Table of Contents
The purpose of this project is an example application of ALA (Abstraction Layered Architecture).
It's not about what the application itself does, it's about how all the code mechanics of an application conforming to this architecture work.
An ALA application expresses user stories directly, but in an executable way. It often uses a diagram because user stories themselves contain a network of relationships.
Consider the diagram below, which is the application's source code. (You can click on the diagram to get a clearer view.)
It describes the user stories of a small desktop application.
An ALA diagram expresses the user stories in terms of instances of domain abstractions, which are the boxes in the diagram. They are wired together by ports, which are programming paradigms implemented as interfaces. An ALA diagram is polyglot in programming paradigms, so some wires represent an event-driven programming paradigm, some represent a UI layout programming paradigm, some represent a data-flow programming paradigm, etc.
This is how the diagram works.
The yellow user stories of the diagram express the base UI layout and the Exit menu item causes the application to close.
The green user stories of the diagram sense if a device is connected to the COM port of the PC and displays "Searching for a device" or "Connected" in the statusbar. SCPSense knows how to send an SCP (Serial Command Protocol) commands to poll for the presence of a device. It does this every 3 seconds when it gets an event from the Timer, which starts when the MainWindow sends an event to start the application. These commands are routed to SCPProtocol, which knows how to add the protocol and serialize them. They are then routed to the COMPort, which knows how to send and receive characters through a hardware serial conenction. Also note that SCPSense uses an instance of an Arbitrator to ensure it has exclusive use of the device while using a command and waiting for a response. The device is considered a resource. When SCPSense gets a response from a device, it outputs a boolean data-flow signal which changes the message on the statusbar, and signals the sessions Grid to start.
The blue user stories display data from the device. When the sessions grid gets its start signal, it requests data on a table-data-flow paradigm. SCPSessions knows how to get columns and rows of data from devices that describe the session files in that device. When the user clicks on one of the session rows, it outputs the index on its rowSelected port, which is wired to SCPSessions. SCPsession knows how to to tell the device to select a specific session. An event is also routed to the data Grid to tell it to start. Note that these two Grids are instances of the same domain abstraction configured with two different instance names. The data Grid now requests data using the table-data-flow programming paradigm, which will come from SCDData. SCPData knows how to get columns and rows of data from a single session in the device.
The brown user stories of the diagram get data off the device and write it to a CSV file. There is an "Import from device" menu item in the File menu, which when clicked outputs an event. This event opens up a Wizard which contains RadioButtons for selecting where the data is to go. When the user makes his selection, the relevant RadioButton outputs an event which opens a SaveFileBrowser. When the filepath is selected it informs a CSVFileReaderWriter. It also causes an event to go to a Transfer domain abstraction that knows how to pull all the rows and columns from a table-data-flow, and push it to another table-data-flow. table-data-flow supports both pull and push modes, but does so in small batches in case there is a lot of data.
Question for thought: How many applications could you read the source code and understand it like we have just done right here in the readme? This diagram is not just documentation, it is source code. In this case we have hand translated the diagram into readable code which you can see in Application.cs. We have done this to connect the dots on how we get the diagram to execute without any magic. However, if the application diagram gets very large, we would want to make things even easier by auto-generating the wiring code.
To see the diagram executing, you can just download and execute the solution in Visual Studio. You will be able to use the application as shown by the gif below. (Not shown in the diagram is that we have wired in a software simulation of a real device to the COM port).
Some knowledge of ALA itself is needed to understand the architecture of the code. A brief explanation of ALA can be read here (Abstraction Layered Architecture.md). An in-depth explanation of ALA and theory can be found in the web site http://www.abstractionlayeredarchitecture.com.
Regardless of the theory behind the organisation of the code into folders, the code itself is really quite simple.
The Application folder contains application-diagram.drawio (the source) and its hand translation into code, application.cs.
The constructor in Application.cs instantiates classes in the DomainAbstractions folder. It configures them with any details from inside the boxes on the diagram via constructor parameters or properties. Then it wires them together using interfaces in the ProgrammingParadigms folder. When a pair of objects is wired, one object has a field of the type of the interface and the other implements the interface. On both instances these are called ports.
The wiring is done by a WireTo operator, which is an extension method in the Foundation folder. WireTo uses reflection to find matching ports. It then assigns the second object, casted as the interface, to the field in the first object. This is like dependency injection without using constructor parameters or setters. The dependency injection is directed by the diagram. We say like dependeny injection, because they are not really dependencies. The domain abstractions do not have dependencies on each other in any way, not even a abstract base class or an API like interface that can have multiple implementations. They only depend on their own ports. The actual dependency is on an abstract interface which we think of as at the abstraction level of a programming paradigm.
