RELAX: Schemas Don't Have to be Hard

RELAX: Schemas Don't Have to be Hard Elliotte Rusty Harold Friday, March 18, 2005 elharo@metalab.unc.edu http://www.cafeconleche.org/

What are Schemas?

• Generically, a document that describes what a correct document may contain

• Specifically, a very complex and baroque W3C Recommendation for an XML-document syntax that describes the permissible contents of XML documents

• But there other options!

About RELAX NG

• Created by James Clark and Murata Makoto

• Based on TREX and RELAX

• Formal theory is hedge automata

• ISO standard: ISO/IEC 19757-2:2002(E), Document Schema Definition Languages (DSDL) — Part 2:Regular-grammar-based validation — RELAX NG, I

What's Wrong with DTDs?

• Unusual, non-XML like syntax

• No data typing, especially for element content

• Limited extensibility

• Only marginally compatible with namespaces

• Cannot use mixed content and enforce order and number of child elements

• Cannot enforce number of child elements without also enforcing order. (i.e. no & operator from SGML)

• Confuses infoset augmentation with validation

What's Wrong with W3C XML Schema Language?

• Complex, hard to understand

• Not sufficiently extensible:

  • Checksums

  • SKUs match database

  • sum of item prices equals total price

• Poorly implemented

• Confuses infoset annotation with validation

greeting.xml

greeting.rng

Validating the document with Jing

$ jing greeting.rng greeting.xml

An Invalid Document

Checking the Invalid Document

$ jing  greeting.rng greeting3.xml
/Users/elharo/Documents/speaking/sd2005west/relaxng/examples/greeting3.xml:
3:6: error: unknown element "P"

Notice how completely decoupled the validation is from the instance document. We can validate any document against any schema. We do not need to specify the schema in the instance document as you must do with DTDs and often need to do with W3C schemas.

RELAX NG Validators

• Jing (Java): http://www.thaiopensource.com/relaxng/jing.html

• libxml2 (C): http://xmlsoft.org/

• Sun's Multischema Validator (Java): http://wwws.sun.com/software/xml/developers/multischema/

• Tenuto (.NET): http://sourceforge.net/projects/relaxng

The compact syntax: greeting.rnc

A More Complex Document

A More Complex Schema

Groups

• zeroOrMore

• oneOrMore

• optional

A More Flexible Schema

A Couple of Notes

• Boundary white space does not need to be declared

Element Content

Nested structures

• All of these structures can nest straight-forwardly

Sharing Content Models

• PRODUCER and COMPOSER are really the same type.

When Order Doesn't Matter

interleave

• Much more powerful than either DTDs or schemas

Mixed Content

• RELAX NG can enforce order and appearance of elements in mixed content.

• Interleave with text element to validate mixed content

Declaring Mixed Content

...
  <define name="personContent">
    <element name="NAME">
      <interleave>
        <text />
        <element name="GIVEN">
          <text/>
        </element>
        <element name="FAMILY">
          <text/>
        </element>
      </interleave>
    </element>
  </define>
...

personContent =
  element NAME {
    text
    & element GIVEN  { text }
    & element FAMILY { text }
  }

Choices

• choice requires exactly one of a group of specified items to appear

• Can be enclosed in optional, oneOrMore, or zeroOrMore

A song must have at least one of ARTIST, COMPOSER, or PRODUCER:

Groups

• group defines a model that can be used as a particle in other models

• Like a sequence in W3C schemas or parentheses in DTDs

Allow a NAME element to contain either plain text or a GIVEN and a FAMILY but not both:

<element name="NAME>
  <choice>
    <text/>
    <group>
       <interleave>
         <element name="GIVEN">
           <text/>
         </element>
         <element name="FAMILY">
           <text/>
         </element>
       </interleave>
    </group>
  </choice>
</element>

A Document with Attributes

Declaring Attributes

• attribute element

• name attribute specifies the name of the attribute

• optional element can make an attribute optional

Enumerations

• List of all allowed values from which one must be chosen

• A choice containing value elements

The publisher must be one of the oligopoly that controls 90% of U.S. music (Warner-Elektra-Atlantic, Universal Music Group, Sony Music Entertainment, Inc., Capitol Records, Inc., BMG Music)

