On 14/08/13 14:34, David Booth wrote:
[Copying Andy Seaborne and Steve Harris for their input.]
Axel is right.
Hi Axel,
The part of the spec that I was trying to bring to your attention was
where it says that the empty group graph pattern "does not bind *any*
variables" (my emphasis). If ?G is not bound then that short form of
listing existing graphs would not work. But I am not sure from reading
the spec whether ?G is supposed to be bound or not. The SPARQL algebra
says:
http://www.w3.org/TR/sparql11-query/#defn_evalGraph
[[
eval(D(G), Graph(var,P)) =
Let R be the empty multiset
foreach IRI i in D
R := Union(R, Join( eval(D(D[i]), P) , Ω(?var->i) )
the result is R
]]
Hmm, there's right parenthesis missing, which I guess I'll report
separately.
It looks like Ω(?var->i) is binding the graph variable, but then it's
doing a "Join", which is defined as:
http://www.w3.org/TR/sparql11-query/#defn_algJoin
[[
Join(Ω1, Ω2) = { merge(μ1, μ2) | μ1 in Ω1and μ2 in Ω2, and μ1 and μ2 are
compatible }
]]
I couldn't readily find the definition of the merge function, but I
think I found the definition of "compatible":
http://www.w3.org/TR/sparql11-query/#defn_algCompatibleMapping
[[
Two solution mappings μ1 and μ2 are compatible if, for every variable v
in dom(μ1) and in dom(μ2), μ1(v) = μ2(v).
]]
which seems to be saying that the variables being joined must be bound
to the same values. But since the empty basic graph pattern does not
bind ?G, I *think* this means that Ω(?var->i) would *not* be compatible
and ?G would therefore not be bound in the result.
It is compatable. It says variables in common must match, not all
variables must match. If dom(μ1) is the empty set, the "forall"
condition is trivially true.
Join(eval({}), X) = X for any pattern X.
One row, no bindings is the join identity.
(Otherwise LeftJoin would not work, nor Union, ...)
try
ASK{}
So, if I have properly followed the SPARQL algebra, I *think* this means
that the short form that you suggested for listing graphs will not work,
and it is not possible to get a list of graphs that includes empty
graphs. (Hence the Sesame 2.7.1 behavior is correct.)
You may find this relevant:
https://openrdf.atlassian.net/browse/SES-1119
Andy or Steve, have I got this right?
Andy
Thanks,
David
On 08/13/2013 03:10 PM, Axel Polleres wrote:
[[ The group pattern: { } matches any graph (including the empty
graph) with one solution that does not bind any variables. ]]
This only means that upon
SELECT ?G WHERE { GRAPH ?G {} }
also empty named graphs should be returned, which would not be the
case for
SELECT ?G WHERE { GRAPH ?G { ?S ?P ?O } }
Obviously, this makes a difference for all graph stores that support
empty named graphs. So, to my understanding at least, this is not a
bug in the spec.
HTH, Axel
On 13 Aug 2013, at 20:21, David Booth <[email protected]> wrote:
Hi Axel,
That doesn't work in Sesame 2.7.1 at least, apparently because ?G
is not bound, even though there is one solution. The SPARQL 1.1
spec says: http://www.w3.org/TR/sparql11-query/#emptyGroupPattern
[[ The group pattern: { } matches any graph (including the empty
graph) with one solution that does not bind any variables. ]]
Is this a bug in the spec?
David
On 08/13/2013 11:48 AM, Axel Polleres wrote:
Hi Quentin,
how about just
SELECT ?G WHERE { GRAPH ?G {} }
(no need to dump all triples, if the only concern is which ?G
exist)
BTW, [email protected] may be the list you wanted to use.
best, Axel
