Debugging and testing

Running widgets as scripts

To run a widget without canvas - for debugging and for nitpicking about the GUI - the widget module must be executable as a script. This is handled by WidgetPreview. It is typically used as follows

if __name__ == "__main__":

where OWMyWidgetName is the widget’s class.

We can also pass the data to the widget. For instance,

if __name__ == "__main__":

passes the Iris data set to the widget. Passing data in this way requires that there is a single or default signal for the argument’s data type. Multiple signals can be passed as keyword arguments in which the names correspond to signal handlers:

if __name__ == "__main__":
    data ="iris")

If the signal handler accepts multiple inputs, they can be passed as a list, like in the following method in the Table widget.

if __name__ == "__main__":
        [(Table("iris"), "iris"),
        (Table("brown-selected"), "brown-selected"),
        (Table("housing"), "housing")

Preview ends by tearing down the widget and calling sys.exit with the widget’s exit code. This can be prevented by adding a no_exit=True argument. We can also prevent showing the widget and starting the event loop by using no_exec=True. This, together with some previewers method described below, can be used for debugging the widget. For example, OWRank’s preview,

if __name__ == "__main__":
    from Orange.classification import RandomForestLearner
        set_learner=(RandomForestLearner(), (3, 'Learner', None)),

can be temporarily modified to

if __name__ == "__main__":
    from Orange.classification import RandomForestLearner
    previewer = WidgetPreview(OWRank)""), no_exit=True)
        set_learner=(RandomForestLearner(), (3, 'Learner', None)))

which shows the widget twice, allows us a finer control of signal passing, and offers adding some breakpoints.

Unit-testing Widgets

Orange provides a base class WidgetTest with helper methods for unit testing.