Xml Matters #27: Relax Ng Forms

Compact Syntax and XML Syntax

David Mertz, Ph.D.
Facilitator, Gnosis Software, Inc.
May, 2003

The RELAX NG compact syntax provides a much less verbose, and easier to read, format for describing the same semantic constraints as RELAX NG XML syntax. This installment looks at tools for working with and transforming between the two syntax forms.

Emphasizing Readability

Readers of my earlier installments on RELAX NG will have noticed that I chose to provide many of my examples using compact syntax rather than XML syntax. Both formats are semantically equivalent, but the compact syntax is, in my opinion, far easier to read and write. Moreover, readers of this column in general will have a sense of how little enamored I am of the notion that everything vaguely related to XML technologies must itself use an XML format. XSLT is a prominent example of this XML-everywhere tendency, and its pitfalls--but that is a rant for a different column.

In the later part of this article, I will discuss the format of the RELAX NG compact syntax in more detail than the prior installments allowed.

Tool Support

On the down side, since the RELAX NG compact syntax is newer--and not 100% settled at its edges--tool support for compact syntax is less complete than for the XML syntax. For example, even though the Java tool trang supports conversion between compact and XML syntax, the associated tool jing will only validate against XML syntax schemas. Obviously, it is not overly difficult to generate the XML syntax RELAX NG schema to use for validation, but direct usage of the compact syntax schema would be more convenient. Likewise, the Python tools xvif and 4xml validate only against XML syntax schemas.

To help remedy the gaps in direct support for compact syntax, I have produced a Python tool for parsing RELAX NG compact schemas, and for outputting them to XML format. While my rnc2rng tool only does what trang does, Eric van der Vlist and Uche Ogbuji have expressed their interest in including rnc2rng in xvif and 4xml, respectively. Hopefully, in the near future direct validation against compact syntax schemas will be included in these tools.

Writing rnc2rng proved more difficult than I anticipated; and there is probably a lesson in that. While RELAX NG compact syntax is quite readable--as we will see below--there are enough variations in the arrangement of tokens between instances that a parser was non-trivial to write. For better or worse, I use PLY's lex module to tokenize the schema, but gave up on using yacc for the parsing, using application-specific massaging of the token stream instead. Debugging declarative grammars is often more difficult than incrementally adjusting imperative code. Despite my frequent concern for the unfriendliness of XML, the task of parsing an XML syntax schema would have been far simpler, since I could have let a framework like SAX or DOM do most of the work for me.

More On Relax Ng Editors

Since the last installment, tool support for RELAX NG has gotten a little bit better. The XML editor oXygen has come out with a version 2.0 that incorporates trang as a plugin, and thereby some support for RELAX NG. While this is not the place for a full review, I found oXygen 2.0--which I liked in version 1.2 to start with--has gained a number of nice features and general polish. I would like to see RELAX NG integrated at a deeper level into various editors--to a degree similar to DTD and W3C XML Schema. With a bit more time, I think greater RELAX NG integration into tools is likely.

Syntax Features - Namespaces

A compact syntax RELAX NG schema may begin with any of several optional namespace declarations. Each of these looks a lot like an assignment statement in a programming language. A default namespace for schema tags may be specified with:

default namespace = "http://relaxng.org/ns/structure/version"

When converted to XML syntax, use of this declaration appends a "ns" attribute to the root element of the schema. If this namespace is not explicitly specified, the "default default" namespace is used, and is declared with the root attribute, e.g.:

<root-tag xmlns="http://relaxng.org/ns/structure/1.0">

You may also declare an external namespace for elements or attributes:

namespace foo = "http://some.path.to/foo"

This lets you describe elements like:

element foo:bar { ... }

When converted to XML syntax, the namespace URL is added to the root tag as an extra attribute:

<root-tag xmlns="http://relaxng.org/ns/structure/1.0"

The namespace "a" is a bit special here. RELAX NG allows annotations, which are basically just tags with the "a" namespace. In compact syntax, you can avoid thinking about namespaces, and add an annotation with initial double hash marks:

## An annotation

Converted to XML syntax, this annotation appears as:

<a:documentation>An annotation</a:documentation>

By the way, a single leading hash introduces a comment instead of an annotation. The following compact syntax form:

# This is a comment

Corresponds to the XML form:

<!-- This is a comment -->

You can also use a slightly odd compact syntax form to specify other annotations within the "a" namespace:

[ a:defaultValue = "foo" ]

A root attribute "xmlns:a" will be specified automatically in the XML syntax if annotations are used, but since "a" is just another namespace, you can specfify you own URL if you want. The default attribute is equivalent to specifying:

namespace a = "http://relaxng.org/ns/compatibility/annotation/1.0"

One more special namespace is specified differently, in both syntax forms. Datatypes rely on a modular specification, usually using W3C XML Schema datatypes. You may specify these with compact syntax:

datatypes xsd = "http://www.w3.org/2001/XMLSchama-datatypes"

Or XML syntax:

<root-tag xmlns="http://relaxng.org/ns/structure/1.0"

Syntax Features - Nested And Context Free

The main body of a RELAX NG grammar may have either of two styles. In some way, the more direct style is to simply nest elements and attributes where they should occur in a valid instance. Generally it is good form to use indentation much as you would in a programming language, but as in C-family languages, curly-braces are the actual block delimiters. A moderately complete schema looks like, e.g.:

A nested compact syntax schema

# A library patron example
default namespace = "http://some.other.url/ns"
namespace foo = "http://home.of.foo/ns"
datatypes xsd = "http://www.w3.org/2001/XMLSchema-datatypes"
## Annotation here
element patron {
  element name { xsd:string { pattern = "\w{,10}" } }
  & element id-num { xsd:string }
  & element book {
      ( attribute isbn { text }
      | attribute title { text }
      | attribute anonymous { empty })

