Digest Values for DOM (DOMHASH) Proposal

Revision History
23 Apr 1998Initial version
25 May 1998Clarify definition for text node handling
12 Jun 1998Clean up for external release
19 Aug 1998Some modification for DOM Level 1 Proposed Recommendation
22 Jan 1999Namespace support incorporated (Internet Draft version)
24 Feb 1999Changes about Comment and ProcessingInstruction
Authors:
Hiroshi MARUYAMA, IBM Research, Tokyo Research Laboratory
Kent TAMURA, IBM Research, Tokyo Research Laboratory
Naohiko URAMOTO, IBM Research, Tokyo Research Laboratory

Status of this Document

This document is intended to contribute discussions how digest (hash) values should be defined for general DOM structures. See Document Object Model (DOM) Level 1 Specification Version 1.0 for the specifications of W3C DOM 1.0.


Contents

  1. Introduction
  2. Hash Value Calculation
  3. Proposed API
  4. Discussions
  5. References

1. Introduction

The purpose of this document is to give a clear and unambiguous definition of digest (hash) values of the XML objects. In particular, we propose to add a new API getDigest() to the interface Node that returns a digest value, a fixed length value (normally 128 bits or 160 bits) representing an entire subtree. Two subtrees are considered identical if their hash values are the same, and different if their hash values are different.

There are at least two usage scenarios of DOMHASH. One is as a basis for the Digital Signature for XML (XMLDSIG) proposal. Digital signature algorithms normally require hashing a signed content before signing. DOMHASH provides a concrete definition of the hash value calculation.

The other is to use DOMHASH when synchronizing two DOM structures. Suppose that a server program generates a DOM structure which is to be rendered by clients. If the server makes frequent small changes on a large DOM tree, it is desirable that only the modified parts are sent over to the client. A client can initiate a request by sending the root hash value of the structure in the cache memory. If it matches with the root hash value of the current server structure, nothing needs be sent. If not, then the server compares the client hash with the older versions in the server's cache. If it finds one that matches the client's version of the structure, then it locates differences with the current version by recursively comparing the hash values of each node. This way, the client can receive only an updated portion of a large structure without requesting the whole thing. A similar idea of minimizing the network communication for data replication was proposed in The HTTP Distribution and Replication Protocol.

One way of defining digest values is to take a surface string as the input for a digest algorithm. However, this approach has several drawbacks. The same internal DOM structure may be represented in may different ways as surface strings even if they strictly conform to the XML specification. Treatment of white spaces, selection of character encodings, entity references (i.e., use of ampersands), and so on have impact on the generation of a surface string. If the implementations of surface string generation are different, the hash values would be different, resulting in unvalidatable digital signatures and unsuccessful detection of identical DOM structures. Therefore, it is desirable that digest of DOM is defined in the DOM terms -- that is, as an unambiguous algorithm using the DOM API. This is the approach we take in this proposal.

Introduction of namespace is another source of variation of surface string because different namespace prefixes can be used for representing the same namespace URI. In the following example, the namespace prefix "edi" is bound to the URI "http://ecommerce.org/schema" but this prefix can be arbitrary chosen without changing the logical contents as shown in the second example.
<?xml version="1.0"?>
<root xmlns:edi='http://ecommerce.org/schema'>
    <edi:order>
        :
    </edi:order>
</root>
<?xml version="1.0"?>
<root xmlns:ec='http://ecommerce.org/schema'>
    <ec:order>
        :
    </ec:order>
</root>

The DOMHash defined in this document is designed so that the choice of the namespace prefix does not affect the digest value. In the above example, both the "root" elements will get the same digest value.


2. Digest Calculation

2.1. Overview

Hash values are defined on the DOM type Node. We consider the following five node types that are used for representing a DOM document structure:

  1. Element
  2. Attr
  3. ProcessingInstruction
  4. Text (including the subtype CDATASection)

Comment nodes and Document Type Definitions (DTDs) do not participate in the digest value calculation. This is because DOM does not require a conformant processor to create data structures for these. DOMHash is designed so that it can be computed with any XML processor conformant to the DOM or SAX specification.

Nodes with the node type EntityReference are assumed to be expanded before digest calculation.

The digest values are defined recursively on each level of the DOM tree so that only a relavant part needs to be recalculated when a small portion of the tree is changed.

Below, we give the precise definitions of digest for these types. We describe the format of the data to be supplied to a hash algorithm using a figure and a simple description, followed by a Java code fragment using the DOM API and the JDK 1.1 Platform Core API only. Therefore, the semantics should be unambiguous.

As the rule of thumb, all strings are to be in UTF-16 in the network byte order (Big Endian) with no byte order mark. If there is a sequence of Text nodes without any element nodes inbetween, these text nodes are merged into one by concatenating them. A zero-length Text node is always ignored.