Once all the instances of domain abstractions are wired according to the diagram, they can communicate at run-time because they have ordinary fields with references to each other. For example IDataflow is used everywhere that a piece of data needs to be pushed from one instance to another in the diagram. One example is the wire connecting the instance of a SaveFileBrowser to the instance of a CSVFileReaderWriter. When the SaveFileBrowser has the filepath from the user, it pushes it out via its port called simply "output", which has type IDataflow. It arrives at the CSVFileReaderWriter because CSVFileReaderWriter implements IDataflow as its port.
The IUI interface in the programming paradigms folder is used to wire parent UI elements to their children. The IUI interface has a method that asks each child for its underlying WPF (Windows Presentation Foundation) widget. So after the wiring code has all run, and MainWindow is told to run, the first thing it does is to ask its children for their WPF objects. Those children do the same, and so the WPF tree structure is built according to the wiring of the diagram.
After that is done, the WPF UI starts responding to user events, but also MainWindow outputs a appStart event which is wired to an instance of a domain abstraction called Timer, which then generates events every 3 seconds. These are routed by the diagram to SCPSence.
Peruse the code to see how each of the programming paradigm interfaces work, and how some of the domain abstractions work. You will find the domain abstractions are like independent programs to read and understand because they depend only on their own ports.
You may notice that some of the programming paradigms have a class called Connector as well as an interface. Connectors solve a range of wiring scenarios such as fanout, multiple ports of the same interface type (C# doesn't allow implementing the same interface twice, even though it makes perfect sense to do so in ALA), and some other situations.
- Clone this repository or download as a zip.
- Open the solution in Visual Studio 2019 or later
- When the application runs, you will see data already loaded from a simulated real device.
- Click on the row on the left, and it will show data from different sessions on the right.
- Click on File, Import from Device, to open a Wizard, then select Local CSV File, then enter a filename to save data to a CSV file.
- Click on the Toolbar icon to do the same thing.
C#, Visual Studio 2019
GALADE (Graphical ALA Development Environment) see Future work below. A GALADE version is a work-in-progress in a branch of this repository.
Contributions are what make the open source community such an amazing place to be learn, inspire, and create. Any contributions you make are greatly appreciated.
- Fork the project using the button at top right of the main Github page or (https://github.com/johnspray74/ALAExample/fork)
- Create your feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -am 'Add AmazingFeature') - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
Swift, Java, Python and Rust versions needed.
We have a graphical tool called GALADE (Graphical ALA development environment) written by Arnab Sen here https://github.com/arnab-sen/GALADE.
GALADE automatically generates code of course. GALADE also features automatic layout when drawing the diagram so that the developer can concentrate on expressing user stories and inventing the needed abstractions - a feature that we find critcally important because most drawing packages ask you to spend too much effort on layout. It also presents the boxes representing instances of domain abstractions with their configuration constructor parameters and properties ready to be filled in. It also shows all the ports ready to be wired. Importantly, it acheives this by reading the code in the classes in the DomainAbstractions folder. This is really powerful integration with your C# code, and supports round trip of both the application wiring code and the domain abstractions code.
We would love help to further develop GALADE to support ALA development, for example, a routing algorithm for its wiring, or to show errors for illegal dependencies according to the ALA fundmental rules.
We would also like to see other graphical tools based on Visual Studio and Eclipse graphical environments to generally support ALA and specifically to generate the wiring code from the diagram.
ALAExample is a cut-down example of an application used by farmers to get data to/from their EID readers, livestock weighing devices, etc.
The full Windows PC desktop application was a research project for a software reference architecture called ALA (Abstraction Layered Architecture). ALA is theoretically optimized for the maintainabilty quality attribute by telling you how to organise code. This research was to measure if this is true in paractice.
The project was done by a masters student and internship students over two internships. By using the architecture, they were able to write quality code to replace a legacy degenerate application with approximately 2 man-years of effort compared with approximately 12 man-years of effort for the legacy application. To give an idea of size, the ALA version contains 55 KLOC and the legacy version 70 KLOC for approximately the same functionality.
Rosman Cheng, John Spray, Roopak Sinha, Arnab Sen
John Spray - [email protected]
This project is licensed under the terms of the MIT license. See License.txt
