oslc-core message
[Date Prev]
| [Thread Prev]
| [Thread Next]
| [Date Next]
--
[Date Index]
| [Thread Index]
| [List Home]
Subject: Re: [oslc-core] Re: Blank nodes & deprecation of LocalResource. Re: [oslc-core] Dialog discovery: triple and Link header for discovery of dialogs won't be on same resources
- From: David Honey <david.honey@uk.ibm.com>
- To: "Jim Amsden" <jamsden@us.ibm.com>
- Date: Mon, 28 Sep 2015 18:03:48 +0100
Hi Jim,
There's the potential confusion about
the meaning "inline" versus "in graph".
From an OSLC point of view, I think
we need to distinguish between a resource that is described in the same
graph versus another graph that the client has to GET. For the latter,
a client must support doing a subsequent GET for that referenced resource.
For the former, clients should be able to accept blank nodes, or URIs of
any value (whether they contain a hash fragment based on a common URI root,
or uses a different URI altogether. From that aspect, I don't find the
term "inline" particularly helpful from a client aspect. Turtle
can express in-graph references for both blank nodes and URIs.
Best regards,
David
From:
"Jim Amsden"
<jamsden@us.ibm.com>
To:
Martin P Pain/UK/IBM@IBMGB
Cc:
OASIS <oslc-core@lists.oasis-open.org>
Date:
2015-09-28 17:16
Subject:
[oslc-core]
Re: Blank nodes & deprecation of LocalResource. Re: [oslc-core] Dialog
discovery: triple and Link header for discovery of dialogs won't be on
same resources
Sent by:
<oslc-core@lists.oasis-open.org>
Martin,
You are correct re: RDF abstract syntax and blank nodes - I realized that
right after pressing send. I was thinking more about common use than RDF
abstract syntax rules.
As far as I can see, you are correct, Turtle doesn't support inlined resources
except for blank nodes:
What are the implications for OSLC? I suppose a resource that has a URI
could still be included inline in Turtle using a blank node if instructed
to do so by a particular resource shape. The client wouldn't know what
that URI is, which is at odds with HATEOAS, but would work. However is
this semantically correct? Is a blank node in some resource representation
required to be a blank node in the RDF repository - i.e., do blank nodes
need to be consistent between the persistent data and any representation
of that data?
If so, then it appears that Turtle couldn't support inlined resources.
That would seem to be a problem.
If we removed oslc:LocalResource, then there would be no way of ensuring
the resources are inline in Turtle? And the resource couldn't be local
in one representation and a referencable URI in another?
Jim Amsden, Senior Technical Staff Member
OSLC and Linked Lifecycle Data
919-525-6575
From: Martin
P Pain <martinpain@uk.ibm.com>
To: Jim
Amsden/Raleigh/IBM@IBMUS
Cc: OASIS
<oslc-core@lists.oasis-open.org>
Date: 09/28/2015
10:19 AM
Subject: Blank
nodes & deprecation of LocalResource. Re: [oslc-core] Dialog discovery:
triple and Link header for discovery of dialogs won't be on same resources
Blank nodes are part of the "RDF abstract syntax" (mathematical/conceptual
structure), not just RDF/XML. (Even if that distinction was made retrospectively,
I don't know).
However, "Blank node identifiers" are just part of RDF/XML (and
are only local to a representation, or store) - as the abstract concept
of blank nodes is precisely that they have no identifier.
This is the definition of blank nodes in the abstract syntax, and it is
followed by a note referring to "blank node identifiers": http://www.w3.org/TR/rdf11-concepts/#section-blank-nodes(although
the link from that note to the definition of "concrete RDF syntaxes"
conveniently doesn't mention RDF/XML... which is only mentioned once in
the entire spec).
But I agree that they tend to be used as a syntactical mechanism in the
concrete syntaxes, rather than used for their "existential statement"
intentions. Therefore I agree we should discourage them. It's a shame Turtle
doesn't seem to be able to nest a resource without making it a blank node
(please correct me if I'm wrong).
So... does that mean we don't want to encourage shapes to use oslc:LocalResource,
as it requires resources to be blank nodes? So we could deprecate that
in favour of oslc:AnyResource with a oslc:representation of oslc:Inline.
And therefore the suggestion that we remove LocalResource from the shapes
in the spec could still stand.
Arthur, Nick & Jim, any thoughts on whether we should make either of
those changes?
