Currently, we have two injection methods to handle method calls: beforeMethodCall and beforeAtomicMethodCall.
This solution has several problems:
Code duplication.
We do not handle classes inheriting from Atomic* classes propertly, because their methods are not detected as "atomic" statically by the MethodCallTransformer bytecode transformer that injects beforeMethodCall and beforeAtomicMethodCall invocations. This is because this transformer decides what method to insert statically based on the class name of the invoked method. It checks only for j.u.c.Atomic* method names, and do not handles subclasses of atomics.
This PR removes beforeAtomicMethodCall injection, letting the managed strategy to decide how to handle the method dynamically at runtime, based on the actual type of the method receiver.
I've also added a new test AtomicSubclassTest.kt checking that subclasses of Atomic* classes are properly supported.
Please note, that with this PR the memory access methods of Atomic*, Atomic*FieldUpdater, VarHandle and Unsafe APIs are treated as atomic. Therefore, this PR also closes #288.
The latter also implies that Atomic*Array::length() method is no longer considered to be atomic method.
Because of this, it has disappeared from the trace output of some tests.
Currently, we have two injection methods to handle method calls:
beforeMethodCallandbeforeAtomicMethodCall. This solution has several problems:Code duplication.
We do not handle classes inheriting from
Atomic*classes propertly, because their methods are not detected as "atomic" statically by theMethodCallTransformerbytecode transformer that injectsbeforeMethodCallandbeforeAtomicMethodCallinvocations. This is because this transformer decides what method to insert statically based on the class name of the invoked method. It checks only forj.u.c.Atomic*method names, and do not handles subclasses of atomics.This PR removes
beforeAtomicMethodCallinjection, letting the managed strategy to decide how to handle the method dynamically at runtime, based on the actual type of the method receiver.I've also added a new test
AtomicSubclassTest.ktchecking that subclasses ofAtomic*classes are properly supported.Please note, that with this PR the memory access methods of
Atomic*,Atomic*FieldUpdater,VarHandleandUnsafeAPIs are treated as atomic. Therefore, this PR also closes #288.The latter also implies that
Atomic*Array::length()method is no longer considered to be atomic method. Because of this, it has disappeared from the trace output of some tests.