upgrading to mavericks – problems including broke eclipse/ant java for jmockit – and how to fix

I didn’t upgrade to Mavericks right away because I was getting FIOS.  If there were internet problems, I wanted to know they were Verizon’s fault and not be in a position where they could claim it was my computer’s problem.  I’ve had Mavericks long enough that I can write about the experience.

The problems describe here:

  • git
  • Eclipse/Java/Ant/JMockIt
  • OpenOffice crashing

How long it took

Not counting the download, the install took 2 hours.  Much of which was Apple saying there were “17 minutes” left.  I left it alone and it didn’t finish though.  I’ve seen reports of significantly faster installs though.

Problem #1 – git

The first problem I ran into was the error message The “git” command requires the command line developer tools.  Not a big deal.  It was a two minute install.  I was just surprised as I was expecting what was installed before the upgrade to  be installed after.

Problem #2 – Java and Eclipse

And now for the meat of this.  Being able to develop again in Eclipse was by no means seemless.

Error: Ant can’t find Java

Exact message: Specified VM install not found: type MacOS X VM, name Java SE 6 (MacOS X Default)

Solution – for each Ant build:

  • Run > External Tools Configuration
  • Run in same JRE as the workspace

Error: JUnit tests don’t run in Eclipse view

When running from Eclipse, I got the stack trace:


java.lang.IllegalStateException: Unable to load Java agent; please add lib/tools.jar from your JDK to the classpath

at mockit.internal.startup.JDK6AgentLoader.getVirtualMachineImplementationFromEmbeddedOnes(JDK6AgentLoader.java:102)

at mockit.internal.startup.JDK6AgentLoader.loadAgent(JDK6AgentLoader.java:74)

at mockit.internal.startup.AgentInitialization.initializeAccordingToJDKVersion(AgentInitialization.java:41)

at mockit.internal.startup.Startup.initializeIfNeeded(Startup.java:271)

at org.junit.runner.Runner.<clinit>(Runner.java:25)

at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)

at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:57)

at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)

at java.lang.reflect.Constructor.newInstance(Constructor.java:525)

at org.junit.internal.builders.AnnotatedBuilder.buildRunner(AnnotatedBuilder.java:35)

at org.junit.internal.builders.AnnotatedBuilder.runnerForClass(AnnotatedBuilder.java:24)

at org.junit.runners.model.RunnerBuilder.safeRunnerForClass(RunnerBuilder.java:57)

at org.junit.internal.builders.AllDefaultPossibilitiesBuilder.runnerForClass(AllDefaultPossibilitiesBuilder.java:29)

at org.junit.runners.model.RunnerBuilder.safeRunnerForClass(RunnerBuilder.java:57)

at org.junit.internal.requests.ClassRequest.getRunner(ClassRequest.java:24)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestReference.<init>(JUnit4TestReference.java:33)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestClassReference.<init>(JUnit4TestClassReference.java:25)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestLoader.createTest(JUnit4TestLoader.java:48)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestLoader.loadTests(JUnit4TestLoader.java:38)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.runTests(RemoteTestRunner.java:452)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.runTests(RemoteTestRunner.java:683)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.run(RemoteTestRunner.java:390)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.main(RemoteTestRunner.java:197)

java.lang.NoClassDefFoundError: org.junit.runner.Runner

at org.junit.runners.model.RunnerBuilder.safeRunnerForClass(RunnerBuilder.java:59)

at org.junit.internal.requests.ClassRequest.getRunner(ClassRequest.java:24)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestReference.<init>(JUnit4TestReference.java:33)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestClassReference.<init>(JUnit4TestClassReference.java:25)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestLoader.createTest(JUnit4TestLoader.java:48)

at org.eclipse.jdt.internal.junit4.runner.JUnit4TestLoader.loadTests(JUnit4TestLoader.java:38)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.runTests(RemoteTestRunner.java:452)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.runTests(RemoteTestRunner.java:683)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.run(RemoteTestRunner.java:390)

at org.eclipse.jdt.internal.junit.runner.RemoteTestRunner.main(RemoteTestRunner.java:197)

Solution:

First I tried switching creating a new installed JVM of type MacOS X JVM.  That didn’t help directly.  I left it like that since I didn’t want to muck around with my original configuration.  Then I added tools.jar to the installed JRE.

  1. Eclipse > Preferences
  2. Java > Installed JREs
  3. Click target JRE
  4. Edit
  5. Add external jars
  6. Add  /Library/Java/JavaVirtualMachines/1.7.0.jdk/Contents/Home/lib/tools.jar

Warning: The first time I tried this procedure, Eclipse hung.  I had it to kill it.  When I opened my workspace, I saw “unable to load plugin x” in every single view.  I used Time Machine to restore to before I tried changing the JREs and then re-did that change.  The second time it was successful.

