[antlr-interest] disambiguating sempred in a closure?

[Terence Parr]

On Dec 21, 2007, at 2:58 PM, J Chapman Flack wrote:
> Ok, here is the simplified grammar that really does
> demonstrate my dilemma. The predicates can't do what
> I need because of the grouping parentheses added to
> factor out the common something_else suffix.
>
> I can get the code I want if I leave off the grouping
> parens and duplicate the something_else suffix on
> each branch. Is that the best I can do?  I hate to
> duplicate the something_else (which is followed by
> a nontrivial action).

Hi Chap!  long time no chat ;)

Heh, so the code gen is not right in v2?  I can't remember how it  
works :(

> Is the story the same in ANTLR 3?
> class CelebParser extends Parser;
>
> prod : something
>       ( options { greedy=true; }
>       : ( { muppet }? ( KERMIT | FOZZIE )
>         | { turtle }? ( LEONARDO | DONATELLO )
>         )
> 	 something_else
>       )*
>     ;

v3 is way better at this stuff.  You use a gated semantic pred, which  
always get hoisted into decision:

prod : something
       ( ( { muppet }?=> ( KERMIT | FOZZIE )
         | { turtle }?=> ( LEONARDO | DONATELLO )
         )
	 something_else
       )*
     ;

results in DFA predictor:

-------------- next part --------------
A non-text attachment was scrubbed...
Name: Celeb.dec-2.pdf
Type: application/pdf
Size: 16459 bytes
Desc: not available
Url : http://www.antlr.org/pipermail/antlr-interest/attachments/20071221/ee983471/attachment-0001.pdf 
-------------- next part --------------




The preds are hoisted out of the subrule inside (...)* and used to  
control exit.  3rd alt (exit) is implicit.

Ter