Namespaces

• ns attribute specifies the namespace of the element it matches

• Only applies in schema

• Applies to that element element and its descendants

• Until overridden

• Validates solely based on URI; irrelevant whether instance document uses prefixes or not

Default Namespace

A Schema for a Document that uses Namespaces

A Prefixed Schema for a Document that uses Namespaces

• Can use prefixed names instead if you prefer

• Usual namespace prefix binding attributes

• Convenient in rare cases when namespaces shift from one element to the next (two widely intermixed vocabularies like XSLT)

Namespaces on Attributes

Attaching a namespace to an attribute

• attribute element can have an ns attribute to specify the namespace

• attribute element does not inherit ns attribute

Data Typing

Consider this document:

<foo>
    <value>45.67</value>
</foo>

What is the type of value?

Possible types

• A longitude or latitude

• A decimal monetary type, as in COBOL

• A fixed point number

• An infinitely precise floating point number such as represented by the java.math.BigDecimal class

• An IEEE754 double

• A Java double

• An IEEE 754 float

• A VAX Fortran REAL

• An imprecisely known decimal number with 4 significant digits that's plus or minus 1 in the last place.

• An imprecisely known decimal number with 4 significant digits that's plus or minus 5 in the last place.

• Build 67 of version 45 of Microsoft Word

• A regular expression matching all strings that begin with the two characters '4' and '5', followed by a single character, followed by the two characters '6' and '7'.

• A string of characters a monkey typed on a keyboard

Other interpretations are doubtless possible, and even make sense in particular contexts. There's no guarantee that the string 45.67 in fact represents any particular type.

RELAX NG type libraries

• RELAX NG defines no data types beyond text

• But implementations are free to implement other type libraries

• W3C XML Schema Data Types are commonly available

• User defined type libraries

data element

• data element replaces text element in content models

Primitive Data Types for Schemas

• Boolean

• String

• URIs

• Numeric types

• Time types

• XML types

Numeric Data Types for Schemas

Time Data Types for Schemas

XML Data Types for Schemas

Assorted Data Types for Schemas

Restricting types

• param children of data specify the constraints on the type to create a subset of the normally accepted values

• For example, this data element restricts a date to be any year from 1877 (the year Edison invented the phonograph) on:

  <element name="YEAR">
    <data type="gYear">
      <param name="minInclusive">1877</param/>
    </data>
  </element>

• In the compact syntax:

  element YEAR {xsd:gYear {minInclusive = "1877"}}

Facets

• Facets include:

  • length

  • minLength

  • maxLength

  • pattern

  • maxInclusive

  • maxExclusive

  • minInclusive

  • minExclusive

  • totalDigits

  • fractionDigits

• Not all facets apply to all types.

• Facets not allowed/applicable in RELAX NG:

  • enumeration

  • whiteSpace

Length Facets: length, minLength, maxLength

• The number of units allowed in a value

• For strings (string, normalizedString, token, QName, NCname, ID, IDREF, language, anyURI, ENTITY, NOTATION, and NMTOKEN) the units are characters

• For lists (IDREFS, ENTITIES, and NMTOKENS) the units are tokens

• For binary types (hexBinary, base64Binary) the units are bytes after decoding

• Must be a non-negative integer

• For example, to say that all names and titles must contain between 1 and 255 characters:

• In compact syntax:

Facets for ordered items: minExclusive, maxExclusive, minInclusive, maxInclusive

• Determines the minimum and maximum allowed values

• Applies to ordered simple types including byte, unsignedByte, integer, positiveInteger, negativeInteger, nonNegativeInteger, nonPositiveInteger, int, unsignedInt, long, number, unsignedLong, short, unsignedShort, number, float, double, time, dateTime, duration, date, gMonth, gYear, gDay, and gMonthDay.

• For example, to say that the year must be between 1877 and 2100:

• In the compact syntax:

Facets for decimal numbers: totalDigits and fractionDigits

• totalDigits facet specifies the maximum number of decimal digits in a number as a positive integer

• fractionDigits facet specifies the maximum number of decimal digits to the right of the decimal point as a non-negative integer

• Applies to all types derived from decimal including byte, unsignedByte, integer, positiveInteger, negativeInteger, nonNegativeInteger, nonPositiveInteger, int, unsignedInt, long, unsignedLong, short, and unsignedShort.

