- From: Dave Reynolds <der@hplb.hpl.hp.com>
- Date: Tue, 03 Feb 2009 23:50:55 +0000
- To: RIF WG <public-rif-wg@w3.org>
As Jos pointed out, without a specific definition of
guaranteed-terminating safeness it is hard to construct a clear cut example.
However, here's a simple example which I believe would fail for any such
definition.
Length of an RDF list:
length(rdf:nil, 0).
length(?list, ?len) :- ?list[rdf:rest->?next],
length(?next, ?l), ?len = ?l + 1 .
For well-formed RDF data then lists are linear and finite and this will
terminate with either a forward or backward chaining proof strategy. RDF
does not preclude circular lists so it would be possible to construct a
RIF+RDF combination which would not terminate with this rule set, hence
no reasonable definition of terminating-safeness would admit this rule
set. Yet I assert that counting of finite things (e.g. the number of
items in an invoice, or number of authors on a paper) is a reasonable
thing to want to do within Core.
Note that the original proposal [1] did not allow for assignment and had
no distinction between recursive and non-recursive rule sets so in fact
a simple "business" rule like:
finalPrice(?x, ?p) :- price(?x, ?r), discount(?x ?d), ?p = ?r * ?d.
would, I believe, have been unsafe.
Does this help?
Dave
[1] http://lists.w3.org/Archives/Public/public-rif-wg/2008Sep/0178.html
Received on Tuesday, 3 February 2009 23:51:40 UTC