Tag Archives: Java

Effective OpenAL with LWJGL 3

Effective OpenAL with LWJGL 3

Jesus Bloody Christ it’s been a while.

So, a lot of you are likely interested in developing on Java with LWJGL3 instead of LWJGL 2.9.*; as you should be. LWJGL3 has support for a lot of modern industry trends with older versions did not; such as multi-monitor support without back flipping through flaming hoops, or basically anything involving GLFW. It’s still in beta, I know, but it’s a solid piece of work and the team on it is dedicated enough to make it a reliable and standing dependency for modern projects.

Except for every now and then, when it happens to be missing some minor things. Or, more importantly, when there’s a dearth of documentation or tutorials on a new trick you’re pulling.

I can contribute, at least in part, to both of those.

OpenAL is the audio world’s equivalent to OpenGL; it’s a sophisticated and sleek interface to sound hardware. Many common effects and utilities, such as 3D sound, are built into it directly; and it interfaces sublimely with code already designed for OpenGL. Additionally, it’s also a very tight interface that does not take long at all to learn.

In the past, I would suggest using JavaSound for Java game audio, which is also a tight API, but it lacks these features. Most major audio filters have to be built into it rather directly and often by your own hand; and there’s no official guarantee of hardware optimization. However, what LWJGL3’s OpenAL interface now lacks can easily be supported by readily-present JavaSound features; such as the audio system’s file loader.

This entry is on, step by step, how one would do such a thing.

Let’s start with a basic framework. I’ve tried to keep a balance between minimal dependencies and staying on-topic, so I’ll suggest that you have both LWJGL3 (most recent version, preferably), and Apache Commons IO, as dependency libraries.

class Lesson {
    public Lesson() throws Exception {
        //Start by acquiring the default device
        long device = ALC10.alcOpenDevice((ByteBuffer)null);

        //Create a handle for the device capabilities, as well.
        ALCCapabilities deviceCaps = ALC.createCapabilities(device);
        // Create context (often already present, but here, necessary)
        IntBuffer contextAttribList = BufferUtils.createIntBuffer(16);

        // Note the manner in which parameters are provided to OpenAL...


        // Don't worry about this for now; deals with effects count

        //create the context with the provided attributes
        long newContext = ALC10.alcCreateContext(device, contextAttribList);
        if(!ALC10.alcMakeContextCurrent(newContext)) {
            throw new Exception("Failed to make context current");
        //define listener
        AL10.alListener3f(AL10.AL_VELOCITY, 0f, 0f, 0f);
        AL10.alListener3f(AL10.AL_ORIENTATION, 0f, 0f, -1f);
        IntBuffer buffer = BufferUtils.createIntBuffer(1);
        //We'll get to this next!
        long time = createBufferData(buffer.get(0));
        //Define a source
        int source = AL10.alGenSources();
        AL10.alSourcei(source, AL10.AL_BUFFER, buffer.get(0));
        AL10.alSource3f(source, AL10.AL_POSITION, 0f, 0f, 0f);
        AL10.alSource3f(source, AL10.AL_VELOCITY, 0f, 0f, 0f);
        //fun stuff
        AL10.alSourcef(source, AL10.AL_PITCH, 1);
        AL10.alSourcef(source, AL10.AL_GAIN, 1f);
        AL10.alSourcei(source, AL10.AL_LOOPING, AL10.AL_FALSE);
        //Trigger the source to play its sound
        try {
            Thread.sleep(time); //Wait for the sound to finish
        } catch(InterruptedException ex) {}
        AL10.alSourceStop(source); //Demand that the sound stop
        //and finally, clean up



The beginning is not unlike the creation of an OpenGL interface; you need to define an OpenAL context and make it current for the thread. Passing a null byte buffer to alcOpenDevice will provide you with the default device, which is usually what you’re after. (It is actually possible to interface with, say, multiple sets of speakers selectively, or the headphones instead of the speaker system, if you would like; but that’s another topic.)

Much like graphics devices, every audio device has its own set of capabilities. We’ll want a handle on those, as well. It’s safe to say that if a speaker can do it, OpenAL is capable of it; but not all speakers (or microphones) are created the same.

After this, OpenAL will want to know something of what we’re expecting it to manage. Note that it’s all passed over as a solid int buffer. We’re providing it with a notion of what features it will need to enact, or at least emulate; with a sequence of identifiers followed by parameters, terminated with a null. I haven’t begun to touch all that is possible here, but this attribute list should be enough for most uses.

After that, create the context, make it current, check to see that it didn’t blow up in your face, and register the capabilities. (Feel free to play with this once you’ve got the initial example going.)

So, before I get to the part where JavaSound comes in, let’s start with the nature of how OpenAL views sound. Sound, in its view, has three components: a listener, a source, and an actual buffer.

The listener would be either you or your program user; however, the program would want to know a little about your properties. Are you located something to the left or right? Are you moving (or virtually moving)? I usually set this first as it is likely to be constant across all sounds (kind of like a graphics context).

Next, we have a method of my own creation that builds and registers the audio file. Forgive me for the delay, but that’s where JavaSound’s features (in the core JKD) come in, and I’m deferring it to later in the discussion. You will note that the audio buffers have to be registered with OpenAL; as it needs to prepare for the data. There’s a solid chance that you will have sound-processor-local memory, much like graphics memory, and it will have to be managed accordingly by that processor before you can chuck any data at it.

Let’s look at that audio buffer creator.

     private long createBufferData(int p) throws UnsupportedAudioFileException, IOException {
        //shortcut finals:
        final int MONO = 1, STEREO = 2;
        AudioInputStream stream = null;
        stream = AudioSystem.getAudioInputStream(Lesson3.class.getResource("I Can Change — LCD Soundsystem.wav"));
        AudioFormat format = stream.getFormat();
        if(format.isBigEndian()) throw new UnsupportedAudioFileException("Can't handle Big Endian formats yet");
        //load stream into byte buffer
        int openALFormat = -1;
        switch(format.getChannels()) {
            case MONO:
                switch(format.getSampleSizeInBits()) {
                    case 8:
                        openALFormat = AL10.AL_FORMAT_MONO8;
                    case 16:
                        openALFormat = AL10.AL_FORMAT_MONO16;
            case STEREO:
                switch(format.getSampleSizeInBits()) {
                    case 8:
                        openALFormat = AL10.AL_FORMAT_STEREO8;
                    case 16:
                        openALFormat = AL10.AL_FORMAT_STEREO16;
        //load data into a byte buffer
        //I've elected to use IOUtils from Apache Commons here, but the core
        //notion is to load the entire stream into the byte array--you can
        //do this however you would like.
        byte[] b = IOUtils.toByteArray(stream);
        ByteBuffer data = BufferUtils.createByteBuffer(b.length).put(b);
        //load audio data into appropriate system space....
        AL10.alBufferData(p, openALFormat, data, (int)format.getSampleRate());
        //and return the rough notion of length for the audio stream!
        return (long)(1000f * stream.getFrameLength() / format.getFrameRate());

We’re hijacking a lot of the older JavaSound API utilities for this. OpenAL, much like OpenGL, isn’t really “open”, nor is it technically a “library”. So, having something around for handling audio data is helpful, and why bother writing our own when it’s already built into the JDK?

