[antlr-interest] Grammars that uses Keyword as Identifiers

[Alexandre Porcelli]

Hi All...

  I'm constructing a T-SQL grammar and this grammar accepts keywords
as identifiers. So I started to implement as Terr suggested on his
book (page 287 - Keyword as Variables):

address_key
	: {input.LT(1).getText().equalsIgnoreCase("address")}? IDENTIFIER
	;

But... the generated code for the parser has a problem in
specialStateTransition method.. it has a semantic predicate use..  :

if ( ((input.LT(1).getText().equalsIgnoreCase("constraint")||input.LT(1).getText().equalsIgnoreCase("identity")||input.LT(1).getText().equalsIgnoreCase("collate")))
) {s = 3;}

else if ( ((input.LT(1).getText().equalsIgnoreCase("create")||input.LT(1).getText().equalsIgnoreCase("primary")||input.LT(1).getText().equalsIgnoreCase("alter")||input.LT(1).getText().equalsIgnoreCase("constraint")))
) {s = 2;}

I could solve this problem... just externalizing with a function:

@parser::members {
	private boolean validateIdentifierKey(String text){
		return input.LT(1).getText().equalsIgnoreCase(text);
	}
}

address_key
	: {validateIdentifierKey("address")}? IDENTIFIER
	;

Now it woks... but now the parser behavior looses most of its LL(*)
advantages (as Terr advices in the book when he talks about the DFA
evaluation)...

I'd like to ask/suggest/propose if it would be possible to create a
way to identify a special rule with some specific option should be
treated as an identifier but codes related to LL(*) still, something
like:

address_key
options {keywordText = "address";}
	: IDENTIFIER
	;

or something like fragment syntax:

keyword
address_key
	: 'address'
	;

There are several languages that needs this kind of behavior (I can
remember for now: COBOL, Natual/Adabas, JCL, T-SQL, PL-SQL).

Best Regards,
Alexandre Porcelli