- From: Andy Seaborne <andy@apache.org>
- Date: Thu, 9 Jul 2020 18:06:50 +0100
- To: public-shacl@w3.org
A few comments: I haven't tried experimental code here except code I
already have for incremental processing.
There are good arguments from everyone here on all the points.
1) The name "targetShape"! The shape isn't the target - it's the nodes
in the data graph described by the shape in the shapes graph so
"targetByShape" to be like "targetSubjectsOf".
2) I do not believe that performance on its own should be a blocker for
a feature. It is important, and SHACL's focus on shapes is useful to get
efficient implementation without a high burden, but it is another step
to keep a feature out just because it's inefficient sometimes. Not all
data is large; expressivity is also important.
4) I support keeping the sh: namespace conservative. In fact, if AF is a
living document, there is a case to have shx:, then gather
implementation experience from users, before promoting to sh:
Implementation experience takes time to gather and is best if the
details, not just reports, are public.
5) About examples 2, 3 and 4 and efficiency:
Both points of view are valid.
If validating a difference stream, they can be efficient for a large
subset of constraints.
If on the whole dataset, it would need "compilation" of the shape to be
better than the definition that is a step (small? large?) on the road to
raising the implementation barrier.
Andy
Received on Thursday, 9 July 2020 17:07:04 UTC