review fix(annotation): bug when using an annotation targeting declaration and types

[surli]

Fix #1773

In this case the JLS says that: "It is possible for @foo to be both a declaration annotation and a type annotation simultaneously." So we should attach the annotation to the field/method AND to the type in Spoon model.
Then in this case we don't want to print it twice: we print it only for the field/method.

[surli]

This PR is now ready to be reviewed @pvojtechovsky @monperrus

Please note that the PR is potentially breaking.

When using a code like:

@MyAnnotation
String myField;

@Target({ElementType.TYPE_USE, ElementType.FIELD})
public @interface MyAnnotation {}

the model will now contain MyAnnotation annotation on the field AND on the type reference String.
This has been done according to the following sentence of the JLS:

@foo int f;
@foo is a declaration annotation on f if Foo is meta-annotated by @target(ElementType.FIELD), and a type annotation on int if Foo is meta-annotated by @target(ElementType.TYPE_USE). It is possible for @foo to be both a declaration annotation and a type annotation simultaneously.

So this is not true only for fields, but for all declaration.
Moreover, in case of the usage of a an annotation without @Target the JLS says that

If an annotation of type java.lang.annotation.Target is not present on the declaration of an annotation type T, then T is applicable in all declaration contexts except type parameter declarations, and in no type contexts.

So I considered those exceptions for adding the annotation or not.
It means that in no classpath, if we have no information about the annotation type, MyAnnotation would be only attached to the field on the example below: this is the previous Spoon behaviour.

[monperrus]

explain this complex condition with a fat comment?

[monperrus]

use an anonymous scanner to avoid multiple instanceof?

[surli]

I don't see how to use easily an anonymous scanner here: I must return a variable and I cannot store it as a local variable if I use the anonymous scanner. And I don't have any method to call to setup the value.

[monperrus]

I agree, it's not straightforward.

I must return a variable and I cannot store it as a local variable if I use the anonymous scanner

Unless you create an inline class

class Result { CtVariable v }
final Result result = new Result();
// then the anonymous class

[seintur]

don't have -> have not

[seintur]

"can be sure" or "cannot be sure"?

[monperrus]

merge?

[surli]

merge?

ok for me