For JavaSound, you work with either Clips, or (more frequently) AudioInputStreams. You can read most audio file formats directly via AudioSystem.getAudioInputStream(…); in this case, I’ve elected to use a WAV format of LCD Soundsystem’s “I Can Change”, because James Murphy is a god damned genius. However, you can use anything you would like; to get it to work with this just drop it in the same source directory.

Next up, grab the format of the sound with AudioStream.getFormat(). This will provide you with a lot of valuable information about the stream. If it’s a big endian stream (which most wave files are not), you might need to convert it to little endian or make proper alterations to OpenAL. I’ve glossed over this, as endian-ness is not really a part of the tutorial and there are plenty of good byte-management tutorials out there.

I’ve elected to use format to check for the mono/stereo status (more are actually possible), and whether the sound is 8-bit or more frequently 16-bit. (Technically 32- or even 64- bit sound is possible; but there is actually a resolution to the cochlea of the ear, and you’re not going to bump into that outside of labs with very funny looking equipment. Even Blu-ray doesn’t go above 24-bit. Seriously, there’s generally just no point in bothering.)

Afterward, we load the stream into a byte array (I’m using IOUtils for this for brevity, but you can do it however you like), and the byte array into a ByteBuffer. Flip the buffer, and punch it over to OpenAL, which will take care of the rest of the work with it. Afterwards, we will eventually need the length of the audio stream, so calculate it as shown and send it back to the calling method.

After the buffer’s been created and the length of it is known, we’ve got to create a source for it! This is where most of the cooler built-in effects show up. alGenSources() creates a framework for the source; alSourcei(source, AL10.AL_BUFFER, buffer.get(0)) ties it to the sound buffer. You’ll also see that I set up AL_GAIN and AL_PITCH, which are fun to play with.

You’re almost done!

To actually play the buffer, you use the source. alSourcePlay(source) starts it. After that, I have the Thread sleep for the calculated length of the sound, just so we have time to hear it. At the end, I call alSourceStop(source) to demand an end to the source.

Lastly, I delete all sources. You might also want to delete devices, if you’ve done anything silly with them; this is very low-level access. You now have everything you need to load audio into your games and programs, and if you happen to bump into an SPI for a preferred format, it will now also be enough to get you going on OpenAL as well.

Leave a comment

Posted by on July 4, 2016 in Java, Programming


Tags: , , , , , , , , , ,

Apache Commons DecompositionSolvers

Apache Commons DecompositionSolvers

Jesus it’s been too long since I got back to this!

Anyway, right, the


Intro / Why-I-Need-To-Worry-About-This

Linear algebra exists for a reason; namely, us. Suppose we’re attempting to find the coordinate values, which under a certain transform, become a specific value. Let’s keep it simple and call it:

 x + 2y + 3z = 1
2x + 4y + 3z = 2
4x           = 3

As I’m sure you can remember from grade school, you have the same number of equations as unknowns, so it is almost certainly solvable. We just subtract two of the first equation from the second, four of the first equation form the third, four of the second from the third, one of the second from the first, and a quarter of the third from the first. Then we maybe divide the third by eight and the second by three, and presto,

x = 3/4
y = 1/8
z = 0

Unfortunately, as programmers, we both know that this is much easier done in practice than in theory; and when you’re automating  a task, a working theory is the only thing that really counts.

So, those of you who have already taken linear algebra (quite possibly all of you) may be familiar with a much easier way of representing this problem:

┌1 2 3┐┌x┐   ┌1┐
│2 4 3││y│ = │2│
└4 0 0┘└z┘   └3┘

A decomposition basically solves this, through a sequence of steps on both sides that reduces the original matrix to an identity matrix, while having the right-hand matrix undergo the same operations. This is commonly written as an augmented matrix, like so:

┌1 2 3│1┐
│2 4 3│2│
└4 0 0│3┘

Matrix reduction is a heck of a lot more straightforward than the nonsense I spouted a few paragraphs back, though going into its details is a bit off topic here. Our final matrix, after the reduction, looks like this:

┌1 0 0│3/4┐
│0 1 0│1/8│
└0 0 1│ 0 ┘

How Do We Do This in Java?

Not just Java, actually; this is specifically about the Apache Commons Math3 decomposition solver interface.

One of the tricks with reduction is that there are a lot of different, equally effective, ways to go about it; and like any other algorithm, the efficiency depends, in large part, on the initial state of your matrix. My personal favorite is the LU Decomposition. (Or, if you prefer a link that isn’t a video, look here.)

First I recommend making a Maven project out of your Java project, presuming that it isn’t already fitting that form factor. Afterwards, open up pom.xml, and add this:


right after the close of the build tag. Your project is now pulling classes from across the internet, on Apache Commons Math3. Later on, you may want the version number to be a bit higher; for now I’m using version 3.5.

So, you’ll note that you have access to a host of new classes, all in some subpackage of org.apache.commons.math3. Import org.apache.commons.math3.linear.* into your class file.

We can solve the above problem by creating a RealMatrix of the initial matrix, potentially like so:

RealMatrix matrix = new Array2DRowRealMatrix(new double[][]{
    {1.0, 2.0, 3.0},
    {2.0, 4.0, 3.0},
    {4.0, 0.0, 0.0}

But don’t get me wrong, there are literally dozens of ways to create a RealMatrix.

Next, create a RealVector, describing the other side of the equation, perhaps like so:

RealVector vector = new ArrayRealVector(new double[]{

We now have a matrix and vector representation of the two sides of our equation.

Working with RealMatrix and RealVector

If you’re an experienced programmer, you probably expect some kind of Command Pattern to show up next. It’s certainly what I would do, if I needed to duplicated the exact operations in the exact order on more than one piece of base data. Fortunately, something like it has already been implemented by Apache.

If you look up the Apache Commons Math3 javadocs, you’ll notice that while RealMatrix has a lot of handy operations, they generally just involve polling for data, not actually operating on it. Commons has made the wise move to encapsulate operations in their own classes, rather than just their own methods. There are many dozen other classes, such as MatrixUtils (remember that one!), which both generate and operate on RealMatrix and RealVector classes.

In this instance, turn to DecompositionSolver. It’s meant for tasks just like our own, and there are many subclasses. As I said, my preference is LUDecomposition, but that is only capable of handling square matrices. Since our matrix is square, that’s fine; in other cases when your matrix doesn’t fit the profile, look through EigenDecomposition, SingularValueDecomposition, or some other utility.

For LUDecomposition, we’ll want to do something like this:

DecompositionSolver solver = new LUDecomposition(matrix).getSolver();

The work has been done, as one initialization, LUDecomposition doesn’t just store the matrix as a property; it determines from it the exact sequence of operations necessary to turn it into an identity matrix.

