What are the benefits of using runtime proxies with an ORM provider like Hibernate or EclipseLink compared to bytecode instrumentation/enhancement?
I know that bytecode instrumentation helps to optimize dirty checking. Because it can intercept modifications, ORM provider doesn't need to traverse all the objects in the persistence context and compare the current values with the snapshot taken when the objects were loaded (thus avoiding additional memory consumption as well, I assume).
Also, I know that it can help in situations in which additional queries have to be executed because the ORM provider doesn't know whether a proxy needs to be created or what the primary key value in the proxy is, like in lazy one-to-one associations, thus making the associations effectively eager in some situations.
Some other proxy pitfalls have been described in this blog, like direct field access (in
hashCode methods) and using of
It seems to me that bytecode instrumentation is a better alternative. Does it have any pitfalls of its own compared to runtime proxies approach? Is there any rule of thumb which should be followed when choosing one instead of the other?