Home / FAQ / Java Tools / ANTLR

Terence Parr

Well, I do lots of imaginary tokens as roots. For example,

class LangParser extends Parser;

options {
	buildAST=true;
}

tokens {
	EXPR;	// imaginary token
}

block
        :       LCURLY! ( statement )* RCURLY!
                // add imaginary BLOCK node on top of statement list
                {#block = #([BLOCK, "BLOCK"], #block);}
        ;

This operation should be done as the last operation in a rule where you have auto tree generation on lest you interfere with the tree construction. If you use '!' to turn off auto construction for the alternative or rule, then it is ok to add children and so on to various nodes during the rule's execution.

As for adding imaginary children nodes to existing trees, you can label the node (not a rule reference unless you know the rule only returns a node) and then use the #(...) tree operator to add children:

rule	:	A b:B C
		{#rule = #(#b, [IMAG, "IMAG"]);}
	;

If you want the operation to occur during the recognition, turn of auto construction:

rule	:!	A b:B {#rule = #(#b, [IMAG, "IMAG"]);} C
		
	;

You don't want to interfere with ANTLR's construction code. Note that this "add imaginary child" operation will be rare. "Add imaginary parent" will be more common. Further, the above rule really should be written as:

rule	:!	A b:B C
		{#rule = #(null, A, #(#b, [IMAG, "IMAG"]), C);} 
	;