[antlr-interest] Warning(200) for k=1

[Gavin Lambert]

At 12:10 15/06/2008, Aurelian Melinte wrote:
 >I am forced to use k=1 for a large C++ grammar - while the 
grammar
 >has been designed for k=2, Antlr 3.0.1 will run out of memory 
for
 >anything else but k=1. I

I'd start by fixing up your lexer.  And write some unit tests for 
it.

 >OCTALINT
 >	:
 >		'0' ('0'..'7')* (IntegerSuffix)?
 >	;
 >
 >HEXADECIMALINT
 >	:
 >		'0' ('x' | 'X') (HexadecimalDigit)+ (IntegerSuffix)?
 >	;
 >
 >FLOATONE
 >	:
 >		( (Digit)+ ('.' | 'e' | 'E') )=>
 >		(Digit)+
 >		( '.' (Digit)* (Exponent)?
 >		| Exponent	// should be FLOATTWO?
 >		)
 >	;
 >
 >FLOATTWO
 >	:
 >		'.' (Digit)+ (Exponent)? (FloatSuffix)?
 >	;

At k=1 (and when encountering loops, the lexer always behaves like 
it's k=1), these rules are mutually ambiguous.  You should merge 
them into a single rule with appropriate lookahead disambiguation.

Remember, predicates don't choose between rules, they choose 
between alts within a rule.  Or that's how they seem to behave in 
the lexer, at least.

 >IntegerSuffix
 >	:
 >		LongSuffix (UnsignedSuffix)
 >		UnsignedSuffix (LongSuffix)
 >	;
 >
 >FloatSuffix
 >	:
 >		('f' | 'F' | 'l' | 'L' )
 >	;

Surely these should be fragments?