The presentation of this document has been augmented to identify changes from a previous version. Three kinds of changes are highlighted: new, added text, changed text, and deleted text.


Architecture of the World Wide Web, First Edition

W3C Editor'sWorking Draft 16 DraftAugust 28 September 2004

This version:
Latest editor's draft:
Previous version:
Latest version:
Ian Jacobs, W3C
Norman Walsh, Sun Microsystems, Inc.
See acknowledgments (§8).


The World Wide Web is an information space of interrelated resources. This information space is the basis of, and is shared by, a number of information systems. In each of these systems, people and software retrieve, create, display, analyze, relate, and reason about resources.

Web architecture defines the information space in terms of identification of resources, representation of resource state, and the protocols that support the interaction between agents and resources in the space. Web architecture is influenced by social requirements and software engineering principles. These lead to design choices for and constraints on the behavior of systems that use the Web, aimed at achieving the desired properties of the shared information space: efficiency, scalability, and the potential for indefinite growth across languages, cultures, and media. Good practice by agents in the system is also important to the success of the system. This document reflects the three bases of Web architecture: identification, interaction, and representation.

Status of this document

This section describes the status of this document at the time of its publication. Other documents may supersede this document. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at

This is NOT the 16 August 2004 Last Call Working Draft of "Architecture of the World Wide Web, First Edition." The Last Call review period ends 17 September 2004, at 23:59 EDT. Please send Last Call review comments on this document before that date to the public W3C TAG mailing list (archive). Last Call Working Draft status is described in section 7.4.2 of the W3C Process Document. To the extent possible, please provide a separate email message for each distinct comment.

The TAG has used the last call comments page to track the status and discussion of comments on the 9 December 2003 Draft. This draft has been informed by a large number of comments made on the on the 9 December 2003 Last Call Working Draft. Because the text has changed substantially, it is not always clear whether or not comments made on the previous draft still apply. The TAG expects a future revision of this document to become a W3C Recommendation.

This document has been developed by W3C's Technical Architecture Group (TAG) (charter). A complete list of changes to this document since the first public Working Draft is available on the Web.

The TAG charter describes a process for issue resolution by the TAG. In accordance with those provisions, the TAG maintains a running issues list. The First Edition of "Architecture of the World Wide Web" does not address every issue that the TAG has accepted since it began work in January 2002. The TAG has selected a subset of issues that the First Edition does address to the satisfaction of the TAG; those issues are identified in the TAG's issues list. The TAG intends to address the remaining (and future) issues after publication of the First Edition as a Recommendation.

This document uses the concepts and terms regarding URIs as defined in draft-fielding-uri-rfc2396bis-06, preferring them to those defined in RFC 2396. The IETF Internet Draft draft-fieldi ng-uri-rfc2396bis-06 is expected to obsolete RFC 2396, which is the current URI standard. The TAG is tracking the evolution of draft-fielding-uri-rfc2396bis-06.

Publication as a Working Draft does not imply endorsement by the W3C Membership. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than "work in progress."

The patent policy for this document is expected to become the 5 February 2004 W3C Patent Policy, pending the Advisory Committee review of the revision to the W3C Technical Architecture Group Charter. Patent disclosures relevant to this specification may be found on the Technical Architecture Group's patent disclosure page. An individual who has actual knowledge of a patent which the individual believes contains Essential Claim(s) with respect to this specification should disclose the information in accordance with section 6 of the W3C Patent Policy.

Table of Contents

List of Principles, Constraints, and Good Practice Notes

The following principles, constraints, and good practice notes are discussed in this document and listed here for convenience. There is also a free-standing summary.

Data Formats
General Architecture Principles

1. Introduction

The World Wide Web (WWW, or simply Web) is an information space in which the items of interest, referred to as resources, are identified by global identifiers called Uniform Resource Identifiers (URI).

Examples such as the following travel scenario are used throughout this document to illustrate typical behavior of Web agents—people — people or software (on behalf of a person, entity, or process) acting on this information space. A user agent acts on behalf of a user. Software agents include servers, proxies, spiders, browsers, and multimedia players.


While planning a trip to Mexico, Nadia reads "Oaxaca weather information: ''" in a glossy travel magazine. Nadia has enough experience with the Web to recognize that "" is a URI and that she is likely to be able to use software to retrieve associated information (in this case, about the weather).weather, as advertised). When Nadia enters the URI into her browser:

  1. The browser performs an information retrieval action in accordance with its configured behavior for resources identified via the "http" URI scheme.
  2. The authority responsible for "" provides information in a response to the retrieval request.
  3. The browser displays the retrieved information, which includes hypertext links to other information. Nadia can follow these hypertext links to retrieve additional information.

This scenario illustrates the three architectural bases of the Web that are discussed in this document:

  1. Identification (§2). URIs are used to identify resources. In this travel scenario, the resource is a periodically updated report on the weather in Oaxaca, and the URI is "".
  2. Interaction (§3). Web agents communicate the information state of a resource using protocols. Protocols define the syntax and semantics of messages exchanged by agents over a network. ByWeb agents typing the URIinformation state of into her browser, orprotocols. By clicking on a hypertext link (§4.4), Nadia tells her browser to request the information state of the resource referenced by the URI in the link. In this example, the browser sends an HTTP GET request to the server at "" and the server sends back a representation of the information state of the resource. In this example, the representation includes XHTML data and metadata such as the Internet media type of the data, "application/xhtml+xml". Note: In this document, the noun "representation" means "octets that encode resource state information". These octets do not necessarily describe the resource, or portray a likeness of the resource, or represent the resource in other senses of the word "represent".
  3. Formats (§4). Representations are built from a non-exclusive set of data formats, used separately or in combination (including XHTML, CSS, PNG, XLink, RDF/XML, SVG, and SMIL animation). In this scenario, the representation is built primarily in XHTML. While interpreting the XHTML representation data, the browser retrieves and displays weather maps identified by URIs within the XHTML. Some of those maps are built in SVG.

The following illustration shows the relationship between identifier, resource, and representation.

A resource (Oaxaca Weather Info) is identified by a particular URI and is represented by pseudo-HTML content

In the remainder of this document, we highlight important architectural points regarding Web identifiers, protocols, and formats. We also discuss some important general architectural principles (§5) in the context of the Web.

1.1. About this Document

This document describes the properties we desire of the Web and the design choices that have been made to achieve them. It promotes re-use of existing standards when suitable, and gives guidance on how to innovate in a manner consistent with Web architecture.

The terms MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY are used in the principles, constraints, and good practice notes in accordance with RFC 2119 [RFC2119]. However, this document does not include conformance provisions for these reasons:

1.1.1. Audience of this Document

This document is intended to inform discussions about issues of Web architecture. The intended audience for this document includes:

  1. Participants in W3C Activities
  2. Other groups and individuals designing technologies to be integrated into the Web
  3. Implementers of W3C specifications
  4. Web content authors and publishers
Readers will benefit from familiarity with the Requests for Comments (RFC) series from the IETF, some of which define pieces of the architecture discussed in this document.

Note: This document does not distinguish in any formal way the terms "language" and "format." Context determines which term is used. The phrase "specification designer" encompasses language, format, and protocol designers.

1.1.2. Scope of this Document

This document presents the general architecture of the Web. Other groups inside and outside W3C also address specialized aspects of Web architecture, including accessibility, internationalization, device independence, and Web Services. The section on Architectural Specifications (§7.1) includes references to these related specifications.

This document strives forstrikes a balance between brevity and precision while including illustrative examples. TAG findings are informational documents that complement the current document by providing more detail about selected topics. This document includes some excerpts from the findings. Since the findings evolve independently, this document also includes references to approved TAG findings. For other TAG issues covered by this document but without an approved finding, references are to entries in the TAG issues list.

Many of the examples in this document that involve human activity suppose the familiar Web interaction model where a person follows a link via a user agent, the user agent retrieves and presents data, the user follows another link, etc. This document does not discuss in any detail other interaction models such as voice browsing (see, for example, [VOICEXML2]). For instance, when a graphical user agent running on a laptop computer or hand-held device encounters an error, the user agent can report errors directly to the user through visual and audio cues, and present the user with options for resolving the errors. On the other hand, when someone is browsing the Web through voice input and audio-only output, stopping the dialog to wait for user input may reduce usability since it is so easy to "lose one's place" when browsing with only audio-output. This document does not discuss how the principles, constraints, and good practices identified here apply in all interaction contexts.

1.1.3. Principles, Constraints, and Good Practice Notes

The important points of this document are categorized as follows:

An architectural principle is a fundamental rule that applies to a large number of situations and variables. Architectural principles include "separation of concerns", "generic interface", "self-descriptive syntax," "visible semantics," "network effect" (Metcalfe's Law), and Amdahl's Law: "The speed of a system is limited by its slowest component."
In the design of the Web, some design choices, like the names of the p and li elements in HTML, the choice of the colon (:) character in URIs, or grouping bits into eight-bit units (octets), are somewhat arbitrary; if paragraph had been chosen instead of p or asterisk (*) instead of colon, the large-scale result would, most likely, have been the same. This document focuses on more fundamental; these are the fundamental design choices: design Design choices that lead to constraints, i.e., restrictions in behavior or interaction within the system. Constraints may be imposed for technical, policy, or other reasons to achieve desired properties of the system, such as accessibility,accessibility and global scope, and non-functional properties, such as relative ease of evolution, re-usability of components, efficiency, and dynamic extensibility.
Good practice
Good practice—by software developers, content authors, site managers, users, and specification designers—increases the value of the Web.

2. Identification

In order to communicate internally, a community agrees (to a reasonable extent) on a set of terms and their meanings. Since its inception, one goal of the Web has been to build a global community in which any party can share information with any other party. To achieve this goal, the Web makes use of a single global identification system: the URI. URIs are a cornerstone of Web architecture, providing identification that is common across thesystem. Web. The global scope of URIs promotes large-scale "network effects": the value of an identifier increases the more it is used consistently (for example, the more it is used in hypertext links (§4.4) [section 4.4] ).

Principle: Global Identifiers

Global naming leads to global network effects.

This principle dates back at least as far as Douglas Engelbart's seminal work on open hypertext systems; see section Every Object Addressable in [Eng90].

2.1. Benefits of URIs

The choice of syntax for global identifiers is somewhat arbitrary; it is their global scope which is important. The Uniform Resource Identifier, [URI], currently being revised) has been successfully deployed since the creation of the Web. There are substantial benefits to participating in the existing network of URIs, including linking, bookmarking, caching, and indexing by search engines, and there are substantial costs to creating a new identification system that has the same properties as URIs.

Good practice: Identify with URIs

To benefit from and increase the value of the World Wide Web, agents should provide URIs as identifiersengines. for resources.

A resource should have an associated URI if another party might reasonably want to create a hypertext link to it, make or refute assertions about it, retrieve or cache a representation of it, include all or part of it by reference into another representation, annotate it, or perform other operations on it. Software developers should expect that sharing URIsa URI across applications will be useful, even if that utility is not initially evident. The TAG finding "URIs, Addressability, and the use of HTTP GET and POST" discusses additional benefits and considerations of URI addressability.

Good practice: Identify with URIs To benefit from and increase the value of the World Wide Web, agents should provide URIs as identifiers for resources. Other resource identification systems (see the section on future directions for identifiers) may expand the Web as we know it today. However, there are substantial costs to creating a new identification system that has the same properties as URIs.

