[ubl-ndrsc] Minutes for 11 September 2002 UBL NDR SC meeting (jointmtg w/ LC SC)

["Eve L. Maler" <eve.maler@sun.com>]

Minutes for 11 September 2002 UBL NDR SC meeting

PLEASE NOTE: The NDR SC is now planning to meet on September 25, rather 
than skipping that meeting.  See the bottom of these minutes for details.

1. Roll call (quorum is 8)
      * Bill Burcham        YES
      * Mavis Cournane      YES
      * Mark Crawford       no
      * Fabrice Desré       no
      * Matt Gertner        no
      * Arofan Gregory      YES (joined y:12, left y:??)
      * Jessica Glace       no
      * Michael Grimley     YES
      * Eduardo Gutentag    regrets
      * Eve Maler           YES
      * Sue Probert         YES (joined y:00)
      * Lisa Seaburg        YES
      * Gunther Stuhec      YES (joined y:00, left y:31)
      * Paul Thorpe         regrets
      * Kris Ketels         no

     Others:
      * Tim McGrath (LC)    yes
      * Bill Meadows (LC)   regrets
      * Joe Chiusano (LC)   regrets
      * Jon Bosak (UBL)     yes
      * Ray Seddigh (LC)    yes

     NDR quorum not achieved.  We proceeded informally.  We achieved
     quorum at y:12, but lost it again at y:31.

1'. Approval of agenda

     Approved, noting that we can end early if we want!

2. Joint session begins: containership

    - Review Tim's paper:
       http://lists.oasis-open.org/archives/ubl-ndrsc/200209/msg00016.html

      Tim summarized the paper and added some side comments.

      Any vocabulary needs to have some way of logically grouping
      business information entities into aggregates.  This paper
      describes a process of rules by which we can establish this
      logical grouping.  The main concept around which grouping is
      done is "dependencies", and the process is called "normalization".
      (These come from the relational world.)  The paper has an appendix
      on how to apply these logical processes to XML and XSD.

      Tim believes that any process we come up with will still have to
      allow for other factors, such as subjective judgment based on
      domain knowledge and so on.  A process simply lets us to come up
      with somewhat consistent results.

      "Structure" and "ABIE" and "BIE" are terms to describe the
      logical model and its process.  "Container" and "element" are
      terms used to describe the physical model (the XML result).

      A (functional) dependency is the organizing principle that we
      often use intuitively already.  "Dependency means that if the
      value of an attribute changes when another attribute value
      changes, then the former set is dependent on the latter."  A name
      depends on a date of birth , since when you change one, the other
      changes -- they vary together, per person.  So "person" would make
      sense as a group structure (entity, in relational-speak).

      The normalization technique was evolved to help analysts figure
      out real dependencies and get rid of false ones.  The different
      "<n>th normal forms" reflect different qualities or stages of
      normalization.  Third normal form is what Tim believes we should
      target.

      We shouldn't have constituent BIEs inside an ABIE that are
      dependent on each other in addition to the ABIE.  This means
      that they should be separated into their own ABIE.  If transport
      providers vary per vehicle, vehicle should be a group structure.
      (To take Arofan's example in his containership paper.)

      In the case of persons, a name might in fact be useful as a
      unique identifier for each instance of a person.  (So sometimes
      people use a shorthand, saying that dates of birth vary "per
      name".)  In UBL modeling, we do have a lot of BIEs that are
      "Identifiers", but they're partial only -- they cover only one
      segment of what would be needed to *uniquely* identify the BIE
      in question.

      Each of the line items in an order can be partially identified
      by the item number, but to make it truly unique, an identifier
      would also have to refer to the order number.  (However, it's
      also possible to create a singular line item identifier that
      incorporates the order number within it somehow.)

      First normal form involves removing sets of repeating structures
      an making them their own structure.  Second normal form involves
      the name/date of birth types of analysis, checking each of the
      BIEs within an ABIE and ensuring that they are dependent
      on it (or, as shorthand, dependent on the BIE that can serve as
      a unique identifier within the ABIE).  In third normal form, the
      final step (as recommended by Tim), the non-key (non-identifier)
      elements are scrutinized to find and remove anything that varies
      independently of the rest of the structure.

