 |
 |
Karen, just some short inline comment in addition to what I answered to Tom's mail: > > At first glance, Holger's "left side" diagram looks okay to me. I'd > > quibble about the terminology---I think what he's really talking about > > is the SPARQL Semantics (as opposed to a SPARQL Engine, which seems like > > an implementation detail which can vary) interfacing with a > > Dataset---but without digging too deep, this seems like a good approach. > > It avoids the need for a new query semantics except with respect to the > > Shapes Graph, but allows SHACL functions to sit in a layer above. > > Holger is indeed talking about a SPARQL engine, as I understand his > statements, and he has already implemented it as a SPARQL Engine. > However, note that the group is pretty evenly split on this issue, with > some feeling that SHACL must be solely an extension to SPARQL, with > Holger's approach, allowing SHACL to be independent of SPARQL, > considered too liberal to work well. If you put it this way we definitely should support Holger! Making one technology that dependent on another is a fast way to disaster... > I sent a link to Active Triples to another person in the group and the > answer I got was that this would require a different SHACL engine than > the one being envisioned, and that engine would not be able to implement > all of SHACL, and therefore: why would we not want to implement a SPARQL > engine? I don't have an answer to that, but essentially a non-SPARQL > approach is seen as counter to the spec in at least some areas. Because there can/should/must be many different ways to implement a specification. Exactly how would a non-SPARQL approach be counter to the spec? > If we think this is important, then someone other than me needs to step > in and make the case. Issue-47[1] is at the crux of this in some way, > which is why we took a straw vote on it (inconclusive). I'm at a loss to > say which approach we prefer because I don't understand the issue. It > seems to be fundamental to how the standard would work. This seems to be > related to another issue[2] which asks whether the SHACL graphs and the > instance data must be in the same graph or not (also no conclusion from > the group). My gut feeling is that they cannot mandate the number of graphs. If I want to perform streaming validation they are definitely two different graphs. > If you can do some reading and lend a hand, that would be great. > Otherwise, we'll just get what we get -- it depends on how much it > matters to us. I'm on extremely limited time right now, but if you can say exactly what I need to do, I can try to shove it in... Best, Lars