2.2. URI/Resource Relationships

By design a URI identifies one resource. We do not limit the scope of what might be a resource. The term "resource" is used in a general sense for whatever might be identified by a URI. A significant class of resources, information resources, are discussed in Information Resources and Representations (§3.1)[section 3.1].

Constraint: URIs Identify a Single Resource

Assign distinct URIs to distinct resources.

Since the scope of a URI is global, the resource identified by a URI does not depend on the context in which the URI appears (see also the section about indirect indirect identification (§2.2.3)).

[URI] is an agreement about how the Internet community allocates names and associates them with the resources they identify. URIs are divided into schemes and URI Scheme specifications define the protocols by which scheme scheme-specific URI are associated with resources and take on meaning. For example, the HTTP URI scheme (RFC2616) uses DNS so that names such as “” take on meaning by way of HTTP GET response messages from the domain holder (or an agent they delegate to). HTTP GET is not the only way to obtain information about a resource. A third party, for example, might publish a document that purports to define the meaning of a particular URI. TheseWhile other sources of informationcommunications may suggest meanings for such names, but it's a local policy decision whether those suggestions should be heeded, whereas the result obtained through HTTP GET is, by Internet-wide agreement, authoritative.

Just as one might wish to refer to a person by different names (by full name, first name only, sports nickname, romantic nickname, and so forth), Web architecture allows the association of more than one URI with a resource. URIs that identify the same resource are called URI aliases. The section on URI aliases (§2.3.1) discusses some of the potential costs of creating multiple URIs for the same resource.

The following sections address other questions about the relationship between URIs and resources, including:

2.2.1. URI allocation

To avoid URI collision (§2.2.2), it is important to avoid assigning the same URI to different resources. URI ownership

URIOne ownership isway a relation betweenURI acollision URI andthrough delegated aownership. It social entity, such asfor a person or an organisation. URI Ownership givesunique relationship the relevant social entity rights to:

  1. passa on ownership of some or all owned URIfor the to another"http", owner—URI allocation; and
  2. to associate a resource with anfor owned URI—URI assignment.
  3. This

Byrelationship social convention, URI ownership. delegatedIn this from the IANA URIphrase scheme registry [IANASchemes],for itself a socialX" entity, to IANA registered URIentity owns scheme specifications.

SomeURIs URI scheme specifications further delegate ownership to subordinate registries or to other nominated owners, who may further delegate ownership. Forownership may example, the URNdelegated Syntax scheme [RFC2141] delegatesparties, ownership of portions of URNserver manager space to URN Namespace specifications which themselves are registered in an IANA maintainedURI registry ofon a URN Namespace Identifiers.

Other URI schemes or nominated owners retain ownership within the pattern relevantwhereby the scheme specification (or some other specification registered in a subordinate registry)[IANASchemes] such that that specification and the community that maintain it serve in the role of URI owners. For example, the specification consequence of the data URL the scheme [RFC2397] servesheavy as owner ofthe central all dataregistry. A scheme URI; itowner does not further delegate ownership of portionsthe resource identified of data URI space. It associatesexample, each data URI with an resource,the HTTP a representation of which is embeddedrepresentations, with each data URIserver (defined (the representation[RFC2616]) is infactthe software the resource).

URIacting Owners have a responsibility to avoid multiple URI assignments that associaterepresentations for equivalent URI with multiple resources. Thus,URI. The it is also necessary that any approach to URI allocation, which passes ownershipthe resource of individualidentified or organised sets of URI through delegation, ensures that ultimate ownership of a particular URI is vested in a single social entity (which may be aconstraints, specification, a person or an organization).

URIRecall that Owners may organise or deploy infrastruture to create URI aliases. This ensure that representationsmultiple parties of associated resources are available, and where appropriate interaction with the resourceURI is possible through the exchange of representations.interaction.

There are social expectations for responsible representation management (§3.6) [section 3.6] by URI owners, discussed below. Additional social implications of URI ownership are not discussed here. However, the success or failure of these different approaches depends on the extent to which there is consensus in the Internet community on abiding by the defining defining specifications.

See TAG issue siteData-36, which concerns the expropriation of naming authority. Other allocation schemes

Some schemes use techniques other than delegated ownership to avoid overloading. Some schemes have construction rules that work to guarantee uniqueness. Other schemes (such as "news:comp.text.xml") rely on a social process.

2.2.2. URI collision

As discussed above, a URI identifies one resource. Using the same URI to directly identify different resources produces a URI collision. Collision often imposes a cost in communication due to the effort required to resolve ambiguities.

Suppose, for example, that one organization makes use of a URI to refer to the movie The"The Sting,Sting", and another organization uses the same URI to refer to a discussion forum about The"The Sting. Sting." This collision creates confusion about what the URI identifies, undermining the value of the URI. If one wanted to talk about the creation date of the resource identified by the URI, for instance, it would not be clear whether this meant "when the movie created" or "when the discussion forum about the movie was created."

The section below on URI assignment (§2.2.1) [section 2.2.1] examines approaches for establishing the authoritative source of information about what resource a URI identifies.

2.2.3. Indirect Identification

To say that the URI "" identifies both an Internet mailbox and Nadia, the person, introduces a URI collision. However, we can use the URI to indirectly identify Nadia. Identifiers are commonly used in this way.

Listening to a news broadcast, one might hear a report on Britain that begins, "Today, 10 Downing Street announced a series of new economic measures." Generally, "10 Downing Street" identifies the official residence of Britain's Prime Minister. In this context, the news reporter is using it (as English rhetoric allows) to indirectly identify the British government. Similarly, URIs identify resources, but they can also be used in many constructs to indirectly identify arbitrary entities. Globally adopted assignment policies make some URIs appealing as general-purpose identifiers. Local policy establishes what they indirectly identify.

For example, the URI "" identifies an Internet mailbox (as specified by the "mailto" URI scheme). Suppose this that nadia@example.comURI is Nadia's email address. The organizers of a conference attended by Nadia attends might use "" to refer indirectly to her (e.g., using the URI as a database key in their database of conference participants). This doesn’t introduce a URI collision.

2.3. URI Comparisons

URIs that are identical, character-by-character, refer to the same resource. Since Web Architecture allows the association of multiple URIs with a given resource, two URIs that are not character-by-character identical may still refer to the same resource. Different URIs do not necessarily refer to different resources but there is generally a higher computational cost to determine that multiple different URIs refer to the same resource.

To reduce the risk of a false negative (i.e., an incorrect conclusion that two URIs do not refer to the same resource) or a false positive (i.e., an incorrect conclusion that two URIs do refer to the same resource), some specifications specify equivalence tests in addition to character-by-character comparison. For example, for "http" URIs, the authority component (the part after "//" and before the next "/") is defined to be case-insensitive. Thus, the "http" URI specification allows agents to conclude that authority components in two "http" URIs identify the same resource when those strings are character-by-character equivalent or differ only by case. Agents that reach conclusions based on comparisons that are not licensed by the relevant specifications take responsibility for any problems that result; see the section on error handling (§5.3) [section 5.3] for more information about responsible behavior when reaching unlicensed conclusions. Section 6 of [URI] provides more information about comparing URIs and reducing the risk of false negatives and positives.

See the section below on approaches other than string comparison that allow different agents to conclude that two URIs identify the same resource (§2.7.2) [section 2.7.2] .

2.3.1. URI aliases

Although there are benefits (such as naming flexibility) to URI aliases, there are also costs. URI aliases are harmful when they cause bifurcation in the web of related resources. A corollary of Metcalfe's Principle (the "Network Effect") is that the value of a given resource can be measured by the number and value of other resources inthat link to it its network neighborhood,neighborhood of the measured resource). This type of valuation is commonly used that is, the relative value of search results because people tend to create links relating a given topic to those resources that they feel best reflect that topic, and hence the number of inbound references are a reflection of the link to it. the community values a resource.

The problem with aliases is that if half of the neighborhood points to one URI for a given resource, and the other half points to a second, different URI for that same resource, the neighborhood is divided. Not only is the aliased resource undervalued because of this split, the entire neighborhood of resources loses value because of the missing second-order relationships that should have existed among the referring resources by virtue of their references to the aliased resource.

Good practice: Avoiding URI aliases

A URI owner SHOULD NOT associate arbitrarily different URIs with the same resource.

URI consumers also have a role in ensuring URI consistency. For instance, when transcribing a URI, agents should not gratuitously percent-encode characters. The term "character" refers to URI characters as defined in section 2 of [URI]; percent-encoding is discussed in section 2.1 of that specification.

Good practice: Consistent URI usage

An agent that receives a URI SHOULD refer to the associated resource using the same URI, character-by-character.

When a URI alias does become common currency, the URI owner should use protocol techniques such as server-side redirects to relate the two resources. The community benefits when the URI owner supports redirection of an aliased URI to the corresponding "official" URI. For more information on redirection, see section 10.3, Redirection, in RFC2616. See also [CHIPS] for a discussion of some best practices for server administrators.

2.3.2. Representation reuse


Dirk is editing a web page and inserts a link to ""resource identified by labeled ‘Current Oaxaca Weather Forecast’. Nadia spots the error and explains that the resource"". He Dirk has identified isit matters the weather forecast forbookmarks. Nadia explains that it certainly does, 03 August 2004. Theone URI for the current forecast is "", even though that URI andidentifies the one heresource used return the same representations2004," even today because it is the third of August.

URI aliasing only occurs when more than one URI is used to identify the same resource. The fact that different resources sometimes have the same representation does not make the URIs for those resources aliases. This occurs most commonly when a URI owner provides a URI for both the "current version" of something and a URI for the "version at a point in time," as the Oaxaca weather Web site is doing, but it can occur in other ways as well.

The distinguishing characteristic of representation reuse, as opposed to aliasing, is that the underlying resources are different.

2.4. URI Schemes

In the URI "", the "http" that appears before the colon (":") names a URI scheme. Each URI scheme has a specification that explains the scheme specific details of how scheme identifiers are allocated and become associated with a resource. The URI syntax is thus a federated and extensible naming system wherein each scheme's specification may further restrict the syntax and semantics of identifiers within that scheme.

Examples of URIs from various schemes include:

While Web architecture allows the definition of new schemes, introducing a new scheme is costly. Many aspects of URI processing are scheme-dependent, and a large amount of deployed software already processes URIs of well-known schemes. Introducing a new URI scheme requires the development and deployment not only of client software to handle the scheme, but also of ancillary agents such as gateways, proxies, and caches. See [RFC2718] for other considerations and costs related to URI scheme design.

Because of these costs, if a URI scheme exists that meets the needs of an application, designers should use it rather than invent one.

Good practice: Reuse URI schemes

A specification SHOULD reuse an existing URI scheme (rather than create a new one) when it provides the desired properties of identifiers and their relation to resources.

Consider our travel scenario: should the agent providing information about the weather in Oaxaca register a new URI scheme "weather" for the identification of resources related to the weather? They might then publish URIs such as "weather://". When a software agent dereferences such a URI, if what really happens is that HTTP GET is invoked to retrieve a representation of the resource, then an "http" URI would have sufficed.

If the motivation behind registering a new scheme is to allow a software agent to launch a particular application when retrieving a representation, such dispatching can be accomplished at lower expense via Internet media types. When designing a new data format, the appropriate mechanism to promote its deployment on the Web is the Internet media type. Media types also provide a means for building new information space applications (§4.6) [section 4.6] , described below.

Note that even if an agent cannot process representation data in an unknown format, it can at least retrieve it. The data may contain enough information to allow a user or user agent to make some use of it. When an agent does not handle a new URI scheme, it cannot retrieve a representation.