Once you have your solver, you can get your final right-hand vector via:


which will provide you with:

└ 0 ┘

Final Source Code

Here’s a working example of how such a program might work.

 package oberlin.math3;

import java.util.*;

import org.apache.commons.math3.linear.*;

public class MatrixReducer {
    public static void main(String...args) {
        new MatrixReducer();
    public MatrixReducer() {
        try(BufferedWriter writer = new BufferedWriter(new OutputStreamWriter(System.out));
                Scanner scanner = new Scanner( {
            writer.write("\nEnter first row of three numbers: ");
            RealVector vector1 = new ArrayRealVector(new double[]{scanner.nextDouble(), scanner.nextDouble(), scanner.nextDouble()});
            writer.write("\nEnter second row of three numbers: ");
            RealVector vector2 = new ArrayRealVector(new double[]{scanner.nextDouble(), scanner.nextDouble(), scanner.nextDouble()});

            writer.write("\nEnter third row of three numbers: ");
            RealVector vector3 = new ArrayRealVector(new double[]{scanner.nextDouble(), scanner.nextDouble(), scanner.nextDouble()});
            //create matrix
            RealMatrix matrix = MatrixUtils.createRealIdentityMatrix(3);
            matrix.setRowVector(0, vector1);
            matrix.setRowVector(1, vector2);
            matrix.setRowVector(2, vector3);
            //get other side
            writer.write("\nEnter vector on right side (3 entries):");
            RealVector vector = new ArrayRealVector(new double[]{scanner.nextDouble(), scanner.nextDouble(), scanner.nextDouble()});
            DecompositionSolver solver = new LUDecomposition(matrix).getSolver();
            matrix = solver.solve(matrix);
            vector = solver.solve(vector);
            writer.write("Solution: \n");
            writer.write("┌" + matrix.getEntry(0, 0) + " " + matrix.getEntry(0, 1) + " "
                    + matrix.getEntry(0, 2) + "┐┌x┐   ┌" + Double.toString(vector.getEntry(0)) + "┐\n");
            writer.write("│" + matrix.getEntry(1, 0) + " " + matrix.getEntry(1, 1) + " "
                    + matrix.getEntry(1, 2) + "││y│ = │" + Double.toString(vector.getEntry(1)) + "│\n");
            writer.write("└" + matrix.getEntry(2, 0) + " " + matrix.getEntry(2, 1) + " "
                    + matrix.getEntry(2, 2) + "┘└z┘   └" + Double.toString(vector.getEntry(2)) + "┘\n");
        } catch (IOException e) {
Leave a comment

Posted by on June 30, 2015 in Java, Programming


Tags: , , , , , , , , , ,

The NIO.2 Watcher

So, I’ve been working on a side project involving the Builder tutorial. It roughly (not entirely, but roughly) works out as a machine-operated interpreter, that is, code altered by machine before being translated. After that it does something even more awesome, but it’s only capable of triggering the compilation, after alteration, through a utility that isn’t as well known as it should be.

The Watcher Utility

As of Java 7, we got the NIO.2 classes. These included Path (which most of you are probably familiar with), Files, FileSystem, asynchronous channels, and a host of other goodies. One of them was the Watch Service API.

What Watch ultimately amounts to is a device that can trigger an event any time an arbitrary subset of data is altered in some way. The easiest possible example is monitoring a directory for changes, but this is, gloriously, not exclusive. In classic Java nomenclature, one might think of it as a sort of PathEventListener, in a way; but it’s capable of a bit more than that particular name implies. It doesn’t have to be associated with Paths, and unlike most listeners, it’s less about monitoring for user-generated interrupts, and more about monitoring for system-wide circumstances, including secondary effects.

Using a Watcher

Watchers keep internal collections of keys, each one associated with a source object. This registration is typically located on the source object, at least directly. The best, and most correct, way to do this is through direct implementation of the Watchable interface. Many JDK classes, such as Path, already implement this. Once implemented, you would use the method:

Watchable.register(WatchService watchService, Kind<?>

This method registers all events, of the specified types, on the provided WatchService object. Every time one of them occurs, the key is flagged as signaled, and in its own time the WatchService will retrieve the data from that key and operate.

Note that a Path can be many things. It could be a path to a directory on your machine, which is of program concern. It could be a path to a printer tray, or a server, or even a kitchen appliance (think polling the status of an automated espresso machine). In this example, I will be showing a manner in which a directory path can register to be watched for alterations.

WatchEvent.Kind interface

This can be thought of, for old school Java programmers, as the class type of a Watch event. Most of the frequently used keys are in java.nio.file.StandardWatchEventKinds, but as an interface, it is fully customizable. They only require two methods to be overridden, that is,, which simply returns a String value representing the type of event; and WatchEvent.Kind.type(), which returns a Class that describes the location context of the event.

WatchEvent.Kind.type() may return null, and it won’t break anything; but after getting a feel for the results of StandardWatchEventKinds, you might consider implementing it. As an example, for ENTRY_CREATE, ENTRY_MODIFY, and ENTRY_DELETE, the context is a relative path between the Path being watched, and the item that has changed. (Knowing that a random item was deleted is of little if any use, without knowing which one.)

Implementing a WatchService

Most of the WatchServices you are likely to use are stock in the JDK. I’m going to start with one of them; in a later blog, I’ll probably create one from scratch, but it really is better to start simple.

For the common case of monitoring a directory, FileSystem.newWatchService() covers everything you need. It is important to get a watcher for the correct type of FileSystem, though; as many of you know, Java is capable, as of version 7, of taking advantage of the numerous file system-specific capabilities. The safest way to do it is through:

WatchService watcher = FileSystem.getDefault().newWatchService();

But there may be many points in which you intend to grab a watcher from a file system of a specific, or even custom, type. This is fine, but be aware of the extra layer of debugging.

Afterward, each path can be registered with the watch service through its Path.register(…) method. Be certain to include every variety of WatchEvent.Kind that you want to watch for. It may be tempting to simply register for every single standard type every time, but I encourage you, as a matter of practice, to consider whether you’re really concerned about each Kind before including it. They do, technically, cost a small amount of system resources; and while it may not be noticeable for small projects, when you’re dealing with massive file hierarchies it can become a concern.

When polling for changes, it is mildly more complicated than it is with Listeners. The watcher must be polled for WatchKey objects. WatchKeys are generated when a watched alteration occurs. They all have a state, which is continuously either ready, meaning the associated Watchable is valid but without events; signaled, meaning that at least one event has occurred and been registered with this WatchKey; and invalid, meaning that it is no longer sensible to consider the associated Watchable a candidate for events.

There’s more than one way to get the next signaled WatchKey, but one of the most efficient methods is WatchService.take(). This will always return a signaled WatchKey. It is a blocking method, so use it with that in mind; if no WatchKeys are yet signaled, it will wait until one is before returning.

Once you have a WatchKey, a secondary loop examines every sequential change that has occurred. (If you’re curious, if a WatchEvent occurs for a WatchKey that is already signaled, it is added to the stack and no other alterations are made; if it occurs while the WatchKey is ready, it initiates the stack and WatchKey is flipped to signaled). This is done via WatchKey.pollEvents(). For each event, you may examine the WatchEvent, and act on it accordingly.

After all is said and done, and the WatchKey has zero events left to parse, call WatchKey.reset(). This attempts to flip the WatchKey back to the ready state; if it fails (if the key is now invalid), the method returns false. This might signal, as an example, that the watched path no longer exists.


