[antlr-interest] How to use predicate or workaround the bug ?

[Terence Parr]

 From the book:

Another limitation on semantic predicates is that they should not
reference local variables, parameters, or anything else that is not
visible to all rules just in case they are hoisted out of one rule
into another's prediction DFA. If you are positive that the predicate
will not be hoisted out of the rule, you can use a parameter, for
example, to alter prediction in a rule:

method[boolean allowConcrete]
     :   {allowConcrete}?=> methodHead body
     |   methodHead ';'
     ;

Workaround is a field or dynamically scope var.

Ter
On Jan 2, 2007, at 6:34 AM, Ilia Kantor wrote:

>
> I'm using antlr 3b6 build from Tue Jan  2 14:28:20 UTC 2007. Also  
> tried 3b5.
>
> Trying to make use of predicate, but fail.. Is there a workaround ?
>
> Here I try to rewrite
> 'LCURL  exprs? RCURL' into 'TEXT exprs? TEXT' in common case and strip
> LCURL/RCURL if this LCURL/RCURL block is outmost, that is: if LCURL  
> is first
> token.
>
> ==================== rule with predicate=====
> curly_block
> 	:	
> 	 {input.index()==0}? LCURL exprs? RCURL -> exprs? |
>      LCURL exprs? RCURL -> TEXT[$LCURL] exprs? TEXT[$RCURL]
> ===================== error  ===========
>
>
> ANTLR Parser Generator  Version 3.0b6 (??)  1989-2007
> Exception in thread "main" java.lang.StackOverflowError
>         at java.util.HashMap$KeyIterator.<init>(HashMap.java:875)
>         at java.util.HashMap$KeyIterator.<init>(HashMap.java:875)
>         at java.util.HashMap.newKeyIterator(HashMap.java:889)
>         at java.util.HashMap$KeySet.iterator(HashMap.java:921)
>         at java.util.HashSet.iterator(HashSet.java:154)
>         at
> org.antlr.analysis.DFAState.getGatedPredicatesInNFAConfigurations 
> (DFAState.java:695)
>         at
> org.antlr.codegen.CodeGenerator.canGenerateSwitch 
> (CodeGenerator.java:1119)
>         at
> org.antlr.codegen.ACyclicDFACodeGenerator.walkFixedDFAGeneratingStateM 
> achine(ACyclicDFACodeGenerator.java:72)
>         at
> org.antlr.codegen.ACyclicDFACodeGenerator.walkFixedDFAGeneratingStateM 
> achine(ACyclicDFACodeGenerator.java:146)
> =..................... the latest line repeats many times ..........=
>
>
>
> ======= Full grammar ===========
>
>
> /* putting this to tokens { .. } causes antlrworks bugs like "...no  
> lexer rule
> for IF..." */
> IF	:	{false}? .;
> THEN	:	{false}? .;
> ELSE	:	{false}? .;
> EVAL	:	{false}? .;
> TEXT	:	{false}? .;
> RULE_PROPERTY	:	{false}? .;
> RULE_PART	:	{false}? .;
>
> LCURL	:	'{';
> RCURL	:	'}';
>
> RULE_CALL
> 	:	'#';
>
> CONTROL
> 	:	'~';
> 	
> SEMI:	';';
>
> DOT	:	'.';
>
> MINUS
> 	:	'-';
>
> GT	:	'>';
>
> SPECIAL_CHAR
> 	:	'`' | '!' | '@' | '$' | '%' | '^' | '&' | '*' | '(' | ')' |
> 	'+' | '=' | '[' | ']' | ':' | '\'' | '"' | '\\' | '|' | ',' | '<' |
> 	 '/' | '?';
> 	
>
>
> fragment WS_CHAR  :
>        (' '|'\r'|'\t'|'\u000C'|'\n' )
>     ;
>
> WS	:	WS_CHAR+;
>
> // not SPECIAL_CHAR not WS_CHAR
> WORD:	
> {
> 	// check if previous token was CONTROL
> 	Boolean afterControl = input.index()>0 && input.LT(-1)=='~';
> }
> 	(~ ('`' | '!' | '@' | '$' | '%' | '^' | '&' | '*' | '(' | ')' |
> 	'+' | '=' | '[' | ']' | ':' | '\'' | '"' | '\\' | '|' | ',' | '<'  
> | '/' | '?'
> 	| '{'
> 	| '~'
> 	| '}'
> 	| '#'
> 	| ';'
> 	| '.'
> 	| '-'
> 	| '>'	
> 	| ' '|'\r'|'\t'|'\u000C'|'\n' ) )+
> {
>     if (afterControl) {
>     	// output WORDs as commands if after control
> 		if (getText().equalsIgnoreCase("if")) $type=IF;
> 		else if (getText().equalsIgnoreCase("else")) $type=ELSE;
> 		else if (getText().equalsIgnoreCase("then")) $type=THEN;
> 	}
> }
> ;
>
>
>
> document
> 	:	 exprs;
>
>
> exprs:
>         (expr)+
> ;
>
>
> expr:
> (
>         expr_no_semi -> expr_no_semi |
>         SEMI -> TEXT[$SEMI]
> )
> ;
>
> exprs_no_semi
>         :       (expr_no_semi)+
> ;
>
>
> expr_no_semi
>         :
>                 (expr_simple) | (curly_block)
>         ;
>
>
>
> exprs_simple
>         :       (expr_simple)+
>         ;
>
> /* expression without curl at start and ; */
> expr_simple
>         :
>         command
>         | rule_call
>         | text
> ;
>
>
> command:
>          command_eval |
>          command_if
> ;
>
>
> command_if
>        :
>         CONTROL IF LCURL exprs? command_if_part_then  
> command_if_part_else?
> RCURL ->
>         ^(IF exprs? command_if_part_then command_if_part_else?)
> ;
>
>
> command_if_part_else:
>         CONTROL ELSE exprs? -> ^(ELSE exprs?)
> ;
>
> command_if_part_then
>         : CONTROL THEN exprs? -> ^(THEN exprs?)
> ;
>
>
>
> command_eval:
>         CONTROL LCURL exprs? RCURL -> ^(EVAL exprs?)
> ;
>
>
>
> rule_call: RULE_CALL WORD -> ^(RULE_CALL WORD) |
> 	RULE_CALL LCURL WORD rule_long_def_part* RCURL -> ^(RULE_CALL WORD
> rule_long_def_part*)
> ;
>
> rule_long_def_part
> 	:	MINUS GT WORD -> ^(RULE_PROPERTY WORD) |
> 		DOT WORD -> ^(RULE_PART WORD)
> 	;
> 	
>
> curly_block
> 	:	
> 	 {true}? LCURL exprs? RCURL -> exprs?
> 	 |  LCURL exprs? RCURL -> TEXT[$LCURL] exprs? TEXT[$RCURL]
> ;
>
> text:
> 	DOT -> TEXT[$DOT] | MINUS -> TEXT[$MINUS] | WORD -> TEXT[$WORD]
> 	| WS -> TEXT[$WS] | GT -> TEXT[$GT] | SPECIAL_CHAR -> TEXT 
> [$SPECIAL_CHAR]
> ;
>
>
>