NoSuchMethodError on Decapsulation.setField with JDK9 build 118

[russgold]

We get:

java.lang.NoSuchMethodError: sun.reflect.ReflectionFactory.newFieldAccessor(Ljava/lang/reflect/Field;Z)Lsun/reflect/FieldAccessor;
at weblogic.server.ServerLifecycleRuntimeTest.testShutdownRequestTimeout(ServerLifecycleRuntimeTest.java:244)
at weblogic.server.ServerLifecycleRuntimeTest.testNewDefaultShutdownRequestTimeout(ServerLifecycleRuntimeTest.java:222)

when running with JMockit 1.23 on the latest version of JDK9, as JMockit is using internal Sun APIs to access internals. These were removed in build 115

[rliesenfeld]

Is the test calling Deencapsulation.setField on a static final field? If so, it won't work on JDK 9 for the reason exposed, and the test should be changed. However, apart from that there is no problem in keeping this method for use on JDK 8 or older.

[russgold]

Yes, the call from which this stack trace was taken does set a static final field. It works in JDK8 just fine. It doesn't work in JDK9, as of build 115. Apparently, we have about a thousand such calls in our tests, so changing all of these tests isn't particularly feasible. Are you saying that you are unable to make it work in the latest JDK9?

[rliesenfeld]

As far as I know, there is no way to make it work in JDK 9, because they removed the "sun.reflect.ReflectionFactory" class, with no equivalent replacement. I did some google searches, but couldn't find anything useful on the topic.

I am inclined to remove this feature (the ability to set a static final field), as it has always been of dubious value anyway. And if it won't be possible anymore in Java 9+, people will have to change their tests sooner or later.

[russgold]

Take a look at the method setPrivateStaticField in this class

I have confirmed that this works in JDK9, other than for primitives and strings, which the compiler might optimize away.

It turns out to be quite useful in a lot of cases, at least with regard to test written with SimpleStub. With JMockit, you could probably do it a different way; unfortunately, a lot of our developers have chosen to do it this way.

[rliesenfeld]

Yes, it does work. For some reason, I thought that setting the field through "ReflectionFactory#newFieldAccessor(...)" was needed, but it's not.