[APACHE DOCUMENTATION]

Apache HTTP Server Version 1.3

Content Negotiation

Apache's support for content negotiation has been updated to meet the HTTP/1.1 specification. It can choose the best representation of a resource based on the browser-supplied preferences for media type, languages, character set and encoding. It is also implements a couple of features to give more intelligent handling of requests from browsers which send incomplete negotiation information.

Content negotiation is provided by the mod_negotiation module, which is compiled in by default.


About Content Negotiation

A resource may be available in several different representations. For example, it might be available in different languages or different media types, or a combination. One way of selecting the most appropriate choice is to give the user an index page, and let them select. However it is often possible for the server to choose automatically. This works because browsers can send as part of each request information about what representations they prefer. For example, a browser could indicate that it would like to see information in French, if possible, else English will do. Browsers indicate their preferences by headers in the request. To request only French representations, the browser would send

  Accept-Language: fr

Note that this preference will only be applied when there is a choice of representations and they vary by language.

As an example of a more complex request, this browser has been configured to accept French and English, but prefer French, and to accept various media types, preferring HTML over plain text or other text types, and preferring GIF or JPEG over other media types, but also allowing any other media type as a last resort:

  Accept-Language: fr; q=1.0, en; q=0.5
  Accept: text/html; q=1.0, text/*; q=0.8, image/gif; q=0.6,
        image/jpeg; q=0.6, image/*; q=0.5, */*; q=0.1
Apache 1.2 supports 'server driven' content negotiation, as defined in the HTTP/1.1 specification. It fully supports the Accept, Accept-Language, Accept-Charset and Accept-Encoding request headers.

The terms used in content negotiation are: a resource is an item which can be requested of a server, which might be selected as the result of a content negotiation algorithm. If a resource is available in several formats, these are called representations or variants. The ways in which the variants for a particular resource vary are called the dimensions of negotiation.

Negotiation in Apache

In order to negotiate a resource, the server needs to be given information about each of the variants. This is done in one of two ways:

Using a type-map file

A type map is a document which is associated with the handler named type-map (or, for backwards-compatibility with older Apache configurations, the mime type application/x-type-map). Note that to use this feature, you've got to have a SetHandler some place which defines a file suffix as type-map; this is best done with a


  AddHandler type-map var

in srm.conf. See comments in the sample config files for details.

Type map files have an entry for each available variant; these entries consist of contiguous RFC822-format header lines. Entries for different variants are separated by blank lines. Blank lines are illegal within an entry. It is conventional to begin a map file with an entry for the combined entity as a whole (although this is not required, and if present will be ignored). An example map file is:


  URI: foo

  URI: foo.en.html
  Content-type: text/html
  Content-language: en

  URI: foo.fr.de.html
  Content-type: text/html; charset=iso-8859-2
  Content-language: fr, de
If the variants have different source qualities, that may be indicated by the "qs" parameter to the media type, as in this picture (available as jpeg, gif, or ASCII-art):
  URI: foo

  URI: foo.jpeg
  Content-type: image/jpeg; qs=0.8

  URI: foo.gif
  Content-type: image/gif; qs=0.5

  URI: foo.txt
  Content-type: text/plain; qs=0.01

qs values can vary between 0.000 and 1.000. Note that any variant with a qs value of 0.000 will never be chosen. Variants with no 'qs' parameter value are given a qs factor of 1.0.

The full list of headers recognized is:

URI:
uri of the file containing the variant (of the given media type, encoded with the given content encoding). These are interpreted as URLs relative to the map file; they must be on the same server (!), and they must refer to files to which the client would be granted access if they were to be requested directly.
Content-type:
media type --- charset, level and "qs" parameters may be given. These are often referred to as MIME types; typical media types are image/gif, text/plain, or text/html; level=3.
Content-language:
The languages of the variant, specified as an Internet standard language code (e.g., en for English, kr for Korean, etc.).
Content-encoding:
If the file is compressed, or otherwise encoded, rather than containing the actual raw data, this says how that was done. For compressed files (the only case where this generally comes up), content encoding should be x-compress, or x-gzip, as appropriate.
Content-length:
The size of the file. Clients can ask to receive a given media type only if the variant isn't too big; specifying a content length in the map allows the server to compare against these thresholds without checking the actual file.