I could have added tools.jar to my classpath for the project instead.  I choose not to because I was hoping updating it in the JVM install would solve the Ant problem (described below) as well.  It didn’t.  I also didn’t want to change the project .classpath as that is shared in Subversion and I didn’t want to make a change that affects others.

This whole exercise felt like a hack until I read the JMockIt instructions.  This leads me to believe I was running with a different JDK before the Mavericks upgrade.

If you are developing on a JDK of version 1.6 or newer on Mac OS X, add <jdkDir>/lib/tools.jar to the classpath, where <jdkDir> is the home directory for your local JDK 1.6 installation.

Error: JUnit tests don’t run from Ant

When running from Ant, every test failed with:

Caused an ERROR</span>
<pre>
null

java.lang.reflect.InvocationTargetException

at java.lang.reflect.Constructor.newInstance(Constructor.java:525)

What I learned:

  • I tried setting JAVA_HOME at the command line to no avail.  I confirmed javac is on the path. (/usr/bin/javac); it was.
  • I tried adding tools.jar to the classpath in launch configuration in Eclipse for running Ant.
  • If I change to fork=”no” in the <junit> task in the Ant script, most of the tests pass.  (Some of them require the JVM be forked which is why it was in the first place.) Showing the problem is when a new JVM is created, it doesn’t inherit the tools.jar from the previous step.
  • If I hard code the classpath entry for tools.jar in the Ant build file, all of the tests pass.

Solution:

If you need to fork the JVM, add the following insider you <junit> task.

<classpath location="/Library/Java/JavaVirtualMachines/1.7.0.jdk/Contents/Home/lib/tools222.jar" />

I don’t like this solution.  I also didn’t like the hack for Eclipse so my unease is pretty uniform here.  Luckily Ant doesn’t choke if you include a file (or disk or filesystem) in the classpath that doesn’t exist.  Which means this won’t hurt others working on the same project.

Problem #3 – Open Office crashes when opening a document

The solution was to upgrade from Open Office 3.3.0 to the latest (4)

My favorite new feature

It’s a minor one, but I like the feature to synchronization of “Read Later” between my Mac and iPad.  (I haven’t tried it yet so it may not be that nice in practice.)  Most of the new features are things I don’t need.

why parens aren’t like html

I’m currently taking the programming languages course on Coursera.  The instructor defends the parens LISP/Scheme/Racket by saying HTML takes the same approach but longer:

  • (foo maps to <foo>
  • ) maps to </foo>

And that people don’t complain about HTML.  The problem isn’t the number of characters though.  It’s the clarity. </foo> gives you context.  I think a better example would be:

<div id="one"><div id="two"><div id="three"></div></div></div>

Quick – which div is which?  I do agree with the instructor that indenting properly avoids this problem.  And that editors can tell you what the parens go with.

I’ll end with the cartoon he mentions in the lecture:  (http://imgs.xkcd.com/comics/lisp_cycles.png)

java puzzlers at the java sig

I got to see Josh Bloch and Bob Lee perform 9 Java Puzzlers.  That’s right.  I said perform.  They were funny, dynamic and had great banter.  It felt so natural.  And it was educational on top of all that.

I’m not going to try to transcribe the puzzlers.  You can read the book for many like this.  What I am going to do is mention my favorite 3 things I learned.

1) Double.MIN_VALUE isn’t what you think

Integer.MIN_VALUE is the smallest possible Integer.  So clearly Double.MIN_VALUE is the smallest possible Double.  Nope!  Double.MIN_VALUE is the smallest positive number that can be represented as a double.  4.9E-324.  Double.NEGATIVE_INFINITY is the smallest possible Double.

2) Compile time “constants” can change

System.out.println(Const.A + ” ” + Const.B);

If you compile and run this println against

interface Const {

String A = “a”;

String B = null;

}

it prints “a null”.

Then if you change it to run against

interface Const {

String A = “A”;

String B = “B”;

}

it prints “a B”.  That’s right.  The “a” is remembered because compile time constants are compiled in.  Whereas null isn’t a compile time constant so the new value is used.  Same for enums – not a compile time constant.

3) Autoboxing is a minefield

Ok.  So this one I knew already.  But it’s really important.  When comparing Integer objects, == is used.  Not autoboxing.  And JDKs are free to have the cutoff for where Integer objects are cached for autoboxing in different places.  So it is important not to rely on this.