W3C home > Mailing lists > Public > www-annotation@w3.org > July to December 2002

attributions, create document names and message ids

From: Eric Prud'hommeaux <eric@w3.org>
Date: Tue, 17 Dec 2002 20:12:51 -0500
To: www-annotation@w3.org
Message-ID: <20021218011128.GC23830@w3.org>

INTENDED CURRENT IMPLEMENTATION
  Statements table:
    #  subject      predicate  object        attribution
    3 a:Comment rdfs:subclass a:Annotation    2
    4 annot1    rdf:type      a:Comment       7
    5 annot1    a:annotates   www.w3.org/...  7
    6 annot1    rdf:type      a:Annotation    8

  Attributions table:
    # type       uri(document name)       parent       authName
    2 Source     htt...AttributionA          -         swick@w3.org
    7 Source     htt...AttributionE          -         eric@w3.org
    8 Rule       htt...AttributionF htt...AttributionE  -

  AttributionClosures
    child parent generation
     2     2      0
     7     7      0
     8     8      0
     8     7      1

The Attributions code as a couple problems:
  0 Attributions have multiple parents. The inference
      annot1 rdf:type a:Annotation.
    comes from
      annot1 rdf:type a:Comment.            -- attribution 7
      a:Comment rdfs:subclass a:Annotation. -- attribution 2
    and the rule
      this log:forAll :ob, :super, :sub.
      {:ob rdf:type :sub.
       :super rdfs:subclass :sub} log:implies {:ob rdf:type :super}.
    which has no attribution, though should.
  1 The full lineage of 8 is not represented in Attributions or
    AttributionClosures.
  2 The one parent expressed in Attributions is duplicated in
    AttributionClosures.

Attributions are currently thought of as documents. When you tell an
agent with a triple store to look at a document, it looks for an
attribution with a uri of that name. If it exists, ie, the server is
refreshing its view of a document, it deletes all of the Statements
with that attribution or a child attribution (as found in
AttributionClosures) in preparation for parsing the incoming
assertions. If it does not exist, a new attribution is created. All
parsed RDF statements are stored with the found or created
attribution.

The annotation server is a specialized application that actually
creates attributions. When a new annotation is stored, the server
needs a name for that document and invents a new attribution URI. The
server also works hard to make it so that the invented document looks
just like any other document on the web, clients may HEAD, GET, PUT or
DELETE the attribution URI.
  GET - Select all the triples from the triple store with that
        attribution and pass them to the RDF serializer.
  DELETE - Delete all the triples from the triple store with that
        attribution.
  PUT - Delete all the triples from the triple store with that
        attribution. Parse the incoming data and attribute the
        resulting statements to the PUT attribution. This has
        the same affect as telling the server to refresh its view
        of a document.

FUTURE DESIGN
All the current tables exist; one new table is added:
  History table:
    #  subject      predicate  object        attribution

When a set of statements are "deleted", they are moved to the History
table. Deleting the statements with attribution 7 (and attribution 8
as it is a child of 7) will result in the following statements being
moved to the History table:
    #  subject      predicate  object        attribution
    4 annot1    rdf:type      a:Comment       7
    5 annot1    a:annotates   www.w3.org/...  7
    6 annot1    rdf:type      a:Annotation    8

I have no plans for how to use algae to query history at this
point. Since algae allows one to specify a query database on any ask,
I will probably implement the historical access as another
database. History is simply an added feature and represents no changes
to groupthink (1984 reference to the common knowledge).

Changing the attributions from recording document characteristics to
recording message characteristics requires a little mindset change and
a little code change. The main change is that the name of an
attribution represents a message that caused data to be inserted into,
or deleted from the Statements table. If a new annotation is POSTed, a
new attribution will be created to store the attributes of that POST.
If the same annotation is replaced, the statements from the first POST
are moved to History and a new attribution is created to group all the
new statements. The tricky part is the name of this message.

REPLICATION CASE We start the mother of all annotation servers and
want to replicate RDF data from the hundreds of annotation servers
that have, due to uncanny popularity, sprung up all around the
world. If we have an interface that allows an agent to query for
statements from a particular message, we need to provide an identifier
for that message.  Do we want the user to be able to use the same
identifier on a server where an annotation was POSTed as on a server
that has scooped up all that rich, creamy annotation data? If we say
no; the message identifiers should be different on the replicated
server. This allows the message identifier to be a function of the
primary key for the table. If we say yes; a new field must be added to
identify the message identifier.

DOCUMENT HOSTING CASE
IF we say that message identifiers are constant across replicated
hosts, the appliation (for instance, the annotation server) would be
responsible for making up the message name as well as the document
name. I suspect it would be unfortunate to make them be the same
identifier; it would blur the distinction between a message and the
document it created/retrieved.

UPDATE ADVISORY CASE
I think we still want an identifier for the document for the case
where the application gets a notification to refresh its view of a
document with URI u1. The update request would get a message
identifier, but we want to record the identifier u1 to tie the
statements in the database to that resource.


I'm leaning towards using a message identifier not based on the
primary key of the attribution. This means marginally more work for
the application but enables one to use the same identifier for the
same message from one triple store to another.

What do y'all think?
-- 
-eric

office: +1.617.258.5741 NE43-344, MIT, Cambridge, MA 02144 USA
cell:   +1.857.222.5741

(eric@w3.org)
Feel free to forward this message to any list for any purpose other than
email address distribution.
Received on Tuesday, 17 December 2002 20:12:51 GMT

This archive was generated by hypermail 2.2.0 + w3c-0.30 : Friday, 25 March 2005 11:19:18 GMT