Custom ANTLR grammar not working for every input

Question

I am trying to write a grammar for our custom rule engine which uses ANTLR (for parsing) and Pentaho Kettle (for executing the rules)

Valid inputs for the parser would be of the type:
(<Attribute_name> <Relational_Operator> <Value>) AND/OR (<Attribute_name> <Relational_Operator> <Value>)
i.e. PERSON_PHONE = 123456789

Here is my grammar:

grammar RuleGrammar;
options{
language=Java;
}

prog                : condition;

condition
                                :  LHSOPERAND RELATIONOPERATOR RHSOPERAND
                                ;

LHSOPERAND
                                :  STRINGVALUE
                                ;

RHSOPERAND
                                :  NUMBERVALUE    |
                                   STRINGVALUE
                                ;


RELATIONOPERATOR
                                :   '>'    |
                                     '=>'  |
                                     '<'   |
                                     '<='  |
                                     '='   |
                                     '<>'
                                ;

fragment NUMBERVALUE
                              : '0'..'9'+
                              ;

fragment STRINGVALUE
                              :  ('a'..'z' | 'A'..'Z' | '_') ('a'..'z' | 'A'..'Z' | '_')*
                              ;


fragment LOGICALOPERATOR
                              :  'AND' |
                                 'OR'  |
                                 'NOT'
                              ;

The issue I am facing is comparing against string value i.e. PERSON_NAME=1 would pass the grammar, but the value PERSON_NAME=BATMAN does not work. I am using ANTLRWorks and on debugging for PERSON_NAME=BATMAN, I get a MismatchTokenException for the RHS value.

Can anyone please guide me where I am going wrong?

Answer 1

BATMAN is being tokenized as a LHSOPERAND token. You must realize that the lexer does not take into account what the parser "needs" at a particular time. The lexer simply tries to match as much as possible, and in case 2 (or more) rules match the same amount of characters (LHSOPERAND and RHSOPERAND in your case), the rule defined first will "win", which is the LHSOPERAND rule.

EDIT

Look at it like this: first the lexer receives the character stream which it converts in a stream of tokens. After all tokens have been created, the parser receives these tokens which it then tries to make sense of. Tokens are not created during parsing (in parser rules), but before it.

A quick demo of how you could do it:

grammar RuleGrammar;

prog
 : condition EOF
 ;

condition
 : logical
 ;

logical
 : relational ((AND | OR) relational)*
 ;

relational
 : STRINGVALUE ((GT | GTEQ | LT | LTEQ | EQ | NEQ) term)?
 ;

term
 : STRINGVALUE
 | NUMBERVALUE
 | '(' condition ')'
 ;

GT          : '>';
GTEQ        : '>=';
LT          : '<';
LTEQ        : '<=';
EQ          : '=';
NEQ         : '<>';
NUMBERVALUE : '0'..'9'+;
AND         : 'AND';
OR          : 'OR';
STRINGVALUE : ('a'..'z' | 'A'..'Z' | '_')+;
SPACE       : ' ' {skip();};

(note that EQ and NEQ aren't really relational operators...)

Parsing input like:

PERSON_NAME = BATMAN OR age <> 42

would now result in the following parse: