[Meeting Minutes] SQL Schema and WSDL Spec review special meeting1/27/2005

[Farrukh Najmi <Farrukh.Najmi@Sun.COM>]

Date: 1/27/2005
Who: Farrukh, Kathryn, Paul, Joe, Goran, Carl, Monica, Prathap Gangula 
(IBM), Peter
What: Finish review of 3.0 SQL Schema and WSDL files

Thanks to all that attended this special meeting. We started and 
finished reviewing the SQL Schema.
Thanks to Goran for sending issues in advance of meeting. This made the 
meeting productive and focused.

The following comments were logged:

    * Goran: home and lid are URIs. Is their length 128 or 256?
          o Farrukh: URN and URI are 256 chars per RIM. Since id is now
            a URN both home and id MUST be 256 chars. home already is.
          o Resolution: Is a bug. Will fix in draft 02
    * Goran: What is the reason that VersionInfo is flattened in all
      RegistryObject-based tables?
          o Farrukh: There are many cases in the schema where we flatten
            columns for some composed classes into columns in the
            containing class' table.See for example Person and
            PersonName. It avoids a join. This is an optimization we
            have always done where it makes sense inthe relational binding.
          o Resolution: No change
    * Goran: AffectedObject does not match rim.xsd definition (eventId
      is not in rim.xsd, createReplica is not in AffectedObject)
          o Farrukh: That is because AffectedObjects are composed within
            AuditableEvent and in XML Schema and UML their composite
            object is implicit.
          o Resolution: No change
    * Goran: Can we add MimeType as a canonical ClassificationScheme. If
      we do so, mimeType's datatype in ExtriniscObject will become
      varchar(64)
          o Farrukh: The spec is complicated for that. It is hard to get
            it right. Would it be a flat or multi-level hierarchy? Which
            mimeTypes should we include canonically? Sometimes
            specifying too much is not desirable and I think this is one
            of those cases. Also it is a fair amount of work that I do
            not think is worth the effort for version 3.
          o Resolution: No change
    * Goran: Can we also add Lang and Charset as canonical
      ClassificationSchemes. If we do so, all lang and charset columns
      will have varchar(64) datatype
          o Farrukh: Same as previous comment.
          o Resolution: No change
    * Goran: createReplica is missing in ObjectRef
          o Farrukh: That is because createReplica is only needed for
            request processing when submitting an ObjectRef and serves
            no purpose once the request has been processed.
          o Resolution: No change
    * Goran: PostalAddress, EmailAddress need id because of Slots
          o Farrukh: This is a bug. We have two options:
                + a) Could make them derive from Identifiable and remove
                  the Slots within their definition since that would be
                  inherited.
                + b) Not make them extensible and remove slots from
                  these types
          o Resolution: Do (a) for PostalAddress and (b) for others
            because PostalAddress has known use cases for requiring
            extensibility.
    * Goran: operator's length in Registry should be 256
          o Resolution: Is a bug. Will fix in draft 02. Also now that id
            is a URN all reference attributes including operator MUST be
            made 256 chars. 
    * Goran: specificationVersion in Registry should be not null
          o Resolution: Is a bug. Will fix in draft 02.
    * Goran: Slot table should be transformed into two tables: Slot and
      SlotValue (Second Normal Form). Slot columns should be: name,
      slotType and parent with (parent,name) primary key. SlotValue
      columns should be: name, value and parent with (parent,name,value)
      primary key
          o Farrukh: That would be an extra join and cost a performance hit?
          o Resolution: No change
    * Goran: subscription is missing in Notification 
          o Farrukh: Agreed. It is also missing the ObjectRefList
            mapping. This will require a separate table like
            AffectedObjects probably called NotificationObjects.
          o Resolution: Is a bug. Will fix in draft 02.
    * Goran: person_firstName and person_lastName in User_ should be not
      null
          o Farrukh: In 3.0 these attributes have become option so no
            change is needed. rim.xsd has been checked for consistency.
          o Resolution: No change
    * Goran: What is Person table used for? 
          o Farrukh: To record persons. Just as Organization records
            Organizations. Persona and Organization are an important and
            commonly used metadata. This class was added with TC
            approval based upon Paul Spencer's request for enhancement
            based on user feedback. Note that Person class is also a
            base class for User class.
          o Resolution: No change
    * Goran: Informal RDBMS naming standard is that multi-word names for
      database objects (tables, indexes, etc.) have underscore between 
      "the words". For example, classification_node.
          o Farrukh: There is no strong convention on this. It is better
            to align with our classnames as a matter of convention
          o Resolution: No change
    * Goran: Suggested we make MimeType, Language, Charset  as schemes
          o Resolution: Add reference to existing (IETF etc.) specs on
            these.
    * Farrukh: Specify that index definitions in SQL schema are
      non-normative
          o Resolution: Agreed to add verbage.

In tomorrow's meeting we will review the WSDL files for 3.0. Please 
review and send comments in advance.

-- 
Regards,
Farrukh