GUI Design Guidelines

Summary

This page describes guidelines that should be followed when implementing an interface in MantidPlot. The aim is to encourage a consistent approach to developing interfaces.

MVP (Model View Presenter)

GUIs in Mantid aim to use the MVP pattern. The MVP pattern is a generic concept for how to structure GUI code. MVP allows components of the GUI to be tested separately and automatically. It also allows for greater flexibility. Decoupling the model and view means that if the developer wants to experiment with, for example, a different GUI toolkit, or a different method of doing their calculations, then it is easy and safe to swap out components. A description of each component is given below.

To illustrate MVP, a simple example of a calculator GUI has been created using Python (the concepts of MVP can be applied to any programming language). This example can be found in MVP Calculator GUI Example, and you can run it with python Calculator.py.

It is good practice to have model, view or presenter (as appropriate) at the end of the name for each file (e.g. FFTView, FFTModel, FFTPresenter), and each component should be a class in its own right. Within the MVP pattern the model and view never exchange any information directly.

Model

The model is where the ‘hard sums’ take place within GUI. Any Mantid algorithms should be run in the model, as well any other calculations that may need to be performed.

It is possible that a presenter may have multiple models. For example if two GUIs require the same calculation (e.g. mean) but not all of the model (one GUI may need standard deviation and the other the median), then it would be sensible for there to be three models (with the mean model being shared). This prevents code duplication and makes maintenance easier.

It is important to note that the values used in the calculations should be received from the presenter (more of which below).

View

The view determines the look of the GUI. In passive-view MVP, there will generally be very little logic in the view. A view should define the following sections:

  • The look of the GUI (often this will be defined in a Qt .ui file instead)
  • Get methods to return values from the widgets (text input, checkbox etc)
  • Set methods to update the output from the GUI (eg. plot some data, fill in some text boxes)

A view will probably also contain connections. A detailed explanation of signals and slots can be foud here. Briefly, a widget may emit signals. For example QPushButton emits the signal clicked when it is clicked. In order to handle the button being clicked, the view will implement a slot method. This method does whatever we need for a button click. To ensure that this method is called whenever the button is clicked, we connect the clicked signal of our button to the handleButtonClick slot of our view.

The view should have a parent - this will be the widget containing it. An example of a parent would be a main window containing tabs - the children of the main window would be the tabs, and the children of the tabs would be the widgets contained within the tabs.

Presenter

The presenter acts as a ‘go-between’. It receives data from the view, passes it to the model for processing, receives it back from the model and passes it to the view to be displayed to the user. The presenter generally should contain relatively simple logic (though it will be more complex than the view).

The model and the view are stored as members of the presenter class. These should be passed into the presenter at initialisation.

It is important to note that the model and view should have as little access as possible to the presenter. Presenter-model communication is simple - the presenter generally just calls methods on the presenter. Presenter-view communication is slightly more involved. There are two ways of doing it:

  • Connections - the presenter may contain connections as well as the view. You may choose to define custom signals in your view, such as a plotRequested signal to announce that the user has asked to plot some data, probably by clicking a button. The presenter will need to implement a slot (let’s call it handlePlotRequested) to handle this, which gets the relevant data from the model and passes it to the view. We then need to connect the signal to the slot in the presenter’s constructor. It is also possible for a signal emitted by a view to be caught in the presenter of a parent view. In order to communicate by connections using Qt in C++ the presenter must inherit from QObject. It’s generally considered good practice to avoid having Qt in the presenter, so this method works best for GUIs written in Python (or another language with a more relaxed type system).
    • Note that is good practice to handle all signals in the presenter if you can, even if it is possible to just handle them in the view. This is because by going through the presenter we can unit test the handling of the signals.
  • Notify - the presenter may instead allow the view to ‘notify’ it. This can be achieved by implementing a set of possible notifications (in C++ an enum class works well) and a method notify(notification) on the presenter. In the above example, handlePlotRequested is still needed, but now notify invokes it whenever it is passed a plotRequested notification. This method requires the view to have a pointer to the presenter, which introduces a circular dependency and leaks information about the presenter to the view. The leak can be resolved by having the presenter implement an interface which exposes only the notify method, and having the view keep a pointer to this.

Doing presenter-view communication with connections is the cleaner of the two, so this method should be used unless writing a GUI in C++. You’ll notice that, in both cases, the view never passes data (for example, the input from a text box) directly to the presenter, instead it justs tells the presenter that something needs to be done. In passive-view MVP the presenter, in handling this, gets any data it needs from the view using the view’s get methods.

Testing MVP Components

MVP allows us to write automated tests for a large amount of the GUI. We can write independent tests for the presenter and model, but usually not the view (for this reason, the view should be as simple as possible, ideally containing no logic at all).

