[antlr-interest] Yet another bug in ANTLRv4

[Jan Finis]

Hi Ter,

I tried to change the "expr with concatenation" grammar to work without 
using a doubly left-recursive concat operation ( expr : expr expr;)
in case that Honey Badger will never allow such rules again.

The resulting grammar is accepted by latest Honey Badger (the one pulled 
from parrt/antlr). However, the resulting parser contains errors: an 
unused semPred method with errors is produced. Since it is unused, I 
simply deleted it. However, the parser still does not match the input 
correctly. I guess both things have are symptoms of the same bug.

Here is (a simplified version of) the grammar producing the erroneous 
parser:

rule returns [String r]
   :  t=ID  ':' alts+=expr ('|' alts+= expr)* ';'
       {  $r = "x"; }
   ;

strongExpr returns [String r]
   : ID                      { $r = "x"; }
   | '(' ep=parenExpr ')'       		{ $r = "x"; }
   | e1=strongExpr '?'             { $r = "x"; }
   ;

expr returns [String r]
   : e=strongExpr (e2=expr)? { $r = "x"; }
   | e1=expr '->' ID {$r = "x"; }
   ;

parenExpr returns [String r]
   : e1=expr '|' e2=parenExpr { $r = "x"; }
   | expr              {$r = $expr.r; }
   ;

ID  :	('A'..'Z'|'a'..'z'|'_') ('a'..'z'|'A'..'Z'|'0'..'9'|'_')*
     ;

the resulting unused sempred with errors:

public boolean sempred(RuleContext _localctx, int ruleIndex, int 
predIndex) {
switch ( ruleIndex ) {
case 1 : return strongExpr_sempred((StrongExprContext)_localctx, 
predIndex);	
case 2 : return expr_sempred((ExprContext)_localctx, predIndex);
		}
		return true;
	}

the problems are the casts (StrongExprContext)_localctx and 
(ExprContext)_localctx, since _localctx is a RuleContext here.
As mentioned, the sempred is not called anyway.

Best regards,
Jan