Minimal Python 2 & 3 shim around all Qt bindings - PySide, PySide2, PyQt4 and PyQt5.
Python Shell
Clone or download
Latest commit cc5e730 May 19, 2018

Build Status PyPI version Anaconda-Server Badge Gitter enables you to write software that runs on any of the 4 supported bindings - PySide2, PyQt5, PySide and PyQt4.

Date Version Event
Jan 2018 1.1.0 Adds new test suite, new members
Mar 2017 1.0.0 Increased safety, backwards incompatible
Sep 2016 0.6.9 Stable release
Sep 2016 0.5.0 Alpha release of --convert
Jun 2016 0.2.6 First release of
Table of contents

Project goals

Write once, run in any binding. was born in the film and visual effects industry to address the growing need for software capable of running with more than one flavor of the Qt bindings for Python - PySide, PySide2, PyQt4 and PyQt5.

Goal Description
Support co-existence should not affect other bindings running in same interpreter session.
Build for one, run with all Code written with should run on any binding.
Explicit is better than implicit Differences between bindings should be visible to you.

See for more details.

Install is a single file and can either be copy/pasted into your project, downloaded as-is, cloned as-is or installed via pip or conda.

# From PyPI
$ pip install
# From Anaconda
$ conda config --add channels conda-forge
$ conda install
  • Pro tip: Never use the latest commit for production. Instead, use the latest release. That way, when you read bug reports or make one for yourself you will be able to match a version with the problem without which you will not know which fixes apply to you nor would we be able to help you. Installing via pip or conda as above ensures you are provided the latest stable release. Unstable releases are suffixed with a .b, e.g. 1.1.0.b3.
  • Pro tip: Supports vendoring


Use as you would use PySide2.


import sys
from Qt import QtWidgets

app = QtWidgets.QApplication(sys.argv)
button = QtWidgets.QPushButton("Hello World")


All members of Qt stem directly from those available via PySide2, along with these additional members.

Attribute Returns Description
__version__ str Version of this project
__binding__ str A string reference to binding currently in use
__qt_version__ str Reference to version of Qt, such as Qt 5.6.1
__binding_version__ str Reference to version of binding, such as PySide 1.2.6


>>> from Qt import __binding__
>>> __binding__

Compatibility also provides compatibility wrappers for critical functionality that differs across bindings, these can be found in the added QtCompat submodule.

Attribute Returns Description
loadUi(uifile=str, baseinstance=QWidget) QObject Minimal wrapper of PyQt4.loadUi and PySide equivalent
translate(...) function Compatibility wrapper around QCoreApplication.translate
wrapInstance(addr=long, type=QObject) QObject Wrapper around shiboken2.wrapInstance and PyQt equivalent
getCppPointer(object=QObject) long Wrapper around shiboken2.getCppPointer and PyQt equivalent


>>> from Qt import QtCompat
>>> QtCompat.loadUi

Class specific compatibility objects

Between Qt4 and Qt5 there have been many classes and class members that are obsolete. Under Qt.QtCompat there are many classes with names matching the classes they provide compatibility functions. These will match the PySide2 naming convention.

from Qt import QtCore, QtWidgets, QtCompat
header = QtWidgets.QHeaderView(QtCore.Qt.Horizontal)
QtCompat.QHeaderView.setSectionsMovable(header, False)
movable = QtCompat.QHeaderView.sectionsMovable(header)

This also covers inconsistencies between bindings. For example PyQt4's QFileDialog matches Qt4's return value of the selected. While all other bindings return the selected filename and the file filter the user used to select the file. Qt.QtCompat.QFileDialog ensures that getOpenFileName(s) and getSaveFileName always return the tuple.

Environment Variables

These are the publicly facing environment variables that in one way or another affect the way is run.

Variable Type Description
QT_PREFERRED_BINDING str Override order and content of binding to try.
QT_VERBOSE bool Be a little more chatty about what's going on with
QT_SIP_API_HINT int Sets the preferred SIP api version that will be attempted to set.

