Avoid exceptions when evaluating validation hints

[dreis2211]

Hi,

we noticed in one of our loadtests that a substantial amount of "hidden" NoSuchMethodExceptions is being thrown when using @javax.validation.Valid on controller methods.

@PostMapping("test")
public void test(@RequestBody @Valid @NotNull TestRequest request) {
     // Some code
}

This is caused by calling AnnotationUtils.getValue() on this annotation, which has no such property. This PR bypasses the annotation utility call. I also took the freedom to refactor this logic into a common place in ValidationUtils to avoid copy-paste throughout the different modules.

An isolated JMH benchmark shows the following results when dealing with @Valid:

Benchmark                                         Mode  Cnt     Score     Error   Units
MyBenchmark.testNew                               avgt   10     6,253 ±   0,312   ns/op
MyBenchmark.testOld                               avgt   10  1761,995 ±  92,262   ns/op

As a workaround you can use org.springframework.validation.annotation.Validated of course, but I would prefer not to have such a drawback when using javax.validation.Valid.

Let me know what you think.

Cheers,
Christoph

[dreis2211]

That would be 5.3.7 now - let me know if I should add that.

[jhoeller]

Hey @dreis2211 it's indeed worth handling the common javax.annotation.Valid case explicitly there, and also worth extracting the evaluation logic. However, it'll have to be a new org.springframework.validation.annotation.ValidationAnnotationUtils class in order to avoid a package cycle, since it refers to the Validated annotation in the subpackage there. If you could refactor it that way with a since marker for 5.3.7, we'd be happy to roll it in...

[dreis2211]

@jhoeller Thanks for the review. I just committed the requested changes.

[snicoll]

@dreis2211 I can see ModelAttributeMethodProcessorTests is failing. Can you please have a look?

[dreis2211]

@snicoll Should be fixed - I was a bit overly ambitious and wanted to optimize the custom @Valid annotations case as well, which broke when the annotation was defined as an inner class.

[snicoll]

Thanks for the follow-up Christoph.