[PATCH] Add support for named values in the parser.

[Samuel Benzaquen]

================
Comment at: lib/ASTMatchers/Dynamic/Parser.cpp:431
@@ +430,3 @@
+        S->getNamedValue(Tokenizer->peekNextToken().Text);
+    if (!NamedValue.isNothing()) {
+      Tokenizer->consumeNextToken();  // Actually consume it.
----------------
Peter Collingbourne wrote:
> What if the user defines a named value with the same name as a matcher? Yes, probably a bad idea, but it's easy to disambiguate between a matcher expression and a named value by checking whether the next token is a '('. That would also let us give better diagnostics (i.e. "no such named value" instead of "no such matcher").
> 
> It would also be nice (but not required for this change) if it were possible to bind against named values.
I thought about it.
What happens if you have: "expr". What error should that give?
Right now it gives "missing (". If I change it like you say it will give "no such value", which might be confusing for anyone that doesn't know about named values.

Another alternative would be to accept "expr" as a matcher with no arguments. That way I try a named value, and then a matcher with no args. But what happens if it can't find any? which error should it give then?

wrt binding named values, the user can always wrap it in a bindable thing, like expr(my_conditions).bind("foo").


http://llvm-reviews.chandlerc.com/D3229

COMMIT
  http://llvm-reviews.chandlerc.com/rL205419