Re ACTION-201 http://www.w3.org/2001/tag/group/track/actions/201
JAR email: http://lists.w3.org/Archives/Public/www-tag/2010Mar/0070.html
12/09 draft: http://www.w3.org/2001/tag/awwsw/http-semantics-report.html


Premise (following Roy Fielding): The HTTP protocol *expresses*
some state of affairs.  There may be other ways to express it, too.
(e.g. cache APIs, HTTP-over-SPARQL)


Logical relations

  W(R,Z,t)
      Z "corresponds to" R
      Z is a [data] "representation" of [the state of] R
  redirectsPermanentlyTo(R1,R2,t)
  redirectsTemporarilyTo(R1,R2,t)
  seeOther(R1,R2,t)
  exists(R,t)

  R resides at URI, R located at host  ?


The "meaning" of the protocol

  GET U / 200 Z 
     expresses
  W(<U>,Z,t)

  GET U / 30{1,7,3} Z 
    expresses
  redirectsPermanentlyTo(R1,R2,t)
  redirectsTemporarilyTo(R1,R2,t)
  seeOther(R1,R2,t)


The "representation" is selected or composed by the server, 
NOT the resource

  there may be infinitely many [acceptable] "representations"
  they don't necessarily "come from" the resource in detail
  can be selected from a set or constructed according to 
    what is required for that resource


Time intervals are expressible using Last-modified: / Expires:

  for all t in [t1,t2)
    W(R,Z,t)

  maybe write this as
    W_over_interval(R,Z,[t1,t2))
  or just
    W(R,Z,[t1,t2))


Redirection means some shared properties

  redirectsTemporarilyTo(R1,R2,t)
  Prop(R2,t)
  ---------
  Prop(R1,t)


e.g. shared representations, shared seeOthers

  redirectsTemporarilyTo(R1,R2,t)
  W(R2,Z,t)
  ---------
  W(R1,Z,t)

  redirectsTemporarilyTo(R1,R2,t)
  seeOther(R2,R3,t)
  ---------
  seeOther(R1,R3,t)


Compare Fielding's formalism:

  resource = mapping from time to set of value
    (value = representation or URI)

  Mapping(R) = 
    lambda t. {Z | W(R,Z,t)} union {...}


Not all information resources are REST resources

  REST:
    representation of state of resource
    change in representations results from change in resource
  Not required by HTTP, not true of all [information] resources


Goal: theory of metadata about things-on-the-web

  This is in some sense the reason for the httpRange-14 decision: why
  would you care that the URI names the description and not the thing,
  if you weren't going to use the URI as a name for the description
  *for some reason* - if there weren't consequential statements you
  would make about the description resource?


E.g.

  <http://example.org/pathayes> dc:creator
     [is foaf:primaryTopic of <http://example.org/pathayes>].

  is consequential metadata that couldn't be written if 
  <http://example.org/pathayes> named Pat.


This is a bidirectional problem:

  (1) Enable metadata, given that web behavior is fixed.

  (2) Enable web behavior, given that metadata is fixed.


(1) Enable metadata, given that web behavior is fixed.

  Suppose resource is named by a http: URI U
  W(<U>,Z,t) is "authoritative" if it comes from domain owner

  Possible sources of metadata
  (a) entailments of W(R,Z,t)
        can you infer anything about the resource from a representation?
  (b) independently asserted by someone  
        (e.g. Dublin Core provided through <link> or Link:)
  (c) hypotheses
        reasonable assumptions that are not "authorized"
  (d) purpose inferred from manner of use
        <a href="http://example.org/about">
           This page tells you about example.org</a>


Representation tells you what about the resource?

  - depends on resource-specific "equivalence" relation
       e.g. content-length vs. dc:creator
  - content, creator, etc. may differ depending on session info
  - so probably no rigorous entailments unless you know 
    something about the resource already
  - the URI for the description resource isn't doing us much 
    good, lacking other information about it


But surely the representation can tell you what's *not* the case?
(falsifiability)

  - W(R,Z,t) where Z is Jabberwocky by Dodgson
      is incompatible with
    R = Hiawatha by Longfellow


Intersection theory vs. union theory
  - resource has properties that are common to all representations
  - resource has properties of all representations
  - not all properties are equal


(2) Enable web behavior, given that metadata is fixed.

  FRBR
    Is R rdf:type frbr:Work. compatible with W(R,Z,t)?
    Is R rdf:type frbr:Expression. compatible with W(R,Z,t)?
    Is R rdf:type frbr:Manifestation. compatible with W(R,Z,t)?
    Is R rdf:type frbr:Item. compatible with W(R,Z,t)?
  IAO
    Is R rdf:type iao:InformationContentEntity. compatible with W(R,Z,t)?
    Is R rdf:type bfo:PlanSpecification. compatible with W(R,Z,t)?
  CItO
  BIBO


Nasty issues around "equivalence"

  Truncation
    e.g. http://dx.doi.org/10.12345/foo  => abstract (402?)
    e.g. table of contents instead of whole document
  Additional quasi-content (e.g. advertising)


Conservative approach:
  - maybe content is wrong or misleading
  - use #, 303, blank nodes
      (who is authoritative for W ?)

Liberal view:
  - I'm competent & authorized to say whether W(R,Z,t)
  - use 200 - as long as server thinks it's OK, it's OK
      (contradicts spirit of httpRange-14 / 
       supposed consequential nature of W)


What is a [information] resource?

  Fielding: "The only thing that is required to be static for a resource
  is the semantics of the mapping [from time to sets of values], since
  the semantics is what distinguishes one resource from another."

  - Does context (role of resource, how we encoutered it) rescue us?


Goal: definition of "information resource" (or guidelines, etc.)
should be a *consequence* of this analysis