Spring AOP @Aspect pointcut ignores generic type params and matches to many join points [SPR-3628]

[spring-projects-issues]

Oliver Hutchison opened SPR-3628 and commented

I raised this issue on the forum but received no responses:

http://forum.springframework.org/showthread.php?t=40559

This is a critical issue for us as we've been forced off AspectJ because of instability there and now we find that Spring AOP does not implement the same pointcut model.

---

Issue Links:

• Spring AOP doesn't process generics correctly [SPR-3556] #8239 Spring AOP doesn't process generics correctly ("depends on")

[spring-projects-issues]

Juergen Hoeller commented

Ramnivas, can you please try to check this case as well? It may be solved as part of the recent #8239 fix, but we should double-check.

Juergen

[spring-projects-issues]

Ramnivas Laddad commented

Here is preliminary finding:

Return type specification with an after-returning advice plays a part in determination of a join point match. However, there is a difference in how the AspectJ weaver and Spring AOP use this information. In AspectJ, the statically-determinable portion is computed at weave-time itself (so it can determine that certain advice cannot be applied to a join point based on statically available information alone). If there is a residue it leads to the compiler issuing an "unchecked match" warning. In Spring, this match is performed entirely dynamically in AspectJAfterReturningAdvice.shouldInvokeOnReturnValueOf() that does not use generics type information. So a fix would involve keeping generic type info associated with the return type of the advice method and matching it with the return type of the advised method

Working on it...

[spring-projects-issues]

Ramnivas Laddad commented

Fixed.

After advice now considers generic type information to align semantics with that of AspectJ. Tests includes assignment to raw types, lower and upper bounded types.