Yesterday we migrated Playing With Wire from our old FreeBSD Jail to a brand new VMware Virtual Server. The new server has a much faster CPU and much more RAM, which we hope will decrease the load-time.

The migration itself went smooth without any downtime for PWW. If you discover something that appears to be missing or malfunctioning, please notify us at admin@wireload.net.

We had a couple of hours of downtime this morning as a new diesel generator was installed at Playing With Wire’s hosting location. Sorry about that.

SwtCallback 0.1 was released only a little more than a week ago, but that’s not stopping us from releasing 0.2 today! In line with the original philosophy, SwtCallback has been extended to again reduce clutter and deep nesting in your Java code. This time SwtCallback attacks the Runnable interface.

As I wrote in my release post about SwtCallback, SwtCallback is a java library for SWT applications which adds an alternative way to handle events: callbacks. Instead of clunky listener interfaces or messy anonymous classes, SwtCallback enables one line solutions that make it easy to keep your event handling code in a single flat class – this while still being more readable and (surprisingly) improving encapsulation in some cases.

This latest release extends this functionality from just listeners to anything that takes a runnable. Prime examples in SWT are display.syncExec(), display.asyncExec() and timers.

For example, maybe you wish to refresh some UI components to reflect a change in data. But you’re not on the SWT thread so you have to use syncExec. Normally, you’d do something like,

display.syncExec(new Runnable() {
  public void run() {
    doUpdateSwtLabels();
  }
});

With a Callback you can turn this into one highly readable line:

display.syncExec(new Callback(this, "doUpdateSwtLabels"));

The most useful application may be for timers. In SWT, a timer needs to be reestablished at the end of timer execution if you wish it to recur. You may end up with unwieldy code such as,

void initWidgets() {
  /* ... */

  // Prime the timer.
  display.timerExec(INITIAL_DELAY, new Runnable() {
    public void run() {
      refresh();
    }
  }); 
}

void refresh() {
  /* ... update stuff ... */
  
  // Set the timer again.
  display.timerExec(TIMER_DELAY, new Runnable() {
    public void run() {
      refresh();
    }
  }); 
}

With a Callback you can now do this:

void initWidgets() {
  /* ... */

  // Prime the timer.
  display.timerExec(INITIAL_DELAY, 
    new Callback(this, "refresh")); 
}

void refresh() {
  /* ... update stuff ... */
  
  // Set the timer again.
  display.timerExec(TIMER_DELAY, 
    new Callback(this, "refresh")); 
}

This is much easier to read and follow, at least in my personal opinion. If you agree, go ahead and grab the download below – SwtCallback is released under the BSD license so you should be able to integrate it with whatever you’re working on.

Download the source and binary here:

Download SwtCallBack. More information is available here.

There’s some extra javadoc in this release to make it easier to get started. Still, there isn’t much in the way of documentation – we’ll go back and write some if there’s demand for it.

           

Today WireLoad is releasing SwtCallback, a Java library for SWT that enables callbacks for event handling. SwtCallback is inspired by the Swing enhancer Buoy by Peter Eastman.

If you know you want this, just skip to the bottom of this post for the download link. Otherwise, read on and I’ll explain the reason we wrote this library.

I’m a big fan of readable and concise code. Another thing I’m a big fan of is user interfaces that look and feel like what users expect. Given this, I’ve always had an aversion to UI programming in Java. Because I think it’s very telling, I’ll let this flash video illustrate: totally gridbag. Bottom line is that in Java, user interfaces look weird and often attempts to alleviate this weirdness result in enormous amounts of hard to read code.

One solution is to use the Standard Widget Toolkit (SWT) instead. SWT enables easier UI building with resulting UIs that are both faster and look and feel better for the end user. What’s more, the code is more concise thanks to FormLayouts, sane widget sizing and the availability of standard functionality.

There’s one thing I think can be improved on though: readability. SWT code has a tendency to, like Swing, become littered with hundreds of listener classes, or worse, faceless listener implementations. For example, your code may look like this:

public class Demo implements SelectionListener {

...
   void initialize() { 
      Button myButton = new Button(shell, SWT.NONE);
      myButton.setText("Click Me"); 
      myButton.addSelectionListener(this);
   }
...

   public void widgetDefaultSelected(SelectionEvent e) {
      // I don't care about this event but I have to 
      // have a handler to satisfy the interface.
   }

   public void widgetSelected(SelectionEvent e) {
      // Do something
   }
}

Notice how the method named ‘widgetSelected’ has a very generic name: it has to since that’s the name it was given in the SelectionListener interface. Also, if there are many widgets that listen for selection events, then this single method has to handle all of them with a big if statement or some such. Not very readable. We also had to thumb on conciseness by adding one method which we didn’t even want: widgetDefaultSelected.