Martin
Pain
Software Developer - Green Hat
Rational Test Virtualization Server, Rational Test Control Panel |
IBM United Kingdom Limited Registered in England and Wales with number
741598 Registered office: PO Box 41, North Harbour, Portsmouth, Hants.
PO6 3AU
From: "Jim
Amsden" <jamsden@us.ibm.com>
To: OASIS
<oslc-core@lists.oasis-open.org>
Date: 25/09/2015
19:25
Subject: Re:
[oslc-core] Dialog discovery: triple and Link header for discovery of dialogs
won't be on same resources
Sent by: <oslc-core@lists.oasis-open.org>
I agree with Arthur on the use of blank nodes. But I think blank nodes
have nothing to do with RDF or its meaning - rather they are syntactical
mechanism to deal with URIs in an RDF/XML resource representation for which
the creator is either not interested in, does not know, or doesn't care
about assigning a unique URI to an object. That flies in the face of web
best practices.
Jim Amsden, Senior Technical Staff Member
OSLC and Linked Lifecycle Data
919-525-6575
From: Arthur
Ryman <arthur.ryman@gmail.com>
To: Martin
P Pain <martinpain@uk.ibm.com>
Cc: Jim
Amsden/Raleigh/IBM@IBMUS, OASIS <oslc-core@lists.oasis-open.org>
Date: 09/25/2015
12:09 PM
Subject: Re:
[oslc-core] Dialog discovery: triple and Link header for discovery of dialogs
won't be on same resources
Sent by: <oslc-core@lists.oasis-open.org>
Martin,
My responses are inlined below:
- "I agree that we should re-introduce
olsc:representation into v3. " - I'm sure you will agree with this.
Agreed.
- "I am pretty sure that valueType
is never the same as range" - I' confident that you will agree with
this.
Agreed, for two reasons.:
1) rdfs:range defines an inference rule which lets you infer a type triple
when the given property used as the predicate of a triple. In contrast
Shapes define constraints, so oslc:valueType defines a check on the value
of any RDF node that is the object of a triple whose predicate is the given
property. This distinction between inference rules and constraints is the
source of a lot of confusion in developers coming from an OO background,
which is basically all of us. The W3C chartered the Data Shapes WG to define
what most developers wanted, namely a way to describe and check constraints
in RDF data.
2) The effect that developers want from rdfs:range can be achieved by using
oslc:valueType with a value of oslc:Resource, oslc:LocalResource, or oslc:AnyResource
and oslc:range, i.e. if the value of a property is non-literal then you
can provide information about the value's rdf:type using oslc:range.
oslc:valueType is primarily used to constrain the syntactic type of an
RDF node, i.e. IRI, blank node, or literal. In the case of literal, we
decided to be prescriptive and restrict the choice to a small number of
XSD types, which are enumerated in the spec.
- "I also agree that, if we were
starting from scratch, then we should only have one valueType for RDF resources"
- I expect you would disagree with this. I think it would be useful to
have some scenarios/examples (covering at least GET and POST cases, probably
PUT as well) showing in what cases requiring or prohibiting blank nodes
would be useful (and when requiring or prohibiting inlining would be useful)
to show the value of having these three valueTypes for resources.
I
do disagree based on the way people use RDF in practice, i.e. as not being
very strongly typed. Multiple values for oslc:valueType are primarily useful
for literal types. The meaning of having multiple values is that the node
must match at least one of them. RDF is intrinsically permissive. So this
boils down to a policy decision for OSLC, i.e. to impose strong typing
or not. The tradeoff is ease-of-implementation (with one type) versus flexibility.
Perhaps this could be handled with a SHOULD requirement, i.e. state that
properties SHOULD use just one type of RDF node.
- "Any references in the spec to
oslc:LocalResource would become oslc:AnyResource with an oslc:representation
of oslc:Inline." - even if we don't deprecate any of the resource
valueTypes, it feels like when talking about GETs (which the vast majority
of the shapes in the spec do.... I think.... at least that's how I read
them) there's no need to force something to be a blank node - as long as
it is inlined when it should be, as far as I can see that's what matters
to a client doing a GET. When POSTing I can understand that a server doesn't
want the client to force certain IRIs in the body, so requiring blank nodes
might make more sense. This suggests to me that perhaps we ought to have
separate shapes for POSTing and getting...not that we have the time to
make those sort of changes. This is very much getting into the scenarios
discussion that I suggest above - showing what the value & motivations
are for requiring or prohibiting blank nodes, and also looking at the difference
between GETs and POSTs.
