[antlr-interest] uh oh...trouble in meaning of (..)=> pred!!!

Terence Parr parrt at cs.usfca.edu
Wed Mar 14 16:05:58 PDT 2007

Hi.  Harmut submitted a bug report, which I have converted to a parser:

grammar T;

x : (a d) => a
   | (b d) => b
   | ('a'|'b')+

a       :       'a' 'a' ;
b       :       'a' 'a' 'b' ;
digit   :       '0'|'1' ;

Basically, in the book and in my intentions, predicates order the  
alts.  The problem is that ANTLR's analysis doesn't consider  
syntactic predicates if it can figure out what to do w/o them.   
That's an optimization.  The problem is that you are often specifying  
the lookahead in the predicate that must be evaluated.  Crap.  ANTLR  
is not forcing those predicates in there.

For semantic predicates, we have {...}? and {...}?=> where the latter  
forces backtracking.  Perhaps (...)=> should always force  
backtracking.  BUT, for backtracking=true, I add a predicate to every  
alt!  I guess for that backtracking mode, those predicates should be  
analogous to {...}? and manually specified (...)=> should operate  
like {..}?

I have exactly 4 days to resolve this issue before the book goes to  
copy editing.  Anybody wanna help me think about this?


More information about the antlr-interest mailing list