Designing XML Formats: Versioning vs' Extensibility

1
Designing XML Formats Versioning vs.
Extensibility

• Dare Obasanjo
• Program Manager
• Microsoft Corporation
• dareo_at_microsoft.com

2
Definition Versioning

• The process for modifying an XML format over
  subsequent releases v1.0 -gt v1.1 -gt v2.0 -gt v3.0
• Types of versioning
• Backward compatible v2.0 format -gt v1.0 consumer
• Forward compatible v1.0 format -gt v2.0 consumer
• Types of changes
• Breaking change a change that is not backward or
  forward compatible
• Non-breaking change a change that is backward
  and forward compatible

3
Why is Versioning Important?

• Need Ways to Migrate Users from Services
• SOAP 1.1 -gt SOAP 1.2 Google, Amazon, eBay
• Atom 0.3 -gt Atom 1.0 Blogger
• Poor Versioning Story Hurts Migration to New
  Formats
• HTML -gt XHTML
• XML 1.0 -gt XML 1.1

4
Message Transfer Negotiation

• On closed networks like LANs or where formal
  agreements made before message exchange occurs
• Versioning can be done out of band
• Versions of format used can be fixed

5
Versioning Message Payloads

• On WANs like the WWW message exchange done
  without a priori agreements
• Its hard or impossible to upgrade all clients
  and servers at the same time
• Messages should carry versioning information
• This is the scenario we care about

6
Definition Extensibility

• A mechanism that enables new data to show up side
  by side (or concurrently) with data for a given
  format
• Typically version-independent
• Primarily benefits third parties who want to
  extend the format

7
Why is Extensibility Important?

• Enables evolution without central control of the
  format
• Not limited to XML formats. Consider extensible
  headers in HTTP, email and USENET.
• Simplifies the format by pushing niche use cases
  out of the core
• RSS 2.0 Extensions for reading posting
  comments
• XSD Embedded Schematron and SQLXML annotated
  schemas

8
Example Person 1.0

• ltperson version"1.0" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• lt/persongt

9
Version Numbers vs. Namespace Names

• The namespace name is part of the identity
• Changing namespace name breaks namespace-aware
  applications such as XSLT, XPath, etc
• Use version numbers for minor revisions and
  change namespace name for major revisions
• Version numbers primarily for human readers
• Practice is inconsistent today
• SOAP 1.1 -gt SOAP 1.2 (different namespace)
• XSLT 1.0 -gt XSLT 2.0 (same namespace)

10
Example Person 1.1

• ltperson version"1.1" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltmiddle-initialgtJlt/middle-initialgt
• lt/persongt

11
Example Person 2.0

• ltperson version"2.0" xmlns"http//www.example.or
  g/person2" gt
• ltnamegt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltmiddle-initialgtJlt/middle-initialgt
• lt/namegt
• ltssngt123-45-6789lt/ssngt
• lt/persongt

12
The Unique Particle Attribution problem I

• ltperson version"1.0" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• lt/persongt

• ltxsschema xmlnsxs"http//www.w3.org/2001/XMLSch
  ema" targetNamespace"http//www.example.com/perso
  n" elementFormDefault"qualified"gt
• ltxselement name"person"gt
• ltxscomplexTypegt
• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxsany minOccurs"0" maxOccurs"unbounded"
  namespace"targetNamespace" /gt
• lt/xssequencegt
• ltxsattribute name"version" type"xsstring"
  /gt
• lt/xscomplexTypegt
• lt/xselementgt
• lt/xsschemagt

13
The Unique Particle Attribution problem II

• ltperson version"1.1" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltmiddle-initialgtJlt/middle-initialgt
• lt/persongt

• ltxsschema xmlnsxs"http//www.w3.org/2001/XMLSch
  ema" targetNamespace"http//www.example.com/perso
  n" elementFormDefault"qualified"gt
• ltxselement name"person"gt
• ltxscomplexTypegt
• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxselement name"middle-initial"
  type"xsstring" minOccurs"0" /gt
• ltxsany minOccurs"0" maxOccurs"unbounded"
  namespace"targetNamespace" /gt
• lt/xssequencegt
• ltxsattribute name"version" type"xsstring"
  /gt
• lt/xscomplexTypegt
• lt/xselementgt
• lt/xsschemagt

