Issues running on Linux Wayland

[laundmo]

Trying this on Linux with a Wayland compositor (KDE Plamsa) i've run into a few issues.

For context: I installed with uv tool install pyimagecuda-studio instead of pip to install this in its own venv, and because global pip installation is blocked on linux due to that using the system-managed python installation.

1. numpy dependency was missing, so i had to run uv tool install pyimagecuda-studio --with numpy to install it.
2. QEGL Error 3009:

   $ pics
   [GUI] Allocated U8 buffer: 2048x2048 for preview
   QEGLPlatformContext: Failed to create context: 3009
   QRhiGles2: Failed to create temporary context
   QEGLPlatformContext: Failed to create context: 3009
   QRhiGles2: Failed to create context
   Failed to create QRhi for QBackingStoreRhiSupport
   QEGLPlatformContext: Failed to create context: 3009
   QOpenGLWidget: Failed to create context
   QEGLPlatformContext: Failed to create context: 3009
   QOpenGLWidget: Failed to create context
   QEGLPlatformContext: Failed to create context: 3009
   QOpenGLWidget: Failed to create context
   

   fixed by running: QT_QPA_PLATFORM="xcb" pics instead
3. Lack of PNG support (?)

   [LOAD_IMAGE] Image loading image from /home/a/test.png
   Traceback (most recent call last):
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/gui/properties/properties_factory.py", line 263, in browse_file
       notify_change(file_path)
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/gui/properties/properties_factory.py", line 23, in notify_change
       on_change_callback(param.name, value)
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/gui/properties/node_properties.py", line 181, in on_param_changed
       self.current_node.update_param(param_name, value)
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/nodes/core.py", line 29, in update_param
       self.update_param_child(param_name, value)
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/nodes/node_generator.py", line 59, in update_param_child
       self.generate_image()
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda_studio/nodes/generators/generators.py", line 45, in generate_image
       self.original_image = load(self.path)
                             ^^^^^^^^^^^^^^^
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyimagecuda/io.py", line 82, in load
       vips_img = pyvips.Image.new_from_file(filepath, access='sequential', autorotate=autorotate)
                  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyvips/vimage.py", line 329, in new_from_file
       return pyvips.Operation.call(name, filename,
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
     File "/home/a/.local/share/uv/tools/pyimagecuda-studio/lib/python3.12/site-packages/pyvips/voperation.py", line 285, in call
       raise Error(f'{operation_name} does not support optional '
   pyvips.error.Error: VipsForeignLoadPngFile does not support optional argument autorotate
   

   solution: Convert to jpeg with magick test.png test.jpeg
4. Preview does not render. From what little i could tell, it seems like an issue with the far more strict OpenGL validation on Linux. A bit of testing led me to the OpenGL Preview widget, and i found that errors raised in initializeGL are silently swallowed. wrapping everything in a try-except and printing the error message, it seems that a context isn't enabled when something is called that needs a context. For testing this, i've worked around it by exporting the image - which works.

[offerrall]

@laundmo
Thanks for the detailed bug report! Let me address each issue:

1. NumPy dependency missing

This is surprising - I don't use NumPy directly in PyImageCUDA Studio at all. This is likely an undeclared dependency from either PyVips or PyOpenGL that they're not managing properly in their package metadata. I've never seen this happen before in any installation.

I'll investigate adding NumPy as an explicit dependency if this is a recurring issue, but it shouldn't be needed.

2. QEGL Error 3009 (Wayland/XCB context issue)

Thanks for finding the QT_QPA_PLATFORM="xcb" workaround! This is a known Qt6 + Wayland + OpenGL compatibility issue on some systems. I'll add this to the Linux troubleshooting documentation.

For a permanent fix, you could add an alias:

echo 'alias pics="QT_QPA_PLATFORM=xcb pics"' >> ~/.bashrc

3. PNG Support Issue - Please reinstall with v0.1.6

This error is because you're running an older version. The issue you encountered was introduced in v0.1.5 when I added automatic EXIF orientation handling. The autorotate parameter was made required in that version, which broke PNG loading since PNGs don't support EXIF rotation.

This was fixed in v0.1.6 (current version) by making autorotate optional again.