      This process formalizes things as much as possible, but it's
      a craft rather than a science.  And the output of this process
      still needs to be turned into XSD, which isn't entirely trivial.
      (Of course, the model could also be mapped to relational tables
      and object representations as well -- a relational E/R diagram
      could as easily be turned into a class diagram without methods.)

      The choices we make in deciding how to "containerize" the XML
      version of the model might reflect a path through, or view
      into, the data.  This might make a difference particularly when
      it comes to dynamic assembly, but even in the meantime, it
      would be important to capture why each choice was made.  It would
      be possible to avoid association by containment to a very large
      degree (a series of independent XML elements, associated only by
      linking).  If we did this, we might not be leveraging XML
      technology in the way that our guiding principles suggest, since
      a higher-level application layer would have to do all the
      association work.  However, it would be taking the rest of our
      work to its logical conclusion, and that's interesting and may
      be productive when we get to dynamic assembly.

      Rough reaction: Eve and Mike feel that Tim's paper is a good
      way to proceed and will result in high-quality XML.  Jon notes
      that he likes the tone of discussion, but cautions that the
      people who *use* what we come up with will need very simple
      descriptions.  Mavis thinks it jells, but wants to study the
      notes a bit more and cautions that the devil is in the details.
      Tim suggests that people work through their own examples and
      see how well it hangs together.

    - Next steps

      ACTION: Tim agreed to put together a draft position paper that
      all can review before the F2F (including ubl-comment!), with
      the plan that we will resolve all outstanding containership
      issues at the F2F.

      We won't continue to meet jointly next week.

3. NDR-only session begins: Acceptance of minutes of previous meeting

     4 September 2002
     http://lists.oasis-open.org/archives/ubl-ndrsc/200209/msg00008.html

     Deferred until we have quorum.

4. Schedule planning

    - Do we want to publish the NDR document before the F2F, or after?
      (Mavis gives regrets for the next two NDR meetings.)  We don't
      see an imperative to get a new snapshot out very soon, although
      there are good reasons to do so very soon after the meeting.  We
      will take that as our goal (unless we decide otherwise at the
      F2F).

    - Do we still want to skip the September 25 meeting?  No!  We'd
      rather meet on the 25th than on October 9th!  So let's do it.

    - What items do we consider high-priority right now?  Let's decide
      this next week, on the assumption that we'll consider late November
      to be our last chance to deliver new work.  Below is our current
      plan of record; it's subject to heavy change.

      A+ Containership IN PROGRESS
      A+ Embedded documentation NEARLY DONE
      A Code lists DONE
      A Dates and times IN PROGRESS
      A Nested supplementary components IN PROGRESS
      A Identifier references and whether to pass content by reference
      A- Local vs. global elements
      B Updating guiding principles
      B Modnamver URN/schema location
      B Referencing of content, e.g. for attachments
      C Facets
      C Wildcards/open content
      C Nillability
      C Aggregation of similar information for XPath V1.0 addressing

5. Review open action items

    Let's reassess all these items when we decide what the top
    priorities are next week.

      Gunther:
      - Write content referencing paper. IN PROGRESS
      - Send date/time NDR snippets to Mavis. IN PROGRESS
      - With Arofan, prepare samples of how to handle second-tier
        attributes. IN PROGRESS
      - Bring the donkey to Burlington!

      Bill:
      - Update modnamver paper by September 11.
      - Start an email thread proposing a schema location solution.
      - Start a thread on RTs in aggregates and possible NDR
        document bugs in this area.

      Arofan:
      - NEW: Update the embedded documentation writeup.

6. Adjourn

    Adjourned z:25.

-- 
Eve Maler                                        +1 781 442 3190
Sun Microsystems                            cell +1 781 883 5917
XML Web Services / Industry Initiatives      eve.maler @ sun.com