Mocking is very useful tool for testing the presenter. Mocking allows us to return a predefined result from a method of either the view or the model.

It is useful to mock out the model because, providing that we’ve written adequate tests for it, we don’t care what the output is in the tests for the presenter - we just care that the presenter handles it correctly. The model may perform time-consuming calculations, such as fitting, so by returning a dummy value from the fitting method we cut down the time our tests take to run. We can also potentially change how the model works - if the GUI uses an algorithm which undergoes some changes, such as applying a different set of corrections, the tests for the presenter will be unaffected.

It’s useful to mock out the view because we don’t want to have to manually input data every time the unit tests are run - instead we can mock the get methods to simulate the user entering data.

Using GMock in C++, or unittest.mock in Python, we can set expectations in the unit tests for certain methods to be called, and with certain arguments.

Visual Design

Qt Designer

The layout of all interfaces and reusable widgets should be done by using the Qt’s Designer tool. This has several advantages:

  • immediate visual feedback of what the widget/interface will look like
  • far easier to maintain, e.g. moving a control is a simple drag and drop
  • reduces the amount of hand-written code required

If it is felt that the design must be hand coded then this should be discussed with a senior developer.

Reusable Widgets

Many interfaces will require similar functionality. For example, the ability to enter a filename string to search for a file along with a ‘Browse’ button to select a file from the filesystem. This type of behaviour should be captured in a new composite widget that can be reused by other components.

The new widget should be placed in the MantidWidgets plugin and a wrapper created in the DesignerPlugins plugin so that the new widget type can be used from within the Qt Designer.

The current set of reusable items are:

Class Name Parent Class Abiltity
AlgorithmSelectorWidget QWidget A text box and tree widget to select an algorithm
CatalogSearch QWidget An interface interface to the catalog system
CatalogSelector QWidget Displays the available catalog services
CheckBoxHeader QHeaderView Enables checkboxes to exist in the table header
ColorBarWidget QWidget Show a color bar that can accompany a colored bidimensional plot
DataSelector MantidWidget A box to select if input is from a file or workspace along with the appropriate widget to choose a workspace or file.
DisplayCurveFit MantidWidget A plot to display the results of a curve fitting process
FindReplaceDialog QDialog A dialog box to find/replace text within a ScriptEditor
FitPropertyBrowser QDockWidget Specialisation of QPropertyBrowser for defining fitting functions
FunctionBrowser QWidget Provides a wiget to alter the parameters of a function
InstrumentSelector QCombobox A selection box populated with a list of instruments for the current facility
LineEditWithClear QLineEdit A QLineEdit with a button to clear the text
MessageDisplay QWidget Display messages from the logging system
MWRunFiles MantidWidget Provides a line edit to enter filenames and a browse button to browse the file system
MWView QWidget A colored, bidimensional plot of a matrix workspace
ProcessingAlgoWidget QWidget A composite widget that allows a user to select if a processing step is achieved using an algorithm or a Python script. It also provides a script editor.
ScriptEditor QsciScintilla The main script editor widget behind the ScriptWindow
WorkspaceSelector QComboBox A selection box showing the workspaces currently in Mantid. It can be restricted by type.

Icons

Icons are a contentious subject as they can in some cases cause more confusion and hinder more than they help. The NHS came up with a good set of rules for what icons should be used and this could be useful to designers, check out this article.. It may fit a situation more to have a text button instead of an icon.

Whilst having too many icons will confuse the average user there are cases where many cases where it would help, for example if a button does a similar thing to another button somewhere else in the program then it should have the same icon. Have a look to see if the need you has an icon in Mantid by look at this handy Mantid Icon Table.

Python

Interfaces can also be created in Python using the qtpy package. The code for the interface should be placed in a Python package under the Code/Mantid/scripts directory. It should be named after the interface name (without spaces). The code within the package should be structured to avoid placing all of the code in a single file, i.e. separate files for different classes etc. Sub packages are recommended for grouping together logical sets of files.

For the interface to appear from within MantidPlot create a startup python file under the Code/Mantid/scripts directory. Assuming the code for the interface is in a directory called foo_app then the startup file would look like:

from foo_app import FooGUI

app = FooGUI()
app.show()

where FooGUI is the MainWindow for the interface. Some more detailed documentation on creating GUIs in Python can be found at Qt Designer for Python.

Designer

As with the C++ GUI the Qt Designer should be used for layouts of all widgets and the main interface. It is recommended that the .ui files be placed in a ui subdirectory of the interface package. To generate PyQt code from the UI xml you will need to run the pyuic4 program that ships with PyQt4. It is also recommended that the output file is named, using the -o argument, ui_[widgetname].py and placed in the ui subdirectory.