Notice: this Wiki will be going read only early in 2024 and edits will no longer be possible. Please see: https://gitlab.eclipse.org/eclipsefdn/helpdesk/-/wikis/Wiki-shutdown-plan for the plan.
Difference between revisions of "Higgins XDI Harmonization"
(→References) |
(→Issues to Resolve) |
||
Line 22: | Line 22: | ||
Kinds of issues to address: | Kinds of issues to address: | ||
− | + | # Semantics are undefined in PDX - need to add the semantics to PDX | |
− | + | # Semantics are undefined in Higgins - need to add the semantics to Higgins | |
− | + | # Semantics are different but in a shallow or unimportant way - need to adopt one or the other | |
− | + | # Semantics are incompatible | |
Reference | Reference |
Revision as of 13:24, 10 June 2010
{{#eclipseproject:technology.higgins|eclipse_custom_style.css}}
This document summarizes some of the notes and conversations that have happened today at the EIC conference. These are still rough notes, but I felt it would be good to share them all the same. The memo is organized by comparing Higgins CDM, HDM and PDM in turn with XDI/PDX.
Contents
Introduction
Higgins speaks an XDI dialect that we'll call Higgins XDI. The XDI community is developing a dialect of XDI called PDX XDI. There is no need or reason to have Higgins XDI "standardized" and in fact over time we hope to gradually adjust the Higgins code (mostly in the Attribute Service and client components, but possibly one or two minor changes to IdAS itself) so that it is conformant with the PDX XDI dialect.
Two kinds of graphs
The PDM/HDM/CDM models (see Personal_Data_Store_Overview under data models) taken together define the types of graphs that will be found in production Higgins-based systems. In Higgins these graphs are separated first into multiple top-level contexts--the meta context and 0..n others. Within these contexts there may be sub-contexts [In the RDF-backed context providers these are implemented by RDF code that uses named graphs to model sub-contexts].
On the other hand the XDI defines the types of graphs that will be found in XDI conformant systems. So we have two kinds of graphs with differences between them. Our goal is to ensure that these graphs can be losslessly transformed bi-directionally. Our goal is NOT to make them the same graphs.
In order to make the graphs transformable we need to ensure that the semantics are equivalent between the two.
With that as background, we now need to focus on semantic differences between what the two kinds of graphs can represent.
Issues to Resolve
Kinds of issues to address:
- Semantics are undefined in PDX - need to add the semantics to PDX
- Semantics are undefined in Higgins - need to add the semantics to Higgins
- Semantics are different but in a shallow or unimportant way - need to adopt one or the other
- Semantics are incompatible
Reference
- Links to CDM, HDM and PDM are found here:
http://wiki.eclipse.org/Personal_Data_Store_Overview#Data_Models
Issues
References (links)
- URI conventions. An entityId is is a URI UDI that is either absolute or relative (# fragment) by syntactic inspection.
- Resolution: Non issue. XDI == CDM
- Global URIs, if resolvable, resolve to exactly one external description.
- Resolution: XDI == CDM
- Higgins represents a single person as a meta context + 0..N other contexts. XDI represents a single person as a single XDI document which can have multiple personas.
- Resolution: AS will map between representations. [It already does this]
CDM
4) CDM allows multiple values of an attribute. XDI does not.
Resolution: This can be mapped into a sub-context (in XDI) and back out losslessly. This is already implemented in the Higgins Attribute Service.
5) In PDX you can identify values with a persistent identifier that won't change when the value changes (and you can have order). In CDM you cannot.
Resolution: TBD Perhaps IdAS needs to support at least ordered values.
HDM 1) For privacy reasons h:correlations are directed. The semantics are reversible except for the issue of privilege.
Resolution: h:correlation is the same as $is (although there is another wrinkle mentioned below about h:correlation)
2) A single local entity may be linked to a single external (global) entity via an h:correlation link. [I believe this is different from PDX wherein an interstitial LinkContract entity is inserted.]
Resolution: TBD.
3) Link Contracts. In Higgins we consider Link Contracts to be attributes associated with a context. If a single entity is being shared, then it would be placed in its own context and the link contract metadata attached to the context (not the entity).
4) Class descriptions. In HDM we define a number of attributes on classes to enable software to dynamically construct UIs, etc. Some of these are taken from the SPIN vocabulary. Others are from OWL and RDF.
