Per Hibernate Core docs, accessing a lazy property on a bytecode-instrumented
Entity returns an unproxied target property. From what I can tell, however, in
the case where the target property is an Entity as well, this is implemented by
(in hibernate 3.3.1.GA):
-generating a proxy for target entity
-setting unwrap flag to true
-immediately unwrapping it.
In (both javassist and cglib) FieldInterceptorImpl:
public Object readObject(Object target, String name, Object oldValue) {
Object value = intercept( target, name, oldValue );
if (value instanceof HibernateProxy) {
LazyInitializer li = ( (HibernateProxy) value
).getHibernateLazyInitializer();
if ( li.isUnwrap() ) {
value = li.getImplementation();
}
}
return value;
}
In our test, The implicit call to
DefaultLoadEventListener.createProxyIfNecessary() contributes 13% of exec time,
9% coming AbstractEntityPersister.createProxy(). (our tests execute read-only
methods that hit 2nd level cache. For context, fetching actual data from 2nd
level cache takes only 6% of exec time). In other words, proxy generation isn't
cheap.
Given the stated intent to return an unproxied entity why not ensure that
AbstractEntityPersister.initializeLazyProperty returns unproxied values? Better
yet, why not introduce another version of EntityType.resolveIdentifier that
bypasses proxy creation (via immediateLoad maybe)?
If there are some really good reason to create and immediately discard proxies
in this case, createProxy() should be made cheaper. 1/3 of that method is spend
doing ReflectHelper.overridesEquals(Class) in BasicLazyInitializer constructor.
Any reason not to cache results of that method (and, actually, other
ReflectHelper methods)?
I'm more than happy to write contribute the code given a little guidance.
Thanks
-nikita
_______________________________________________
hibernate-dev mailing list
hibernate-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev
