...
That does not say there is no use for tree to tree transformations. ANTLR does so itself. I believe that the time to use such transformation is when the transformation results in a tree representing the same language, such as C to C or ANTLR to ANTLR. For example, it's extremely appropriate to normalize or optimize input by rewriting the tree. x+0 can be converted to just x, which means converting AST (+ x 0) to x or parse tree (expr x + 0) to (expr x). ANTLR transforms left recursive grammars to non-left recursive grammars using tree rewriting, but it does so manually at the moment. (boo)
ANTLR does not have tree transformation support at this time and it will not look like it did in v3 when it arrives. Many powerful rewriting systems lets let you perform rewrites in the concrete syntax of the language you are translating. E.g., see ASF+SDF by example. I imagine we'll do something very similar and allow a series of declarative rules like:
...
Sam Harwell and I are also talking about an xpath-like or other query mechanism to specify paths and nodes within parse trees. That way you could say: ``go get me all variable declarations'' and so on. There are lots of possibilities, given all of the research that has been done in this area.