pijyoi

I had noticed this before on the Raspberry Pi 3B. This failure occurs on aarch64 but not arm32. On the recently released Raspberry Pi OS 32-bits "Bullseye", the test passes....

As I recall, the pixel differences occurred where there was multiple objects being painted at the same pixel. So in the example image that you provided, I seem to recall...

For example, there are several arm64-specialized codepaths here: https://code.woboq.org/qt5/qtbase/src/gui/painting/qdrawhelper_neon.cpp.html (search for Q_PROCESSOR_ARM_64)

This is the output with ```PYQTGRAPH_AUDIT=1``` 

>I had noticed this before on the Raspberry Pi 3B. This failure occurs on aarch64 but not arm32. On the recently released Raspberry Pi OS 32-bits "Bullseye", the test passes....

I compared the `closed_init` reference image against the outputs from x86_64 (Windows PyQt6), armhf (Raspberry Pi OS Bullseye PyQt5), and aarch64 (Raspberry Pi OS Bullseye PyQt5). For x86_64 and armhf,...

I created a standalone (pyqtgraph-free) script to see if I could reproduce the pixel differences. On amd64 Windows and WSL2, I got the md5sum `09d0b9c9e54eade1fae3a317ba337612`. (Qt5 or Qt6 doesn't matter)...

Would it not be possible for you to do something like the following since you are already inheriting from ```PlotDataItem```? ```python class MyCustomItem(pg.PlotDataItem): def setData(self, x, y, *args, **kwargs): self._my_raw_data...

This seems to be a known problem on Windows platform: https://bugreports.qt.io/browse/QTBUG-51093 https://doc.qt.io/qt-5/windows-issues.html#fullscreen-opengl-based-windows However, the workaround class is not available in the Python bindings. https://riverbankcomputing.com/pipermail/pyqt/2020-December/043458.html What I did was to insert...

Do any of the things listed here https://doc.qt.io/qt-5/qopenglwidget.html#limitations and here https://doc.qt.io/qt-5/qopenglwidget.html#alternatives apply to your use-case?