ACORDs Experiences using W3C Schemas

1
ACORDs Experiences using W3C Schemas

• Dan Vint
• Senior Architect
• dvint_at_acord.org

2
Background

• ACORD
• Not-for-profit Insurance industries standards
  body, we have standards for Life, PC and RLC
  lines. Used in US, Canada, England/London Market,
  and Europe. Growing interest in Japan/China
• Three standards started in 3 different ways
• PC based upon an earlier EDI standard.
  Originally designed for DTD use, but added schema
  support when it was available (currently using
  both)
• Life started as a MS initiative then brought
  into ACORD. DTD based originally, now only using
  schema
• RLC started in a European standards group, also
  based upon an EDI standard. Using DTDs currently
  trying to add schema support
• Business Issue How to bring the 3 standards into
  one design
• For ACORD the best thing would be a common model,
  design and naming conventions
• Cant get there because we are on different
  cycles. Life and RLC cannot break backwards
  compatibility, PC is currently redesigning
• Each as a level of adoption and entrenchment.

3
Things that we like

• Having the schema format be an XML document has
  been very useful
• We generate schemas and DTDs from a meta-format
  that is maintained in a database
• We slice the schema into smaller subsets via XSLT
• Support for basic data types was crucial for our
  needs.
• Having a built-in documentation standard is very
  nice.
• Support for namespaces and import/include/redefine
  is useful, but rough around the edges.
• Can we get this sort of support reversed
  engineered into XML and DTDs?

4
Things we dont like about W3C Schema

• Took a simple 30 page standard and blew it out of
  proportion.
• XML took a simplifying approach, schema reversed
  that and put in experimental features and
  complexity
• Fallout
• You need tools to manage a schema
• No two tools agree on their implementations
• Incomplete implementations
• Should have left out xsdall, DTDs did and Schema
  implementation is not generally usable anyway.
• I can teach DTD syntax in an hour or less it
  takes a day for schema
• Entities requiring the use of DTDs.
• Yuch! DTDs, thats old technology and a kludge
  Surely if the schema guys had wanted entities to
  be used they would have defined a new method (or
  they wouled have at least documented this use.

5
Things we dont like about W3C Schema (cont)

• No way to clearly specify the Schema
• Hint attributes!
• Confused world of support and no support for the
  hints, inconsistent on what should happen when
  they appear
• No way to version schema that is supported by
  tools
• xsdversion attribute has no use or support
• Namespaces only way to force recognition of a
  change
• Namespace designation of extension want a
  different namespace rather than the original
  ACORD. (namespace pollution)
• Extensions, did content really have to be added
  outside the original choice or sequence?
• Extensibility of enumerated lists
• Probably 1/3 of the elements in ACORD are based
  upon code lists, and we cant extended them.
• Need a way to identify the extended code values.
  We have our tpe based upon QName, but this
  doesnt work for external lists.

6
Things we dont like about W3C Schema (cont)

• No easy way to both validate or ignore extensions
• We need extensions, but not all trading partners
  will know what the extensions are, so they need
  to at least be able to validate the basic ACORD
  information and structure
• We have two designs within our standard
• One uses global elements and groups with
  restriction and extension
• Another uses an element of ltExtensiongt that is of
  type xsdany and is added at the end of our
  aggregates
• How about a processContents"strict or
  processContentslax"
• We could use co-constraints and/or business
  rules
• We publish a huge schema 700 messages, 29,000
  lines need help in modularizing to make the
  smallest schema possible
• Business level validation just because a date
  is the wrong format, should I reject a 50
  million transaction?
• Dont like substituionGroups because it is not
  easy to determine the final content model
• I like the old SGML notion of no surprises

7
Where we need help

• Need an integrated approach that supports all
  tools
• Code generators my biggest headache currently
  dont support all features, dont expect to and
  have decided what subset of XML they think is
  appropriate
• Groups
• Redefine
• Attributes on data level elements
• Too many levels of inheritance in classes
  create complex code
• abstract
• Possible solution profiles
• XML DTD level support plus data types (import,
  include, namespaces)
• DTD with base types (abstracts, redefine)
• Schema extreme (key, unique, substitution groups)
• Should the W3C consider specifying a language
  that uses XML as a data type? Scripting or
  compiled
• XML should become a native data type
• Consider the adoption of a catalog approach to
  schema location, general processing, and
  versioning
• One standard has a wide content model for
  messages and uses a type code to designate the
  type of message being sent
• Everything is optional in the core design of
  elements
• We need a very rich method for managing
  co-constraints based upon several hundred type
  codes
• Would really like to somehow get auto-magic
  schema validation of these real message
  requirements
• We want to standardize on the resulting data
  stream when it comes to extensions

8
Where we need help (cont)

• Reuse of other schemas
• Good intentions, but impossible to do unless
  designed at the same time
• Design principals and modeling techniques
• Based upon different datatypes with/without id
  attributes
• I want to share part C of an outside schema,
  outside schema has a definition for Party and it
  is used in C, I also have a Party definition and
  I need to substitute it in part C to make it
  useful
• No easy way to take a part of the other schema
  unless it is in a subset file
• All I want is codelist A
• All I want is the Party definition
• Need to promote a standard way to define
  codelists and encourage standards producers to
  publish them in XML based upon that standard. See
  previous bullet.
• Want extensions to be identified against the
  extender, not the originator
• Propose a new attribute "hideBase" to manage
  namespace pollution

9
Why ask why?

• Why do xml special attributes require definition
  in schema, but not DTD?
• xmlspace requires and include file, but xmlns is
  automagically there!
• How will the xmlid attribute be implemented?
• Why is a simpleType not always simple?
• Why does adding an attribute make a simpleType
  become complex with a simple content?