Any WachService manager must be running continuously. The antipattern approach is to simply use a while-true block; but in general, it is less hazardous to make it its own thread.

import java.nio.*;
import java.nio.file.*;
import java.nio.file.WatchEvent.Kind;

public class DirectoryWatcher implements Runnable {
    private WatchService watcher;
    private volatile boolean isRunning = false;
    public DirectoryWatcher() throws IOException {
        watcher = FileSystems.getDefault().newWatchService();
     * Begins watching provided path for changes.
     * @param path
     * @throws IOException 
    public void register(Path path) throws IOException {
        //register the provided path with the watch service
        path.register(watcher,    StandardWatchEventKinds.ENTRY_CREATE,

    public void run() {
        isRunning = true;
        while(isRunning) {
            //retrieve the next WatchKey
            try {
                WatchKey key = watcher.take();
                key.pollEvents().stream().forEach(event -> {
                    final Kind<?> kind = event.kind();
                    if(kind != StandardWatchEventKinds.OVERFLOW) {
                        final Path path = ((WatchEvent<Path>)event).context();
                        System.out.println(kind + " event occurred on '" + path + "'");
                if(!key.reset()) {
                    //the key should be valid now; but if it is not,
                    //then the directory was likely deleted.
            } catch (InterruptedException e) {
    public void stop() {
        this.isRunning = false;

Simple enough, yes?

The register(…) method may be a little redundant; however, the run() method is where the meat is. WatchKeys are retrieved with WatchService.take(); afterward, in a parallel stream, each WatchEvent associated with that key is looped through. (When an event is of type OVERFLOW, it usually means that data on the event has been lost; not optimal, but the best course of action here is to continue to the next key.)

In this instance, the event is simply reported to the terminal, but this lambda expression is where you would take arbitrary actions according to the event. It is also possible to use an external iteration to do this, if you need to change values or perform another non-lambda-kosher action.

After all events have been iterated through, WatchKey.reset() is called, and checked. In the event that it returns false, something has happened to our directory, and the thread has become a potential resource leak; so it is shut down automatically. Otherwise, the thread then yields to other threads, and repeats itself.

Here’s a small Main class that I’ve built to use this. A single path parameter will be the directory to monitor; or it will simply watch for $HOME/watchtest.

import java.nio.file.Path;
import java.nio.file.Paths;

public class Main {

    public static void main(String[] args) throws IOException {
        final String location = (args.length > 0) ? args[1] : 
            System.getProperty("user.home") + "/watchtest";
        final Path path = Paths.get(location);
        DirectoryWatcher watcher = new DirectoryWatcher();
        (new Thread(watcher)).start();
        //wait a set amount of time, then stop the program
        try {
        } catch (InterruptedException e) {


Try running it, and making a few changes to your select directory. See what it does.

And That’s It!

The next real question is how to create your own WatchService; which is totally doable. Generally, though, it isn’t necessary. The next time I come back to this subject, I’ll be going over that, possibly starting with WatchKey.Kinds. First, though, I need to get back to the project that I started this for, and I need to continue the Build Tool tutorial, so it might be a bit.

Good coding!

Leave a comment

Posted by on March 11, 2015 in Java, NIO.2, Programming


Tags: , , , , , , , , , , , , , , , , ,

A Case Against Using Null. (For Almost Anything.)

Java is my usual language, but this goes for everything.

I promise that this is not going to be another rant about NullPointerExceptions and their kin in other languages. This is not to say that such rants are not warranted, and even cheered; but I’m going to be a bit more academic about it. I’m also going to provide solutions, not only the ones available in Java 8, but what I used to do beforehand.

What Does “Null” Actually Mean?

Great question. Null as an adjective, according to the dictionary, means without value, effect, or significance. It also means, lacking, and nonexistent. It also means, empty. And, lastly, also probably most recently, it means zero. This is most likely a linguistic artifact, as everything is ultimately expressed in the binary on a computer. In C, null actually does equate to zero. However, this necessity has led all of us to a lot of abuse, because symbolically it isn’t what null is for. I’ll come back to that.

Null’s etymological origin comes from the latin nullus, meaning none, as in, it-has-not-been-set. While zero is reasonable, zero is an actual number. If you were enumerating the entries of a set of numbers, and you wanted to count the length of that set, you would not skip every entry that was zero, would you? However, you likely would for null, as it denotes no-entry in the set. Therein lies the critical difference.

In Java, objects are initialized to null, before they are set to any value. Object instances are the Java equivalent of C’s pointers; and while they cannot be without a value, they initialize to a language constant that reflects the absence of an intended one. Null is typically represented as zero, but not always, and I am unsure of the case with Java. However, a null pointer is a symbol, it is not reasonably a pointer to the zero position in memory. This position does, actually, exist; but on the reference of such a pointer, the virtual machine (or platform) throws up a red flag.

The nasty habit of using null as a return value when something goes wrong in an operation is almost ubiquitous, but unless this literally represents that no value has been set, it is a dangerous move. I’ve even seen it in the JDK. The response to such an ill-though-out method is usually a few lines of defensive programming, checking to see whether the object is null, and acting accordingly.

Java 8 Solutions

If you aren’t a Java programmer, you may wish to skip this section.

As it turns out, the defensive programming response to returned nulls is so similar, in every instance, that it can be encapsulated into an object itself. This would be Java 8’s Optional. Optional represents a possible value, that is, a value which cannot be guaranteed to exist. However, the Optional itself is never null.

On initialization of an Optional, it is best to set it to Optional.empty(), that is, an Optional with no contents. If a value is being wrapped in an Optional, use Optional.of(). If the presence of the value is unknown, use Optional.ofNullable(), it will do the defensive work for you. The rest of the methods of Optional apply Java 8’s influences from functional programming. What used to require complex if statements is now done primarily through ifPresent(…) and orElse(…).

This might seem like an overreaction to you. However, compared to the work that I used to have to do just to catch every wrench-in-gears value that might pass by, it is a miracle. If you disagree, you need only ask yourself how frequently you have been getting NullPointerExceptions. Adopt Optionals, and you won’t get them anymore.

Older Java Solutions

In previous versions, several further techniques have been added. The biggest problem with “!= null” is that it is an operation, and a mandatory operation, which will slow down code very slightly. This is imperceptible for the vast majority of programs, but if you need something to run searing fast, then it can be unacceptable.

If you are writing an API, I might suggest funnelling all input through a defensive checking method before passing it along to the meat methods; but if you are writing code that only you will access, there is a simpler solution: assertions. This is particularly true for unit testing with programs like JUnit.

This is exclusively functional during development, as in order to enable assertion testing, you need to pass the -ea parameter to the java compiler. Unless you can force this on users, it is exclusively meant to help you identify routes by which null, or any other unacceptable value, can make it to your methods.

The syntax is simple. Given parameter “x”:

assert x != null : "[error message]";

If the provided boolean expression evaluates to false, an AssertionError is thrown, with a message of the toString() value of whatever was passed on the right (for me, most typically an actual string).

I don’t generally like to see assertions making their way into production code today, as I am inclined toward Optionals; but this is quite effective for debugging. Additionally, such statements can be considered an essential part of JUnit tests. If you are in a rush, it is possible to ignore all assertions remaining in a slice of code by removing the “-ea” parameter from the compiler; but on the human end, this is bad practice and worth avoiding.

As an alternative, Apache Spring has a class called Assert which handles more or less the same tasks as the assert keyword.

Broader Solutions for Object Oriented Languages

At last, in the most general sense, there is the Null Object Pattern. This is, still, my ultimate preference when building a set of classes, as there is no need for Optional when null never enters the equation.