2.4.1. URI Scheme Registration

The Internet Assigned Numbers Authority (IANA) maintains a registry [IANASchemes] of mappings between URI scheme names and scheme specifications. For instance, the IANA registry indicates that the "http" scheme is defined in [RFC2616]. The process for registering a new URI scheme is defined in [RFC2717].

Unregistered URI schemes SHOULD NOT be used for a number of reasons:

  • There is no generally accepted way to locate the scheme specification.
  • Someone else may be using the scheme for other purposes.
  • One should not expect that general-purpose software will do anything useful with URIs of this scheme beyond URI comparison.comparison; the network effect is lost.

Note: Some URI scheme specifications (such as the "ftp" URI scheme specification) use the term "designate" where the current document uses "identify."

2.5. URI Opacity

It is tempting to guess the nature of a resource by inspection of a URI that identifies it. However, the Web is designed so that agents communicate resource information state through representations, not identifiers. In general, one cannot determine the Internet media type of representations of a resource by inspecting a URI for that resource. For example, the ".html" at the end of "" provides no guarantee that representations of the identified resource will be served with the Internet media type "text/html". The HTTP protocol does not constrain the Internet media type based on the path component of the URI; the URI owner is free to configure the server to return a representation using PNG or any other data format.

Resource state may evolve over time. Requiring a URI owner to publish a new URI for each change in resource state would lead to a significant number of broken references. For robustness, Web architecture promotes independence between an identifier and the state of the identified resource.

Good practice: URI opacity

Agents making use of URIs SHOULD NOT attempt to infer properties of the referenced resource except as specified by relevant specifications.

The example URI used in the travel scenario ("") suggests that the identified resource has something to do with the weather in Oaxaca. A site reporting the weather in Oaxaca could just as easily be identified by the URI "". And the URI "" might identify the resource "my photo album."

On the other hand, the URI "" indicates that the URI refers to a mailbox. The "mailto" URI scheme specification authorizes agents to infer that URIs of this form identify Internet mailboxes.

Some URI assignment authorities document and publish their URI assignment policies. For more information about URI opacity, see TAG issue metaDataInURI-31.

2.6. Fragment Identifiers


When browsingnavigating within the XHTML document that Nadia receives as a representation of the resource identified by "", she finds that the URI "" refers to the part of the representation that conveys information about thetomorrow's weather weekend outlook. This URI includes the fragment identifier "weekend" (the string after the "#").

The fragment identifier component of a URI allows indirect identification of a secondary resource by reference to a primary resource and additional identifying information. The secondary resource may be some portion or subset of the primary resource, some view on representations of the primary resource, or some other resource defined or described by those representations. The terms "primary resource" and "secondary resource" are defined in section 3.5 of [URI].

The interpretation of fragment identifiers is discussed in the section on media types and fragment identifier semantics (§3.3.1).

See TAG issue abstractComponentRefs-37, which concerns the use of fragment identifiers with namespace names to identify abstract components.

2.7. Future Directions for Identifiers

There remain open questions regarding identifiers on the Web. The following sections identify a few areas of future work in the Web community.

2.7.1. Internationalized identifiers

The integration of internationalized identifiers (i.e., composed of characters beyond those allowed by [URI]) into the Web architecture is an important and open issue. See TAG issue IRIEverywhere-27 for discussion about work going on in this area.

2.7.2. Assertion that two URIs identify the same resource

Emerging Semantic Web technologies, including the "Web Ontology Language (OWL)" [OWL10], define RDF properties such as sameAs to assert that two URIs identify the same resource or inverseFunctionalProperty to imply it.

3. Interaction

Communication between agents over a network about resources involves URIs, messages, and data. The Web's protocols (including HTTP, FTP, SOAP, NNTP, and SMTP) are based on the exchange of messages. A message may include data as well as metadata about a resource (such as the "Alternates" and "Vary" HTTP headers), the message data, and the message itself (such as the "Transfer-encoding" HTTP header). A message may even include metadata about the message metadata (for message-integrity checks, for instance). Two important classes of message are those that request a representation of an Information Resource (§3.1), and those that return the result of such a request.


Nadia follows a hypertext link labeled "satellite image" expecting to retrieve a satellite photo of the Oaxaca region. The link to the satellite image is an XHTML link encoded as <a href="">satellite image</a>. Nadia's browser analyzes the URI and determines that its scheme is "http". The browser configuration determines how it locates the identified information, which might be via a cache of prior retrieval actions, by contacting an intermediary (such as a proxy server), or by direct access to the server identified by a portion of the URI. In this example, the browser opens a network connection to port 80 on the server at "" and sends a "GET" message as specified by the HTTP protocol, requesting a representation of the resource.

The server sends a response message to the browser, once again according to the HTTP protocol. The message consists of several headers and a JPEG image. The browser reads the headers, learns from the "Content-Type" field that the Internet media type of the representation is "image/jpeg", reads the sequence of octets that make up the representation data, and renders the image.

This section describes the architectural principles and constraints regarding interactions between agents, including such topics as network protocols and interaction styles, along with interactions between the Web as a system and the people that make use of it. The fact that the Web is a highly distributed system affects architectural constraints and assumptions about interactions.

3.1. Information Resources and Representations

The term Information Resource refers to resources that convey information. Any resource that has a representation is an information resource. A representation consists logically of two parts: data (expressed in one or more formats (§4) used separately or in combination) and metadata (such as the Internet media type (§3.3) of the data).

The Information Resource provides the foundation for the familiar hypertext Web, where agents use representations to modify as well as retrieve information state. Much of this document describes architecture specific to Information Resources. For instance, the techniques of caching and content negotiation, and the social processes of publishing, apply to Information Resources.

3.2. Using a URI to Access a Resource

Agents may use a URI to access the referenced resource; this is called dereferencing the URI. Access may take many forms, including retrieving a representation of the resource (for instance, by using HTTP GET or HEAD), adding or modifying a representation of the resource (for instance, by using HTTP POST or PUT, which in some cases may change the actual state of the resource if the submitted representations are interpreted as instructions to that end), and deleting some or all representations of the resource (for instance, by using HTTP DELETE, which in some cases may result in the deletion of the resource itself).

There may be more than one way to access a resource for a given URI; application context determines which access method an agent uses. For instance, a browser might use HTTP GET to retrieve a representation of a resource, whereas a hypertext link checker might use HTTP HEAD on the same URI simply to establish whether a representation is available. Some URI schemes set expectations about available access methods, others (such as the URN scheme [RFC 2141]) do not. Section 1.2.2 of [URI] discusses the separation of identification and interaction in more detail. For more information about relationships between multiple access methods and URI addressability, see the TAG finding "URIs, Addressability, and the use of HTTP GET and POST".

Although many URI schemes (§2.4) are named after protocols, this does not imply that use of such a URI will necessarily result in access to the resource via the named protocol. Even when an agent uses a URI to retrieve a representation, that access might be through gateways, proxies, caches, and name resolution services that are independent of the protocol associated with the scheme name.

3.2.1. Details of retrieving a representation

Dereferencing a URI generally involves a succession of steps as described in multiple specifications and implemented by the agent. The following example illustrates the series of specifications that govern the processinvolved when a user instructs a user agent to follow a hypertext link (§4.4) that is part of an SVG document. In this example, the URI is "" and the application context calls for the user agent to retrieve and render a representation of the identified resource.

  1. Since the URI is part of a hypertext link in an SVG document, the first relevant specification is the SVG 1.1 Recommendation [SVG11]. Section 17.1 of this specification imports the link semantics defined in XLink 1.0 [XLink10]: "The remote resource (the destination for the link) is defined by a URI specified by the XLink href attribute on the 'a' element." The SVG specification goes on to state that interpretation of an a element involves retrieving a representation of a resource, identified by the href attribute in the XLink namespace: "By activating these links (by clicking with the mouse, through keyboard input, voice commands, etc.), users may visit these resources."
  2. The XLink 1.0 [XLink10] specification, which defines the href attribute in section 5.4, states that "The value of the href attribute must be a URI reference as defined in [IETF RFC 2396], or must result in a URI reference after the escaping procedure described below is applied."
  3. The URI specification [URI] states that "Each URI begins with a scheme name that refers to a specification for assigning identifiers within that scheme." The URI scheme name in this example is "http".
  4. [IANASchemes] states that the "http" scheme is defined by the HTTP/1.1 specification (RFC 2616 [RFC2616], section 3.2.2).
  5. In this SVG context, the agent constructs an HTTP GET request (per section 9.3 of [RFC2616]) to retrieve the representation.
  6. Section 6 of [RFC2616] defines how the server constructs a corresponding response message, including the 'Content-Type' field.
  7. Section 1.4 of [RFC2616] states "HTTP communication usually takes place over TCP/IP connections." This example does not address that step in the process, or other steps such as Domain Name System (DNS) resolution.
  8. The agent interprets the returned representation according to the data format specification that corresponds to the representation's Internet Media Type (§3.3) (the value of the HTTP 'Content-Type') in the relevant IANA registry [MEDIATYPEREG].

Precisely which representation(s) are retrieved depends on a number of factors, including:

  1. Whether the URI owner makes available any representations at all;
  2. Whether the agent making the request has access privileges for those representations (see the section on linking and access control (§3.6.2));
  3. If the URI owner has provided more than one representation (in different formats such as HTML, PNG, or RDF; in different languages such as English and Spanish; or transformed dynamically according to the hardware or software capabilities of the recipient), the resulting representation may depend on negotiation between the user agent and server.
  4. The time of the request; the worldinformation changes over time, and so representations representations of resourcesthat information are also likely to change over time.change.

AssumingNote also that a representation has been successfully retrieved, the expressive power of the representation’sa format will affect how how precisely the representation provider communicates resource state. If the representation communicates the stateuse of the resource inaccurately,language this inaccuracy or ambiguity may lead to URIambiguity collision (§2.2.2). Effortswhat like the Semantic Web areresource seeking to provide a framework for accurately communicating the semantics of awhich resource in a machine readable way. Machine readableturn semantics may alleviate someURI of the ambiguity associated with naturalcollision. language descriptions of resources.

3.3. Internet Media Type

The Internet media type [RFC2046]) of a representation determines which data format specification(s) provide the authoritative interpretation of the representation data (including fragment identifier syntax and semantics (§3.3.1), if any). The IANA registry [MEDIATYPEREG] maps media types to data formats (§4). The TAG finding "Internet media type registration, consistency of use" provides more information to W3C groups about media type registration.

The Internet media type mechanism does have its limitations. For instance, media type strings do not support versioning (§4.2.1) or other parameters. See TAG issue mediaTypeManagement-45, which concern the appropriate level of granularity of the media type mechanism.

3.3.1. Media types and fragment identifier semantics


In one of his XHTML pages, Dirk creates a hypertext link to an image that Nadia has published on the Web. He creates a hypertext link with <a href="">Nadia's hat</a>. Emma views Dirk's XHTML page in her Web browser and follows the link. The HTML implementation in her browser removes the fragment from the URI and requests the image "" Nadia serves an SVG representation of the image (with Internet media type "image/svg+xml"). Emma's Web browser starts up an SVG implementation to view the image. It passes it the original URI including the fragment, "" to this implementation, causing a view of the hat to be displayed rather than the complete image.

