[antlr-interest] [bug?] AST rewrite rule with no (imaginary) token yields C runtime compile error

[Gavin Lambert]

At 20:07 30/05/2008, David Benn wrote:
 >predefined_type_def
 >                  : primitive_type -> ^(primitive_type)
 >                  | IDENT IDENT -> ^(IDENT IDENT)
 >                  ;
 >
 >i.e. it had no imaginary root node in the AST rewrite rule.
[...]
 >Assuming the rewrite rule is legal (but perhaps not desirable), 
I
 >guess this is a bug in the C runtime code generation step.

I vaguely remember a discussion on something similar a while 
back.  I think the general idea was that this sort of thing isn't 
legal, since 'primitive_type' might expand to zero or more tokens, 
and exactly one token must be the root node.

You could always just not try to wrap it into another subtree 
level though -- 'primitive_type -> primitive_type' should work 
fine.  (In fact I think just 'primitive_type' alone should also 
work, but I vaguely recall something odd with that as well.)