In our JVM (see Comments section) we do some extra validity testing in our
debug version and assert when running:

-Xverify -Xmx128m -Xgc:gencon -cp c:\testing\jck\JCK-runtime-14a\classes
-Djava.security.policy=c:\testing\jck\JCK-runtime-14a\lib\jck.policy
javasoft.sqe.tests.api.java.awt.java2d.font.LineBreakMeasurer.ConstructorTest
-TestCaseID testCase4


Java stack-trace:
    at java.text.Bidi.nativeBidiChars(Native Method)
    at java.text.Bidi.<init>(Bidi.java:189)@0BB0CB6B
    at java.awt.font.TextMeasurer.initAll(TextMeasurer.java:190)
    at java.awt.font.TextMeasurer.<init>(TextMeasurer.java:149)
    at java.awt.font.LineBreakMeasurer.<init>(LineBreakMeasurer.java:292)
    at javasoft.sqe.tests.api.java.awt.java2d.font.LineBreakMeasurer.Constructor
Test.testCase4(ConstructorTest.java:237)

The problem is in nativeBidiChars (in jbidi.c) it does:

   jchar *cText = (jchar*)(*env)->GetPrimitiveArrayCritical(env, text, NULL);
   ..
   ..
   .. resetBidi(); <-----
   ..
   (*env)->ReleasePrimitiveArrayCritical(env, text, cText, JNI_ABORT);

The problem is in resetBidi which does
   g_bidi_class = (*env)->NewGlobalRef(env, cls);

while holding the PrimitiveArrayCriticial(cText)

The JNI specification clearly dictates that no other JNI-calls may be
made between GetPrimitive and ReleasePrimitive.