Re: Test cases for blank node label scope - and a case for discussion

In the mail archive the attachments have been inlined.

syn-bad-39.rq is

PREFIX : <http://example.org/>
SELECT *
WHERE
{
   _:a ?p ?v .  FILTER(true) . [] ?q _:a
}


Seaborne, Andy wrote:
> Attached are some tests (positive and negative) for reuse of blank nodes 
> labels only within BGPs.
> 
> There is one case to point out: syn-bad-39.rq
> 
> {
>    _:a ?p ?v .  FILTER(true) . <x> ?q _:a
> }
> 
> 1/ By the principle that FILTERs apply to the whole group, it could be 
> taken as like:
>   { _:a ?p ?v . <x> ?q _:a . FILTER(true) }
> which is one BGP.
> 
> [Filter true
>   [BGP
>     _:b0 ?p ?v
>     <x> ?q _:b0
>   ]]
> 
> 2/ By the principle that BGPs are adjacent triple patterns, it has 
> appearance as if it were two BGPs like:
> 
>   { BGP( _:a ?p ?v . ) FILTER(true) BGP( [] ?q _:a . ) }
> 
> so it would be:
> [Filter true
>   [Join
>     [BGP _:a ?p ?v]
>     [BGP <x> ?q _:a]
>   ]]
> which is illegal (_:a spans BGPs).
> 
> Blank node labels can't appear in FILTERs so this also makes sense by 
> saying FILTERs end the blank node label scope.
> 
> 
> 
> Just at the moment, I think I prefer (2), emphasizing the immediate 
> appearance, and the fact blank node labels can't appear in filters.
> 
> (This example, either way round, will affect optimizations moving the 
> filter around if BGPs are sent to an external systems like a DL reasoner.)
> 
>     Andy
>