- (Aside: you said in issue 34 that "
In RDF, IRI are opaque identifiers so you really should not talk about
fragments since it's not meaningful from an RDF viewpoint." - however,
when POSTing, especially to create a resource, it seems that there is a
distinction. It seems to me that it should be perfectly valid and acceptable
for a representation being POSTed in a creation request to define its own
relative fragment identifiers [e.g. <#me> - with no base IRI defined],
which will be resolved relative to the IRI the server mints for the new
resource [if the server allows relative graphs]. However if the representation
contains triples whose subject is another resource entirely, specifying
its absolute IRI on the same server, then that is ambiguous as to whether
it wants the server to create that resource or just store triples about
it in this resource's graph/representation - it depends what sort of model
the server is using, and I expect most implementations would want ot avoid
that sort of thing. I'm not sure I made myself clear here, but what I wastrying
to say is that in GETs I can see that IRIs are opaque, whether they contain
fragment identifiers or not, but that in POSTs there is a difference between
an IRI to another resource and relative fragment identifiers.)
Personally, I think OSLC should actively discourage the use of blank nodes,
because they are not used correctly. In RDF, the meaning of a blank node
is as a placeholder when the IRI of a resource is not known. The interpretation
of triples is in terms of existentially statements, e.g. Somebody likes
Martin. However, developers use blank nodes when they want the RDF to be
compact or they don't realize that they could use hash IRIs to identify
parts of a resource.
When you POST a resource, the server in effect copies it into a new resource.
The server is free to coin new IRIs, including fragment IRIs. This behaviour
should be described in the API spec for the POST service.
Yes, there can be different Shapes for POST vs GET. The Creation Factory
should link to a shape used for POST. The Query Capability or the resource
itself should like to its GET shape.
-- Arthur
I withdraw my suggestion from that email to deprecate oslc:Resource and
oslc:LocalResource. However I'd like to see examples (which hopefully already
exist) of when requiring or prohibiting blank nodes is useful, so we can
be confident of what should be in the shapes in the spec.
Thanks,
Martin
Pain
Software Developer - Green Hat
Rational Test Virtualization Server, Rational Test Control Panel |
IBM United Kingdom Limited Registered in England and Wales with number
741598 Registered office: PO Box 41, North Harbour, Portsmouth, Hants.
PO6 3AU
From: Arthur
Ryman <arthur.ryman@gmail.com>
To: Jim
Amsden <jamsden@us.ibm.com>
Cc: OASIS
<oslc-core@lists.oasis-open.org>
Date: 24/09/2015
23:14
Subject: Re:
[oslc-core] Dialog discovery: triple and Link header for discovery of dialogs
won't be on same resources
Sent by: <oslc-core@lists.oasis-open.org>
Jim,
Still catching up after vacation. Your comments contain some misinterpretations.
I've replied in other several places. Is this still an issue. fyi, I tried
to clarify the meaning of these vocabulary terms in the W3C submission.
Please refer to http://www.w3.org/Submission/2014/SUBM-shapes-20140211/#valueType.
-- Arthur
On Wed, Sep 2, 2015 at 5:12 PM, Jim Amsden <jamsden@us.ibm.com>
wrote:
I'm trying to understand the point of ResourceShape valueType and
representation properties.
oslc:valueType description is: "A URI that indicates the value type,
for example XML Schema or RDF URIs for literal value types, and OSLC-specified
for others. If this property is omitted, then the value type is unconstrained."
That is defining the type of the property, not its representation
in a resource, and covers what would be captured in the RDFS/OWL range
property of the property.
The description for property oslc:representation is "Should be http://open-services.net/ns/core#Reference,
http://open-services.net/ns/core#Inlineor
http://open-services.net/ns/core#Either".
This is addressing whether the representation of the referenced object
must or may be in the same resource representation as the subject of that
property. This is probably included in order to limit the number of GETs
required to do discovery. Its a resource representation optimization that
has no semantic meaning.
Later on in the ResourceShape vocabulary, there is a comment: "<!--
********** Property: oslc:valueType allowed values ********** -->"
which is followed by the enumeration types for Resource, LocalResource
and AnyResource. This seems completely orthogonal to the definition of
oslc:valueType and its common use to specify expected type of the value
of a property, and appears to overlap with the values for oslc:representation.