Note that the HTML implementation in Emma's browser did not need to understand the syntax or semantics of the SVG fragment (nor does the SVG implementation have to understand HTML, WebCGM, RDF ... fragment syntax or semantics; it merely had to recognize the # delimiter from the URI syntax [URI] and remove the fragment when requesting the resource). This orthogonality (§5.1) is an important feature of Web architecture; it is what enabled Emma's browser to provide a useful service without requiring an upgrade

The semantics of a fragment identifier are defined by the set of representations that might result from a retrieval action on the primary resource. The fragment's format and resolution are therefore dependent on the media type [RFC2046] of a potentially retrieved representation, even though such a retrieval is only performed if the URI is dereferenced. If no such representation exists, then the semantics of the fragment are considered unknown and, effectively, unconstrained. Fragment identifier semantics are orthogonal to URI schemes and thus cannot be redefined by URI scheme specifications.

Interpretation of the fragment identifier is performed solely by the agent that dereferences a URI; the fragment identifier is not passed to other systems during the process of retrieval. This means that some intermediaries in Web architecture (such as proxies) have no interaction with fragment identifiers and that redirection (in HTTP [RFC2616], for example) does not account for fragments.

As with any URI, use of a fragment identifier component does not imply that a retrieval action will take place. A URI with a fragment identifier may be used to refer to the secondary resource without any implication that the primary resource is accessible or will ever be accessed. One may compare URIs with fragment identifiers without a retrieval action. Parties that draw conclusions about the interpretation of a fragment identifier based solely on a syntactic analysis of all or part of a URI do so at their own risk; such interpretations are not authoritative because they are not licensed by any specification (specifically [URI]).

Please note the following about primary and secondary resources:

  1. A resource may be both a primary and secondary resource since more than one URI may identify the resource.
  2. One cannot carry out an HTTP POST operation using a URI that identifies a secondary resource. (The semantics of HTTP GET provide for client-side interpretation of the secondary resource, but those semantics do not apply to POST, PUT, or any of the other HTTP operations)

3.3.2. Fragment identifiers and content negotiation

Content negotiation refers to the practice of making available multiple representations via the same URI. Negotiation between the requesting agent and the server determines which representation is served (usually with the goal of serving the "best" representation a receiving agent can process). HTTP is an example of a protocol that enables representation providers to use content negotiation.

Individual data formats may define their own rulesrestrictions on, foror structure use of the fragment identifier syntax for specifying different types of subsets, views, or external references that are identifiable as secondary resources by that media type. Therefore, representation providers must manage content negotiation carefully when used with a URI that contains a fragment identifier. Consider an example where the owner of the URI "" uses content negotiation to serve two representations of the identified resource. Three situations can arise:

  1. The interpretation of "zicatela" is defined consistently by both data format specifications. The representation provider decides when definitions of fragment identifier semantics are are sufficiently consistent.
  2. The interpretation of "zicatela" is defined inconsistently by the data format specifications.
  3. The interpretation of "zicatela" is defined in one data format specification but not the other.

The first situation—consistent semantics—poses no problem.

The second case is a server management error: representation providers must not use content negotiation to serve representation formats that have inconsistent fragment identifier semantics. This situation also leads to URI collision (§2.2.2).

The third case is not a server management error. It is a means by which the Web can grow. Because the Web is a distributed system in which formats and agents are deployed in a non-uniform manner, Web architecture does not constrain authors to only use "lowest common denominator" formats. Content authors may take advantage of new data formats while still ensuring reasonable backward-compatibility for agents that do not yet implement them.

In case three, behavior by the receiving agent should vary depending on whether the negotiated format defines fragment identifier semantics. When a received data format does not define fragment identifier semantics, the agent should not perform silent error recovery unless the user has given consent; see [CUAP] for additional suggested agent behavior in this case.

See related TAG issue RDFinXHTML-35.

3.4. Inconsistencies between Representation Data and Metadata

Successful communication between two parties depends on a reasonably shared understanding of the semantics of exchanged messages, both data and metadata. At times, there may be inconsistencies between a message sender's data and metadata. For instance, examples that have been observed in practice of inconsistencies between representation data and metadata include:

On the other hand, there is no inconsistency in serving HTML content with the media type "text/plain", for example, as this combination is licensed by specification. Receiving agents should detect protocol inconsistencies and perform proper error recovery.

Principle: Data-metadata inconsistency

Agents MUST NOT ignore message metadata without the consent of the user.

Thus, for example, if the parties responsible for "" mistakenly label the satellite photo of Oaxaca as "image/gif" instead of "image/jpeg", and if Nadia's browser detects a problem, Nadia's browser must not ignore the problem (e.g., by simply rendering the JPEG image) without Nadia's consent. Nadia's browser can notify Nadia of the problem or notify Nadia and take corrective action.

Furthermore, representation providers can help reduce the risk of inconsistencies through careful assignment of representation metadata (especially that which applies across representations). The section on media types for XML (§4.5.7) presents an example of reducing the risk of error by providing no metadata about character encoding when serving XML.

The accuracy of metadata relies on the server administrators, the authors of representations, and the software that they use. Practically, the capabilities of the tools and the social relationships may be the limiting factors.

TheIn all cases the accuracy of these and other metadata fields is just as important for dynamic Web resources, where a little bit of thought and programming can often ensure correct metadata for a huge number of resources.

Often there is a separation of control between the users who create representations of resources and the server managers who maintain the Web site software. Given that it is generally the Web site software that provides the metadata associated with a resource, it follows that coordination between the server managers and content creators is required.

Good practice: Metadata association

Server managers SHOULD allow representation creators to control the metadata associated with their representations.

In particular, content creators need to be able to control the content type (for extensibility) and the character encoding (for proper internationalization).

The TAG finding "Authoritative Metadata" discusses in more detail how to handle this type of inconsistency and how server configuration can be used to avoid it.

3.5. Safe Interactions

Nadia's retrieval of weather information (an example of a read-only query or lookup) qualifies as a "safe" interaction; a safe interaction is one where the agent does not incur any obligation beyond the interaction. An agent may incur an obligation through other means (such as by signing a contract). If an agent does not have an obligation before a safe interaction, it does not have that obligation afterwards.

Other Web interactions resemble orders more than queries. These unsafe interactions may cause a change to the state of a resource and the user may be held responsible for the consequences of these interactions. Unsafe interactions include subscribing to a newsletter, posting to a list, or modifying a database. Note: In this context, the word "unsafe" does not necessarily mean "dangerous"; the term "safe" is used in section 9.1.1 of [RFC2616] and "unsafe" is the natural opposite.


Nadia decides to book a vacation to Oaxaca at "" She enters data into a series of online forms and is ultimately asked for credit card information to purchase the airline tickets. She provides this information in another form. When she presses the "Purchase" button, her browser opens another network connection to the server at "" and sends a message composed of form data using the POST method. This is an unsafe interaction; Nadia wishes to change the state of the system by exchanging money for airline tickets.

The server reads the POST request, and after performing the booking transaction returns a message to Nadia's browser that contains a representation of the results of Nadia's request. The representation data is in XHTML so that it can be saved or printed out for Nadia's records.

Note that neither the data transmitted with the POST nor the data received in the response necessarily correspond to any resource identified by a URI.

Safe interactions are important because these are interactions where users can browse with confidence and where agents (including search engines and browsers that pre-cache data for the user) can follow hypertext links safely. Users (or agents acting on their behalf) do not commit themselves to anything by querying a resource or following a hypertext link.

Principle: Safe retrieval

Agents do not incur obligations by retrieving a representation.

For instance, it is incorrect to publish a URI that, when followed as part of a hypertext link, subscribes a user to a mailing list. Remember that search engines may follow such hypertext links.

The fact that URI retrieval is safe does not imply that all safe interactions must be done through URI retrieval. At times, there may be good reasons (such as confidentiality requirements or practical limits on URI length) to conduct an otherwise safe operation using a mechanism generally reserved for unsafe operations (e.g., HTTP POST).

For more information about safe and unsafe operations using HTTP GET and POST, and handling security concerns around the use of HTTP GET, see the TAG finding "URIs, Addressability, and the use of HTTP GET and POST".

3.5.1. Unsafe interactions and accountability


Nadia pays for her airline tickets online (through a POST interaction as described above). She receives a Web page with confirmation information and wishes to bookmark it so that she can refer to it when she calculates her expenses. Although Nadia can print out the results, or save them to a file, she would also like to bookmark them.

Transaction requests and results are valuable resources, and like all valuable resources, it is useful to be able to refer to them with a persistent URI (§3.6.1). However, in practice, Nadia cannot bookmark her commitment to pay (expressed via the POST request) or the airline company's acknowledgment and commitment to provide her with a flight (expressed via the response to the POST).

There are ways to improve the situation. For transaction requests, user agents can provide an interface for managing transactions where the user agent has incurred an obligation on behalf of the user. For transaction results, HTTP allows representation providers to associate a URI with the results of an HTTP POST request using the "Content-Location" header (described in section 14.14 of [RFC2616]).

3.6. Representation Management


Since Nadia finds the Oaxaca weather site useful, she emails a review to her friend Dirk recommending that he check out ''. Dirk clicks on the resulting hypertext link in the email he receives and is frustrated by a 404 (not found). Dirk tries again the next day and receives a representation with "news" that is two-weeks old. He tries one more time the next day only to receive a representation that claims that the weather in Oaxaca is sunny, even though his friends in Oaxaca tell him by phone that it in fact it is raining.raining (and he trusts them more than he trusts the Web site in question). Dirk and Nadia conclude that the URI owners are unreliable or unpredictable. Although the URI owner has chosen the Web as a communication medium, the owner hashave lost two customers due to ineffective resource management.

A URI owner may supply zero or more authoritative representations of the resource identified by that URI. There is a benefit to the community in providing representations.

Good practice: Available representation

A URI owner SHOULD provide representations of the resource itidentified identifies

For example, owners of XML namespace URIs should use them to identify a namespace document (§4.5.4).

Just because representations are available does not mean that it's always desirablecorollary to retrieve them. In fact, in some cases the opposite isgood true. Dereferencingpractice a URI has a (potentially significant) cost in computing and bandwidth resources, may have security implications, and may impose significant latency on the dereferencing application. Dereferencing URIs should be avoided except when necessary.application developers.

Principle: Reference does not imply dereference

An application developer or specification author SHOULD NOT require networked retrieval to representations each time they are referenced.

Dereferencing a URI has a cost, potentially a significant cost, perhaps in terms of security, network latency, or other factors. Attempting to retrieve representations of resources when such retrieval is not necessary should be avoided.

The following sections discuss some aspects of representation management, including promoting URI persistence (§3.6.1), managing access to resources (§3.6.2), and supporting navigation.

3.6.1. URI persistence

As is the case with many human interactions, confidence in interactions via the Web depends on stability and predictability. For an Information Resource (§3.1), persistence generally depends directly on the consistency of information conveyed by a series of representations. The representation provider decides when representations are sufficiently consistent (although that determination generally takes user expectations into account).

Although persistence in this case is observable as a result of representation retrieval, the term URI persistence is used to describe the desirable property that, once associated with a resource, a URI should continue indefinitely to refer to that resource.

Good practice: Consistent representation

A URI owner SHOULD provide representations of the identified resource consistently and predictably.

URI persistence is a matter of policy and commitment on the part of the URI owner. The choice of a particular URI scheme provides no guarantee that those URIs will be persistent or that they will not be persistent.

HTTP [RFC2616] has been designed to help manage URI persistence. For example, HTTP redirection (using the 3xx response codes) permits servers to tell an agent that further action needs to be taken by the agent in order to fulfill the request (for example, a new URI is associated with the resource).

In addition, content negotiation also promotes consistency, as a site manager is not required to define new URIs when adding support for a new format specification. Protocols that do not support content negotiation (such as FTP) require a new identifier when a new data format is introduced. Improper use of content negotiation can lead to inconsistent representations.