A Nullary Object is an object extending the appropriate interface, with defined behavior, denoted as equivalent to null. This has its ups, and its downs. As an example, suppose we had this interface:

public interface Animal {
    public String speak();

with these implementations:

public class Dog implements Animal {
    public String speak() { return "bark"; }

public class Cat implements Animal {
    public String speak() { return "meow"; }

public class Bird implements Animal {
    public String speak() { return "tweet"; }

And we had one further class that requires one unknown animal, which will indubitably call “speak()”. Which animal is beyond our control, and we don’t want our program to crash on a NullPointerException simply because no animal was specified. The solution is one further class:

public class NullaryAnimal implements Animal {
    public String speak() { return "…"; }

In the case of abstract classes, it is often helpful to have the nullary class be a member of the class itself. This is also particularly helpful when there are multiple behaviors which might, otherwise, be implemented as “null”. The potential down side is for people who were actually looking for an exception to be thrown; in such a case, simply fill speak() with an Apache Commons NotImplementedException or something relatable.

One extension of this pattern is such:

public abstract class Sequence {
    public static final Sequence ANY = new Sequence(…);
    public static final Sequence ALL = new Sequence(…);
    public static final Sequence NONE = new Sequence(…);

In this instance, a new Sequence can be initialized to Sequence.NONE, ALL, or ANY, and be replaced if a new value is provided. Additionally, since these are actual objects and constant values, they respond appropriately to equals checks.

There may be a name for this pattern, I’m honestly not sure. I came up with it on my own, but I very much doubt that I’m the first.


Hardly. However, you now hopefully have a new set of tools to keep unfinished declarations and, even worse, “= null” statements out of your program. I hope I’ve made your life easier!

Leave a comment

Posted by on January 10, 2015 in Programming


Tags: , , , , , , , , , , , , , , , ,

Visual Feedback on an Abstract Parsing Tree with JavaFX

I honestly didn’t expect to be writing this, but it seems fair.

In the past few editions, I’ve been discussing the AST. It can be overwhelmingly complicated for a complete program; so I’ve been using a simple, single line equation as the sample. Unfortunately, that isn’t very realistic; and it would be very helpful to have a procedurally generated visual tree available. That tree is what this lesson is all about.

At first I considered using a graphical tree style, like javax.swing.JTree; but that can be painfully over-simplistic in itself. I would prefer to outline the material the same way I would draw it on a white board (which, if you’re wondering, I do). The best way to do this? JavaFX.

JavaFX whiteboard abstract syntax tree

Graphical AST tree rendering, through JavaFX/F3

If you aren’t familiar with JavaFX, please do me a favor and tolerate the name. It was originally F3, for Form-Follows-Function. I kind of liked F3, until some marketer decided that “JavaFX” sounded better. Functionally speaking, it’s an excellent revision on how user interfaces are designed in Java. I fully stand behind it. It allows for XML structuring and CSS styling, just like a web page, to more hard-coded controls. This is much, much faster; and it allows for significant beauty in user interfaces. However, it works very differently from things like Swing and AWT; and while I’m certain that it isn’t the first API to do so, it takes some getting used to.

I fully intend to write a true tutorial on all of JavaFX on some point. Do you need to understand it to understand translators? Absolutely not. However, this code does work. It is not part of the Github repository, as it is technically a tangential project; but the same license (GNU GPL) applies to it and you are welcome to copy it token for token. I’ll put it up on Github as I get the chance. I’ll make a few minor comments along the way to help you follow it.

1. The Basic Application

We have exceedingly few needs for our app. It simply reads a program from a stream, parses it, and feeds the parse tree to a custom node, which displays it graphically. Accordingly, the program code is rather small. I’ll begin by displaying it, then I’ll spend a moment piecing it together in English for you.

package oberlin.builder.gui;

import oberlin.algebra.builder.AlgebraicBuilder;
import oberlin.builder.parser.ast.AST;
import javafx.application.*;
import javafx.scene.*;
import javafx.scene.layout.*;
import javafx.stage.*;

public class GUIMain extends Application {

    public static void main(String...args) {

    public void start(Stage primaryStage) throws Exception {
        Pane root = new Pane();
        Scene scene = new Scene(root);
    private void populate(Pane p) {
         * This is simply an example, so I've ignored input for now.
         * In theory, you would replace the line below (containing
         * hard code) with an input loop.
        AST program = (new AlgebraicBuilder()).getParseTree("1+2");
        p.getChildren().add(new GUITree(program));

All JavaFX/F3 programs begin with Application.launch(String…args). JavaFX programs run in what is effectively their own thread, and more so than with Swing-based programs. Launch parses arguments and stores them in their own object, appropriately called Parameters. They can be accessed, at any point later on, via Application.getParameters(). Our available overloads and customizations cut out for a moment, then come back in in the start(Stage) method.

Stage is basically where Frame would be; but it’s a little more complicated than that. Unlike Swing and AWT, which were designed to be platform independent, JavaFX is designed to be hardware context independent. What you are writing here will work equally well on a PC, tablet, and smart phone; as well as anything else built (now or later) that maintains a JavaFX compatibility standard. Thus, what might otherwise be called a frame or window is referred to as a stage, as it might be neither of those things.

You’ll notice that the instantiated Pane is given a style class. If you aren’t familiar with CSS, a style class is what’s used to differentiate between one element and any number of others which, otherwise, would look exactly like it. Thus, it allows CSS to pick and choose which elements of the layout it is styling at a given moment. I’ve chosen “backing” as the name for this element, as it is the backboard of our tree. You will also note that, two lines later, the CSS file itself is loaded.

Next, a Scene is created. Scenes are critically important, and distinct from stages. While a stage represents the context that the layout is drawn in, the scene represents the actual controls and constraints within that space. Thus, while many aspects of Stage are immutable (and unknowable), Scene allows for greater flexibility. JavaFX sees to it that they correspond, so don’t worry about that.

Scene is styled through its root element, which in this case is our pane. You’ll notice that instead of the stricter setWidth() and setHeight() that you might be familiar with from Swing, we are setting a minimum on these bounds. That minimum is not guaranteed, as the display may not be capable of it, but it is treated as a general rule to be followed if at all possible. In this case, I’m going for classic analog low-def TV resolution, 640 width by 480 height. (Looking back, those numbers might be inadequate, but for now they’re quite functional.) If this is too small for you, the frame—if it is a frame, anyway—is easily resizable.

Populate() is a method I wrote to add the paraphernalia to the scene; but note that afterwards we call show(). This is very important, as otherwise our stage will be constructed in memory, but never displayed to the screen. Additionally, there will be no way to kill the JavaFX thread save for a hard interrupt. Once shown, the closing of the primary stage will flag the program to terminate.

1.1. Populate

It’s a generally good habit, but not a necessary one, to populate your frame in a separate and dedicated method. This is what I do here, even though for the moment, I only have one control to add.

The AST method should be old news; it’s a stand-in, for the moment, for an actual code-reading portion. (I’m assuming that you’re looking to compile more than just “1+2”.) GUITree is a custom JavaFX node, which I will explain next. Note that to add a node to a program, you must take some structure (not yet visible) in the scene graph (stemming from your chosen root), and get its Children as a modifiable list. Then, you must add that node to the list.

