Question

In Java.g, we can find the following rule (whitespace removed for clarity):

statement 
     :   block
         |   ('assert') expression (':' expression)? ';'
         |   'assert'  expression (':' expression)? ';'            
         |   'if' parExpression statement ('else' statement)?          
         (...)
         ;

I’ve two questions:

1. Are two ‘assert’ subrules really needed?
2. If not, why doesn’t ANTLR complain of ambiguity, since they seem to define exactly the same input?

Thanks in Advance.

Answer 1

Pedro F. wrote:

Are two ‘assert’ subrules really needed?

No, they’re equivalent. However, by looking at the way the grammar Java.g is formatted:

statement 
    :   block

    |   ('assert'
        )
        expression (':' expression)? ';'
    |   'assert'  expression (':' expression)? ';'     
    |   ...

I suspect (by looking at the empty spaces) that the grammar originally had embedded code in it:

statement 
    :   block     { /* embedded code */
        }    
    |   ('assert' { /* embedded code */ }
        )
        expression (':' expression)? ';'
    |   'assert'  expression (':' expression)? ';'     
    |   ...

and that perhaps more rules/tokens were stripped (all guesswork, of course!).

Pedro F. wrote:

If not, why doesn’t ANTLR complain of ambiguity, since they seem to define exactly the same input?

Because global backtracking (the backtrack=true; in the options) is enabled.