- From: Andy Seaborne <andy.seaborne@epimorphics.com>
- Date: Sat, 15 Dec 2012 17:39:19 +0000
- To: public-ldp-wg@w3.org
On 15/12/12 17:19, Arnaud Le Hors wrote: > Hi Bart, > If we want to model LDP after a filesystem I think we should consider > the unix filesystem which provides for "hard links" that are constrained > to one system and symbolic links that can cross boundaries. Hard links are also restricted so that an additional link is only to a file, not a directory. This stops cycles; it's a more restrictive condition than is necessary for that. They require reference counting and they are not what I understand "containment" to be (= one owner) from previous discussions. Deletion of containers with delete of sub-tree becomes quite interesting (doable, but needs spec'ing). If we have container management, I suggest we do not allow additional hard links unless someone makes a strong case for it. Andy (who hasn't seen how we do soft links but I'm guessing its POST to a container of a some specific RDF document as a container entry. The link is strongly managed by the container.) > > Regards. > -- > Arnaud Le Hors - Software Standards Architect - IBM Software Group > > > Bart van Leeuwen <Bart_van_Leeuwen@netage.nl> wrote on 12/14/2012 > 05:54:29 AM: > > > From: Bart van Leeuwen <Bart_van_Leeuwen@netage.nl> > > To: public-ldp-wg@w3.org, > > Date: 12/14/2012 05:55 AM > > Subject: Container membership alternative approach ACTION-27 > > > > Hi, > > > > Since our meeting at TPAC I've been thinking about our discussion of > > ISSUE-25 [1]. It resulted in the end in ACTION-27 which is a rewrite > > of the spec to comply with the resolution of the issue. > > As Steve Speicher stated in his email [2] there are already some > > issues raised with a more detailed description of what seems to be > > missing in the spec. > > > > I've always liked the idea of a resource being member of multiple > > containers. So I've tried to go over all the emails and discussion > > summaries I could find and see where a possible solution might lie > > or come up with one. > > One approach that immediately popped up in my head was the way that > > the long forgotten OS/2 Workplace Shell used to place objects in > > multiple folders. Instead of creating short cuts, it created a thing > > called a shadow. > > The shadow of a object was exactly what its name tells you, a direct > > reference to the original, not a pointer. you modify the shadow, you > > modify the original object and vice versa, the only exception to > > this is a deletion operation. If you delete the object, all the > > shadows disappear with it, if you delete the shadow, it will just > > delete the shadow not the original object. > > > > This would work perfectly well for a LDPR since the direct reference > > is the URI, PUT and DELETE will work on the original LDPR. > > For creating and deleting a shadow LDPR in a LDPC I still have to > > figure out the mechanics in a REST way, but there are probably > > enough bright minds to help with that. > > A GET on the container could return a list of real resources and > > maybe shadow resources through a subclass of rdfs:member? > > I did not yet figure out how to use this method with the > > membershipSubject / membershipPredicate construction > > In the overal spec I miss is what is addressed in ISSUE-21 the > > reverse membership, and in this case shadow, predicates on a LDPR. > > > > I'm pretty sure I overlook something very obvious, but at least I've > > made a brain dump now ;) > > > > > > [1] http://www.w3.org/2012/ldp/track/issues/25 > > [2] http://lists.w3.org/Archives/Public/public-ldp-wg/2012Nov/0198.html > > > > Met Vriendelijke Groet / With Kind Regards > > Bart van Leeuwen > > @semanticfire > > > > ############################################################## > > # netage.nl > > # http://netage.nl <http://netage.nl/> > > # Enschedepad 76 > > # 1324 GJ Almere > > # The Netherlands > > # tel. +31(0)36-5347479 > > ##############################################################
Received on Saturday, 15 December 2012 17:39:49 UTC