So, I’ve just got my laptop back after a brief separation. What does one do?

Compile the JDK, of course!

For those unfamiliar with the situation, Apple hate Java developers. Well, one suspects they do. Apple being Apple, one never knows. We could get Java 1.6.0_12 with 32bit support tomorrow, or it could never come. You’ll never know until it appears.

Luckily, Landon Fuller loves us. Which Apple play with their iPhones, he beat the OpenJDK into submission on OS X. Only drawback – Swing is X11 based. But still, it’s JDK6 for 10.4, 10.5 and in both 32 and 64bit, and available as SoyLatte.

This work has all found its way back into the trunk, and so JDK7 is buildable on the Mac (usually). The only catch: the instructions are scattered, and often vary slightly. So I’m documenting them here for tonight’s build, mostly for my reference.

Thanks go to Landon Fuller and Stephen Bannasch for putting these together.

1. Get version 3.1 (or above, presumably) of XCode.

2. Download SoyLatte. I found the JDK7 build Landon released in 2008 did not work as a bootstrap.

3. Get Mercurial

sudo port install mercurial +bash_completion

4. Install the Forest extension

hg clone http://bitbucket.org/pmezard/hgforest-crew

You’ll need to point Mercurial at the hgforest-crew directory, by adding the following to ~/.hgrc:

[extensions]
hgext.forest=/opt/hgforest-crew/forest.py

5. Grab Kurt Miller’s BSD binary plugs:

wget http://www.intricatesoftware.com/distfiles/jdk-7-icedtea-plugs-1.6b.tar.gz

6. Get the JIBX libraries (version 1.1.5) from:

http://sourceforge.net/project/showfiles.php?group_id=69358&package_id=68290

7. Check out the OpenJDK:

hg fclone http://hg.openjdk.java.net/bsd-port/bsd-port

8. Place the following in build.sh in the bsd-port directory (modify as appropriate, depending on where you placed downloaded items):

LC_ALL=C
LANG=C
unset CLASSPATH
unset JAVA_HOME
make \
  ALT_BOOTDIR=/opt/soylatte16-i386-1.0.3/ \
  ALT_BINARY_PLUGS_PATH=/opt/jdk-7-icedtea-plugs \
  ALT_FREETYPE_HEADERS_PATH=/usr/X11R6/include \
  ALT_FREETYPE_LIB_PATH=/usr/X11R6/lib \
  ALT_JIBX_LIBS_PATH=/opt/jibx/lib \
  ALT_CUPS_HEADERS_PATH=/usr/include \
  ANT_HOME=/usr/share/ant \
  NO_DOCS=true \
  HOTSPOT_BUILD_JOBS=2

9. Run, and cross your fingers!

10. I ran into a ld error (archive has no table of contents). Should you hit this, try the following and then rerun build.sh:

ranlib build/bsd-i586/tmp/java/fdlibm/obj/*.a

11. Copy bsd-port/build/bsd-i586/j2sdk-image somewhere useful – and you’re done!

The problem is that getting familiar is a path strewn with brambles and holes full on punji sticks. One of the bigger holes is the event dispatch thread (EDT) – everything Swing related should take place on the EDT (even initialisation, under the latest Sun guidelines). When you’re trying to keep the UI fluid it’s all too easy to break the rule – hence, aspects to the rescue!

This topic has been covered by many before, including Alexander Potochkin and Anders Prisak – however, I found their solution needed a little tweaking to be used in our environment. In particular, they had missed two cases we cover – SwingUtilities and SwingWorker.

So, here’s the tweaked aspect. safeMethods now includes a few extras.

package org.infernus.swing.aspects;

import org.aspectj.lang.JoinPoint;
import org.aspectj.lang.annotation.Aspect;
import org.aspectj.lang.annotation.Before;
import org.aspectj.lang.annotation.Pointcut;

import java.awt.*;

@Aspect
public class EDTCheck {

    @Pointcut("call (* javax.swing..*+.*(..)) || "
            + "call (javax.swing..*+.new(..))")
    public void swingMethods() {
    }

    @Pointcut("call (* javax.swing..*+.add*Listener(..)) || "
            + "call (* javax.swing..*+.remove*Listener(..)) || "
            + "call (* javax.swing..*+.getListeners(..)) || "
            + "call (* javax.swing..*+.revalidate()) || "
            + "call (* javax.swing..*+.invalidate()) || "
            + "call (* javax.swing..*+.repaint()) || "
            + "target (javax.swing.SwingWorker+) || "
            + "call (* javax.swing.SwingUtilities+.invoke*(..)) || "
            + "call (* javax.swing.SwingUtilities+.isEventDispatchThread()) || "
            + "call (void javax.swing.JComponent+.setText(java.lang.String))")
    public void safeMethods() {
    }

    @Before("swingMethods() && !safeMethods() && !within(EDTCheck)")
    public void checkCallingThread(final JoinPoint.StaticPart thisJoinPointStatic) {
        if (!EventQueue.isDispatchThread()) {
            System.err.println("Swing EDT violation: " + thisJoinPointStatic.getSignature()
                    + " (" + thisJoinPointStatic.getSourceLocation() + ")");
            Thread.dumpStack();
        }
    }
}

Once it’s built, we just need to weave it – I’ve already got compile-time weaving configured for Spring @Configurable support, so just add the JAR containing the aspect as a weaveDependency and then the magic happens.

Now, if a Swing call is made off of the EDT, you’ll get complaints:

Swing EDT violation: String javax.swing.JTextArea.getText() (YourSwingClass.java:98)
java.lang.Exception: Stack trace
        at java.lang.Thread.dumpStack(Thread.java:1224)
        at org.infernus.swing.aspects.EDTCheck.checkCallingThread(EDTCheck.java:55)
     ... and so on