• Does not apply to float and double

• You can specify at most two fractional digits or at most seven decimal digits, but not at least two fractional digits or exactly seven decimal digits

Adding a Price

The pattern facet

• Suppose you want a money type to specify that the PRICE element content must look like $1.35 or ¥11000

• Use the pattern facet to specify a regular expression instances must match

Regular Expressions

• More or less Perl-like including the Unicode extensions introduced in Perl 5.6

• The money regular expression:

  \p{Sc}\p{Nd}+(\.\p{Nd}\p{Nd})?

  \p{Sc}

  Any Unicode currency indicator; e.g. $, ¥, £, ¤, etc.

  \p{Nd}

  A Unicode decimal digit character

  \p{Nd}+

  One or more Unicode decimal digit characters

  \.

  The period character

  (\.\p{Nd}\p{Nd})

  A string with a period followed by two decimal digits like .35

  (\.\p{Nd}\p{Nd})?

  Zero or one strings of the form .35

The Price Schema

Lists

• Matches a list of words separated by white space

• Each word has a type

• Can have multiple types

Allow multiple years in the YEAR element:

<element name="YEAR">
  <list>
    <oneOrMore>
      <data type="gYear"/>
    </oneOrMore>
  </list>
</element>

element SONG {
  element TITLE     { text },
  element COMPOSER  { xsd:string }+,
  element PRODUCER  { xsd:string }*,
  element PUBLISHER { xsd:string }?,
  element LENGTH    { xsd:string }?,
  element YEAR {
    list { xsd:gYear+ }
  }?,
  element ARTIST    { xsd:string }+
}

User defined type libraries

• Available in Java (1.2 or later) and .NET

• Implement the org.relaxng.datatype.DatatypeLibraryFactory interface

Prime Datatype

Prime Datatype Library

Prime Datatype Library Factory

Locating type libraries

• Put the datatype library in a JAR archive

• Include a META-INF/services/org.relaxng.datatype.DatatypeLibraryFactory file that contains the name of the class that implements the org.relaxng.datatype.DatatypeLibraryFactory interface

• Add the .jar file to the CLASSPATH

• Add the MSV/Jing .jar file to the CLASSPATH (Runnable archives don't work)

An invalid document

A Prime schema

include element

• Breaks schema into multiple parts.

externalref element

• Matches any pattern defined in an external schema

• href attribute points to the external schema

Pointing to the XHTML schema

Annotations

• All elements/attributes from namespaces other than http://relaxng.org/ns/structure/1.0 are ignored

• Annotate with XHTML, Schematron or any other namespaced vocabulary

• Use RELAX NG div element to group multiple elements in a grammar together with a single annotation

RELAX NG in Java 1.5/JAXP 1.3

• Not bundled; must install third party library

Other features

• Wild cards with anyName and nsName

• Exclusions with except

• Redefining external content models when importing

• Attribute or child element

What RELAX NG doesn't do

• Cannot declare entities

• Parent models

• Extra-document validation

• ID-IDREF/key-keyref^*

• PSVI

^* Can be added

RELAX NG is being used for

• XHTML 2.0

• DocBook 4.4/5.0

• SVG 1.2

• XBL

• OpenOffice

• and many more...

Trang

• Converts between:

  • RELAX XML syntax

  • RELAX Compact syntax

  • DTDs

  • W3C XML Schema Language

$ trang http://www.w3.org/Graphics/SVG/1.2/rng/Full-1.2/Full-1.2.rng full.xsd
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/tiny-structure.rng:401:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/tiny-structure.rng:420:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/tiny-structure.rng:438:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Full-1.2/structure.rng:24:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/script.rng:38:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/tiny-flow.rng:22:15: warning: cannot represent an optional group of attributes; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Tiny-1.2/handler.rng:44:13: warning: choice between attributes and children cannot be represented; approximating
http://www.w3.org/Graphics/SVG/1.2/rng/Full-1.2/style.rng:81:13: warning: choice between attributes and children cannot be represented; approximating

To Learn More

• RELAX NG

  • Eric van der Vlist

  • O'Reilly & Associates, 2004

  • ISBN 0-596-00421-4

  • http://books.xmlschemata.org/relaxng/

  
  

• This presentation: http://www.cafeconleche.org/slides/sd2005west/relaxng/

• RELAX NG Tutorial: http://www.oasis-open.org/committees/relax-ng/tutorial.html