This model-view-viewmodel framework is intended for large Qt based scientific applications written in C++. Project was created as a playground toward GUI refactoring of BornAgain project.
Main features of the framework are:
- Application model to store arbitrary data of GUI session. Qt-independent.
- Serialization of application models to json.
- Undo/redo based on command pattern.
- View model to show parts of application model in Qt widgets. Depends on Qt.
- Scientific plotting based on qcustomplot.
- Automatic generation of widgets from model content.
- Property editors.
- Flexible layout of Qt's trees and tables.
- C++-17
- CMake 3.14
- Qt 5.12
git clone --recurse-submodules https://github.com/gpospelov/qt-mvvm.git
mkdir <build-dir>; cd <build-dir>
cmake <source>; make -j8; ctest
# run one of examples
<build-dir>/bin/collidingmice
If you are familiar with Qt's reach example section you might saw
it's funny collidingmice
example showing basics of
QGraphicsScene
.
To demonstrate the idea behind qt-mvvm
library the code of the example was
slightly modified. The mice data has been moved into the dedicated model,
the content of the model was shown both in the QGraphicsScene
and in QTreeView
.
It is possible now to save the application state in json file and later load
the session back with saved mice positions. Additionally, it is possible to go
back in time and watch how mice are moving in opposite directions by dragging a
slider:
The demo shows that qt-mvvm
library allows
to equip the GUI with the serialization and undo/redo and to provide proper
model/view relations via relatively small modifications to the original code.
Implementing similar features from the scratch in bare metal Qt would take
much more time and the resulting code wouldn't be easily transferable to another project.
This and other examples can be found in examples sub-directory of qt-mvvm package.
Qt, naturally, has model view to manage the data and its presentation. According to some, Qt is misusing the model view terminology and more correct naming would be
- Data -> Model
- Model -> ViewModel
- Delegate -> Controller
- View (that's ok)
Whatever the right terminology is, Qt doesn't tell much about the architecture of
complex applications, how to structure the Data
and where to put business logic.
Qt's model view is rather a way to adapt the user
data for Qt's trees and tables and adjust their behavior with delegates.
Given library is an attempt to understand how to deal with the application data.
The framework consists of three libraries: libmvvm_model.so
, libmvvm_viewmodel.so
, and libmvvm_view.so
.
libmvvm_model.so
defines tree-like structure SessionModel
to store
any data of the GUI session. This part of the framework
is intentionally made Qt-independent. The idea behind is the following:
In large Qt applications, the business logic gets quickly spoiled with presentation
logic. Qt classes like QModelIndex
start to appear everywhere,
even in places that have nothing to do with Qt graphics. Attempt to store GUI
session data in QAbstractItemModel
leads to inflexible layout in Qt trees and tables. Attempt to fix this with QAbstractProxyModel
leads to the appearance of
overwhelmingly complicated proxy models. Removing Qt from dependencies allows
focusing more on common needs (i.e. objects construction, property editing, etc) of GUI applications rather than on presentation details.
SessionModel
has a concept of properties, undo/redo,
serialization, and it's own minimal signaling to handle business logic.
Thus, the intention here is to build an application model to handle data and logic of GUI while being independent on any particular GUI library.
Strictly speaking,
libmvvm_model.so
still relies onQVariant
but eventually, it will be replaced with std::variant.
Second library, libmmv_viewmodel.so
, contains ViewModel
and serves
as a thin counterpart of SessionModel
in the Qt world.
ViewModel
doesn't own the data but simply acts
as a proxy
to different parts of SessionModel
.
It is derived from QAbstractItemModel
and intended to work together with Qt's trees and tables. The layout of ViewModel
(i.e. parent/child relationships) doesn't follow the layout of the original SessionModel. It is generated on the fly
by using strategy who-is-my-next-child
provided
by the user. In practice, it allows generating Qt tables and trees with arbitrary layouts, based on a common data source, without diving into the nightmare of
QAbstractProxyModel
.
Particularly, the aforementioned machinery allows having something in the line of the ancient Qt property browser framework.
Third library, libmmv_view.so
, contains few widgets for plotting and property editing.
- 20k loc of libraries (
libmvvm_model.so
,libmmv_viewmodel.so
andlibmmv_view.so
) - 15k loc of tests
- 10k of user examples
The library is intended for large GUI applications.
The definition of large
is quite arbitrary and means something
in the range 20k - 200k lines of code.
The main logic here is that using the additional library for smaller Qt applications is redundant, Qt has everything that may be required. If small GUI becomes
messy with time, it can always be refactored or even rewritten from scratch.
However, when the number of views to show the same data is getting large, and the GUI enters the range 20k - 200k, this is were a given library might help in proper separation of data, logic, and UI. When the GUI grows even further, well, developers of such large GUI know already what they need and probably have already implemented similar machinery.
The project is under active development.