For more discussion about URI persistence, see [Cool].

3.6.2. Linking and access control

It is reasonable to limit access to a resource (for commercial or security reasons, for example), but merely identifying the resource is like referring to a book by title. In exceptional circumstances, people may have agreed to keep titles or URIs confidential, otherwise they are free to exchange them.

As an analogy: The owners of a building might have a policy that the public may only enter the building via the main front door, and only during business hours. People who work in the building and who make deliveries to it might use other doors as appropriate. Such a policy would be enforced by a combination of security personnel and mechanical devices such as locks and pass-cards. One would not enforce this policy by hiding some of the building entrances, nor by requesting legislation requiring the use of the front door and forbidding anyone to reveal the fact that there are other doors to the building.


Nadia sends to Dirk the URI of the current article she is reading. With his browser, Dirk follows the hypertext link and is asked to enter his subscriber username and password. Since Dirk is also a subscriber to services provided by "," he can access the same information as Nadia. Thus, the authority for "" can limit access to authorized parties and still provide the benefits of URIs.

The Web provides several mechanisms to control access to resources; these mechanisms do not rely on hiding or suppressing URIs for those resources. For more information, see the TAG finding "'Deep Linking' in the World Wide Web".

3.7. Supporting Navigation

It is a strength of Web Architecture that links can be made and shared; once one user has found an interesting part of the Web, they can share this experience just by republishing a URI.


Nadia and Dirk want to visit the Museum of Weather Forecasting in Oaxaca. Nadia goes to "", locates the museum, and mails the URI ";lon=-96.716;scale=6" to Dirk. Dirk goes to "", locates the museum, and mails the URI ";userID=Dirk" to Nadia. Dirk reads Nadia's email and is able to follow the link to the map. Nadia reads Dirk's email, follows the link, and receives an error message 'No such session/user'. Nadia has to start again from "" and find the museum location once more.

For resources that are generated on demand, machine generation of URIs is common. For resources that might usefully be bookmarked for later perusal, or shared with others, server managers should avoid needlessly restricting the reusability of such URIs. If the intention is to restrict information to a particular user, as might be the case in a home banking application for example, designers should use appropriate access control (§3.6.2) mechanisms.

Interactions conducted with HTTP POST (where HTTP GET could have been used) also limit navigation possibilities. The user cannot create a bookmark or share the URI with others since the URI seen by the user does not change as the user moves from page to page.

3.8. Future Directions for Interaction

There remain open questions regarding Web interactions. The TAG expects future versions of this document to address in more detail the relationship between the architecture described herein, Web Services, peer-to-peer systems, instant messaging systems (such as [XMPP]), streaming audio (such as RTSP [RFC2326]), and voice-over-IP (such as SIP [RFC3261]).

4. Data Formats

Data formats (for example, XHTML, RDF/XML, SMIL, XLink, CSS, and and PNG) are agreementsspecifies on the correct interpretation of representation data. The first data format used on the Web was HTML. Since then, data formats have grown in number. Web architecture does not constrain which data formats content providers can use. This flexibility is important because there is constant evolution in applications, resulting in new data formats and refinements of existing formats. Although Web architecture allows for the deployment of new data formats, the creation and deployment of new formats (and agents able to handle them) is expensive. Thus, before inventing a new data format (or "meta" format such as XML), designers should carefully consider re-using one that is already available.

For a data format to be usefully interoperable between two parties, the parties must agree (to a reasonable extent) about its syntax and semantics. Shared understanding of a data format promotes interoperability but does not imply constraints on usage; for instance, a data sender of data cannot count on being able to constrain the behavior of a data receiver.

Below we describe some characteristics of a data format that facilitate integration into Web architecture. This document does not address generally beneficial characteristics of a specification such as readability, simplicity, attention to programmer goals, attention to user needs, accessibility, nor internationalization. The section on architectural specifications (§7.1) includes references to additional format specification guidelines.

4.1. Binary and Textual Data Formats

Binary data formats are those in which portions of the data are encoded for direct use by computer processors, for example 32 bit little-endian two's-complement and 64 bit IEEE double-precision floating-point. The portions of data so represented include numeric values, pointers, and compressed data of all sorts.

A textual data format is one in which the data is specified in a defined encoding as a sequence of characters. HTML, Internet e-mail, and all XML-based formats (§4.5) are textual. Increasingly, internationalized textual data formats refer to the Unicode repertoire [UNICODE] for character definitions.

IfText (i.e., a data format is textual, as defined in this section, that does nota imply that it should be served with a media type beginning with with "text/". Although XML-based formats are textual, many XML-based formats do not consist primarily of phrases in natural language. See the section on media types for XML (§4.5.7) for issues that arise when "text/" is used in conjunction with an XML-based format.

In principle, all data can be represented using textual formats. In practice, some types of content (e.g., audio and video) are generally represented using binary formats.

The trade-offs between binary and textual data formats are complex and application-dependent. Binary formats can be substantially more compact, particularly for complex pointer-rich data structures. Also, they can be consumed more rapidly by agents in those cases where they can be loaded into memory and used with little or no conversion.

Textual formats are usually more portable and interoperable. Textual formats also have the considerable advantage that they can be directly read by human beings (and understood, given sufficient documentation). This can simplify the tasks of creating and maintaining software, and allow the direct intervention of humans in the processing chain without recourse to tools more complex than the ubiquitous text editor. Finally, it simplifies the necessary human task of learning about new data formats; this is called the "view source" effect.

It is important to emphasize that intuition as to such matters as data size and processing speed is not a reliable guide in data format design; quantitative studies are essential to a correct understanding of the trade-offs. Therefore, designers of a data format specification should make a considered choice between binary and textual format design.

See TAG issue binaryXML-30.

4.2. Versioning and Extensibility

In a perfect world, language designers would invent languages that perfectly met the requirements presented to them, the requirements would be a perfect model of the world, they would never change over time, and all implementations would be perfectly interoperable because the specifications would have no variability.

In the real world, language designers imperfectly address the requirements as they interpret them, the requirements inaccurately model the world, conflicting requirements are presented, and they change over time. As a result, designers negotiate with users, make compromises, and often introduce extensibility mechanisms so that it’s possible to work around problems in the short term. In the long term, they produce multiple versions of their languages, as the problem, and they’re understanding of the problem, evolves. The resulting variability in specifications languages, and implementations introduces interoperability costs.

Extensibility and versioning are strategies to help manage the natural evolution of information on the Web and technologies used to represent that information. For more information about how these strategies introduce variability and how that variability impacts interoperability, see Variability in Specifications.

See TAG issue XMLVersioning-41, which concerns good practices for designing extensible XML languages and for handling versioning. See also "Web Architecture: Extensible Languages" [EXTLANG].

4.2.1. Versioning

There is typically a (long) transition period during which multiple versions of a format, protocol, or agent are simultaneously in use.

Good practice: Version information

A data format specification SHOULD provide for version information.

4.2.2. Versioning and XML namespace policy


Nadia and Dirk are designing an XML data format to encode data about the film industry. They provide for extensibility by using XML namespaces and creating a schema that allows the inclusion, in certain places, of elements from any namespace. When they revise their format, Nadia proposes a new optional lang attribute on the film element. Dirk feels that such a change requires them to assign a new namespace name, which might require changes to deployed software. Nadia explains to Dirk that their choice of extensibility strategy in conjunction with their namespace policy allows certain changes that do not affect conformance of existing content and software, and thus no change to the namespace identifier is required. They chose this policy to help them meet their goals of reducing the cost of change.

Dirk and Nadia have chosen a particular namespace change policy that allows them to avoid changing the namespace name whenever they make changes that do not affect conformance of deployed content and software. They might have chosen a different policy, for example that any new element or attribute has to belong to a namespace other than the original one. Whatever the chosen policy, it should set clear expectations for users of the format.

InFor almost all general, changing the namespace name of an element completely changes the element name. If "a" and "b" are bound to two different URIs, a:element and b:element are as distinct as a:eieio and a:xyzzy. a:xyzzy. Practically speaking, this means that deployed applications will have to be upgraded in order to recognize the new language; the cost of this upgrade may be very high.

It follows that there are significant tradeoffs to be considered when deciding on a namespace change policy. If a vocabulary has no extensibility points (that is, if it does not allow elements or attributes from foreign namespaces or have a mechanism for dealing with unrecognized names from the same namespace), it may be absolutely necessary to change the namespace name. Languages that allow some form of extensibility without requiring a change to the namespace name are more likely to evolve gracefully.

Good practice: Namespace policy

An XMLA format specification SHOULD include information about change policies for XML namespaces.

As an example of a change policy designed to reflect the variable stability of a namespace, consider the W3C namespace policy for documents on the W3C Recommendation track. The policy sets expectations that the Working Group responsible for the namespace may modify it in any way until a certain point in the process ("Candidate Recommendation") at which point W3C constrains the set of possible changes to the namespace in order to promote stable implementations.

Note that since namespace names are URIs, the owner of a namespace URI has the authority to decide the namespace change policy.

4.2.3. Extensibility

Requirements change over time. Successful technologies are adopted and adapted by new users. Designers can facilitate the transition process by making careful choices about extensibility during the design of a language or protocol specification.

Good practice: Extensibility mechanisms

A specification SHOULD provide mechanisms that allow any party to create extensions.

Extensibility introduces variability which has an impact on interoperability. However, languages which have no extensibilityextensions mechanisms will be extended in ad hoc ways that impact interoperabilitydo as well. One key criteria of the mechanisms provided by language designers is that they allow the extended languages to remain conformant, increasing the likelihood of interoperability.

Good practice: Extensibility conformance

Extensibility mechanismsnot MUST NOT interfere with conformance to the original specification.

Application needs determine the most appropriate extension strategy for a specification. For example, applications designed to operate in closed environments may allow specification designers to define a versioning strategy that would be impractical at the scale of the Web.

Good practice: Unknown extensions

A specification SHOULD specify agent behavior in the face of unrecognized extensions.

Two strategies have emerged as being particularly useful:

  1. "Must ignore": The agent ignores any content it does not recognize.
  2. "Must understand": The agent treats unrecognized markup as an error condition.

A powerful design approach is for the language to allow either form of extension, but to distinguish explicitly between them in the syntax.

Additional strategies include prompting the user for more input and automatically retrieving data from available hypertext links. More complex strategies are also possible, including mixing strategies. For instance, a language can include mechanisms for overriding standard behavior. Thus, a data format can specify "must ignore" semantics but also allow for extensions that override that semantics in light of application needs (for instance, with "must understand" semantics for a particular extension).

Extensibility is not free. Providing hooks for extensibility is one of many requirements to be factored into the costs of language design. Experience suggests that the long term benefits of extensibility generally outweigh the costs.

4.2.4. Composition of data formats

Many modern data format include mechanisms for composition. For example:

  • It is possible to embed text comments in some image formats, such as JPEG/JFIF. Although these comments are embedded in the containing data, they arehave little not intended to affect the display of the image.
  • There are container formats such as SOAP which fully expect content to be composed from multiple namespaces but which provide an overall semantic relationship of message envelope and payload.
  • The semantics of combining RDF documents containing multiple vocabularies are well-defined.

In principle, theseThese relationships can be mixed and nested arbitrarily. In arbitrarily. A SOAP message, for example,message can contain an SVG image that contains an RDF comment which refers to a vocabulary of terms for describing the image.

