Real Objects vs. Data Containers

After more than three years of working as backend (and mobile) programmer, mostly in the Java Virtual Machine ecosystem, I have realized that no one of those procedural MVC/MVP/MVVM patterns has made me feel comfortable when implementing new features or big changes in a project. Also, ER-ending classes (Controller, Manager, Helper, etc.), which are well-accepted and used in many frameworks, don’t help with that either: they will get bigger and bigger, and you will have to segregate them without any logical criteria. And when that happens, you’re screwed. Maintainability becomes really hard.

In the last few months, I have been heavily influenced by Yegor Bugayenko and what he claims is “pure” object-oriented programming. Instead of treating objects as simple (and silly) data containers/data structures, we should give them the power and trust them. Convert them into real objects.

Can you see any difference between this Java code:

public class User {
    private String id;
    private String username;
    public String getId() {
        return id;
    }
    public void setId(String id) {
        this.id = id;
    }
    public String getUsername() {
        return username;
    }
    public void setUsername(String username) {
        this.username = username;
    }
}

And this C code:

struct User{
    char  id[50];
    char  username[50];
};

Not at all. Because there is none. However, we write objects like above and are convinced we are object-oriented developers. But we aren’t. We must think of objects as something more than simple in-memory data structures exposing its state to everybody. Objects should wrap its real-life representation. 

Let’s say we need to write a class File (in Java) with a method, content(), that returns its content in a byte array. A first implementation could be something like this:

public class File {
    private String path;
    private byte[] content;
    public File(String path, byte[] content) {
        this.path = path;
        this.content = content;
    }
    //getters and setters
}
public class FileSystem {
    public byte[] getFileContent(String path) {
        return /* read file bytes from disk */;
    }
}

And we would use it this way:

FileSystem fs = new FileSystem();
String path = "/tmp/conf.xml";
File confFile = new File(path, fs.getFileContent(path));

Done.

To be honest, I think this implementation is a mess. Don’t you?

If we change the content of file /tmp/conf.xml, the configFile object becomes inconsistent. The File class is not representing a real file, but a bunch of bytes. It has been reduced to a data container. A C struct. It can’t be considered an object at all.

This is a nice implementation for the same class:

public interface File {
    byte[] content();
 }
public final class FileSystemFile implements File {
    private final String path;
    public FileSystemFile(String path) {
        this.path = path;
    }

    @Override
    public byte[] content() {
        return /* read file bytes from disk */;
    }
}

And we use it this way:

File confFile = new FileSystemFile("/tmp/conf.xml");

Much better, right? Now, configFile is a real object. We have converted Filetype in an interface. This way, we can have more implementations like RemoteFile, if needed.

You may think that this implementation is not efficient because each time we call the content() method, a disk read is performed. It may be slow, depending on the system and the application requirements. You are right. So now,  object composition and the decorator design pattern come into action.

We are going to create a class, CachedFile, that wraps a File and caches its content, so just one disk read will be performed:

public final class CachedFile implements File {
    private final File origin;
    private byte[] cached;
    public CachedFile(File origin) {
        this.origin = origin;
        this.cached = null;
    }
    @Override
    public byte[] content() {
        if (this.cached == null) {
            this.cached = origin.content(); //real disk read
        }
        return this.cached;
    }
}

(This is a very naive cache implementation. Also, you must never use null. I did it just for the example.)

And we use it this way:

File confFile = new CachedFile(new FileSystemFile("/tmp/conf.xml"));

It looks great for me!

What if I tell you that this approach can be applied to database objects, too?

Let’s go back to the User example. A real, stored in-database User would look like this:

public interface User {
    public String id();
    public String username();
}
public DatabaseUser implements User {
    private final String id;
    private final Database db;
    public DatabaseUser(String id) {
        this.id = id;
        this.db = db;
    }
    @Override
    public String id() {
        return this.id;
    }
    @Override
    public String username() {
        return /* SELECT username FROM users WHERE id = this.id */;
    }
}

Now, a User is not just a bunch of data. Its state is not being exposed. It is a real object. Again, we can “decorate” it with a cache or whatever we want for throughput optimization.

If you liked it this approach of OOP, I strongly recommend you to read Yegor’s posts and his book Elegant Objects.