Entity classes: agent, group, organization, person
Attribute classes: authority - no equivalent in PDX correlation - already discussed creator - no equivalent in PDX member - no equivalent in PDX memberOf - no equivalent in PDX part - no equivalent in PDX. Only used to support access control. partOf - no equivalent in PDX timeSpan - no equivalent in PDX. XDI has timestamps synonym - validFrom - no equivalent in PDX, but planned for access control and thus not yet used/implemented in Higgins validTo - no equivalent in PDX, but planned for access control and thus not yet used/implemented in Higgins
Entity Class - rdfs:type - in XDI is $is$a - subclassOf - in XDI is $is$a - rdf:comment - no equivalent in XDI - skos:prefLabel - no equivalent in XDI - skis:prefSymbol: no equivalent is XDI - spin:constraint (points to spl:Attribute) constraints on attributes of a specific Entity class
+ predicate - the attribute we are talking about + maxCount - Markus thinks there is an equivalent in XDI + minCount - Markus thinks there is an equivalent in XDI + defaultValue - need to add to XDI + valueType - Markus thinks this can be done for literals (e.g. +tel / $is$a / $xsd$string). And for object-valued predicates the valueType is implicit
Global Attributes
- rdf:type: URI value must either be owl:DatatypeProperty or owl:ObjectProperty
Resolution: Markus: you can tell in XDI if it is a literal valued because you'll see a $foo $is$a statement
- 0..1 rdfs:domain: URI value must be an Entity Class entity
Resolution: XDI $has is the inverse of rdf:domain; this should suffice
- 0..1 rdfs:range: URI value must be an Entity Class entity, a DataRange, or one of the allowed XML Schema datatypes (e.g. xsd:string, etc.) - 0..1 skos:description: string value that describes the attribute. Used for tooltip text in UIs - 1..1 skos:prefLabel: internationalized display label skos:prefLabel - 0..1 skos:prefSymbol: internationalized display symbol (e.g. icon) skos:prefSymbol - 0..1 skis:example - an example value - ..1 h:category: value must be an instance of skos:Concept. Indicates the skos:Concept category to which this attribute belongs. - 0..N rdfs:subPropertyOf: the value must be another Attribute Class entity - 0..1 rdfs:label: internal (ontology developer) display label - 0..N rdfs:comment: internal (ontology developer) comment
Access Control - lots of predicates and classes designed but not yet implemented in Higgins
5) rdf:type is recommended on each entity instance
Resolution: $is$a is the XDI equivalent to rdf:type. The Higgins AS can map back and forth.
PDM
1) There is a root persona node of fixed name "MetaMe". In XDI there is an "account root" i-number (globally resolvable). This would change from one PDS to another PDS.
Resolution: TBD.
2) In Higgins we use h:correlation from the MetaMe to each/all (sub-)Personas. In PDX $has$a is used sometimes and $is in other cases.
Resolution: TBD.
3) Inheritance. In PDX attributes are inherited "down" the graph. You can override an attribute on a lower Persona. If a lower Persona is shared, then the "upper" / inherited attributes would be pushed to the subscriber.
Resolution: Higgins could adopt the PDX approach.
4) Persona nodes have explicit class Persona. In XDI persona nodes are of various types "+person", "+home+person", etc.
Resolution: TBD
5) Persona nodes have a Persona human readable label string. In XDI this not required.
Resolution: TBD
6) Persona graph: The set of h:correlation and foaf:knows links define the persona graph explicitly. This allows software to know how to navigate and process the graph. This means that software that ONLY knows these two predicates can ignore all others and still walk the graph. In Higgins h:correlation links are ONLY used between persona (not more generally). Code relies on the this fact to be able to navigate the entire graph without understanding deeper structure of the nodes. So a h:correlation will not be used WITHIN a persona node & related sub-graph.
Resolution: TBD
7) PDM uses vCard [w3c 2010 member submission] except for tel URI telephone numbers. A Persona node effectively "is" a vCard (we don't use explicit vCard classes)
Resolution: TBD
8) PDM includes the icard.owl vocabulary. These are a set of attributes and values that are associated with context instances. A context is a card. For personal cards the context contains the entity whose attributes/values are the claims/values of the card. For managed cards the context contains an entity whose attributes/values are the claims/values of the display token retrieved from the card's associated STS.
Resolution: TBD
9) PDM defines some attributes not found in vCard, FOAF, etc. these include notions of p:home, p:work, p:receiving, etc.. These don't exist in PDX.
Resolution: TBD.
10) PDM defines some classes not found elsewhere like PaymentMethod
Resolution: TBD.