You need:

• pyimagecuda-studio 0.1.6
• pyimagecuda 0.1.3 (the underlying CUDA library)

Try a clean reinstall:

uv tool uninstall pyimagecuda-studio
uv tool install pyimagecuda-studio --with numpy  # include numpy since you needed it

Then verify versions:

uv tool run --from pyimagecuda-studio pip list | grep pyimagecuda

You should see:

pyimagecuda         0.1.3
pyimagecuda-studio  0.1.6

After upgrading, PNG loading should work without any conversion needed.

4. Preview not rendering (OpenGL context issues)

This is the most concerning issue. You're right that Linux has stricter OpenGL validation. The fact that export works but preview doesn't suggests the CUDA-OpenGL interop is failing silently.

Could you share the full error message you got when you wrapped initializeGL in try-except? That would help me identify exactly which OpenGL call is failing without a context.

Also, what GPU are you running? (nvidia-smi output would be helpful)

Additionally, could you share your Linux distribution and version? (e.g., cat /etc/os-release). If there's anything special about your setup (custom kernel, specific Qt version, etc.), let me know - I'd like to try replicating your environment on my machine to reproduce and fix these errors.

---

Thank you so much for using PyImageCUDA Studio and taking the time to open these issues! Detailed reports like yours are incredibly valuable for improving Linux compatibility.

Quick question: Have you tried running this on Windows or another Linux distribution? If so, did everything work smoothly there? I'm curious if these issues are specific to KDE Plasma + Wayland or more widespread.

Looking forward to your response! I really hope we can solve all these problems together. PyImageCUDA Studio matters a lot to me, and I'd love for anyone who appreciates it to be able to use it without issues. Any bugs or suggestions you have, I'll do my best to address them as much as possible.

Let me know the results after reinstalling v0.1.6, especially whether PNG loading works and what the OpenGL error message says!

[offerrall]

@laundmo

I've been testing this on a clean Ubuntu 22.04.5 LTS installation to reproduce your issues. Here's what I found:

1. NumPy dependency missing

I couldn't reproduce this - NumPy was not required or installed on my clean Ubuntu setup. This is likely specific to your uv environment or a transitive dependency issue. If it continues to be a problem, let me know and I'll add it as an explicit dependency.

2. QEGL Error 3009 (Wayland/XCB context issue)

This doesn't happen on Ubuntu, so it appears to be specific to certain Linux distributions or desktop environments. I'll add this to the Linux troubleshooting documentation noting that some distros may require this workaround.

3. PNG Support Issue - Will be fixed in ~1 hour

I reproduced this on Ubuntu and identified the root cause. Even though autorotate is optional, passing it as a parameter (even as False) causes PNG loading to fail because VipsForeignLoadPngFile doesn't support it.

The fix is releasing shortly:

• pyimagecuda v0.1.4 - Adds graceful fallback when autorotate is not supported
• pyimagecuda-studio v0.1.7 - Requires pyimagecuda>=0.1.4

In about 1 hour or less, reinstall to get the latest version:

uv tool uninstall pyimagecuda-studio
uv tool install pyimagecuda-studio --with numpy

Verify you have:

pyimagecuda         0.1.4
pyimagecuda-studio  0.1.7

PNG loading should work perfectly after this.

4. Preview not rendering (OpenGL context issues)

I tested on Ubuntu 22.04.5 LTS and did not encounter any OpenGL errors. The preview rendered correctly in all tests.

This is likely specific to KDE Plasma + Wayland or your GPU/driver setup.

[offerrall]

@laundmo

I've just released the fix:

Updated versions available now:

• pyimagecuda-studio v0.1.7
• pyimagecuda v0.1.4

PNG loading is now working perfectly. Please update:

uv tool uninstall pyimagecuda-studio
uv tool install pyimagecuda-studio

Verify you have the correct versions:

uv tool run --from pyimagecuda-studio pip list | grep pyimagecuda

You should see:

pyimagecuda         0.1.4
pyimagecuda-studio  0.1.7

---

About the NumPy warning: I'm also seeing a warning message: unable to import OpenGL.arrays.numpymodule.NumpyHandler: no numpy module present. This is just a PyOpenGL warning and does not affect functionality at all - everything works perfectly despite this message.

Is this the same NumPy error you were seeing, or was yours different? If it was a different error that actually prevented the program from running, please share the full error message.