Multiviews

This is a per-directory option, meaning it can be set with an Options directive within a <Directory>, <Location> or <Files> section in access.conf, or (if AllowOverride is properly set) in .htaccess files. Note that Options All does not set MultiViews; you have to ask for it by name. (Fixing this is a one-line change to http_core.h).

The effect of MultiViews is as follows: if the server receives a request for /some/dir/foo, if /some/dir has MultiViews enabled, and /some/dir/foo does not exist, then the server reads the directory looking for files named foo.*, and effectively fakes up a type map which names all those files, assigning them the same media types and content-encodings it would have if the client had asked for one of them by name. It then chooses the best match to the client's requirements, and forwards them along.

This applies to searches for the file named by the DirectoryIndex directive, if the server is trying to index a directory; if the configuration files specify


  DirectoryIndex index

then the server will arbitrate between index.html and index.html3 if both are present. If neither are present, and index.cgi is there, the server will run it.

If one of the files found when reading the directive is a CGI script, it's not obvious what should happen. The code gives that case special treatment --- if the request was a POST, or a GET with QUERY_ARGS or PATH_INFO, the script is given an extremely high quality rating, and generally invoked; otherwise it is given an extremely low quality rating, which generally causes one of the other views (if any) to be retrieved.

The Negotiation Algorithm

After Apache has obtained a list of the variants for a given resource, either from a type-map file or from the filenames in the directory, it applies a algorithm to decide on the 'best' variant to return, if any. To do this it calculates a quality value for each variant in each of the dimensions of variance. It is not necessary to know any of the details of how negotiation actually takes place in order to use Apache's content negotiation features. However the rest of this document explains in detail the algorithm used for those interested.

In some circumstances, Apache can 'fiddle' the quality factor of a particular dimension to achieve a better result. The ways Apache can fiddle quality factors is explained in more detail below.

Dimensions of Negotiation

Dimension Notes
Media Type Browser indicates preferences on Accept: header. Each item can have an associated quality factor. Variant description can also have a quality factor.
Language Browser indicates preferences on Accept-Language: header. Each item can have a quality factor. Variants can be associated with none, one or more languages.
Encoding Browser indicates preference with Accept-Encoding: header.
Charset Browser indicates preference with Accept-Charset: header. Variants can indicate a charset as a parameter of the media type.

Apache Negotiation Algorithm

Apache uses an algorithm to select the 'best' variant (if any) to return to the browser. This algorithm is not configurable. It operates like this:

  1. Firstly, for each dimension of the negotiation, the appropriate Accept header is checked and a quality assigned to this each variant. If the Accept header for any dimension means that this variant is not acceptable, eliminate it. If no variants remain, go to step 4.
  2. Select the 'best' variant by a process of elimination. Each of the following tests is applied in order. Any variants not selected at each stage are eliminated. After each test, if only one variant remains, it is selected as the best match. If more than one variant remains, move onto the next test.
    1. Multiply the quality factor from the Accept header with the quality-of-source factor for this variant's media type, and select the variants with the highest value
    2. Select the variants with the highest language quality factor
    3. Select the variants with the best language match, using either the order of languages on the LanguagePriority directive (if present), else the order of languages on the Accept-Language header.
    4. Select the variants with the highest 'level' media parameter (used to give the version of text/html media types).
    5. Select the variants with the best encoding. If there are variants with an encoding that is acceptable to the user-agent, select only these variants. Otherwise if there is a mix of encoded and non-encoded variants, select only the unencoded variants. If either all variants are encoded or all variants are not encoded, select all variants.
    6. Select only variants with acceptable charset media parameters, as given on the Accept-Charset header line. Charset ISO-8859-1 is always acceptable. Variants not associated with a particular charset are assumed to be in ISO-8859-1.
    7. Select the variants with the smallest content length
    8. Select the first variant of those remaining (this will be either the first listed in the type-map file, or the first read from the directory) and go to stage 3.
  3. The algorithm has now selected one 'best' variant, so return it as the response. The HTTP response header Vary is set to indicate the dimensions of negotiation (browsers and caches can use this information when caching the resource). End.
  4. To get here means no variant was selected (because non are acceptable to the browser). Return a 406 status (meaning "No acceptable representation") with a response body consisting of an HTML document listing the available variants. Also set the HTTP Vary header to indicate the dimensions of variance.

