ListAutoBean Iterator broken in current snapshot

[Ross Drever]

Calling remove on an iterator from list within an request factory entity proxy results in an UnsupportedOperationException.

I have isolated it to commit 288cc03 Hide Java 8 interfaces methods to generators. If I revert that commit it works properly. Seems like it is picking up the wrong implementation of iterator.remove() when it generates the proxies.

Iterator<SomeProxy> iterator = someEntityProxy.getSomeList().iterator();
iterator.next();
iterator.remove(); //throws UnsupportedOperationException

[Thomas Broyer]

+cc Colin, Goktug

Ah right; this is "actually" because Iterator#remove *became* a `default` method in Java 8:

https://docs.oracle.com/javase/7/docs/api/java/util/Iterator.html#remove%28%29 vs. https://docs.oracle.com/javase/8/docs/api/java/util/Iterator.html#remove--

Really not sure what would be the best outcome here; but I think we need to re-investigate this decision of *hiding* default methods to generators, as there might be other similar cases either in the JDK or in third-party libs moving to Java 8.

In the short-term, we may want to maybe add an annotation to ask GWT to expose a `default` method as if it were 'abstract', only for backwards-compatibility for cases like this.

For AutoBeans specifically, one possibility would be to *not* generate the IteratorAutoBean and instead provide one that implements the remove() method; or to special-case java.util.Iterator to have the generated shim extend a provided base class.

In the medium/long-term, AutoBeans would probably be rewritten as an annotation processor anyway (which could also unify the JVM vs. GWT implementations, instead of using java.lang.reflect.Proxy in the JVM).

[Roberto Lublinerman]

For the short term I think we should handle this by a fix in autobeans. For the long term we could introduce a version of getMethods(flag) that allows to specify which methods you want and the parameterless version defaults to Java 7 behaviour.

[Thomas Broyer]

OK, so I uploaded 3 changes: https://gwt-review.googlesource.com/15420

The first one (linked above) fixes the regression by special-casing j.u.Iterator; this is clearly a hack, we may miss other cases and other (third-party) generators may be impacted as well.

The second one adds getAllMethods / getAllOverloads / getAllInheritableMethos / getAllOverridableMethods to JClassType that return the Java 8 methods (static and default) for interfaces. I started with overloads taking a boolean argument but found new/separate methods to be easier.

The third commit replaces the hack in AutoBean with a proper fix using the new API.