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

[Gavin Lambert]

At 12:27 15/03/2007, Loring Craymer wrote:
 >As a first step, predicates should not specify more
 >tokens that the alternative being selected.  That
 >would rule out the example grammar.

They're often required for disambiguation, though -- I have a few 
rules of similar form in my own grammars (which do appear to work 
so far, although I do recall having to work around some nastyness 
at one point [which IIRC I did by changing the language to make it 
easier to parse, since it was a custom one]).

Regarding the original problem, I think predicates of that form 
will always require backtracking by their very nature, unless you 
have sufficient lookahead cached up already (which may be 
impossible for even fairly simple grammars).

Although it would be nicer if there was a simpler way to express 
them -- since this sort of predicate will usually be of the form 
"(a b c d) => a b" (ie. the matching portion is a strict prefix to 
the predicate), maybe this could be rewritten as something like "a 
b <c d>" (symbols just a suggestion; go with whatever makes more 
sense) to indicate that the tokens/rules at the end need to match 
but not be removed from the input stream and not be part of the 
production.  But that's mostly syntactic sugar, so can probably 
wait.