Fiddling with Quality Values

Apache sometimes changes the quality values from what would be expected by a strict interpretation of the algorithm above. This is to get a better result from the algorithm for browsers which do not send full or accurate information. Some of the most popular browsers send Accept header information which would otherwise result in the selection of the wrong variant in many cases. If a browser sends full and correct information these fiddles will not be applied.

Media Types and Wildcards

The Accept: request header indicates preferences for media types. It can also include 'wildcard' media types, such as "image/*" or "*/*" where the * matches any string. So a request including:

  Accept: image/*, */*
would indicate that any type starting "image/" is acceptable, as is any other type (so the first "image/*" is redundant). Some browsers routinely send wildcards in addition to explicit types they can handle. For example:
  Accept: text/html, text/plain, image/gif, image/jpeg, */*
The intention of this is to indicate that the explicitly listed types are preferred, but if a different representation is available, that is ok too. However under the basic algorithm, as given above, the */* wildcard has exactly equal preference to all the other types, so they are not being preferred. The browser should really have sent a request with a lower quality (preference) value for *.*, such as:
  Accept: text/html, text/plain, image/gif, image/jpeg, */*; q=0.01
The explicit types have no quality factor, so they default to a preference of 1.0 (the highest). The wildcard */* is given a low preference of 0.01, so other types will only be returned if no variant matches an explicitly listed type.

If the Accept: header contains no q factors at all, Apache sets the q value of "*/*", if present, to 0.01 to emulate the desired behavior. It also sets the q value of wildcards of the format "type/*" to 0.02 (so these are preferred over matches against "*/*". If any media type on the Accept: header contains a q factor, these special values are not applied, so requests from browsers which send the correct information to start with work as expected.

Variants with no Language

If some of the variants for a particular resource have a language attribute, and some do not, those variants with no language are given a very low language quality factor of 0.001.

The reason for setting this language quality factor for variant with no language to a very low value is to allow for a default variant which can be supplied if none of the other variants match the browser's language preferences. For example, consider the situation with three variants:

The meaning of a variant with no language is that it is always acceptable to the browser. If the request Accept-Language header includes either en or fr (or both) one of foo.en.html or foo.fr.html will be returned. If the browser does not list either en or fr as acceptable, foo.html will be returned instead.

Note on hyperlinks and naming conventions

If you are using language negotiation you can choose between different naming conventions, because files can have more than one extension, and the order of the extensions is normally irrelevant (see mod_mime documentation for details).

A typical file has a MIME-type extension (e.g., html), maybe an encoding extension (e.g., gz), and of course a language extension (e.g., en) when we have different language variants of this file.

Examples:

Here some more examples of filenames together with valid and invalid hyperlinks:

Filename Valid hyperlink Invalid hyperlink
foo.html.en foo
foo.html
-
foo.en.html foo foo.html
foo.html.en.gz foo
foo.html
foo.gz
foo.html.gz
foo.en.html.gz foo foo.html
foo.html.gz
foo.gz
foo.gz.html.en foo
foo.gz
foo.gz.html
foo.html
foo.html.gz.en foo
foo.html
foo.html.gz
foo.gz

Looking at the table above you will notice that it is always possible to use the name without any extensions in an hyperlink (e.g., foo). The advantage is that you can hide the actual type of a document rsp. file and can change it later, e.g., from html to shtml or cgi without changing any hyperlink references.

If you want to continue to use a MIME-type in your hyperlinks (e.g. foo.html) the language extension (including an encoding extension if there is one) must be on the right hand side of the MIME-type extension (e.g., foo.html.en).

Note on Caching

When a cache stores a document, it associates it with the request URL. The next time that URL is requested, the cache can use the stored document, provided it is still within date. But if the resource is subject to content negotiation at the server, this would result in only the first requested variant being cached, and subsequent cache hits could return the wrong response. To prevent this, Apache normally marks all responses that are returned after content negotiation as non-cacheable by HTTP/1.0 clients. Apache also supports the HTTP/1.1 protocol features to allow caching of negotiated responses.

For requests which come from a HTTP/1.0 compliant client (either a browser or a cache), the directive CacheNegotiatedDocs can be used to allow caching of responses which were subject to negotiation. This directive can be given in the server config or virtual host, and takes no arguments. It has no effect on requests from HTTP/1.1 clients.


Apache HTTP Server Version 1.3

Index