2.2. Namespace Considerations

To avoid the dependence on the namespace prefix, we use "expanded names" to do digest calculation. If an element name or an attribute name is qualified either by a explicit namespace prefix or by a default namespace, the name's LocalPart is prepended by the URI of the namespace (the namespace name as defined in the NameSpace specification) and a colon before digest calculation. In the following example, the default qualified name "order" is expanded into "http://ecommerce.org/schema:order" while the explicit qualified name "book:title" is exapanded into "urn:loc.gov:books:title" before digest calculation.
<?xml version="1.0"?>
<root xmlns='http://ecommerce.org/schema'
         xmlns:book='urn:loc.gov:books'>
    <order>
       <book:title> ... </book:title>
        :
    </order>
</root>

We define an expanded name (either for element or attirbute) as follows:

If a name is not qualified, the exapanded name is the name itself.
If a name is qualified with the prefix "xmlns", the expanded name is undefined.
If a name is qualified either by default or by an explicit namespace prefix, the expanded name is URI bound to the namespace + ":" + LocalPart

In the following definitions, we assume that the getExpandedName() method (which returns the expanded name as defined above) is defined in both Element and Attr interfaces of DOM.

Note that the digest values are not defined on namespace declarations. In other words, the digest value is not defined for an attribute when

In the above example, the two attributes which are namespace declarations do not have digest values and therefore will not participate in the calculation of the digest value of the "root" element.

2.3. Definition with Code Fragments

The code fragments in the definitions below assume that they are in implementation classes of Node. Therefore, a methods call without an explicit object reference is for the Node itself. For example, getData() returns the text data of the current node if it is a Text node. The parameter digestAlgorithm is to be replaced by an identifier of the digest algorithm, such as "MD5" and "SHA".

The computation should begin with a four byte integer that represents the type of the node, such as Node.TEXT_NODE or Node.ELEMENT_NODE.

2.3.1. Text Nodes

The hash value of a Text node is computed on the four byte header followed by the UTF-16 encoded text string.

Node.TEXT_NODE (3) in 32 bit network-byte-ordered integer
Text data in UTF-16 stream (variable length)
  public byte[] getDigest(String digestAlgorithm) {
      MessageDigest md = MessageDigest.getInstance(digestAlgorithm);
      md.update((byte)((Node.TEXT_NODE >> 24) & 0xff));
      md.update((byte)((Node.TEXT_NODE >> 16) & 0xff));
      md.update((byte)((Node.TEXT_NODE >> 8) & 0xff));
      md.update((byte)(Node.TEXT_NODE & 0xff));
      md.update(getData().getBytes("UnicodeBigUnmarked"));
      return md.digest();
  }

Here, MessageDigest is in the package java.security.*, one of the built-in packages of JDK 1.1.

2.3.2. ProcessingInstruction Nodes

A ProcessingIinstruction (PI) node has two components: the target and the data. PI data is from the first non white space character after the target to the character immediately preceding the ?>. A digest value for <?foo param?> is the same as value for <?foo    param?>.

Node.PROCESSING_INSTRUCTION_NODE (7) in 32 bit network-byte-ordered integer
PI target in UTF-16 stream (variable length)
0x00 0x00
PI data in UTF-16 stream (variable length)
  public byte[] getDigest(String digestAlgorithm) {
      MessageDigest md = MessageDigest.getInstance(digestAlgorithm);
      md.update((byte)((Node.PROCESSING_INSTRUCTION_NODE >> 24) & 0xff));
      md.update((byte)((Node.PROCESSING_INSTRUCTION_NODE >> 16) & 0xff));
      md.update((byte)((Node.PROCESSING_INSTRUCTION_NODE >> 8) & 0xff));
      md.update((byte)(Node.PROCESSING_INSTRUCTION_NODE & 0xff));
      md.update(getName().getBytes("UnicodeBigUnmarked"));
      md.update((byte)0);
      md.update((byte)0);
      md.update(getData().getBytes("UnicodeBigUnmarked"));
      return md.digest();
  }

2.3.3. Attr Nodes

The digest value of Attr nodes are defined similarly to PI nodes, except that we need a separator between the expanded attribute name and the attribute value. The '0x0000' value in UTF-16 is allowed nowhere in an XML document, so it can serve as an unambiguous separator. The expanded name must be used as the attribute name because it may be qualified. Note that if the attribute is a namespace declaration (either the attribute name is "xmlns" or its prefix is "xmlns"), the digest value is undefined and the getDigest() method should return null.

Node.ATTRIBUTE_NODE (2) in 32 bit network-byte-ordered integer
Expanded attribute name in UTF-16 stream (variable length)
0x00 0x00
Attribute value in UTF-16 stream (variable length)
  public byte[] getDigest(String digestAlgorithm) {
      if (getNodeName().equals("xmlns") || getNodeName().startsWith("xmlns:"))
          return null;
      MessageDigest md = MessageDigest.getInstance(digestAlgorithm);
      md.update((byte)((Node.ATTRIBUTE_NODE >> 24) & 0xff));
      md.update((byte)((Node.ATTRIBUTE_NODE >> 16) & 0xff));
      md.update((byte)((Node.ATTRIBUTE_NODE >> 8) & 0xff));
      md.update((byte)(Node.ATTRIBUTE_NODE & 0xff));
      md.update(getExpandedName().getBytes("UnicodeBigUnmarked"));
      md.update((byte)0);
      md.update((byte)0);
      md.update(getValue().getBytes("UnicodeBigUnmarked"));
      return md.digest();
  }

2.3.4. Element Nodes

Element nodes are the most complex because they consist of other nodes recursively. Hash values of these component nodes are used to calculate the node's digest so that we can save computation when the structure is partially changed.

