SpEL: problem with static methods conflicting with instance methods [SPR-12502]

[spring-projects-issues]

Andy Clement opened SPR-12502 and commented

This problem came up in Spring XD and I think has come up before. Consider this testcode:

public class User {
  public String getName() { return "Andy"; }
}
...
Expression expression = null;
expression = parser.parseExpression("#root.getClass().getName()");
User u = new User();
assertEquals(NamedUser.class.getName(),expression.getValue(u));

The code in ReflectiveMethodResolver.getMethods() tries to ensure static methods are easily callable so merges two sets of candidates together. This causes problems above because the root object User supports getName() but we want to invoke the getName() on the Class object. Now it sometimes works and picks the right one - depending on the JDK you run it on. That variation is due to the ordering of elements from a collection changing across JDKs. No ordering was specified so that isn't unreasonable...

The solution is to change getMethods() - ensure the ordering is fixed by using a LinkedHashSet rather than a HashSet and ensure the candidates we are most likely interested in are first.

---

Affects: 3.2.12, 4.0.8, 4.1.2

Attachments:

• 0001-Adjust-method-resolution-to-avoid-problems-calling-C.patch (3.67 kB)

Issue Links:

• XD-2409 hdfs-dataset sink with getName() method in Pojo ("is depended on by")
• INT-3585 SpEL PropertyAccessor Order
• SpEL chooses wrong method to invoke when varargs method is present [SPR-12803] #17400 SpEL chooses wrong method to invoke when varargs method is present
• ReflectiveMethodResolver chooses parent class over child for static methods [SPR-13918] #18492 ReflectiveMethodResolver chooses parent class over child for static methods

Backported to: 4.0.9, 3.2.13

[spring-projects-issues]

Juergen Hoeller commented

Andy, let's try to fix this for 4.1.3 still (i.e. by the end of the week)...

Juergen

[spring-projects-issues]

Andy Clement commented

This is a patch that fixes it. I'm a little nervous since this is quite an important area. It does pass the new test as well as all existing tests. It'd be good if someone else cast an eye over it.

[spring-projects-issues]

Andy Clement commented

Juergen - if you let me know what you think of the patch, I can get it in asap.

[spring-projects-issues]

Juergen Hoeller commented

Andy, this looks fine to me. I'm rolling it in right away in a slightly streamlined fashion (also reducing the multi-transformation of the Method set/array/list - another suboptimal arrangement in the old version of that code).

I'm even wondering whether to backport this to 4.0.9 and 3.2.13 at the end of December since such runtime differences between JVMs can be quite unpleasant and hard to track down :-(

Juergen

[spring-projects-issues]

Andy Clement commented

Thanks for rolling it in. It is a nasty issue when you hit it so could be worth back porting, yes.