Broken AppImage after skipping AppRun hook creation on Qt6

[ctlcltd]

With latest builds of linuxdeploy, linuxdeploy-qt-plugin the resulting qt6 AppImage is broken.

Here is the log:

linuxdeploy version 1-alpha (git commit ID cc7b864), GitHub actions build 335 built on 2026-03-01 02:26:01 UTC
[...]
[qt/stdout] -- Creating qt.conf in AppDir -- 
[qt/stdout] Creating Qt conf file: AppDir/usr/bin/qt.conf 
[qt/stdout] 
[qt/stdout] -- Note: skipping AppRun hook creation on Qt  6  -- 
[qt/stdout] 
[qt/stdout] Done! 

Launching bundle.AppImage no longer reads for bin/qt.conf file. Seems the bash script is required to reach the current working directory.

qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in ""
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Aborted                (core dumped) ./test.AppImage

I found this change, issue #155 commit: 4e4d51c

If the apprun-hooks folder doesn't exist with content, linuxdeploy doesn't create the script, linuxdeploy::deployAppRunWrapperIfNecessary() https://github.com/linuxdeploy/linuxdeploy/blob/cc7b86472c3caa3fd729b9dc502fd2aa78394257/src/core/appdir_root_setup.cpp#L268

Maybe the script solution is not very nice, because it requires bash to be installed on the system. I don't know if there's another way to comply this wrap through linuxdeploy directly.

Note: I'm using Qt6 binaries from conda-forge and I don't know if there is implications for current working directory and the logic to locate local qt.conf file.

[TheAssassin]

What happens if you set the env var the script would have set manually?

[ctlcltd]

I'm not sure I understand which env var you're referring to, I always used the tool like this:

export LD_LIBRARY_PATH=/home/runner/miniconda3/envs/test/lib:/home/runner/miniconda3/lib:/usr/local/lib:/usr/lib
export QMAKE=qmake6
linuxdeploy --appdir AppDir --plugin qt --output appimage
export LD_LIBRARY_PATH=/usr/local/lib:/usr/lib

Until October 2025, it produced working builds, without having to touch anything in the finalized bundle.
When I ran the commands a few days ago, I tested the build and I saw it gave the famous "qt.qpa.plugin Qt platform plugin error".
I've tried it both locally and here on the GitHub Actions public runners.
I unsquashed the bundle and the structure has some difference. So I reintroduced the missing bash script and renamed the symbolic link AppRun to AppRun.wrapped, squashed, it works as before.

[TheAssassin]

What OS are you running your AppImage on? Looks like you may be using Wayland. Try export QT_QPA_PLATFORM=wayland, then call your AppImage.

[ctlcltd]

I'm using Arch Linux and X11, Ubuntu 24.04 LTS and Wayland, same results.

I used export QT_DEBUG_PLUGINS=1, it cannot find the paths.

This script line seems to be necessary: https://github.com/linuxdeploy/linuxdeploy/blob/master/src/core/appdir_root_setup.cpp#L313

[ctlcltd]

export QT_DEBUG_PLUGINS=1
export QT_QPA_PLATFORM=wayland
./test.AppImage
qt.core.plugin.factoryloader: checking directory path "/home/user/Downloads/platforms" ...
qt.qpa.plugin: Could not load the Qt platform plugin "wayland" in ""
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Aborted                (core dumped) ./test.AppImage

[TheAssassin]

Is the wayland plugin shipped by conda-forge at all? Please share a complete log.

[ctlcltd]

The wayland plugin is another problem. Since Qt 6.10, the wayland plugin should be in qt6-base, equivalent to qt6-main on conda-forge. Theoretically.

This is the main problem:

qt.core.plugin.factoryloader: checking directory path "/home/user/Downloads/platforms" ...

The problem is fundamental. Qt6 reads proc/self/exe and argv[0], this logic is in the file qcoreapplication.cpp.

The so called appimage/type2-runtime and linuxdeploy-plugin-qt do not correct argv[0], it is passed the original argv[0], which is in the directory from which it is launched, not the mounted one.

• If the binary is wrapped in "AppRun[.sh]", the "$0" argument is handled by bash (or sh) and is correct, it points to the currently mounted directory.
• If the appimage is extracted via --appimage-extract, does it work, the argv[0] argument is correct.

The mounted path is passed into the APPDIR variable, setenv("APPDIR", mount_dir, 1) but it cannot be used by Qt6. The file: https://github.com/AppImage/type2-runtime/blob/main/src/runtime/runtime.c#L1832-L1850

Hope this help.

This is the complete log:

export QT_DEBUG_PLUGINS=1
export QT_QPA_PLATFORM=xcb
VERBOSE=1 ./test.AppImage
Running in verbose mode
Found setuid root executable: /usr/bin/fusermount3
fusermount3 version: 3.18.1
FUSERMOUNT_PROG set to /usr/bin/fusermount3
qt.core.plugin.factoryloader: checking directory path "/home/user/Downloads/platforms" ...
qt.qpa.plugin: Could not load the Qt platform plugin "xcb" in ""
This application failed to start because no Qt platform plugin could be initialized. Reinstalling the application may fix this problem.

Aborted                (core dumped) ./test.AppImage

[TheAssassin]

Sorry for not being as clear as needed. I need to see a bundling log, i.e., when you run linuxdeploy with the Qt plugin. I kind of expect the platform plugins to be missing there. This might be a bug in linuxdeploy if it is not aborting when no such plugin is available.

If you can, please also share the AppImage or a link to your repo so that I can check what's going on.

qt.core.plugin.factoryloader: checking directory path "/home/user/Downloads/platforms" ...

This shouldn't be loaded at all. The Qt plugin writes a qt.conf that should properly resolve the platform plugins dir relative to the binary. I'm not sure why that path prefix plays a role. That's why I need to check out the AppImage or a minimal example (feel free to just ship a main.cpp with hello world or so if it's linked the same way).

[ctlcltd]

I've tried with Qt6 Examples widgets/mainwindows, attached to this comment there is a workflow Github Actions and a full log.

There is also an attempt to fix in the workflow, but I don't know how good...

Regarding wayland and qt6: if the bundle.AppImage is fixed, on Ubuntu 24.04 from terminal, it prints the absence of the wayland plugin (soft warning), but it auto-falls back to xcb, so it just works.

Regarding wayland and conda-forge: I checked the conda-forge::qt6-main version 6.10.2 package and the plugin is there: lib/qt6/plugins/platforms/libqwayland.so

I hope this helps to solve the problem, good luck! 🤞

test-appimage-linuxdeploy.yml

logs_60509334889.zip

[ctlcltd]

@TheAssassin I tried to run the example today, I made a mistake in the fix attempt (last step in the workflow file "Fix artifact proof").

The file AppRun.sh is wrong, this works:

#!/bin/sh

exec "$0".wrapped "$@"

but I don't know how "elegant" is, maybe you or the team can find another way, ciao.

p.s. pay attention to the symlink AppRun.wrapped -> usr/bin/menus, the artifact.zip generated Zip contains a useless squashfs-root directory with resolved symlinks.