Note however, that for general XML there is no semantic model that defines the interactions within XML documents with elements and/or attributes from a variety of namespaces. Each application must define how namespaces interact and what effect the namespace of an element has on the element's ancestors, siblings, and descendants.

See TAG issues mixedUIXMLNamespace-33 (concerning the meaning of a document composed of content in multiple namespaces), xmlFunctions-34 (concerning one approach for managing XML transformation and composability), and RDFinXHTML-35 (concerning the interpretation of RDF when embedded in an XHTML document).

4.3. Separation of Content, Presentation, and Interaction

The Web is a heterogeneous environment where a wide variety of agents provide access to content to users with a wide variety of capabilities. It is good practice for authors to create content that can reach the widest possible audience, including users with graphical desktop computers, hand-held devices and mobile phones, users with disabilities who may require speech synthesizers, and devices not yet imagined. Furthermore, authors cannot predict in some cases how an agent will display or process their content. Experience shows that the separation of content, presentation, and interaction promotes the reuse and device-independence of content; this follows from the principle of orthogonal specifications (§5.1).

This separation also facilitates reuse of authored source content across multiple delivery contexts. Sometimes, functional user experiences suited to any delivery context can be generated by using an adaptation process applied to a representation that does not depend on the access mechanism. For more information about principles of device-independence, see [DIPRINCIPLES].

Good practice: Separation of content, presentation, interaction

A specification SHOULD allow authors to separate content from both presentation and interaction concerns.

Note that when content, presentation, and interaction are separated by design, agents need to recombine them. There is a recombination spectrum, with "client does all" at one end and "server does all" at the other. There are advantages to each: recombination on the server allows the server to send out generally smaller amounts of data that can be tailored to specific devices (such as mobile phones). However, such data will not be readily reusable by other clients and may not allow client-side agents to perform useful tasks unanticipated by the author. When a client does the work of recombination, content is likely to be more reusable by a broader audience and more robust. However, such data may be of greater size and may require more computation by the client.

The design decision about where on this spectrum an application should be placed depends on the power on the client, the power and the load on the server, and the bandwidth of the medium that connects them. If the number of possible clients is unbounded, the application will scale better if more computation is pushed to the client.

Of course, it may be desirable to limit the audience. Designers should consider appropriate technologies, such as encryption and access control (§3.6.2), for limiting the audience.

Some data formats are designed to describe presentation (including SVG and XSL Formatting Objects). Data formats such as these demonstrate that one can only separate content from presentation (or interaction) so far; at some point it becomes necessary to talk about presentation. Per the principle of orthogonal specifications (§5.1) these data formats should only address presentation issues.

See the TAG issues formattingProperties-19 (concerning interoperability in the case of formatting properties and names) and contentPresentation-26 (concerning the separation of semantic and presentational markup).

4.4. Hypertext

A defining characteristic of the Web is that it allows embedded references to other resources via URIs. The simplicity of creating hypertext links using absolute URIs (<a href="">) and relative URI references (<a href="foo"> and <a href="foo#anchor">) is partly (perhaps largely) responsible for the success of the hypertext Web as we know it today.

When one resource (representation) refers to another resource with a URI, this constitutes a link between the two resources. Additional metadata may also form part of the link (see [XLink10], for example). Note: In this document, the term "link" generally means "relationship", not "physical connection".

Good practice: Link identification

A specification SHOULD provide ways to identify links to other resources and to secondaryportions of representation resources (via fragment identifiers).

Formats that allow content authors to use URIs instead of local identifiers promote the network effect: the value of these formats grows with the size of the deployed Web.

Good practice: Web linking

A specification SHOULD allow Web-wide linking, not just internal document linking.

Good practice: Generic URIs

A specification SHOULD allow content authors to use URIs without constraining them to a limited set of URI schemes.

What agents do with a hypertext link is not constrained by Web architecture and may depend on application context. Users of hypertext links expect to be able to navigate links among representations.

Good practice: Hypertext links

A data format SHOULD incorporate hypertext links if hypertext is the expected user interface paradigm.

Data formats that do not allow content authors to create hypertext links lead to the creation of "terminal nodes" on the Web.

4.4.1. URI references

Links are commonly expressed using URI references (defined in section 4.2 of [URI]), which may be combined with a base URI to yield a usable URI. Section 5.1 of [URI] explains different ways to establish a base URI for a resource and establishes a precedence among them. For instance, the base URI may be a URI for the resource, or specified in a representation (see the base elements provided by HTML and XML, and the HTTP 'Content-Location' header). See also the section on links in XML (§4.5.2) [section 4.5.2] .

Agents resolve a URI reference before using the resulting URI to interact with another agent. URI references help in content management by allowing content authors to design a representation locally, i.e., without concern for which global identifier may later be used to refer to the associated resource.

4.5. XML-Based Data Formats

Many data formats are XML-based, that is to say they conform to the syntax rules defined in the XML specification [XML10]. This section discusses issues that are specific to such formats. Anyone seeking guidance in this area is urged to consult the "Guidelines For the Use of XML in IETF Protocols" [IETFXML], which contains a thorough discussion of the considerations that govern whether or not XML ought to be used, as well as specific guidelines on how it ought to be used. While it is directed at Internet applications with specific reference to protocols, the discussion is generally applicable to Web scenarios as well.

The discussion here should be seen as ancillary to the content of [IETFXML]. Refer also to "XML Accessibility Guidelines" [XAG] for help designing XML formats that lower barriers to Web accessibility for people with disabilities.

4.5.1. When to use an XML-based format

XML defines textual data formats that are naturally suited to describing data objects which are hierarchical and processed in a chosen sequence. It is widely, but not universally, applicable for data formats; an audio or video format, for example, is unlikely to be well suited to expression in XML. Design constraints that would suggest the use of XML include:

  1. Requirement for a hierarchical structure.
  2. Immediate need for a wide range of tools on a variety of platforms.
  3. Need for data that can outlive the applications that currently process it.
  4. Ability to support internationalization in a self-describing way that makes confusion over coding options unlikely.
  5. Early detection of encoding errors with no requirement to "work around" such errors.
  6. A high proportion of human-readable textual content.
  7. Potential composition of the data format with other XML-encoded formats.
  8. Desire for data easily parsed by both humans and machines.
  9. Desire for vocabularies that can be invented in a distributed manner and combined flexibly.

4.5.2. Links in XML

Sophisticated linking mechanisms have been invented for XML formats. XPointer allows links to address content that does not have an explicit, named anchor. XLink is an appropriate specification for representing links in hypertext (§4.4) XML applications. XLink allows links to have multiple ends and to be expressed either inline or in "link bases" stored external to any or all of the resources identified by the links it contains.

Designers of XML-based formats should consider using XLink and, for defining fragment identifier syntax, using the XPointer framework and XPointer element() Schemes.

See TAG issue xlinkScope-23.

4.5.3. XML namespaces

The purpose of an XML namespace (defined in [XMLNS]) is to allow the deployment of XML vocabularies (in which element and attribute names are defined) in a global environment and to reduce the risk of name collisions in a given document when vocabularies are combined. For example, the MathML and SVG specifications both define the set element. Although XML data from different formats such as MathML and SVG can be combined in a single document, in this case there could be ambiguity about which set element was intended. XML namespaces reduce the risk of name collisions by taking advantage of existing systems for allocating globally scoped names: the URI system (see also the section on URI assignment (§2.2.1) [section 2.2.1] ). When using XML namespaces, each local name in an XML vocabulary is paired with a URI (called the namespace URI) to distinguish the local name from local names in other vocabularies. All of the globally grounded terms in an XML namespace share the same syntactic prefix: the namespace URI.

The use of URIs confers additional benefits. First, each local name / URI pair can be mapped to another URI, grounding the terms of the vocabulary in the Web. These terms may be important resources and thus it is appropriate to be able to associate URIs with them. One particularly useful mapping in the case of flat namespaces (specified, for example, in [RDFXML]) is to combine the namespace URI, a hash ("#"), and the local name, thus creating a URI for a secondary resource (the identified term). Other mappings are likely to be more suitable for hierarchical namespaces; see the related TAG issue abstractComponentRefs-37.

Designers of XML-based data formats who declare namespaces thus make it possible to reuse those data formats and combine them in novel ways not yet imagined. Failure to declare namespaces makes such re-use more difficult, even impractical in some cases.

Good practice: Namespace adoption

A specification that establishes an XML vocabulary SHOULD place all element names and global attribute names in a namespace.

Attributes are always scoped by the element on which they appear. An attribute that is "global," that is, one that might meaningfully appear on elements of many types, including elements in other namespaces, should be explicitly placed in a namespace. Local attributes, ones associated with only a particular element type, need not be included in a namespace since their meaning will always be clear from the context provided by that element.

The type attribute from the W3C XML Schema Instance namespace "" ([XMLSCHEMA], section 4.3.2) is an example of a global attribute. It can be used by authors of any vocabulary to make an assertion in instance data about the type of the element on which it appears. As a global attribute, it must always be qualified. The frame attribute on an HTML table is an example of a local attribute. There is no value in placing that attribute in a namespace since the attribute is unlikely to be useful on an element other than an HTML table.

Applications that rely on DTD processing must impose additional constraints on the use of namespaces. DTDs perform validation based on the lexical form of the element and attribute names in the document. This makes prefixes syntactically significant in ways that are not anticipated by [XMLNS].

4.5.4. Namespace documents


Nadia receives representation data from "" in an unfamiliar data format. She knows enough about XML to recognize which XML namespace the elements belong to. Since the namespace is identified by the URI "", she asks her browser to retrieve a representation of the identified resource. She gets back some useful data that allows her to learn more about the data format. Nadia's browser may also be able to perform some operations automatically (i.e., unattended by a human overseer) given data that has been optimized for software agents. For example, her browser might, on Nadia's behalf, download additional agents to process and render the format.

Another benefit of using URIs to build XML namespaces is that the namespace URI can be used to identify an Information Resource (§3.1) that contains useful information, machine-usable and/or human-usable, about terms in the namespace. This type of Information Resource is called a namespace document. When a namespace URI owner provides a namespace document, it is authoritative for the namespace.

There are many reasons to provide a namespace document. A person might want to:

  • understand the purpose of the namespace,
  • learn how to use the markup vocabulary in the namespace,
  • find out who controls it and associated policies,
  • request authority to access schemas or collateral material about it, or
  • report a bug or situation that could be considered an error in some collateral material.

A processor might want to:

  • retrieve a schema, for validation,
  • retrieve a style sheet, for presentation, or
  • retrieve ontologies, for making inferences.

In general, there is no established best practice for creating representations of a namespace document; application expectations will influence what data format or formats are used. Application expectations will also influence whether relevant information appears directly in a representation or is referenced from it.

Good practice: Namespace documents

The owner of an XML namespace name SHOULD make available material intended for people to read and material optimized for software agents in order to meet the needs of those who will use the namespace vocabulary.

For example, the following are examples of data formats for namespace documents: [OWL10], [RDDL], [XMLSCHEMA], and [XHTML11]. Each of these formats meets different requirements described above for satisfying the needs of an agent that wants more information about the namespace. Note, however, issues related to fragment identifiers and content negotiation (§3.3.2) if content negotiation is used.

See TAG issues namespaceDocument-8 (concerning desired characteristics of namespace documents) and abstractComponentRefs-37 (concerning the use of fragment identifiers with namespace names to identify abstract components).

4.5.5. QNames in XML

Section 3 of "Namespaces in XML" [XMLNS] provides a syntactic construct known as a QName for the compact expression of qualified names in XML documents. A qualified name is a pair consisting of a URI, which names a namespace, and a local name placed within that namespace. "Namespaces in XML" provides for the use of QNames as names for XML elements and attributes.

