[Mulgara-dev] ITQL

[Paul Gearon]

On Sat, Jul 12, 2008 at 3:51 PM, Seaborne, Andy <andy.seaborne at hp.com> wrote:
>> > Just query with
>> >
>> > SELECT ?g { GRAPH ?g { ?s ?p ?o } }
>> >
>> > All names of graphs.
>>
>> It's interesting you would say this. I wondered about this in the
>> past, and wasn't satisfied that it was allowed. Also, when I wrote my
>> email this morning, then I looked it up again (so I didn't look like
>> an idiot - as I am wont to do) and again, I wasn't satisfied that it
>> was allowed.
>
> Paul,
>
> Examples can't be exhaustive!  We tried to put in as much as we could but not every single case can be an example.

Sure. I suppose I was expecting this one to be in there, as I was
trying to work out how to implement it. From that perspective it
looked like an obvious one that was missing. I didn't see any
discussion on the point either, so I made the (incorrect) call that it
wasn't allowed.

> There are tests in the test suite: graph/graph-02 to -09 or thereabouts
>
> Section 12 gives the definition of the evaluation of a graph pattern.

I take it that 12.5 is the area of most relevance here? Specifically,
the definition of:
  eval(D(G), Graph(var,P)) = .....
???

While this would appear to do what I was looking for in "GRAPH ?x",
the use of named graphs with FROM NAMED appears to restrict it, and I
didn't actually follow where the definitions in section 12 provided
for that difference.

That said, I didn't have as much time on this as I would have liked,
and I confess that I skimmed a lot of things. Rather than trying to
implement everything from scratch, I was trying to map it onto our
internal algebra, and mostly read these sections to justify my
decisions. There was a reasonable correlation between the SPARQL and
TQL languages (after all, TQL was heavily influenced by RDQL and
authors of both languages were involved in SPARQL), so it seemed
reasonable to rely on this... even though I knew it exposed me to
problems in edge cases. I hoped the test suite would help me spot
those problems.

Unfortunately, I don't have enough integration with the SPARQL
protocol to make the test suite easy to run.  (I *can* run them, but
it would take me ages to automate right now). So I haven't caught
everything I want to yet.

Paul