Subset (or "common members")

Members of is a subset of PySide2. Which means for a member to be made accessible via, it will need to (1) be accessible via PySide2 and (2) each of the other supported bindings. This excludes large portions of the Qt framework, including the newly added QtQml and QtQuick modules but guarantees that anything you develop with will work identically on any binding - PySide, PySide2, PyQt4 and PyQt5. If you need to use such excluded modules with, please see

We call this subset "common members" and these can be generated by running the script. The script will output all modules and members of each binding into individual JSON files. These JSON files are then compared and a common_members.json file is generated. The contents of this file is copy-pasted into the _common_members dictionary of Please note that the script will only use the very latest version of our Docker test suite to generate the common members subset, using the most up-to-date set of VFX Platform-stipulated software versions.

⚠️ The version of PySide2 used as reference is the one specified on VFX Platform, currently version is 2.0.x. But unfortunately, the version string of PySide2 is not yet properly maintained and the VFX Platform does not specifiy a explicit commit SHA for PySide2. Therefore, it could be difficult to know exactly which PySide2 is running on your system (unless you built it from source). In layman's terms; as PySide2 is in development and is continuously adding new support for modules, you may see differences between PySide2 built early in the year vs PySide2 built later in the year. The exact commit SHAs of PySide2 used by the test suite can be reviewed in QtC implemented an alternative way to identify which version of PySide2 you are running. You can read more about that here.

Branch binding-specific code

Some bindings offer features not available in others, you can use __binding__ to capture those.

if "PySide" in __binding__:

Override preferred choice

If your system has multiple choices where one or more is preferred, you can override the preference and order in which they are tried with this environment variable.

$ set QT_PREFERRED_BINDING=PyQt5  # Windows
$ python -c "import Qt;print(Qt.__binding__)"

Constrain available choices and order of discovery by supplying multiple values.

# Try PyQt first and then PySide, but nothing else.

Using the OS path separator (os.pathsep) which is : on Unix systems and ; on Windows.

Add or remove members from at run-time.

If you need to expose a module that isn't included in by default or wish to remove something from being exposed in you can do so by creating a module and making it available to Python.

  1. Create a new file
  2. Implement update_members
  3. Expose to Python
def update_members(members):
    """Called by at run-time to modify the modules it makes available.

        members (dict): The members considered by

Finally, expose the module to Python.

$ set PYTHONPATH=/path/to
$ python -c "import Qt.QtCore"
ImportError: No module named Qt.QtCore

Linux and MacOS users, replace set with export

Compile Qt Designer files

See #132 for details.

.ui files compiled via pyside2-uic inherently contain traces of PySide2 - e.g. the line from PySide2 import QtGui.

In order to use these with, or any other binding, one must first erase such traces and replace them with cross-compatible code.

$ pyside2-uic my_ui.ui -o
$ python -m Qt --convert
# Creating ""..
# Successfully converted ""

Now you may use the file as you normally would, with

Loading Qt Designer files

The uic.loadUi function of PyQt4 and PyQt5 as well as the QtUiTools.QUiLoader().load function of PySide/PySide2 are mapped to a convenience function loadUi.

import sys
from Qt import QtCompat

app = QtWidgets.QApplication(sys.argv)
ui = QtCompat.loadUi(uifile="my.ui")

For PyQt bindings it uses their native implementation, whereas for PySide bindings it uses our custom implementation borrowed from the qtpy project.

loadUi has two arguments as opposed to the multiple that PyQt ships with. See here for details - in a nutshell, those arguments differ between PyQt and PySide in incompatible ways. The second argument is baseinstance which allows a ui to be dynamically loaded onto an existing QWidget instance.

QtCompat.loadUi(uifile="my.ui", baseinstance=QtWidgets.QWidget)

uifile is the string path to the ui file to load.

If baseinstance is None, the a new instance of the top-level widget will be created. Otherwise, the user interface is created within the given baseinstance. In this case baseinstance must be an instance of the top-level widget class in the UI file to load, or a subclass thereof. In other words, if you've created a QMainWindow interface in the designer, baseinstance must be a QMainWindow or a subclass thereof, too. You cannot load a QMainWindow UI file with a plain QWidget as baseinstance.

loadUi returns baseinstance, if baseinstance is provided. Otherwise it will return the newly created instance of the user interface.

sip API v2

If you're using PyQt4, sip attempts to set its API to version 2 for the following:

  • QString
  • QVariant
  • QDate
  • QDateTime
  • QTextStream
  • QTime
  • QUrl


The PyQt and PySide bindings are similar, but not identical. Where there is ambiguity, there must to be a clear direction on which path to take.

