[Records] Component annotations not propagated when explicit canonical constructor is given

[Gunnar Morling]

Ah, yes, indeed! I'll track this issue then, thanks! Perhaps it would
even make sense to have it fixed for JDK 14?

--Gunnar

Am Mo., 20. Jan. 2020 um 17:35 Uhr schrieb Jorn Vernee <jorn.vernee at oracle.com>:
>
> Hi,
>
> I remember this issue coming up before. Looks like there was a fix
> targeted at JDK 15: https://bugs.openjdk.java.net/browse/JDK-8236597
>
> Jorn
>
> On 20/01/2020 09:36, Gunnar Morling wrote:
> > Thanks for your reply!
> >
> >> If a canonical ctor / accessor is explicitly specified, the annos on the declaration are used instead.
> > That's where things are still unclear for me, it seems contradictory
> > to "sounds like incorrect behavior to me" above. To make things more
> > concrete, here's an example:
> >
> >    public class RecordAnnos {
> >      public static void main(String... args) {
> >        System.out.println(Arrays.toString(Foo.class.getConstructors()[0].getParameters()[0].getAnnotations()));
> >        System.out.println(Arrays.toString(Bar.class.getConstructors()[0].getParameters()[0].getAnnotations()));
> >      }
> >
> >      public static record Foo (@Deprecated String foo) {}
> >
> >      public static record Bar (@Deprecated String bar) {
> >        public Bar {}
> >      }
> >    }
> >
> > Running this prints:
> >
> >    java --enable-preview --source 14 RecordAnnos.java
> >    [@java.lang.Deprecated(forRemoval=false, since="")]
> >    []
> >
> > Whereas I hoped it'd print this:
> >
> >    [@java.lang.Deprecated(forRemoval=false, since="")]
> >    [@java.lang.Deprecated(forRemoval=false, since="")]
> >
> > In case an explicit canonical constructor is declared without the
> > formal parameter list, there's no way to specify any annotations on
> > the parameters. So shouldn't those from the components be applied?
> > With the current behaviour, if I want to declare any annotation for
> > the constructor itself, I need to repeat all annotations from the
> > components on the annotations of a fully manually implemented
> > constructor:
> >
> >    public static record Bar (@Deprecated String bar) {
> >      @Deprecated
> >      public Bar(@Deprecated String bar) {
> >        this.bar = bar;
> >      }
> >   }
> >
> > Note that @Deprecated is just used here to make the example
> > self-contained; my actual use case is about Bean Validation constraint
> > annotations.
> >
> > Thanks,
> >
> > --Gunnar
> >
> > Am Mo., 20. Jan. 2020 um 01:32 Uhr schrieb Brian Goetz <brian.goetz at oracle.com>:
> >> This sounds like incorrect behavior to me.  The way it should work is:
> >>
> >>   - A declaration annotation is applicable to a record component if it has no @Target meta-anno, or its target includes one or more of PARAMETER, FIELD, METHOD, RECORD_COMPONENT.
> >>   - For _each_ of the applicable types present, the anno is pushed down to the corresponding _implicit_ { ctor parameter, field, accessor method, record component }.
> >>   - If multiple applicable types are present, it is pushed down to all of them.
> >>   - If a canonical ctor / accessor is explicitly specified, the annos on the declaration are used instead.
> >>
> >>
> >>
> >>
> >> On 1/19/2020 4:14 PM, Gunnar Morling wrote:
> >>
> >> Hi,
> >>
> >> I'm observing the following behaviour around annotations on record
> >> components, using JDK 14 b32. I'm putting an annotation to a record
> >> component:
> >>
> >>      public record Person(@NotNull String name) {}
> >>
> >> In this case the @NotNull annotation is propagated to the
> >> corresponding parameter of the generated constructor, from where I can
> >> obtain it using reflection. This is not the case though when I
> >> explicitly declare the canonical constructor:
> >>
> >>      public record Person(@NotNull String name) {
> >>          public Person {
> >>              // ...
> >>          }
> >>      }
> >>
> >> In this case the annotation isn't propagated, and it won't be
> >> retrievable from that constructor's parameter via reflection.
> >>
> >> Is this behaviour intended or is it a bug actually? I lean towards the
> >> latter, as I don't explicitly define the parameter in the constructor,
> >> so I'd expect the annotations given on the component to still be
> >> propagated.
> >>
> >> If it *is* intended, it'd make my use case a bit more complex, as I'd
> >> want to be able to put other annotations to the canonical constructor
> >> *itself*, while still getting all the component annotations propagated
> >> to its parameters.
> >>
> >> Thanks a lot,
> >>
> >> --Gunnar
> >>
> >>