Inconsistent QScreen geometry in multiple screens using different scaling factors

In my 3 screens setup, the behavior of Qt is inconsistent when managing Windows scaling and it's per-screen geometry. Some of their respective QScreens provide correctly the geometry in device-independent pixels and others in native resolution. What's more problematic: for those in which the size remained in native resolution, the device pixel ratio is the one you would expect from scaled geometry and thus, `device-independent pixels` .

To reproduce in my end, I have this screen setup configuration in Windows 10 

• Screen 1:
  • Native resolution: 1920x100
  • Windows scaling: 150%
• Screen 2 ( Main display / primary screen):
  • Native resolution: 1920x1200
  • Windows scaling: 100%
• Screen 3:
  • Native resolution: 1920x1080
  • Windows scaling: 150%

This is my code, unfortunately not easy to reproduce given you need the same screens:

from PySide6 import QtGui, QtCore
app = QtGui.QGuiApplication()
print("Qt version:", QtCore.__version__)
screens = app.screens()
for screen in screens:
    print(screen.name(), screen.geometry(), screen.devicePixelRatio())

and the outcome being:

Qt version: 6.2.1
\\.\DISPLAY8 PySide6.QtCore.QRect(0, 0, 1920, 1200) 1.0    
\\.\DISPLAY9 PySide6.QtCore.QRect(-1920, 0, 1920, 1080) 1.5
\\.\DISPLAY1 PySide6.QtCore.QRect(1920, 121, 1280, 720) 1.5

Note how:

• \\.\DISPLAY8 (Screen 2): is indeed showing the right resolution, and devicePixelRatio
• \\.\DISPLAY1 (Screen 1): is correctly scaling the resolution to 1280x720
  • Note: (1280x720) * 1.5 = 1920x1080
  • Thus this is correct
• \\.\DISPLAY9 (Screen 3): is still keeping the native resolution, but device pixel ratio is 1.5 and not 1.0
  • (1920x1080 ) * 1.5 = 2880 x 1620 (????)
  • Thus this is incorrect and bug

Conda environment being used (yml file):

 
name: qscreen_test
channels:
  - anaconda
  - conda-forge
  - defaults
dependencies:
  - ca-certificates=2021.10.8=h5b45459_0
  - jpeg=9b=vc14h4d7706e_1
  - libblas=3.9.0=5_hd5c7e75_netlib
  - libcblas=3.9.0=5_hd5c7e75_netlib
  - liblapack=3.9.0=5_hd5c7e75_netlib
  - libpng=1.6.37=h2a8f88b_0
  - libsodium=1.0.18=h62dcd97_0
  - libtiff=4.1.0=h56a325e_1
  - lz4-c=1.9.2=hf4a77e7_3
  - m2w64-gcc-libgfortran=5.3.0=6
  - m2w64-gcc-libs=5.3.0=7
  - m2w64-gcc-libs-core=5.3.0=7
  - m2w64-gmp=6.1.0=2
  - m2w64-libwinpthread-git=5.0.0.4634.697f757=2
  - msys2-conda-epoch=20160418=1
  - ninja=1.10.2=h2d74725_1
  - nomkl=1.0=h5ca1d4c_0
  - numpy=1.21.4=py39h6635163_0
  - openssl=3.0.0=h8ffe710_2
  - pip=21.3.1=pyhd8ed1ab_0
  - pybind11=2.8.1=py39h2e07f2f_1
  - pybind11-global=2.8.1=py39h2e07f2f_1
  - pyopengl=3.1.5=py_0
  - python=3.9.7=h900ac77_3_cpython
  - python_abi=3.9=2_cp39
  - pywin32=302=py39hb82d6ee_2
  - setuptools=59.4.0=py39hcbf5309_0
  - sqlite=3.37.0=h8ffe710_0
  - tzdata=2020b=h7b6447c_0
  - vc=14.1=h0510ff6_4
  - vs2015_runtime=14.16.27012=hf0eaf9b_3
  - wheel=0.35.1=py_0
  - xz=5.2.5=h62dcd97_0
  - zlib=1.2.11=vc14h1cdd9ab_1
  - zstd=1.4.4=ha9fde0e_3
  - pip:
    - nuitka==0.6.18
    - pyside6==6.2.1
    - shiboken6==6.2.1

Edit 1: Tested the same setup with Qt 5.15.2

This is the output:

Qt version: 5.15.2
\\.\DISPLAY8 PySide2.QtCore.QRect(0, 0, 1920, 1200) 1.0     
\\.\DISPLAY9 PySide2.QtCore.QRect(-1920, 0, 1920, 1080) 1.0 
\\.\DISPLAY1 PySide2.QtCore.QRect(1920, 121, 1920, 1080) 1.0

Scaling is ignored, but at least it is consistent. But, if I define

set QT_ENABLE_HIGHDPI_SCALING=1
set QT_SCALE_FACTOR_ROUNDING_POLICY=PassThrough

then the output is the correct one.

Qt version: 5.15.2
\\.\DISPLAY8 PySide2.QtCore.QRect(0, 0, 1920, 1200) 1.0
\\.\DISPLAY9 PySide2.QtCore.QRect(-1920, 0, 1280, 720) 1.5
\\.\DISPLAY1 PySide2.QtCore.QRect(1920, 121, 1280, 720) 1.5

 Edit 2: After multiple experiments, this pattern emerged:

• The bug occurs on screens having a scale factor different than 100% (eg. 125% or 150%), whenever said window is not the main display, and the main display does have a scale factor of 100%.