Governing API

The official Qt 5 documentation is always right. Where the documentation lacks answers, PySide2 is right.

For example.

# PyQt5 adheres to PySide2 signals and slots
PyQt5.Signal = PyQt5.pyqtSignal
PyQt5.Slot = PyQt5.pyqtSlot

# PySide2 adheres to the official documentation
PySide2.QtCore.QStringListModel = PySide2.QtGui.QStringListModel


There are cases where is not handling incompatibility issues. Please see for more information.

Known Problems

Send us a pull-request with known problems here!

Who's using

Send us a pull-request with your studio here.

Presented at Siggraph 2016, BOF!


Projects using

Send us a pull-request with your project here.

Projects similar to

Comparison matrix.

Project Audience Reference binding License PEP8 Standalone PyPI Co-existence Film PySide2 MIT X X X X
jupyter Scientific N/A N/A X
QtPy Scientific N/A MIT X X
pyqode.qt Scientific PyQt5 MIT X X
QtExt Film N/A N/A X

Also worth mentioning, pyqt4topyqt5; a good starting point for transitioning to

Send us a pull-request with your project here.

Developer Guide

Tests are performed on each aspect of the shim.

Each of these are run under..

  • Python 2.7
  • Python 3.4
  • Python 3.5
  • Python 3.6

..once for each binding or under a specific binding only.

Each test is run within it's own isolated process, so as to allow an import to occur independently from other tests. Process isolation is handled via nosepipe.

Tests that are written at module level are run four times - once per binding - whereas tests written under a specific if-statement are run only for this particular binding.

if binding("PyQt4"):
    def test_something_related_to_pyqt4():

Code convention

Below are some of the conventions that used throughout the module and tests.

  • Etiquette: PEP8
    • All code is written in PEP8. It is recommended you use a linter as you work, flake8 and pylinter are both good options. Anaconda if using Sublime is another good option.
  • Etiquette: Double quotes
    • " = yes, ' = no.
  • Etiquette: Napoleon docstrings
    • Any docstrings are made in Google Napoleon format. See Napoleon for details.
  • Etiquette: Semantic Versioning
  • Etiquette: Underscore means private
    • Anything prefixed with an underscore means that it is internal to and not for public consumption.

Running tests

Due to the nature of multiple bindings and multiple interpreter support, setting up a development environment in which to properly test your contraptions can be challenging. So here is a guide for how to do just that using Docker.

With Docker setup, here's what you do. Please note this will pull down a ~1 GB image.


# Run nosetests (Linux/OSX)
docker run --rm -v $(pwd):/ -e PYTHON=2.7 fredrikaverpil/
docker run --rm -v $(pwd):/ -e PYTHON=3.4 fredrikaverpil/
docker run --rm -v $(pwd):/ -e PYTHON=3.5 fredrikaverpil/
docker run --rm -v $(pwd):/ -e PYTHON=3.6 fredrikaverpil/

# Run nosetests (Windows)
docker run --rm -v %CD%:/ -e PYTHON=2.7 fredrikaverpil/
docker run --rm -v %CD%:/ -e PYTHON=3.4 fredrikaverpil/
docker run --rm -v %CD%:/ -e PYTHON=3.5 fredrikaverpil/
docker run --rm -v %CD%:/ -e PYTHON=3.6 fredrikaverpil/

# Doctest: test_caveats.test_1_qtgui_qabstractitemmodel_createindex ... ok
# Doctest: test_caveats.test_2_qtgui_qabstractitemmodel_createindex ... ok
# Doctest: test_caveats.test_3_qtcore_qitemselection ... ok
# ...
# ----------------------------------------------------------------------
# Ran 21 tests in 7.799s
# OK

Now both you and Travis are operating on the same assumptions which means that when the tests pass on your machine, they pass on Travis. And everybody wins!

For details on the Docker image for testing, see

See for more of the good stuff.