[antlr-interest] ColonAssign vs Label Colon vs Colon

[Gavin Lambert]

At 06:32 27/06/2008, Michael Fogeborg wrote:
 >COLON	: ':' ;
 >IDENT	: 'A'..'Z' ;
 >ASSIGN	: ID ':=' ID ;
 >LABEL	: IDENT COLON ;
 >
 >The Lexer gets lost on the COLON and I can't get this working...

This is because you're mixing leaf rules (COLON and IDENT) with 
non-leaf rules (LABEL and ASSIGN).  (And ASSIGN is using an 
unlisted rule, but I'm assuming that's a typo.)

ASSIGN definitely looks like it should be in the parser, not the 
lexer, as whitespace ought to be permitted around the ':=' sign, 
and it's much easier to deal with multiple IDs there.  The same 
could also be true for LABEL -- most languages that do labels this 
way do permit whitespace to be between the identifier and the 
colon.

If you really want to match it via a lexer rule, though, then 
you'll need to merge the rules and manually disambiguate:

/* this exists only to create the token name; it'll never be 
called so the content is unimportant */
fragment LABEL: IDENT COLON;

IDENT
   : ('A'..'Z')  /* should that be ('A'..'Z')+ or did you really 
want single-character identifiers? */
     ((COLON) => COLON { $type = LABEL; })?
   ;

(In this case you probably don't actually need the syntactic 
predicate, but I usually prefer to put it in anyway, for symmetry 
with the cases where you do need it.)