AnnotationMetadataReadingVisitor does not honour @AliasFor [SPR-14437]

[spring-projects-issues]

Stéphane Nicoll opened SPR-14437 and commented

See #6337 for some context.

The following code does not behave the same way, considering that FooBar has some @AliasFor declaration:

AnnotationAttributes attributes = AnnotationAttributes.fromMap(
        metadata.getAnnotationAttributes(FooBar.class.getName()));

This code works as expected with StandardAnnotationMetadata while AnnotationMetadataReadingVisitor does not seem to work.

---

Affects: 4.3.1

Referenced from: commits 5ea8c26, 1ca4b81

[spring-projects-issues]

Sam Brannen commented

I'd say this is (almost) to be expected.

We have never claimed that @AliasFor works when using ASM.

In any case, like I mentioned in the Boot issue:

Out of curiosity (since I haven't cloned Boot 1.4 and tried it myself), what happens if you invoke getAliasedStringArray() instead of getStringArray() on the AnnotationAttributes?

[spring-projects-issues]

Stéphane Nicoll commented

We have never claimed that @AliasFor works when using ASM.

It might be me that's far from being obvious. That's the kind of technical detail that I should not worry about when I decide to use @AliasFor.

[spring-projects-issues]

Sam Brannen commented

I understand completely.

It is in fact not intuitive, and one could arguably expect that it works.

But that's unfortunately not the case.

Perhaps Juergen and I are the only ones who actually are aware of this shortcoming, in which case we should at least document @AliasFor support with ASM (or the lack thereof) and potentially add support for it (as much as feasibly possible).

[spring-projects-issues]

Juergen Hoeller commented

I've addressed this through a rather subtle refactoring now, making use of AnnotationAttributes's ability to store the resolved annotation type. Our ASM visitors try to resolve the annotation type upfront and operate against it if available: for default value resolution like before, but now also for alias resolution, all through the standard accessors. As a consequence, there is no need for getAliased* accessors anymore, so I've removed them from 5.0 (and will deprecate them in 4.3.2).

The hardest part is actually the differentiation between default values and user-specified values: We have that for our reflection mode already, and our ASM visitors behave analogously now. Also, for proper error reporting, we store textual references to the class and method in ASM mode now, instead of AnnotatedElement like in reflection mode.