Re: PROV-ISSUE-660 (TomDN): Constraints of PROV-Dictionary [PROV-DICTIONARY]

I don't think it was discussed.  I don't have an objection, but haven't had a chance to think about it very hard.  Either way, I suggest flagging this as a point to revisit if there is any future activity on this (e.g. formalization).  That is, if you remove a constraint that you think should be implied, please mention it in a remark as something that should be checked later.  If you leave it in, leave a remark saying that it appears redundant (which would mean that implementations can skip it).

--James

On Apr 15, 2013, at 10:49 AM, Tom De Nies <tom.denies@ugent.be> wrote:

> I would have liked some feedback on this before we implement it. Any thoughts? 
> Was anything said about this during last week's telecon?
> Thanks.
> - Tom
> 
> 
> 2013/4/11 Tom De Nies <tom.denies@ugent.be>
> Small correction, we need to have enough to guarantee that insertions and removals do not introduce or remove any key-entity pairs, other than those specified.
> 
> I think the two proposed constraints are sufficient for this, unless I'm missing something. 
> 
> 2013/4/11 Provenance Working Group Issue Tracker <sysbot+tracker@w3.org>
> PROV-ISSUE-660 (TomDN): Constraints of PROV-Dictionary [PROV-DICTIONARY]
> 
> http://www.w3.org/2011/prov/track/issues/660
> 
> Raised by: Tom De Nies
> On product: PROV-DICTIONARY
> 
> Luc raised some interesting ideas for the constraints.
> 
> Note that we now have this inference:
> https://dvcs.w3.org/hg/prov/raw-file/default/dictionary/Overview.html#membership-insertion-membership-inference
> Inference D4 (membership-insertion-membership) Here, KV1 is a set of key-entity pairs and K1 is the key-set of KV1.
> 1. IF prov:hadDictionaryMember(d1, e, k) and prov:derivedByInsertionFrom(d2, d1, KV1) and k ∉ K1 THEN prov:hadDictionaryMember(d2, e, k)
> 2. IF prov:hadDictionaryMember(d2, e, k) and prov:derivedByInsertionFrom(d2, d1, KV1) and k ∉ K1 THEN prov:hadDictionaryMember(d1, e, k)
> 
> (2nd part suggested by Luc)
> I do have one immediate question: do we introduce an infinite loop by doing this? (consequent of 1. appears in antecedent of 2., and vice versa)
> Or is this covered by http://www.w3.org/TR/prov-constraints/#overview ?
> 
> This got me thinking. If we have this, do we really need Inference D8? https://dvcs.w3.org/hg/prov/raw-file/default/dictionary/Overview.html#insertion-removal-membership-inference
> 
> Couldn't we just specify the same constraint as D4, but for removal?
> Suggestion:
> Inference D... (membership-removal-membership) Here, K1 is a set of keys.
> 1. IF prov:hadDictionaryMember(d1, e, k) and prov:derivedByRemovalFrom(d2, d1, K1) and k ∉ K1 THEN prov:hadDictionaryMember(d2, e, k)
> 2. IF prov:hadDictionaryMember(d2, e, k) and prov:derivedByRemovalFrom(d2, d1, K1) THEN prov:hadDictionaryMember(d1, e, k)
> Note that in the second case, k ∉ K1 is always true, otherwise constraint D9 is violated.
> 
> Do we then have enough to guarantee that insertions and removals do not introduce any new key-entity pairs, other than those specified? (which is why we had Inference D8)
> I think so, so I'd like to propose this solution. Could we have your support or objections via mail or on today's call?
> 
> Regards,
> Tom
> 
> 
> 
> 
> 
> 
> 
> 
>