- From: Jeremy Carroll <jjc@hplb.hpl.hp.com>
- Date: Wed, 26 Mar 2003 09:32:06 +0000
- To: Jim Hendler <hendler@cs.umd.edu>
- CC: "Peter F. Patel-Schneider" <pfps@research.bell-labs.com>, www-webont-wg@w3.org
This message summarizes the options and gives fall-back positions, and has a proposal that I hope the chairs will give 5 minutes to before the LC vote tomorrow. I will withdraw this proposal if Peter opposes. One technical detail is left to an appendix at the end of the message. Options ======= Problem Statement: How should the proposed recommendation map DisjointClasses and EquivalentClasses with more than two components from the abstract syntax to the concrete syntax. Plan A: Proposal B.1, B.2 from the editors meeting, original made by Peter. With a correct proof. Using current mapping rules. Plan A-: Proposal B.1, B.2 from the editors meeting, original made by Peter. Without a complete proof. (i.e. the proof is basically as is, and it is noted that we failed to finish it due to time pressure). Plan B: The current text - with sufficient conformant implementations. Plan B-: The current text - fudging that implementations diverge from it. Plan C: One of my earler proposals, specifically that these should be mapped by structure copying. Effectively the concrete syntax is limited to the case that n=2 (see appendix for mapping rules). As I understand it - Plan A is universally preferred. Our options for last call are plan A- or plan B. Risks ===== A risk analysis is: plan A- at LC risk that the proof is not possible with the effort that anyone in the group is prepared to put into it. fallback options for plan A- at PR: plan A- at PR risk that the result is actually false plan B at PR unlikely that there would be implementations if we don't have it in LC risk that we may have to delay PR in order to seek implementations. plan B- at PR failing to implement part of our rec is a risk plan C not popular with group, similar risks of delay plan B at LC risk that we will not be able to fully articulate what implementors need to do; risk that they won't do it anyway (it is intricate and not cost effective in implementors time); risk that we put additional unnecessary cost on OWL implementors and users (assuming plan A is in fact feasible). fallback options for plan B at PR: plan B- at PR puts rec at risk as above plan A, A- at PR risk of delay as above plan C not popular, risks of delay I don't think I can quantify any of these risks - other than to say they are all non-trivial. The risk minimization path would be to delay, but I am sure I would in a minority of one if I suggested that! My preference is plan A- at LC aiming for plan A at PR. If the proof is too difficult I think we should then choose between plan A- and the other options depending on technical judgement given the nature of the difficulties with the proof. We currently have plan B aiming for ?guess? plan B at PR. Proposal ======== I propose that we reopen DL Syntax 5.26: - resolve it as in [1] (including B.1 and B.2), - give the S&AS editor discretion to add text, for example as in [2], to make these changes, - and further discretion to add (or not) text as in [3] to articulate DL and Lite graphs as triples, - and then close the issue again. (Note the process document specifically grants the chair the descretion to reopen an issue in light of input from someone who was not at the telecon) Refs: ==== [1] http://lists.w3.org/Archives/Public/www-webont-wg/2003Mar/0066.html [2] http://lists.w3.org/Archives/Public/www-webont-wg/2003Mar/0183.html http://lists.w3.org/Archives/Public/www-webont-wg/2003Mar/0184.html [3] http://lists.w3.org/Archives/Public/www-webont-wg/2003Mar/0161.html http://lists.w3.org/Archives/Public/www-archive/2003Mar/att-0073/m Appendix - Mapping rules for Plan C =================================== These *copy* the relevant descriptions, hence reducng the complexity of the concrete syntax in the opposite fashion to the B.1 B.2 approach. Replace: DisjointClasses(description1 ... descriptionn) => T(descriptioni) owl:disjointWith T(descriptionj) . OR T(descriptionj) owl:disjointWith T(descriptioni) . 1=<i<j=<n T(descriptioni) owl:disjointWith T(descriptionj) . [opt] 1=<i!=j=<n and EquivalentClasses(description1 ... descriptionn) => T(descriptioni) owl:equivalentClass T(descriptioni+1) . 1=<i<n T(descriptioni) owl:equivalentClass T(descriptionj) . [opt] 1=<i,j=<n with DisjointClasses(description1 ... descriptionn) n > 2 => T(DisjointClasses(descriptioni descriptionj)) 1=<i<j=<n T(DisjointClasses(descriptioni descriptionj)) [opt] 1=<i!=j=<n DisjointClasses(description1 description2) => T(description1) owl:disjointWith T(description2) . OR T(description2) owl:disjointWith T(description1) . and EquivalentClasses(description1 ... descriptionn) n > 2 => T(EquivalentClasses(descriptioni descriptioni+1)) 1=<i<n T(EquivalentClasses(descriptioni descriptionj)) [opt] 1=<i,j=<n EquivalentClasses(description1 description2) => T(description1) owl:equivalentClass T(description2) . No other changes anywhere (except in the graph as triples section - I have some text for this somewhere). Jeremy
Received on Wednesday, 26 March 2003 04:32:42 UTC