[CLOSED] Investigate hashcode() and finalzie() non-determinism

[aionbot]

Issue created by yulongaion (on Friday May 04, 2018 at 18:31 GMT)

Approaches:

• code injection
• agent
• method override

[aionbot]

Comment by jeff-aion (on Friday May 04, 2018 at 20:32 GMT)

After initially looking like we couldn't use the Java JVMTI agent to replace a native method in java/lang/Object, through retransformClass (was having a problem with StackOverflowError), it turned out that this was just a special-case I needed to account for, from the ASM library.
It seems like the ASM library won't call visitCode on a native method, even if you strip the native modifier from it. This was leaving a non-native, yet code-less, method in the bytecode. Presumably, this retransformation bypasses verification, so the only problem was when attempting to run this empty method resulted in the StackOverflowError (probably because the empty method defined no stack requirement).
In this case, directly calling the visitCode method on the new node resulted, in the case where we removed the native modifier, seemed to result in a correctly-functioning VM and method.
In my experiments, it appears as though this will work for both Object.hashCode() and String.intern().

[aionbot]

Comment by JunhanHu-aion (on Tuesday May 22, 2018 at 21:39 GMT)

https://stackoverflow.com/questions/2237720/what-is-an-objects-hash-code-if-hashcode-is-not-overridden/32454673#32454673
-XX:hashCode=n is a great option for testing.

[aionbot]

Comment by jeff-aion (on Friday May 25, 2018 at 13:33 GMT)

I think that the last change required, in order to close this issue, is that we should reject any user-provided class which attempts to override finalize since we know that it won't work in our renaming scheme.