We are seeing data race reported by tsan in java.lang.reflect.ProxyGenerator.addProxyMethod(). I did some analysis of the tsan report. It appears there are two different threads read/write a shared (same) ProxyMethod concurrently, while processing two different annotations. One thread is reading a ProxyMethod.exceptionTypes and the other thread is writing the ProxyMethod.exceptionTypes field from the same object.

ProxyGenerator.generateProxyClass() always creates a new ProxyGenerator instance and ProxyGenerator.proxyMethods map is private to each ProxyGenerator instance. In most cases, the ProxyMethod instances handled by generateProxyClass() are private to the invoking thread. 

[~jcking] noticed there are following static fields in ProxyGenerator class. Those could be accessed from different threads. 

```
    /* Preloaded ProxyMethod objects for methods in java.lang.Object */
    private static final ProxyMethod hashCodeMethod;
    private static final ProxyMethod equalsMethod;
    private static final ProxyMethod toStringMethod;
``` 

It would be good for someone who's familiar with the ProxyGenerator area to take a close look to see if the reported race is benign or needs to be fixed.