First, all the attributes except for namespace declarations must be collected. This list is sorted by the expanded attribute names. The sorting is done in ascending order in terms of the UTF-16 encoded expanded attribute names, using the string comparison operator defined as String#compareTo() in Java. The semantics of this sorting operation should be clear (no "ties" are possible because of the unique attribute name constraint).

Child nodes does not contain Comment nodes because SAX applications can not catch comment information.

Node.ELEMENT_NODE (1) in 32 bit network-byte-ordered integer
Expanded element name in UTF-16 stream (variable length)
0x00 0x00
A number of non-namespace-declaration attributes in 32 bit network-byte-ordered unsigned integer
Sequence of digest values of non-namespace-declaration attributes, sorted by String#compareTo() for attribute names
A number of child nodes except Comment nodes in 32bit network-byte-ordered unsigned integer
Sequence of digest values of each child nodes except Comment nodes (variable length)
(A sequence of child texts is merged to one text. A zero-length Text nodes and Comment node are not counted as child)
  public byte[] getDigest(String digestAlgorithm) {
      MessageDigest md = MessageDigest.getInstance(digestAlgorithm);
      ByteArrayOutputStream baos = new ByteArrayOutputStream();
      DataOutputStream dos = new DataOutputStream(baos);
      dos.writeInt(Node.ELEMENT_NODE);    // This is stored in the network byte order
      dos.write(getExpandedName().getBytes("UnicodeBigUnmarked"));
      dos.write((byte)0);
      dos.write((byte)0);
      // Collect all attributes except for namespace declarations
      NamedNodeMap nnm = this.getAttributes();
      int len = nnm.getLength()
              // Find "xmlns" or "xmlns:foo" in nnm and omit it.
      ...
      dos.writeInt(len);    // This is sorted in the network byte order
      // Sort attributes by String#compareTo() on expanded attribute names.
      ...
      // Assume that `Attr[] aattr' has sorted Attribute instances.
      for (int i = 0;  i < len;  i ++)
          dos.write(aattr[i].getDigest(digestAlgorithm));

      Node n = this.getFirstChild();
      // Assume that adjoining Texts are merged and no 0-length Text and no Comment nodes.
      len = this.getChildNodes().getLength();
      dos.writeInt(len);    // This is stored in the network byte order
      while (n != null) {
          dos.write(n.getDigest(digestAlgorithm));
          n = n.getNextSibling();
      }
      dos.close();
      md.update(baos.toByteArray());
      return md.digest();
  }

3. Proposed API

We propose to add a new method to the Node interface as shown below. The getDigest() method takes one string as its parameter that specifies the digest algorithm. We assume that at least two algorithms, "MD5" and "SHA", must be implemented for any DOM processor to be compliant with DOMHASH.

typedef sequence<octet> bytearray;

interface Node {
  // NodeType
  const unsigned short      ELEMENT_NODE       = 1;
  const unsigned short      ATTRIBUTE_NODE     = 2;
  const unsigned short      TEXT_NODE          = 3;
  const unsigned short      CDATA_SECTION_NODE = 4;
  const unsigned short      ENTITY_REFERENCE_NODE = 5;
  const unsigned short      ENTITY_NODE        = 6;
  const unsigned short      PROCESSING_INSTRUCTION_NODE = 7;
  const unsigned short      COMMENT_NODE       = 8;
  const unsigned short      DOCUMENT_NODE      = 9;
  const unsigned short      DOCUMENT_TYPE_NODE = 10;
  const unsigned short      DOCUMENT_FRAGMENT_NODE = 11;
  const unsigned short      NOTATION_NODE      = 12;

  readonly attribute  DOMString            nodeName;
           attribute  DOMString            nodeValue;
                                                 // raises(DOMException) on setting
                                                 // raises(DOMException) on retrieval
  readonly attribute  unsigned short       nodeType;
  readonly attribute  Node                 parentNode;
  readonly attribute  NodeList             childNodes;
  readonly attribute  Node                 firstChild;
  readonly attribute  Node                 lastChild;
  readonly attribute  Node                 previousSibling;
  readonly attribute  Node                 nextSibling;
  readonly attribute  NamedNodeMap         attributes;
  readonly attribute  Document             ownerDocument;
  Node                      insertBefore(in Node newChild, 
                                         in Node refChild)
                                         raises(DOMException);
  Node                      replaceChild(in Node newChild, 
                                         in Node oldChild)
                                         raises(DOMException);
  Node                      removeChild(in Node oldChild)
                                        raises(DOMException);
  Node                      appendChild(in Node newChild)
                                        raises(DOMException);
  boolean                   hasChildNodes();
  Node                      cloneNode(in boolean deep);

  bytearray                 getDigest(in DOMString digestAlgorithm);
};

4. Discussion

The definition described above can be efficiently implemented. XML Parser for Java has a reference implementation with the source code.


5. References


Contacts

Please forward any comments and suggestions to