[jdt-core-dev] TypeReferences vs NameReferences in type cast expressions

[Srikanth S Adayapalam <srikanth_sankaran@xxxxxxxxxx>]

Hello past & present (and future) jdt/core developers,

      I am hoping you can shed some light on this oddity in the parser
code. Thanks in advance for any insights you can share.

Here is a fragment of the java grammar :

      CastExpression ::= PushLPAREN Name PushRPAREN InsideCastExpressionLL1
UnaryExpressionNotPlusMinus

      InsideCastExpressionLL1 ::= $empty
      /.$putCase consumeInsideCastExpressionLL1(); $break ./

org.eclipse.jdt.internal.compiler.parser.Parser.consumeInsideCastExpressionLL1()
 looks like:

protected void consumeInsideCastExpressionLL1() {
      // InsideCastExpressionLL1 ::= $empty
      pushOnExpressionStack(getUnspecifiedReferenceOptimized());
} and

where getUnspecifiedReferenceOptimized() constructs a SingleNameReference
or QualifiedNameReference as the case demands.

Now that is puzzling - why would we construct a NameReference instead of a
TypeReference here ? Intuitively as well per spec the Name
here is guaranteed to be a Typename  (JLS3.0, page 128, 6.5.1)

The 4 other productions for CastExpression use TypeReference and the
production above is the only one that constructs a NameReference.
Do you recall how and why this came about ?

To give some context as to why this matters - JSR308 adds type annotations
to java and so TypeReferences must now be enhanced to capture annotations.
Name here is really a TypeReference, but if it was coded to be a
NameReference for some unalterable reason, we would need to add annotations
fields to these types also - something I would like to avoid very much.

I will make the change to TypeReference and see what breaks, in  the
meantime would appreciate any inputs you have.

Thanks!
Srikanth.