Fix tests for PySide2 and add PySide2 to CI matrix

[kitchoi]

Fix #382
Fix #495

Edited message:

The segfault observed was caused by an incorrect use of the QWheelEvent API. In Qt4, the QWheelEvent's pos and globalpos takes a QPoint type (see doc). However, since Qt5, QWheelEvent API is changed and the pos and globalpos are QPointF instead (see doc, and there are other API changes). For backward compatibility, PySide2/Qt5's QWheelEvent has an overloaded signature to support both, so that code can migrate. If the signature matches that of Qt4, QPoint is expected for pos and globalpos. If the signature matches that of Qt5, QPointF is expected.

This can be demonstrated with the following script:

from PySide2 import QtCore, QtGui

qt4_args = (
    QtCore.QPoint(0, 0),    # pixelDelta
    QtCore.QPoint(0, 200),  # angleDelta
    200,                    # qt4Delta
    QtCore.Qt.Vertical,     # qt4Orientation
    QtCore.Qt.NoButton,     # buttons
    QtCore.Qt.NoModifier,   # modifiers
    QtCore.Qt.ScrollUpdate, # phase
)

# Qt 5 / Qt 6 args
qt_args = (
    QtCore.QPoint(0, 0),    # pixelDelta
    QtCore.QPoint(0, 200),  # angleDelta
    QtCore.Qt.NoButton,     # buttons
    QtCore.Qt.NoModifier,   # keyboard modifiers
    QtCore.Qt.ScrollUpdate, # phase
    False,                  # inverted
)

# Using Qt4 API
print("Using Qt4 API with QPoint")
event = QtGui.QWheelEvent(
    QtCore.QPoint(0, 0), QtCore.QPoint(0, 0), *qt4_args
)

# Using Qt5 API...
print("Using Qt5 API")
event = QtGui.QWheelEvent(
    QtCore.QPointF(0, 0), QtCore.QPointF(0, 0), *qt_args
)

# Using Qt4 API... this segfault!
print("Using Qt4 API with QPointF, this segfault")
QtGui.QWheelEvent(
    QtCore.QPointF(0, 0), QtCore.QPointF(0, 0), *qt4_args
)

Running this script prints the following:

Using Qt4 API with QPoint
Using Qt5 API
Using Qt4 API with QPointF, this segfault
Segmentation fault: 11

Note that in PySide6/Qt6, the backward compatibility layer for Qt4 is removed, so we will have to update these tests again. But I will defer that to a separate PR.

Original message:

I started with trying PySide2 from EDM. Previously segfaults were observed in Pyface (enthought/pyface#765), and I can also reproduce segfault quite consistently on my OSX machine. This is marked as draft so that I can verify the behaviour on CI.
Then I will try to use PySide2 from pip.

[kitchoi]

Yep, got a segfault, at the same test as I have observed on my machine:

test_vertical_mouse_wheel (enable.tests.qt4.test_mouse_wheel.MouseWheelTestCase) ... �[31m�[1mCommand "python -W default -mcoverage run -m unittest discover enable -v" exited with segmentation fault (-11)�[0m

Travis CI log:
enable-segfault-pyside2.txt

I tried PySide2 from PyPI locally and the segfault happens at the same test.
Will investigate.

[kitchoi]

The tests causing the segfault are now fixed. Ready for review.

[rahulporuri]

LGTM