A better method is to associate the handler with only a single widget by using an inner class, and to use an adapter to avoid having to add a dummy method. This would typically look like this:

Button myButton = new Button(shell, SWT.NONE);
myButton.setText("Click Me"); 
myButton.addSelectionListener(new SelectionAdapter() {
   public void widgetSelected(SelectionEvent e) {
      // Do something
   }			
});

Much better. Less code, easier to read and it’s clear what ‘do something’ is associated with. But now we end up with a bunch of anonymous classes and deeply nested code. It’s still not perfect.

Enter SwtCallback. With SwtCallback, our final version of the program may look like so:

void initialize() { 
   Button myButton = new Button(shell, SWT.NONE);
   myButton.setText("Click Me"); 
   Callback.add(myButton, SWT.Selection, this, 
      "myButtonSelected");
}

/**
 * Called when myButton is selected.
 */
@SuppressWarnings("unused")
private void myButtonSelected() {
   // Do something
}

This has everything we wanted: it’s clear what the ‘myButtonSelected’ method does because it’s well named. There are no listeners to implement and no inner class bloat. The definition of myButton is short without being complicated or hard to understand. As an added bonus SwtCallback allowed us to make the event handler method private – this makes sense since the method is only relevant for the corresponding UI class.

UI programmers with experience outside of Swing will hopefully immediately recognize and appreciate the idea of callbacks. If you still need more reasons though, I’ll suggest a couple:

• Less code means fewer bugs and easier maintenance. Callbacks allow you to handle events without lots of extra classes or big if-else statements.
• You can link many events to the same handler. For example, maybe you have a ‘Quit’ button in a window. This button can have the same handler as the standard close event for the window.
• No need for dummy methods. Since you’re not implementing listener interfaces you don’t have to implement handlers for events you don’t care about.
• Natural names for methods. You’re not constrained to some standard name for an event handler. You can name your method ‘handlePasswordFieldGainedFocus’ if you want to.
• Private or protected event handlers. With interfaces your event handling methods have to be public. Most of the time they’re very specific to some UI classes. It doesn’t make sense for other unrelated classes to call your handlePasswordFieldGainedFocus() method.

If you still need more, Peter Eastman, the author of Buoy, expresses the rationale for callbacks much better than I ever could here.

SwtCallback is released under the BSD license. Download the source and binary here:

Download SwtCallBack. More information is available here.

Patches and comments are welcome. Apologies for the lack of documentation – if there’s interest in this library we’ll revisit it. For now, if you find it useful, great!

Update 1: SwtCallback 0.2 has been released.

           

How many times have you made a change to a server to only a couple of days or weeks later discover the change you did broke something? It happened to me a bunch of times. Sure, call me irresponsible, but sometimes I’m in a rush and don’t have time to document the change and write nice comments about it. I try to do so, but sometimes I forget.

For quite some time I’ve thought about a simple way to track the changes to my servers, but without coming up with any perfect solution. My ideas have included everything from wikis to blog-style logging to Tripwire. Sure, a change-blog would work, but it would take quite some extra time to write the entry.

One day when I was working with our development tracker a quite brilliant solution stroke me: why not use CVS/SVN on /etc? It sounds like overkill, but it would do the trick. By using this you would be able to:

* Display when the changes were made
* See what change was made
* Revert to a previous version

There are two different ways to go along using a repository for /etc. One way is to create a CVS/SVN tree on a different server and sync the entire /etc directory to that repository. The benefit with this is that you could have the repository on a different server and be to collect all your servers /etc on the same repository. By combining this with a web-interface, one would be able to easily compare and contrast the configurations of the different servers. Moreover, the change logs would be kept in the repository, and not on the individual servers, creating a cleaner file-structure. Another benefit would be that you could easily sync the /etc filesystem with your local workstation and modify/create your config-files there. The drawback with this solution is that you need to make your local changes first, and then upload the files to the repository. Once you’ve uploaded the files from your local workstation you can go ahead and sync the actual server with the repository to commit the changes.

Another solution would be to use a custom file-system. There are two different filesystem that supports version-tracking: CVSFS and Wayback. Between these two alternatives, Wayback appears to be the most developed alternative. Worth noting though is that both these alternatives only targets the Linux-platform.

The benefit for using such approach is that you don’t need to set up a separate repository to keep the files in, as well as that it enables you to make changes in a traditional way on the server. The drawback is that neither of these two filesystems are very well maintained. It appears that both these programs were last updated back in 2004, rendering them both quite obsolete. The question is if you really want to put the faith of your /etc-folder in a program that no one longer maintain.

Regardless of which one of these two alternatives you find most appealing, the concept is very interesting. I have still not tried to use this myself, but I would certainly argue that having a /etc-folder that allows you to keep track of changes is far superior to the traditional way of keeping /etc.

Edit 1: Since the article was published, I’ve found two more file-systems worth mentioning: CopyFS and ext3cow. Both these filesystems appears to be useful for /etc usage. Out of these two alternatives, ext3cow appears to be more sophisticated. However, CopyFS appears to be easier to set up.