Note that after a stage is visible, precious little of the scene can be changed save for through the constraints built into it. I’m not going to touch on Expressions and Bindings here, but know that if you pull something that doesn’t play by JavaFX’s rulebook, it will throw an ApplicationException and your program will not launch. Thankfully, while exceedingly picky, that rulebook is small. If you call show() and then try and add a child, you will have problems; it must be the other way around.

If you’re curious, hiding a rendered stage does not count for making it modifiable. You must give it your entire concept first, then make it visible. If you’re familiar with OpenGL, you’ll already understand why.

2. The Tree Itself

The tree is a custom JavaFX node, which I admit is rarely necessary. Still, most of the entities that make it work are core to the API.

package oberlin.builder.gui;

import oberlin.builder.parser.ast.AST;
import javafx.geometry.BoundingBox;
import javafx.geometry.Bounds;
import javafx.geometry.Point2D;
import javafx.scene.control.Tooltip;
import javafx.scene.layout.*;
import javafx.scene.paint.Color;
import javafx.scene.shape.CubicCurve;

import java.util.function.IntSupplier;

public class GUITree extends AnchorPane {
    private Bounds bounds = new BoundingBox(0, 0, 640, 480);
    private AnchorPane framing = new AnchorPane();
    private double edgeSize = 0.10;    //ten percent additional length beyond edges of framing
    public GUITree(AST ast) {
        this.setMinWidth(bounds.getWidth() * (1 + edgeSize));
        this.setMinHeight(bounds.getHeight() * (1 + edgeSize));
    private void configureFraming() {
        framing.setLayoutX(edgeSize * (bounds.getWidth() / 2.0));
        framing.setLayoutY(edgeSize * (bounds.getHeight() / 2.0));
    private ASTNode addNode(AST ast) {
        return this.addNode(ast, new Marker(0), new Counter(), 0, null);
    private ASTNode addNode(AST ast, IntSupplier stepsDown, IntSupplier stepsAcross, int index, ASTNode parent) {
        ASTNode node = new ASTNode(ast, stepsDown.getAsInt(), stepsAcross.getAsInt());
        //AnchorPane stuff
        calculateAnchoring(node, parent);
        framing.getChildren().add(index ++, node);
        final StringBuilder tooltipText = new StringBuilder();
        IntSupplier across = new Counter();
        for(AST kid : ast.getContainedNodes()) {
            tooltipText.append(kid.getClass().getSimpleName()).append(" ");
            ASTNode child = addNode(kid,
                    new Marker(stepsDown.getAsInt() + 1),
            CubicCurve cubic = createCubicCurve(node.getNoodleRoot(), child.getTopCenter());
        node.getType().setTooltip(new Tooltip(tooltipText.toString()));
        return node;
    private CubicCurve createCubicCurve(Point2D p1, Point2D p2) {
        CubicCurve curve = new CubicCurve();
        return curve;
    private void calculateAnchoring(ASTNode node, ASTNode parent) {
        node.setOrigin(new Point2D(parent == null ? (bounds.getWidth() - node.getBounds().getWidth())/2.0 :
            justifyX(node, parent), justifyY(node)));
        AnchorPane.setTopAnchor(node, node.getOrigin().getY());
        AnchorPane.setLeftAnchor(node, node.getOrigin().getX());
    private Double justifyX(ASTNode node, ASTNode parent) {
        final double parentCenter = (parent.getOrigin().getX() + (parent.getBounds().getWidth() / 2.0)
                + parent.getNoodleRoot().getX()) / 2.0;
        final double center = parentCenter
                - node.getBounds().getWidth()
                        * (parent.getAST().getElementCount()) / 2.0; 
        return center + node.getOrigin().getX();
    private Double justifyY(ASTNode node) {
        return node.getOrigin().getY();

That was a bit much at once, I know. The central pane, called “framing”, is 640 by 480. Framing is offset in each direction by a 5% inset, via the convenient features of AnchorPane.

AnchorPane is one of the few prepared ways to control where a node is rendered, with precision, in JavaFX. You may often need to keep your own tabs on where it is rendered, as getMinX() and getMaxX() will return zero more often than you will believe. However, through direct layout control, you can still manage them.

The method addNode(…) adds a custom object called ASTNode. I’ll cite it for you here.

package oberlin.builder.gui;

import javafx.collections.FXCollections;
import javafx.collections.ObservableList;
import javafx.geometry.BoundingBox;
import javafx.geometry.Bounds;
import javafx.geometry.Point2D;
import javafx.geometry.Pos;
import javafx.scene.control.Label;
import javafx.scene.layout.StackPane;
import javafx.scene.layout.VBox;
import javafx.scene.text.TextAlignment;
import oberlin.builder.parser.ast.AST;

class ASTNode extends VBox {
    private Bounds bounds = new BoundingBox(0, 0, 100, 40);
    private Point2D origin = new Point2D(0, 0);
    private final double expanse = 1.10;
    private final AST ast;
    private Label type;
    private Label hash;
    private ObservableList<ASTNode> kids = FXCollections.observableArrayList();
    public ASTNode(AST ast) {
        this.ast = ast;
        type = new Label(ast.getClass().getSimpleName().toString());
        hash = new Label(Long.toHexString(ast.hashCode()).toUpperCase());
        VBox vbox = new VBox(new StackPane(type), new StackPane(hash));
        for(AST kid : ast.getContainedNodes()) {
            addKid(new ASTNode(kid));
    public Point2D getNoodleRoot() {
        return new Point2D(getOrigin().getX() + (getBounds().getWidth() / 2),
                getOrigin().getY() + getBounds().getHeight());

    public ASTNode(AST ast, int level) {
        origin = new Point2D(0, level * bounds.getHeight() * expanse);
    public ASTNode(AST ast, int levelDown, int levelAcross) {
        origin = new Point2D(getStepAcrossSize(levelAcross), getStepDownSize(levelDown));
    public double getStepDownSize(int steps) {
        return steps * bounds.getHeight() * expanse;
    public double getStepAcrossSize(int steps) {
        return steps * bounds.getWidth() * expanse;
    public void addKid(ASTNode astNode) {;
    public ObservableList<ASTNode> getKids() {
        return kids;
    public Bounds getBounds() {
        return bounds;
    public Point2D getOrigin() {
        return origin;
    public Point2D getTopCenter() {
        return new Point2D(
                getOrigin().getX() + (getBounds().getWidth()/2),
    public Label getType() {
        return type;
    public void setOrigin(Point2D p) {
        this.origin = p;
    public AST getAST() {
        return this.ast;

ASTNode is a JavaFX Node as well. It simply maintains a reference to the AST itself, and the general presentation of that AST on the tree. There isn’t a lot here. If you’re wondering what VBox is, it’s an abbreviation for “vertical box”. (Naming a class after an abbreviation is bad practice, but it’s long since done by powers above me; I tolerate it as much as I do “AST”.)

Speaking of bad practice, this would ideally actually use Bindings, but I wrote this in a bit of a rush today and will have to correct that in the future. It is also bad practice to repeat data, which is exactly what this program is doing by re-storing the label text in a separate field. All the same…

I’m going to gloss over a lot of the configuration of the labels, as it’s relatively standard. Know that like any other pane in JavaFX, a VBox can be initialized with a list of its bounded nodes; also, a StackPane has the default behavior of centering its own bounded nodes.