Other specifications, starting with [XSLT10], have employed the idea of using QNames in contexts other than element and attribute names, for example in attribute values and in element content. However, general XML processors cannot reliably recognize QNames as such when they are used in attribute values and in element content; for example, the syntax of QNames overlaps with that of URIs. Experience has also revealed other limitations to QNames, such as losing namespace bindings after XML canonicalization.

Good practice: QNames Indistinguishable from URIs

A specification in which QNames represent URI/local-name pairs SHOULD NOT allow both Qnames and URIs in attribute values or element content, where they would be indistinguishable.

For more information, see the TAG finding "Using QNames as Identifiers in Content".

Because QNames are compact, some specification designers have adopted the same syntax as a means of identifying resources. Though convenient as a shorthand notation, this usage has a cost. There is no single, accepted way to convert a QName into a URI or vice versa. Although QNames are convenient, they do not replace the URI as the identification system of the Web. The use of QNames to identify Web resources without providing a mapping to URIs is inconsistent with Web architecture.

Good practice: QName Mapping

A specification in which QNames serve as resource identifiers MUST provide a mapping to URIs.

One particularly useful mapping in the case of flat namespaces is to combine the namespace URI, a hash ("#"), and the local name; see the section on XML namespaces (§4.5.3) for more examples.

See also TAG issues rdfmsQnameUriMapping-6 (concerning the mapping of QNames to URIs), qnameAsId-18 (concerning the use of QNames as identifiers in XML content), and abstractComponentRefs-37 (concerning the use of fragment identifiers with namespace names to identify abstract components).

4.5.6. XML ID semantics

Consider the following fragment of XML: <section name="foo">. Does the section element have what the XML Recommendation refers to as the ID foo (i.e., "foo" must not appear in the surrounding XML document more than once)? One cannot answer this question by examining the element and its attributes alone. In XML, the quality of "being an ID" is associated with the type of an attribute, not its name. Finding the IDs in a document requires additional processing.

  1. Processing the document with a processor that recognizes DTD attribute list declarations (in the external or internal subset) might reveal a declaration that identifies the name attribute as an ID. Note: This processing is not necessarily part of validation. A non-validating, DTD-aware processor can recognize IDs. assignment.
  2. Processing the document with a W3C XML schema might reveal an element declaration that identifies the name attribute as an W3C XML Schema ID.
  3. In practice, processing the document with another schema language, such as RELAX NG [RELAXNG], might reveal the attributes declared to be of ID in the XML Schema sense. Many modern specifications begin processing XML at the Infoset [INFOSET] level and do not specify normatively how an Infoset is constructed. For those specifications, any process that establishes the ID type in the Infoset (and Post Schema Validation Infoset (PSVI) defined in [XMLSCHEMA]) may usefully identify the attributes of type ID.
  4. In practice, applications may have independent means (such as those defined in the XPointer specification, [XPTRFR] section 3.2) of locating identifiers inside a document.

To further complicate matters, DTDs establish the ID type in the Infoset whereas W3C XML Schema produces a PSVI but does not modify the original Infoset. This leaves open the possibility that a processor might only look in the Infoset and consequently would fail to recognize schema-assigned IDs.

The TAG expects to continue to work with other groups to help resolve open questions about establishing "ID-ness" in XML formats; this is the subject of TAG issue xmlIDSemantics-32.

4.5.7. Media types for XML

RFC 3023 defines the Internet media types "application/xml" and "text/xml", and describes a convention whereby XML-based data formats use Internet media types with a "+xml" suffix, for example "image/svg+xml".

There are two problems associated with the “text” mediatwo types: First, for data identified as "text/*", Web intermediaries are allowed to "transcode", i.e., convert one character encoding to another. Transcoding may make the self-description false or may cause the document to be not well-formed.

Good practice: XML and "text/*"

In general, a representation provider SHOULD NOT assign Internet media types beginning with "text/" to XML representations.

Second, representations whose Internet media types begin with "text/" are required, unless the charset parameter is specified, to be considered to be encoded in US-ASCII. Since the syntax of XML is designed to make documents self-describing, it is good practice to omit the charset parameter, and since XML is very often not encoded in US-ASCII, the use of "text/" Internet media types effectively precludes this good practice.

Good practice: XML and character encodings

In general, a representation provider SHOULD NOT specify the character encoding for XML data in protocol headers since the data is self-describing.

4.5.8. Fragment identifiers in XML

The section on media types and fragment identifier semantics (§3.3.1) discusses the interpretation of fragment identifiers. Designers of an XML-based data format specification should define the semantics of fragment identifiers in that format. The XPointer Framework [XPTRFR] provides an interoperable starting point.

When the media type assigned to representation data is "application/xml", there are no semantics defined for fragment identifiers, and authors should not make use of fragment identifiers in such data. The same is true if the assigned media type has the suffix "+xml" (defined in "XML Media Types" [RFC3023]), and the data format specification does not specify fragment identifier semantics. In short, just knowing that content is XML does not provide information about fragment identifier semantics.

Many people assume that the fragment identifier #abc, when referring to XML data, identifies the element in the document with the ID "abc". However, there is no normative support for this assumption. A revision of RFC 3023 is expected to address this.

See TAG issue fragmentInXML-28.

4.6. Data Formats Used to Build New Information Space Applications

Because of their role in defining fragment identifier semantics, data formats enable the creation of new applications to make use of the information space infrastructure. The Semantic Web is one such application, built on top of RDF [RDFXML]. This document does not discuss the Semantic Web in detail; the TAG expects that future editions of this document will. See the related TAG issue httpRange-14.

5. General Architecture Principles

A number of general architecture principles apply to all three bases of Web architecture.

5.1. Orthogonal Specifications

Identification, interaction, and representation are orthogonal concepts, meaning that technologies used for identification, interaction, and representation may evolve independently. For instance:

When two specifications are orthogonal, one may change one without requiring changes to the other, even if one has dependencies on the other. For example, although the HTTP specification depends on the URI specification, the two may evolve independently. This orthogonality increases the flexibility and robustness of the Web. For example, one may refer by URI to an image without knowing anything about the format chosen to represent the image. This has facilitated the introduction of image formats such as PNG and SVG without disrupting existing references to image resources.

Principle: Orthogonality

Orthogonal abstractions benefit from orthogonal specifications.

Experience demonstrates that problems arise where orthogonal concepts occur in a single specification. Consider, for example, the HTML specification which includes the orthogonal x-www-form-urlencoded specification. Software developers (for example, of [CGI] applications) might have an easier time finding the specification if it were published separately and then cited from the HTTP, URI, and HTML specifications.

Problems also arise when specifications attempt to modify orthogonal abstractions described elsewhere. An historical version of the HTML specification added a "Refresh" value to the http-equiv attribute of the meta element. It was defined to be equivalent to the HTTP header of the same name. The authors of the HTTP specification ultimately decided not to provide this header and that made the two specifications awkwardly at odds with each other. The W3C HTML Working Group eventually removed the "Refresh" value.

A specification should clearly indicate which features overlapadvance into with those governed by another specification.

5.2. Extensibility

The information in the Web and the technologies used to represent that information change over time. Extensibility is the property of a technology that promotes both evolution without sacrificingand interoperability. Some examples of successful technologies designed to allow change while minimizing disruption include:

An example of an unsuccessful extension mechanism is HTTP mandatory extensions [HTTPEXT]. The community has sought mechanisms to extend HTTP, but apparently the costs of the mandatory extension proposal (notably in complexity) outweighed the benefits and thus hampered adoption.

Below we discuss the property of "extensibility," exhibited by URIs, some data formats, and some protocols (through the incorporation of new messages).

Subset language: one language is a subset (or "profile") of a second language if any document in the first language is also a valid document in the second language and has the same interpretation in the second language.

Extended language: If one language is a subset of another, the latter superset is called an extended language; the difference between the languages is called the extension. Clearly, extending a language is better for interoperability than creating an incompatible language.

Ideally, many instances of a superset language can be safely and usefully processed as though they were in the subset language. Languages that can evolve this way, allowing applications to provide new information when necessary while still interoperating with applications that only understand a subset of the current language, are said to be "extensible." Language designers can facilitate extensibility by defining the default behavior of unknown extensions—for example, that they be ignored (in some defined way) or should be considered errors.

For example, from early on in the Web, HTML agents followed the convention of ignoring unknown tags. This choice left room for innovation (i.e., non-standard elements) and encouraged the deployment of HTML. However, interoperability problems arose as well. In this type of environment, there is an inevitable tension between interoperability in the short term and the desire for extensibility. Experience shows that designs that strike the right balance between allowing change and preserving interoperability are more likely to thrive and are less likely to disrupt the Web community. Orthogonal specifications (§5.1) help reduce the risk of disruption.

For further discussion, see the section on versioning and extensibility (§4.2). See also TAG issue xmlProfiles-29 and HTML Dialects.

5.3. Error Handling

Errors occur in networked information systems. An error condition can be well-characterized (e.g., well-formedness errors in XML or 4xx client errors in HTTP) or arise unpredictably. Error correction means that an agent repairs a condition so that within the system, it is as though the error never occurred. One example of error correction involves data retransmission in response to a temporary network failure. Error recovery means that an agent does not repair an error condition but continues processing by addressing the fact that the error has occurred.

Agents frequently correct errors without user awareness, sparing users the details of complex network communications. On the other hand, it is important that agents recover from error in a way that is evident to users, since the agents are acting on their behalf.

Principle: Error recovery

Agents that recover from error by making a choice without the user's consent are not acting on the user's behalf.

An agent is not required to interrupt the user (e.g., by popping up a confirmation box) to obtain consent. The user may indicate consent through pre-selected configuration options, modes, or selectable user interface toggles, with appropriate reporting to the user when the agent detects an error. Agent developers should not ignore usability issues when designing error recovery behavior.

To promote interoperability, specification designers should identify predictable error conditions. Experience has led to the following observations about error-handling approaches.

See the TAG issue contentTypeOverride-24, which concerns the source of authoritative metadata.

5.4. Protocol-based Interoperability

The Web follows Internet tradition in that its important interfaces are defined in terms of protocols, by specifying the syntax, semantics, and sequencing constraints of the messages interchanged. Protocols designed to be resilient in the face of widely varying environments have helped the Web scale and have facilitated communication across multiple trust boundaries. Traditional application programming interfaces (APIs) do not always take these constraints into account, nor should they be required to. One effect of protocol-based design is that the technology shared among agents often lasts longer than the agents themselves.

It is common for programmers working with the Web to write code that generates and parses these messages directly. It is less common, but not unusual, for end users to have direct exposure to these messages. It is often desirable to provide users with access to format and protocol details: allowing them to "view source," whereby they may gain expertise in the workings of the underlying system.