14
The Unique Particle Attribution problem contd

• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxselement name"middle-initial"
  type"xsstring" minOccurs"0" /gt
• ltxsany minOccurs"0" maxOccurs"unbounded"
  namespace"targetNamespace" /gt
• lt/xssequencegt

15
XSD Techniques Versioning

• New constructs in new namespace
• Elements and attributes from subsequent versions
  use different namespace

16
New constructs in new namespace I

• ltperson version"1.0" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• lt/persongt
• ltperson version"1.1" xmlns"http//www.example.or
  g/person"
• xmlnsv11"http//www.example.org/personv11
  " gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltv11middle-initialgtJlt/v11middle-initialgt
• lt/persongt

• ltxsschema xmlnsxs"http//www.w3.org/2001/XMLSch
  ema" targetNamespace"http//www.example.com/perso
  n" elementFormDefault"qualified"gt
• ltxselement name"person"gt
• ltxscomplexTypegt
• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxsany minOccurs"0" maxOccurs"unbounded"
  namespace"other" /gt
• lt/xssequencegt
• ltxsattribute name"version" type"xsstring"
  /gt
• lt/xscomplexTypegt
• lt/xselementgt
• lt/xsschemagt

17
New constructs in new namespace II

• Pro
• Straight forward
• Con
• Cant differentiate extensions from core
• Con Additional schema documents needed for
  subsequent versions
• Con not forward compatible in a number of cases

18
XSD Techniques Versioning

• Using extensibility points with sentries
• Extensibility points bounded by delimiters
• Pro Constructs in format use same namespace
• Con Adds complexity to format

19
Using extensibility points with sentries I

• ltperson version"1.0" xmlns"http//www.example.or
  g/person" gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• lt/persongt
• ltperson version"1.1" xmlns"http//www.example.or
  g/person"
• gt
• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltdelimiter /gt
• ltmiddle-initialgtJlt/middle-initialgt
• ltend /gt
• lt/persongt

• ltxsschema xmlnsxs"http//www.w3.org/2001/XMLSch
  ema"
• xmlnsp"http//www.example.org/person"
  targetNamespace"http//www.example.com/person"
  elementFormDefault"qualified"gt
• ltxselement name"delimiter"gt
• ltxscomplexType /gt
• lt/xselementgt
• ltxselement name"end"gt
• ltxscomplexType /gt
• lt/xselementgt
• ltxselement name"person"gt
• ltxscomplexTypegt
• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxssequence minOccurs"0" maxOccurs"1"gt
• ltxssequence minOccurs"0"
  maxOccurs"unbounded"gt
• ltxselement ref"pdelimiter" /gt
• ltxsany namespace"targetNamespace
  local" minOccurs"0" maxOccurs "unbounded"/gt
• lt/xssequencegt
• ltxselement ref"pend" /gt

20
Using extensibility points with sentries I contd

• ltxssequencegt
• ltxselement name"first-name"
  type"xsstring" /gt
• ltxselement name"last-name"
  type"xsstring" /gt
• ltxssequence minOccurs"0" maxOccurs"1"gt
• ltxssequence minOccurs"0"
  maxOccurs"unbounded gt
• ltxselement ref"pdelimiter" /gt
• ltxsany namespace"targetNamespace
  local" minOccurs"0" maxOccurs "unbounded"/gt
• lt/xssequencegt
• ltxselement ref"pend" /gt
• lt/xssequencegt
• lt/xssequencegt

• ltfirst-namegtBoblt/first-namegt
• ltlast-namegtSmithlt/last-namegt
• ltdelimiter /gt
• ltmiddle-initialgtJlt/middle-initialgt
• ltend /gt

21
Using extensibility points with sentries II

• Pro
• Constructs in format use same namespace
• Is backwards and forwards compatible
• Con
• Adds complexity to format

22
XSD Techniques Extensibility

• Using wildcards to create open content models
• Gaining flexibility from substitution groups and
  abstract elements
• Runtime polymorphism via xsitype and abstract
  types
• Using xsredefine to update type definitions

23
General Guidance

• XML formats should be extensible
• Corollary Processing model for extensions MUST
  be defined
• Use the same namespace name for compatible
  versions of the format
• Favor non-breaking changes wherever possible
• You can add, but never remove

24
Questions?