The last thing done in the constructor is the creation of additional ASTNodes for each child node of the abstract syntax tree.  Each of them, in turn, renders their own children. This is not perfect, there is a substantial chance that two lists of nodes will overlap one another; however, it is already excellent for debugging visitor pattern based content. In the end, the GUITree renders each node in an assigned place, with a curved cubic line (technically called a “noodle”) connecting it to its parent and its children.

How does it do that? With IntSuppliers.

3. The IntSuppliers

There are only two of these.

package oberlin.builder.gui;

import java.util.function.IntSupplier;

 * For downward counts; always returns provided number.
 * @author © Michael Eric Oberlin Dec 23, 2014
class Marker implements IntSupplier {
    private int fix;
    public Marker(int fix) {
        this.fix = fix;
    public int getAsInt() {
        return fix;

package oberlin.builder.gui;

import java.util.function.IntSupplier;

 * For counts across; always returns next consecutive number.
 * @author © Michael Eric Oberlin Dec 23, 2014
class Counter implements IntSupplier {
    private int count;
    public int getAsInt() {
        return count++;

IntSuppliers (and really all Suppliers) are part of the java.util.function package, new to Java 8. The great advantage of this package is that a function, or any functional interface, allows you to specify a method that serves as a primitive with conditionally defined values. I know that’s a leap, but I’ve been doing it since long before it was formally adopted into the language and it’s a central totem of functional languages.

We could, in theory and practice, use incrementing and decrementing integers in place of either of these. The problem is that the code gets a lot longer and a lot more cluttered when you do. I prefer the sublime simplicity of packing such behavior into an interface.

Of course, these are not everything. There is one, final, issue.

4. What was that that you said about “CSS”?

The CSS is specific to JavaFX; a complete listing of all of the properties is available here. If you are unfamiliar with the syntax of CSS, you can find an excellent tutorial on it (for HTML, at least) at W3Schools. It isn’t as versatile as Java or C, but its creators pulled many of its properties from C-like languages.


.backing {
    -fx-background-color: lightyellow;
    -fx-insets: 0;
    -fx-padding: 15;
    -fx-spacing: 10;

.node {
    -fx-background-color: lightblue;
    -fx-background-radius: 5.0;
    -fx-border-color: black;
    -fx-border-radius: 5.0; 

Keep this in the same folder as GUIMain, and it will find it as written.

The CSS styling of JavaFX controls is capable of everything HTML 5 is and then some. It’s an excellent fusion of programming and markup. I encourage you to play with the layout of GUIMain’s scene, and the actual program fed to the builder.


Tags: , , , , , , , , , , , , , , , ,

The Easy Way to Import from Guava or Apache Commons

In 1991, Sun Microsystems (specifically, James Gosling) answered a long standing question. The Java programming language, at the time called Oak, was established and released to cut development times into small fractions of what they used to be. The word was “write once, run anywhere”, which for the most part was true. A program could be compiled on one machine, and run on any that supported the same virtual machine.

Don’t get me wrong; Oak was a disaster for efficiency. However, it proved that virtual-machine based productive software wasn’t just an idea, it was actually possible. That was huge. So, Oak became Java (a word already very familiar to programmers), and Sun got to work on expanding the API and improving the compiler. Honestly, Java 1.0 was also crap; but it was very exciting crap. Java 2, in my humble opinion, was where it really took off.

During this time (Oak to now), a lot of features were added. Just-in-time compiling, regular expressions, enumerations, recently lambda expressions, and most importantly a gazillion bazillion classes were added to the JDK. All kinds of solutions to what became a very broad class of problems, shortening development time quite a bit for programmers. Surprisingly the internet, and the population of Java programmers, grew faster than Sun could keep up with. It had the added pressure of improving the compiler, too; which didn’t help.

As a response to this frustrating lag, the Open Source community (you may picture it in a hero cape) took off and created a vast assortment of additional libraries. Many of them, such as LWJGL and JOAL, were domain-specific; some of them weren’t. Apache Commons was the first big guy to come in. It’s actually a collection of libraries, the most important of which (at least for me) was the math (now Math3) library. It offered tried and tested methods for handling complex numbers, Fourier transforms, tuples, and all sorts of awesome stuff. That meant that the people, previously using the vanilla JDK, didn’t have to write it themselves. That saved a boatload of time.

Later, Google came up with Guava, their own contribution to the community (fully compatible with Apache Commons). Guava had neat features like bidirectional maps, and very handy byte conversion methods. Much like Apache Commons, it’s expanding all the time.

In olden days (1990s), it was often necessary to have the entire library as a local resource. That means on-disk. This could be an issue, when you only needed a few methods out of something as large as Math3. It is an enormous library, with a lot of binary data. Then came Apache Maven. I don’t intend to describe how to use Maven manually here, it isn’t something I’m an expert at, and it often isn’t necessary; but there are plenty of wonderful tutorials on the internet. I’m going to describe how to use it quickly.

Maven allowed for the inclusion of libraries from a URL, without the need to download the entire library to disk. As more and more computers were online 24/7, this became increasingly feasible. Through a feature of the Maven build tool, a file called pom.xml, the features of the project could be described, and lazily received as needed. The “POM” in pom.xml stands for “Project Object Model”, which is very accurate.

So How Can I Use Maven to Import these Libraries?

My IDE of choice is Eclipse (which is not to say that there aren’t other good ones out there). There’s almost always a utility native to your environment, which should work similarly to this. Eclipse has a plugin which handles Maven directly. To get it, go to the Help menu, and select “Eclipse Marketplace”. Under the marketplace, look up the keyword “Maven”. You will probably have quite a few “m2e” entries, the central one usually starts with “Maven Integration for Eclipse…”, the rest generally depends on your Eclipse version.

Install it, and restart Eclipse. Next up, assuming that your project already exists, you need to create a Maven project out of it. Right-click it, select “Configure…”, and click “Convert to Maven Project”. (If it nags you about the group ID or the artifact ID, it’s because of a naïve algorithm for generating the identifiers from the project name; just remove any spaces and funny characters and try again. The details of what these identifiers are are better left to more detailed tutorials on Maven.) It will set up your Eclipse project as a Maven project as well, specifically, an M2Eclipse project.

You will have a new file called “pom.xml” located in your project directory. There are other ways to do this, but the dependency information is typically provided in raw XML and copy/pasting it is usually fastest. Enter XML mode on the document (currently by clicking the last lower tab, labelled “pom.xml”), and find the end of the “<build>” entries.

Right below the “</build>” tag, enter “<dependencies>”. Eclipse will often fill in the terminating tag for you. Between these two, you may enter the dependency information typically found on the web for the library you are using; such as, for LWJGL:


Or, for Apache Commons Math3:


Or for Google Guava:


Then, clean your project by going to the Project menu and clicking on the “Clean” option. It’s generally good to recompile a project completely and from the ground, often called cleaning, after making a major change to its dependencies; this is often done automatically, but not always. You’ll now note that you can import any of the packages in Guava, Commons, or whichever library you have imported, without having to download the entire API.

How Does This Change Things?

