For example, there is QPainter.drawPointsNp() which takes 2 numpy-arrays:

https://doc.qt.io/qtforpython-6/PySide6/QtGui/QPainter.html#PySide6.QtGui.PySide6.QtGui.QPainter.drawPointsNp

For OpenGL, the QOpenGLFunctions typically also take numpy arrays.

Nevertheless, creating Python bindings for a Custom widget is also possible with shiboken. It just takes some thought on how to structure it. You might want a C++ library containing the widget, a separate Qt Designer plugin and the bindings.

For generating bindings. see for example
https://doc.qt.io/qtforpython-6/examples/example_scriptableapplication__scriptableapplication.html

QPainter.drawPointsNp

Thank you for your hints. The QPainter.drawPointsNp looked promising, but I need lines :) I even constructed a hack using QXYSeries which also accepts Numpy arrays and can return QPoints, but unfortunately prebuilt PySide6 comes without any Numpy support (why???). Rebuilding PySide and support custom version looks like an overkill for me for now.

I'm really appreciate your help, thank you for the links I will give them a try.

For example, there is QPainter.drawPointsNp() which takes 2 numpy-arrays:

https://doc.qt.io/qtforpython-6/PySide6/QtGui/QPainter.html#PySide6.QtGui.PySide6.QtGui.QPainter.drawPointsNp

For OpenGL, the QOpenGLFunctions typically also take numpy arrays.

Nevertheless, creating Python bindings for a Custom widget is also possible with shiboken. It just takes some thought on how to structure it. You might want a C++ library containing the widget, a separate Qt Designer plugin and the bindings.

For generating bindings. see for example
https://doc.qt.io/qtforpython-6/examples/example_scriptableapplication__scriptableapplication.html

PySide/PyQt are only wrappers around Qt code for Python. Are you sure your "slowness" is in the Python code rather than the time it takes to do whatever Qt drawing or even your algorithm?

I think slowness is in my python draw code. I'm drawing a lot (1000s) of small lines and preparing QPoints lists seems takes some time in Python. I tried obvious optimisations like re-using point objects and had some improvements but anyway it takes up to 100ms to update a frame.

Drawing itself should be quite cheap if I understand correctly, it should directly be mapped to the Open GL draw call.

So the next thing I want to try is moving this code to C++ if possible.

Thanks!

PySide/PyQt are only wrappers around Qt code for Python. Are you sure your "slowness" is in the Python code rather than the time it takes to do whatever Qt drawing or even your algorithm?