The library patron example uses most of the syntax elements. Interpersed "&"s between elements (or attributes) indicate that the several elements must occur, but may do so in any order. In XML syntax, this is the same as the <interleave> tag. Likewise, interpersed "|"s indicate a choice between several items--in XML, <choice>. Notice the "book" element too, the parenthesis indicate a group, but they are redundant in this case. A group (XML: <group>), however, is useful as part of quantification or interpersal, e.g.:

Using groups for quanitfication

element foo {
    ( element bar { text },
      element baz { text } )+,
    element bam { text } }

In this case, a valid document's root <foo> element might have contain several <bar></bar><baz></baz> sequences prior to one final <bam> element. There is no way to express the same concept by only quantifying the indifidual "bar" and "baz" elements.

A nested-style RELAX NG grammar need not describe a single element only. Any well-formed XML document must have a single root element, so clearly an attribute at the top is prohibited. Likewise a sequence or interleave description at the top level could not describe a well-formed XML document, and therefore it could not describe a valid one. But there is nothing wrong with allowing a choice of root elements, e.g.:

Choice as top level grammar

( element foo {text}
| element bar {text} )

A second style of RELAX NG grammar more closely resembles a DTD. A special "production" named "start" is indicated at the beginning, followed by a variety of other named productions. As with namespace declarations, a production is named in the manner of an assignment in a programming language. For example, a library patron schema could also look something like:

A context free compact syntax schema

# A library patron example
default namespace = "http://some.other.url/ns"
namespace foo = "http://home.of.foo/ns"
datatypes xsd = "http://www.w3.org/2001/XMLSchema-datatypes"
## Annotation here
start = patron
patron = name & id-num & book
name = element name { xsd:string { pattern = "\w{,10}" } }
id-num = element id-num { xsd:string }
book = element book {
      ( attribute isbn { text }
      | attribute title { text }
      | attribute anonymous { empty }) }*

Names of productions may occur within other productions, which can avoid repititions, and generally make complex patterns more readable. Beyond readability, naming patterns allows recursive definition of patterns--either direct or mutual recursion. For example, describing HTML--where tables can nest within tables, or lists within lists--is not possible in a strictly nested style. An upshot of recursive XML instance documents is to make DTDs and context free RELAX NG much more natural as descriptions than is W3C XML Schemas (but you can get what is needed out of W3C XML Schemas, just with more work).

It is probably worth looking at an entire XML syntax RELAX NG schema document. For comparison, the below is what rnc2rng produces when processing the above context free library patron schema:

A context free XML syntax schema

<?xml version="1.0" encoding="UTF-8"?>
<!-- A library patron example -->
<grammar xmlns="http://relaxng/ns/structure/1.0"
  <a:documentation>Annotation here</a:documentation>
  <start><ref name="patron"/></start>
  <define name="patron">
      <ref name="name"/>
      <ref name="id-num"/>
      <ref name="book"/>
  <define name="name">
    <element name="name">
      <data type="string"/>
        <param name="pattern">\w{,10}</param>
  <define name="id-num">
    <element name="id-num">
      <data type="string"/>
  <define name="book">
      <element name="book">
          <attribute name="isbn"/>
          <attribute name="title"/>
          <attribute name="anonymous">

I would say this is easier to read than a W3C XML Schema, but it sure does not come close to the compact syntax (prior installments pointed out that this schema is actually impossible to express precisely in W3C XML Schema, or DTDs).


In some of these examples you will have noticed that elements and attributes, in compact syntax, always contain something in curly braces after their name. In XML syntax, you can self-close an attribute tag, but to prevent ambiguity, you need to specify at least {text} or {empty} for an attribute body. You can also use a more complex datatype description if you wish, of course. Also, the only quantification that makes sense for attributes is "?"--attributes might be optional, but they will not be repeated multiple times.

In some corner cases, rnc2rng differs from trang. For example, both tools force an annotation to occur inside a root element in XML syntax, even if the annotation line occurs before the root element in the compact syntax. Since well-formed XML documents are single rooted, this is a necessity. But trang also moves comments in a similar manner, while rnc2rng does not. At minimum, some slightly different use is made of whitespace between the tools. Most likely, a few other variations exist, but hopefully none that are semantically important.


The xvif library itself can be downloaded from:


However, 4Suite is a somewhat more polished tool that incorporates xvif for RELAX NG validation. The command-line tool 4xml will validate against both RELAX NG and DTDs, with various options. 4Suite includes many other tools and libraries for working with many XML-related technologies:


The tools trang and jing are complementary tools for transformation between schemata and validation against RELAX NG schemas. The former depends on the latter, but both can be downloaded in a convenient archive from:


You will need to optain an implementation of the Java API for XML Processing (JAXP) to use trang. If you run a Java 1.4 JVM, you are fine already; otherwise, obtain crimson at:


DTDinst is a Java tool to convert DTDs into an XML instance document format, including handling of parametric entities:


The DTDinst XML format is of limited utility by itself, since nothing else works with it. However, an XSLT stylesheet is available to transform this format into RELAX NG (with a few caveats). You will need an XSLT tool to utilize this:


A collection of documents and tools presented in this series of articles can be found at:


My ealier reviews of XML editors (including oXygen) for this column can be found at:




About The Author

Picture of Author David Mertz thinks that the schema that is real is not the real schema. David may be reached at mertz@gnosis.cx; his life pored over athttp://gnosis.cx/publish/. Suggestions and recommendations on this, past, or future, columns are welcomed.