The JDK is not a small library as it is, it’s actually quite enormous; but if you have ever found yourself struggling to write an extension of a collection or a math utility that could be used in a wide variety of projects, those efforts will now be fewer and further between. You may check the javadocs for these APIs as readily as the javadocs for the JDK, and need not worry about increasing the disk footprint of your development environment or (much worse) your project.

Leave a comment

Posted by on November 22, 2014 in Programming


Tags: , , , , , , , , , , , ,

Software Language Engineering: Terminals, Nonterminals, and Bears (Early Edition)

The Notion of a Terminal AST

So, in the last chapter, I explained what an AST was structurally. There are formally two kinds of extensions to it. I usually implement them in their own classes, extending the base class of AST.

The first is the terminal. If you’ve programmed in Java for even a month, you know that having a method which accepts two different kinds of unrelated classes in the stead of one another is a bad idea for all kinds of reasons.

It is, actually, possible.

That just doesn’t mean that you should do it.

ASTs are formally called trees, but what they are is nodes on a tree. A program is a single AST, with an extension typically called “Program” as the overarching root node. The branch nodes, or nodes with child nodes, are called non-terminal nodes; the end points are called terminal nodes.

Each of those tokens that your Scanner kicked out? Yeah, that’s a terminal node, disguised as a String.

Let me offer you some of my code for ASTs, Terminals, and NonTerminals. (As before, there are major issues that I’m leaving out until later on. See if you can catch them.)

package oberlin.builder.parser.ast;

import oberlin.builder.codegenerator.RuntimeEntity;
import oberlin.builder.visitor.*;

 * Abstract Syntax Tree, capable of representing any sequence of 
 * statements or the entire program.
 * @author © Michael Eric Oberlin Nov 3, 2014
public interface AST {
     * @return number of sub-elements contained in this tree node.
    public int getElementCount();
package oberlin.builder;

import oberlin.builder.parser.ast.AST;

 * Basis of all complete abstract syntax trees. Terminals are basically isolated-tokens known only by their spellings.
 * @author © Michael Eric Oberlin Nov 5, 2014
public class Terminal implements AST {
    private final String spelling;
    public Terminal(String spelling) {
        this.spelling = spelling;
    public final String getSpelling() {
        return this.spelling;

    public String toString() {
        return getSpelling();

    public final int getElementCount() {
        return 1;

Let those soak in a little. Note that each Terminal has a length of one, meaning that it is the only member of its immediate tree. That will be important when we develop our Parser.

A terminal is an instance of an AST, and can be created by simply passing its token to it. The token is stored in the field “spelling”. Terminal is also fully extensible, as even though their token is consistently their only members, there is a significant difference between an equals operator, a binary operator, and numerical data; and nonterminal nodes take that difference quite seriously.

The Notion of a Nonterminal AST

A nonterminal AST is an AST built not from characters in a String, but from a sequence of other ASTs. The constituent ASTs can be terminals, or nonterminals. Remember BNF? Ever listed item before may-consist-of (::=) was a nonterminal. In instances of single-member representation, such as:

Noun ::= ProperNoun

“Noun” is still a nonterminal, as the implication is that (at least in theory) multiple types of items can be its constituents.

The “Program” node is of course always a nonterminal. I’ve written a nice slice of code for them, too.

package oberlin.builder;

import java.util.*;

import oberlin.builder.parser.ast.AST;

public abstract class NonTerminal implements AST {
    private final List<AST> astList;
    public NonTerminal(AST... astList) throws MismatchException {
        if(!checkTypes(astList)) throw new MismatchException("Nonterminal class " + this.getClass() + " does not match " +
        List<AST> list = new ArrayList();
        for(AST ast : astList) {
        this.astList = list; 
    public NonTerminal(List<AST> astList) throws MismatchException {
        try {
            this.astList = resolveTypes(astList);
        } catch(BuilderException ex) {
            throw new MismatchException(ex);
    public abstract List<Class<? extends AST>> getExpectedASTTypes();
     * Check to see that all provided ASTs are some extension of the expected class of AST,
     * and create the internal list of ASTs from it if possible.
     * @param astList current list of program ASTs 
     * @return true if the first ASTs match the expected ones, false otherwise
    private List<AST> resolveTypes(List<AST> astList) throws BuilderException {
        List<AST> ownASTs = new ArrayList<>();
        List<Class<? extends AST>> astTypes = getExpectedASTTypes();
        for(int i = 0; i < astTypes.size(); i++) {
            Class<? extends AST> provided, expected;
            provided = astList.get(i).getClass();
            expected = astTypes.get(i);
            if(!expected.isAssignableFrom(provided)) {
                throw new BuilderException("Cannot get " + expected + " from " + provided.getClass());
        return ownASTs;
     * Check to see that all provided ASTs are some extension of the expected class of AST.
     * @param astList current array of program ASTs 
     * @return true if the first ASTs match the expected ones, false otherwise
    private boolean checkTypes(AST... astList) {
        List<Class<? extends AST>> astTypes = getExpectedASTTypes();
        for(int i = 0; i < astList.length; i++) {
            Class<? extends AST> provided, expected;
            provided = astList[i].getClass();
            expected = astTypes.get(i);
            if(!provided.equals(expected)) {
                return false;
        return true;
    public int getElementCount() {
        return astList.size();

This is where the critical detail is left out of my code. If you already see it, then congratulations! In any case, I’ll let you all in on it at the end of the chapter. It goes back to the advantage of regular expressions over manually checking lists.

In “Programming Language Processors in Java”, David A. Watt and Deryck F. Brown use a visitor pattern to scan for these. That’s a perfectly valid approach. I get the same advantages through my constructor-or-exception pattern, which you have seen before. In fact, if you’re careful, you may notice that it’s another format of the same general pattern; without the traditional Visitor and Element classes. In my github repo, you might notice that I have the outline of a Visitor pattern implemented on these. Pay it no mind, it is proving unnecessary.

Still a good idea. Just unnecessary.

What Do They Mean to a Parser?

A Parser’s job is thus. Iterate through a list of Terminal syntax tree nodes. Compare the beginning of the list to established members of a list of nonterminal nodes. Find a match.

Truncate the matched beginning of the list of ASTs; this is, remove every element that is a member of the new NonTerminal. Now that they are removed, append that NonTerminal to the beginning of the list, in their place. Repeat this process until the list of ASTs is of size one, and return that singular AST.

And of course, if the program cannot be parsed down to a singular AST, throw a Parser Error.


These are meant to be identified through members of an enumeration; just as tokens were. However, you might remember BNF items such as:

Command ::= Command+

That is, a variable number of Command nodes can be condensed into the immediate tree of a single Command node. This is an issue for a simple list of ASTs, as in its provided implementation, NonTerminal simply attempts to match a fixed sequence of AST types to a provided list of ASTs, one by one. This is almost always inadequate for a programming language.

I’m going to step around writing a whole new series on how regular expressions work. (For now.) The important detail is that a minimum and maximum number of items, extending a specific AST class, need to be matched. I was tempted to call these “Clauses”, but that just barely relates to the actual definition of “clause”; so instead, we’ll borrow from regular expression terminology and call them Groups.

The down side? I’m still implementing Groups. They will have their own chapter though.

Now, we learn how to build a Scanner. (Again. Sort of.)


Tags: , , , , , , , , , , , , ,