zed load just fine. So, that's the culprit, somehow - nomachine injects a library into your user shells that I guess monkeypatches some EGL behaviour, which is a bit hacky of them.

It took me a while to find where they were doing this - eventually I found a desktop environment hook source script in cd /etc/xdg/plasma-workspace/env/nx-sourceenv.sh - I'm running KDE plasma

Searching their own support site I found this thread where they provide a command to disable this behaviour, which removes this hook script.

If you don't want to disable the feature, just launching zed in an environment with LD_PRELOAD="" , or running unset LD_PRELOAD before launching it will fix the crashing.

It took me about an hour to fix this so I thought I would drop this info in here in case anyone else was confused. It seems like nomachine are aware they've broken things and I expect they're working on a fix.

This is very much NOT a zed problem really, but it turns out that because of the nature of my machine (a little dell latitude with intel xe graphics) and my workflow (just coding on python and lisp and typescript and stuff), zed is pretty much the only Vulkan using regular software I'm running, so to all intents and purposes, this problem initially looked like a zed crash. Other users might be in the same boat, if they are using nomachine.