In particular, oslc:valueType of LocalResource could only have oslc:representation
oslc:Inline, and all instances of oslc:representation Inline in the OSLC2
specification have oslc:valueType LocalResource.
What seems to have happened is that oslc:valueType got somewhat overloaded.
If the valueType is a resource, then it can have a representation. No other
value type can have a representation. A resource shape might specify multiple
valueTypes for a property, one that is the property's type (i.e., the object
of its range property), the other is a tag indicating how the property
value (that is a resource) should be represented in an HTTP resource -
inlined (blank node or relative URI) or as a (potentially) external GETtable
resource in its own right.
But this overlaps with oslc:representation which says the same thing. So
I think valueTypes of Resource, LocalResource or AnyResource are redundant
and unnecessary (but can't be removed). The value of a property should
simply have a type, and if its type is a non-primitive resource,
then that value should have a representation that MUST or MAY be in the
same resource as the subject URI.
Given the description above, we should be able to decouple resources from
their particular representation. That is, a ServiceProvider resource representation
would expect its Service instances to be inlined in that representation.
But this shouldn't mean the Service can't be an LDPC in its own right,
and certainly a GET on a Service URI would return a resource representation
in which the Service is inlined!.
So I don't think there's a problem here. We should:
1. deprecate the use of oslc:Resource, LocalResource and AnyResource since
they are redundant with oslc:representation. Any value of oslc:representation
applies to any resource, so that doesn't need to be stated. oslc:representation
for a LocalResource has to be Inline. Clients can't make any assumptions
about the representation for AnyResource. So again, all values of oslc:representation
apply.
2. add oslc:representation back into the OSLC3 specifications (it was removed
and doesn't appear in any of the shape .ttl files or generated tables)
3. Treat oslc:representation as a means of specifying what should be included
in resource representations that reference properties, but does not constrain
where servers actually manage those resources. That is, the value of a
property with oslc:representation oslc:Inline could be a blank node, relative
(or hash) URI, or a URI to a resource that that can also be the URI of
a GET request, even though it would be accessed inline in any other referencing
resource representation.
As a result, get on a ServiceProvider would include the publisher and services
inline, but each Service could also be the URI of an LDPC in an OPTIONS,
HEAD, or GET method that provides Link headers for discovery.
Jim Amsden, Senior Technical Staff Member
OSLC and Linked Lifecycle Data
919-525-6575
From: Martin
P Pain <martinpain@uk.ibm.com>
To: OASIS
<oslc-core@lists.oasis-open.org>
Date: 09/02/2015
10:14 AM
Subject: [oslc-core]
Dialog discovery: triple and Link header for discovery of dialogs won't
be on same resources
Sent by: <oslc-core@lists.oasis-open.org>
We have three mechanisms of discovering dialogs:
Link headers & Prefer headers: http://tools.oasis-open.org/version-control/browse/wsvn/oslc-core/jra-editing/specs/dialogs.html#discovery_link
and oslc:selectionDialog/oslc:creationDislog triples in oslc:Service resources:
http://tools.oasis-open.org/version-control/browse/wsvn/oslc-core/jra-editing/specs/discovery.html#dialogs
I had previously expected that these headers and these triples would be
on the same resources, but they cannot be.
The triples are on oslc:Service resources, but OSLC
v2 requires that these resources
be "Local Resources", which it defines to mean blank nodes (although
I need to raise a separate issue to clarify whether the intention was to
allow for hash URIs or not). "Local Resources" cannot have headers
of their own, so oslc:Service resources cannot have Link header. (If they
have hash URIs, technically they can have their own Link headers, but I
don't suggest we go down that route.)
The Link and Prefer headers will be on LDPs themselves. (Although currently
I don't think we have a good way of finding those LDPCs).
Is everyone else ok with the fact that these headers and triples will be
on different resources? I just wanted to make sure we're clear what
the situation is and are ok with it.
(This might make more sense in the context of my previous email and the
wiki page it links to, where I'm thinking about how a server with one or
more LDP containers makes those containers and their capabilities discoverable
using OSLC).
Martin
Pain
Software Developer - Green Hat
Rational Test Virtualization Server, Rational Test Control Panel |
IBM United Kingdom Limited Registered in England and Wales with number
741598 Registered office: PO Box 41, North Harbour, Portsmouth, Hants.
PO6 3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
3AU
Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number
741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6
3AU
[Date Prev]
| [Thread Prev]
| [Thread Next]
| [Date Next]
--
[Date Index]
| [Thread Index]
| [List Home]