6. Glossary

Content negotiation
The practice of providing multiple representations available via the same URI. Which representation is served depends on negotiation between the requesting agent and the agent serving the representations.
Dereference a URI
Access the resource identified by the URI.
Error correction
An agent repairs an error so that within the system, it is as though the error never occurred.
Error recovery
An agent invokes exceptional behavior because it does not correct the error.
Extended language
If one language is a subset of another, the latter is called an extended language.
Fragment identifier
The part of a URI that allows identification of a secondary resource.
Information resource
A resource that conveys information.
A relationship between two resources when one resource (representation) refers to the other resource by means of a URI.
A unit of communication between agents.
Namespace document
The Information Resource identified by an XML Namespace URI
Data and metadata that represents the information state of a resource.
Anything that might be identified by a URI.
Safe interaction
Interaction with a resource where an agent does not incur any obligation beyond the interaction.
Secondary resource
A resource related to another resource through the primary resource with additional identifying information (the fragment identifier).
Subset language
One language is a subset of a second language if any document in the first language is also a valid document in the second language and has the same interpretation in the second language.
Acronym for Uniform Resource Identifier.
URI aliases
Two or more different URIs that that identify the same resource.
URI collision
The use of the same URI to refer to more than one resource in the context of Web protocols and formats.
URI ownership
The relationship between an assigning agent and a URI that is defined by a URI scheme.
URI persistence
The social expectation that once a URI identifies a particular resource, it should continue indefinitely to refer to that resource.
URI reference
An operational shorthand for a URI.
Uniform Resource Identifier (URI)
A global identifier in the context of the World Wide Web.
Unsafe interaction
Interaction with a resource that is not safe interaction.
User agent
One type of Web agent; a piece of software acting on behalf of a person.
View source effect
The result of direct exposure to the underlying protocols which allows users to gain expertise in the workings of a system.
Acronym for World Wide Web.
Shortened form of World Wide Web.
Web agent
A person or a piece of software acting on the information space on behalf of a person, entity, or process.
World Wide Web
An information space in which items of interest are identified by Uniform Resource Identifiers.
XML-based format
One that conforms to the syntax rules defined in the XML specification.

7. References

Common Gateway Interface/1.1 Specification. Available at
Common User Agent Problems, K. Dubost, January 2003. This W3C Team Submission is available at
Cool URIs don't change T. Berners-Lee, W3C, 1998 Available at Note that the title is somewhat misleading. It is not the URIs that change, it is what they identify.
Knowledge-Domain Interoperability and an Open Hyperdocument System, D. C. Engelbart, June 1990.
Mandatory Extensions in HTTP, H. Frystyk Nielsen, P. Leach, S. Lawrence, 20 January 1998. This expired IETF Internet Draft is available at
IANA's online registry of URI Schemes is available at
IETF Guidelines For The Use of XML in IETF Protocols, S. Hollenbeck, M. Rose, L. Masinter, eds., 2 November 2002. This IETF Internet Draft is available at If this document is no longer available, refer to the ietf-xml-use mailing list.
XML Information Set (Second Edition) , J. Cowan, R. Tobin, J. Cowan, Editors, W3C Recommendation, 4 February 2004, . Latest version available at .
IETF Internationalized Resource Identifiers (IRIs), M. Dürst, M. Suignard, Nov 2002. This IETF Internet Draft is available at If this document is no longer available, refer to the home page for Editing 'Internationalized Resource Identifiers (IRIs)'.
IANA's online registry of Internet Media Types is available at
OWL Web Ontology Language Reference , G. Schreiber, M. Dean, G. Schreiber, Editors, W3C Recommendation, 10 February 2004, . Latest version available at .
The Platform for Privacy Preferences 1.0 (P3P1.0) Specification , M. Marchiori, Editor, W3C Recommendation, 16 April 2002, . Latest version available at .
Resource Directory Description Language (RDDL), J. Borden, T. Bray, eds., 1 June 2003. This document is available at
RDF/XMLResource Description Framework (RDF) Model and Syntax Specification (Revised) , D.O. Lassila, R. R. Beckett, Editor, W3C Recommendation, 10 February 2004, . Latest version available at .
The RELAX NG schema language project.
Representational State Transfer (REST), Chapter 5 of "Architectural Styles and the Design of Network-based Software Architectures", Doctoral Thesis of R. T. Fielding, 2000. Designers of protocol specifications in particular should invest time in understanding the REST model and the relevance of its principles to a given design. These principles include statelessness, clear assignment of roles to parties, uniform address space, and a limited, uniform set of verbs. Available at
IETF RFC 2045: Multipurpose Internet Mail Extensions (MIME) Part One: Format of Internet Message Bodies, N. Freed, N. Borenstein, November 1996. Available at
IETF RFC 2046: Multipurpose Internet Mail Extensions (MIME) Part Two: Media Types, N. Freed, N. Borenstein, November 1996. Available at
IETF RFC 2119: Key words for use in RFCs to Indicate Requirement Levels, S. Bradner, March 1997. Available at
IETF RFC 2141: URN Syntax, R. Moats, May 1997. Available at
IETF RFC 2326: Real Time Streaming Protocol (RTSP), H. Schulzrinne, A. Rao, R. Lanphier, April 1998. Available at:
IETF RFC 2616: Hypertext Transfer Protocol &mdash; HTTP/1.1, J. Gettys, J. Mogul, H. Frystyk, L. Masinter, P. Leach, T. Berners-Lee, June 1999. Available at
IETF Registration Procedures for URL Scheme Names, R. Petke, I. King, November 1999. Available at
IETF RFC 2718: Guidelines for new URL Schemes, L. Masinter, H. Alvestrand, D. Zigmond, R. Petke, November 1999. Available at:
IETF RFC 2818: HTTP Over TLS, E. Rescorla, May 2000. Available at:
IETF RFC 3023: XML Media Types, M. Murata, S. St. Laurent, D. Kohn, January 2001. Available at:
IETF RFC 3236: The 'application/xhtml+xml' Media Type, M. Baker, P. Stark, January 2002. Available at:
IETF RFC 3261: SIP: Session Initiation Protocol, J. Rosenberg, H. Schulzrinne, G. Camarillo, et. al., June 2002. Available at:
IETF RFC 977: Network News Transfer Protocol, B. Kantor, P. Lapsley, February 1986. Available at
SOAP Version 1.2 Part 1: Messaging Framework , M. Gudgin, N. Mendelsohn, J. Moreau, M. Hadley, H. Frystyk Nielsen, M. Gudgin, Editors, W3C Recommendation, 24 June 2003, . Latest version available at .
Scalable Vector Graphics (SVG) 1.1 Specification , 藤沢, J. Ferraiolo, 藤沢, D. Jackson, Editors, W3C Recommendation, 14 January 2003, . Latest version available at .
See the Unicode Consortium home page for information about the latest version of Unicode and character repertoires.
Uniform Resource Identifiers (URI): Generic Syntax (T. Berners-Lee, R. Fielding, L. Masinter, Eds.) is currently being revised. Citations labeled [URI] refer to "Uniform Resource Identifier (URI): Generic Syntax."
IAB Technical Comment on the Unique DNS Root, B. Carpenter, 27 September 1999. Available at
Voice Extensible Markup Language (VoiceXML) Version 2.0 , D. C. Burnett, J. Carter, P. Danielsen, S. Tryphonas, J. Ferrans, A. Hunt, B. Lucas, B. Porter, K. G. Rehor, S. Tryphonas, S. McGlashan, D. C. Burnett, J. Carter, P. Danielsen, Editors, W3C Recommendation, 16 March 2004, . Latest version available at .
XHTML™ 1.1 - Module-based XHTML , M. Altheim, S. McCarron, M. Altheim, Editors, W3C Recommendation, 31 May 2001, . Latest version available at .
XML Linking Language (XLink) Version 1.0 , E. Maler, S. J. DeRose, E. Maler, D. Orchard, Editors, W3C Recommendation, 27 June 2001, . Latest version available at .
Extensible Markup Language (XML) 1.0 (Third Edition) , T. Bray, J. Paoli, C. M. Sperberg-McQueen, J. Paoli, E. Maler, T. Bray, F. Yergeau, Editors, W3C Recommendation, 4 February 2004, . Latest version available at .
Namespaces in XML 1.1 , A. Layman, R. Tobin, A. Layman, T. Bray, D. Hollander, Editors, W3C Recommendation, 4 February 2004, . Latest version available at .
XML Schema Part 1: Structures , N.H. S. Mendelsohn, D. Beech, H. S. Thompson, M. Maloney,Mendelsohn, Editors, W3C Recommendation, 2 May 2001, . Latest version available at .
The Extensible Messaging and Presence Protocol (XMPP) IETF Working Group is developing "an open, XML-based protocol for near real-time extensible messaging and presence. It is the core protocol of the Jabber Instant Messaging and Presence technology..."
XPointer Framework , P. Grosso, E. Maler, J. Marsh, P. Grosso, N. Walsh, Editors, W3C Recommendation, 25 March 2003, . Latest version available at .
XSL Transformations (XSLT) Version 1.0 , J. Clark, Editor, W3C Recommendation, 16 November 1999, . Latest version available at .

7.1. Architectural Specifications

Authoring Tool Accessibility Guidelines 1.0 , J. Treviranus, C. McCathieNevile, J. Richards, J. Treviranus, I. Jacobs,Richards, Editors, W3C Recommendation, 3 February 2000, . Latest version available at .
Character Model for the World Wide Web 1.0: Fundamentals , T. Texin, M. J. Dürst, T. Texin, M. Wolf,Yergeau, R. Ishida, F. Yergeau, Editors, W3C Working Draft (work in progress), 25 February 2004, . Latest version available at .
, Editor, W3C (work in progress), NaN , . Latest version available at .
Device Independence Principles , R. Gimson, Editor, W3C Working Group Note, 1 September 2003, . Latest version available at .
Web Architecture: Extensible Languages, T. Berners-Lee, D. Connolly, 10 February 1998. This W3C Note is available at
Principled Design of the Modern Web Architecture, R.T. Fielding and R.N. Taylor, UC Irvine. In Proceedings of the 2000 International Conference on Software Engineering (ICSE 2000), Limerick, Ireland, June 2000, pp. 407-416. This document is available at
QA Framework: Specification Guidelines , L. Henderson, L. Rosenthal, K. Dubost, D. Hazaël-Massieux, L. Henderson, Editors, W3C Working Draft (work in progress), 30 August 2004, . Latest version available at .
IETF RFC 1958: Architectural Principles of the Internet, B. Carpenter, June 1996. Available at
Variability in Specifications , L. Rosenthal, D. Hazaël-Massieux, Editors, W3C Working Draft (work in progress), 30 August 2004, . Latest version available at .
User Agent Accessibility Guidelines 1.0 , I. Jacobs, J. Gunderson, E. Hansen, I. Jacobs, Editors, W3C Recommendation, 17 December 2002, . Latest version available at .
Web Content Accessibility Guidelines 2.0 , B. Caldwell, W. Chisholm, G. Vanderheiden, B. Caldwell, J. White, Editors, W3C Working Draft (work in progress), 30 July 2004, . Latest version available at .
Web Services Architecture , D.H. Haas, F. Orchard, D. Booth, F. McCabe, C. Ferris, H. Haas, E. Newcomer, M. Champion,Booth, Editors, W3C Working Group Note, 11 February 2004, . Latest version available at .
XML Accessibility Guidelines , D. Dardailler, S. B. Palmer, D. Dardailler, C. McCathieNevile, Editors, W3C Working Draft (work in progress), 3 October 2002, . Latest version available at .

8. Acknowledgments

This document was authored by the W3C Technical Architecture Group which included the following participants: Tim Berners-Lee (co-Chair, W3C), Tim Bray (Antarctica Systems), Dan Connolly (W3C), Paul Cotton (Microsoft Corporation), Roy Fielding (Day Software), Mario Jeckle (Daimler Chrysler), Chris Lilley (W3C), David Orchard (BEA Systems), Norman Walsh (Sun), and Stuart Williams (co-Chair, Hewlett-Packard).

The TAG appreciates the many contributions on the TAG's public mailing list, (archive), which have helped to improve this document.