[antlr-interest] Bug in ANTLR-3.1.2 with tree grammar wildcard list label x+= ?

[Gary R. Van Sickle]

> From:  Gavin Lambert
> Sent: Friday, February 27, 2009 6:17 PM
> To: Laurent Caillette; antlr-interest
> Subject: Re: [antlr-interest] Bug in ANTLR-3.1.2 with tree 
> grammar wildcard list label x+= ?
> 
> At 13:07 28/02/2009, Laurent Caillette wrote:
>  >part
>  >  : ( mediumbreak | largebreak )?
>  >
>  >    (   p += levelIntroducer
>  >      | p += paragraph
>  >      | p += blockQuote
>  >      | p += literal
>  >      | p += bigDashedListItem
>  >      | p += cellRowSequence
>  >    )
> [...]
>  >/Users/Shared/Scratch/Novelang/antlrworks/java/NovelangParser.java:
>  >754:
>  >incompatible types
>  >found   : NovelangParser.levelIntroducer_return
>  >required: NovelangParser.cellRowSequence_return
>  >                     p=levelIntroducer();
>  >                                   ^
>  >
>  >It seems that p has taken the type of the first rule  
> >(levelIntroducer). Then attempting to set the result of  
> >cellRowSequence doesn't work.
> 
> I'm surprised that you managed to get it to work as long as 
> you did -- that's been on my list of "don't do that" for most of v3.
> 
> You can fix it by creating a subrule:
> 
> content
>    : levelIntroducer
>    | paragraph
>    | blockQuote
>    | literal
>    | bigDashedListItem
>    | cellRowSequence
>    ;
> 
> And then use p+=content in your original rule.

Bug or feature though?  It seems inconsistent for what appear to be two
semantically equivalent constructs to be totally fine in the one case and a
silent failure in the other.

-- 
Gary R. Van Sickle