- General Model
- Xpointer Forms
- Functions
- Using Xpointers
- Future Developments
- Conclusions
This chapter is from the book
This chapter is from the book
This chapter is from the book
6.2 Xpointer Forms
An XPointer is an identification of a part of an XML resource and is mainly intended to be used as a fragment identifier in a URI reference. As such, it must somehow fit into the framework defined for URIs as described in section 3.2. Basically, it must be represented as a printable string that can be used within URI references and exchanged in the same way URIs are exchanged. (To this end, XPointers must obey a number of character-escaping rules, which are not discussed here but are described in detail in section 6.4.1).
XPointer distinguishes three di8erent forms as follows:
The first one (a bare name) is defined to act mainly as a very concise form and to be backwards compatible with HTML fragment identifiers. This first form is described in section 6.2.1. A second form (a child sequence) still uses an abbreviated syntax but allows more flexibility than the first form. It is described in section 6.2.2. Finally, the full form of XPointer is discussed in section 6.2.3. It is the most complex and most powerful XPointer form.
6.2.1 Bare Names
The simplest form of an XPointer is a bare name. Basically, it consists of the same name as that provided for the argument of a location step using the id function. This function returns the element that has an attribute of type ID14 with the argument's value, so an XPointer bare name does exactly the same thing as an HTML fragment identifier.15 As an example, the URI reference http://www.w3.org/TR/#xptr uses a bare name16 and may consequently be interpreted as pointing to the element that has an attribute of type ID with the value xptr.
There are two reasons XPointer bare names are supported, despite their very limited functionality:
For reasons of backwards compatibility with HTML and as an easy migration to XHTML without the need to rewrite all fragment identifiers when converting a resource from HTML to XHTML. Otherwise, HTML fragment identifiers become invalid if an HTML resource is converted to XHTML.
To encourage the use of IDs, which are the most robust mechanism for pointing into XML documents.
XPointer bare names are very easy to understand and use, but they are limited in two ways.
The first limitation is that they may point only to element nodes, because only elements may be identified via ID attributes. The second limitation is that the element to which an XPointer should point must have an attribute of type ID, otherwise it is impossible to point to this element using a bare name. (If the document is under the control of the creator of the XPointer, then an ID may be created by modifying the document, but this is not always an option.) As an intermediate form of XPointers (between the bare form and full XPointers), we have the child sequence. This form of XPointer, discussed in the next section, is also limited to pointing to element nodes, but it—in contrast to bare names—may point to element nodes that do not carry an attribute of type ID.
6.2.2 Child Sequences
A child sequence is a form of XPointer that selects an element node by navigating through a document's element tree. It resembles an XPath location path in that it uses a path notation, but it is much more limited in that it can select only elements and uses only the child axis. The syntax for a child sequence is as follows:
An example of a child sequence identifying the same element as the http://www.w3.org/TR/#xptr bare name would be http://www.w3.org/TR/#/1/2/17/15/1/1/1. This is nothing more than a navigation path through the element tree of the XHTML page based on child sequences. It could be spelled out as follows:
Select the first child of the document root, which is the <html> element.
Select the second child of the <html> element, which is the <body> element.
Select the seventeenth child of the <body> element, which is a <dl> element.
Select the fifteenth child of the <dl> element, which is a <dt> element.
Select the first child of the <dt> element, which is a <b> element.
Select the first child of the <b> element, which is an <i> element.
Select the first child of the <i> element, which is an <a> element.
This last <a> element is the one carrying the xptr ID, which is why the bare name XPointer and the child sequence XPointer select the same element. In this case, we have selected something that is also accessible through an ID. However, what is interesting about the child sequence mechanism is that it would still be possible to select the element even if was not identified through an ID.
Child sequences have a severe disadvantage in that they are very sensitive to document modifications. It is very unlikely that the example just presented will still work after the W3C's page has been modified since the modification is likely to involve inserting or deleting elements preceding the one we want to select, resulting in breaking the intention of the child sequence.17 Since child sequences are easy to use but also easy to break, there is a second form of child sequence that uses an element identified by an ID as the starting point rather than using the document root. In this case, the XPointer starts with a bare name but then continues with a child sequence,18 navigating the document tree starting from the element with the given ID. An example of this kind of child sequence would be http://www.w3.org/TR/#last-call-list/15/1/1/1. This would work if the W3C had made the <dl> element listing the document in "last call" status accessible through a last-call-list ID (which it has not done). This kind of child sequence would make the XPointer more robust to modifications of the document because only changes within the <dl> element could possibly break the XPointer.
6.2.3 Full XPointers
Both bare names and child sequences are rather limited in their expressiveness, insofar as they can select only element nodes and also in the way they select these nodes. Therefore, in many cases it is necessary to use more complex XPointers (the so-called full XPointer), which provide a much more flexible way of addressing into a document than bare names or child sequences do. Syntactically, a full XPointer is defined as follows:
As an example, now consider the W3C technical reports page. W3C has used IDs to mark individual sections (which are visible as headings and subheadings in Figure 6.1); but unfortunately in XHTML the actual contents of a section are not contained in any specific element. Instead they simply follow a sectioning element, such as the <h3> element used for the "Working Drafts in Last Call" section heading. Consequently, using only child sequences, it is impossible to point to the entry for the XPointer working draft entry. This is because the XPointer entry is part of the <dl> list following the "Working Drafts in Last Call" heading. So if we want to point to the entry for XPointer starting from the heading with the ID, we have to use a full XPointer as follows:
){kind=link}
http://www.w3.org/TR/#xpointer(id('last-call')/following::dl[1]/dt[7]//a)
Here we mainly exploit the fact that the "Working Drafts in Last Call" heading is identified by the last-call ID. The XPointer part of the URI reference could be interpreted as follows (commenting on the individual location steps):
Select the element having the last-call ID. (This is the <h3> element with the "Working Drafts in Last Call" content.) Here we use XPath's id function as described in section 5.4.4.
Select the first definition list (represented by a <dl> element) following the <h3> element. (For an explanation of the XPath following axis, refer to section 5.2.2.)
Select the seventh definition term of this list (i.e., the seventh <dt> child of the <dl> element).
Select the <a> element, which is (directly or indirectly) contained in the <dt> element.
In this example, we see that a normal XPath can be used within an XPointer. Apart from a few exceptions, XPointer allows unlimited use of XPath's expressiveness. Therefore, it is necessary to know XPath if we are to create anything more than trivial XPointers (i.e., bare names and child sequences).
Looking at the example just given, note the presence of a special keyword, xpointer, which, according to rule 4 from the syntactic definition, is the specification of a scheme. XPointer schemes are an important mechanism for building robust XPointers. According to rule 3 of the syntax, full XPointers use schemes to define any number of scheme-specific parts. XPointer specifies that scheme parts must be evaluated left to right, so the interpretation of the scheme parts is well defined. Currently, there are only two schemes defined for XPointer (as indicated by rule 4):
xpointer – This is by far the most widely used scheme, and it indicates that the expression that follows (contained in parentheses) is an XPointer expression. This scheme has been used in the example just given.
xmlns – This second scheme is for the initialization of namespaces. In XPath (and, therefore, in XPointer), an expression may contain a qualified name (i.e., a name containing a namespace prefix and a local name). In order to make the interpretation of qualified names possible, there must be a mechanism for the initialization of namespaces (i.e., for the assignment of a namespace URI to a namespace prefix). This scheme will be discussed in detail in section 6.4.2.
While these two schemes are the only ones defined in the XPointer specification, it is possible for applications to define their own schemes, which can then be used to access subresources in an application-specific way. This would, however, make the XPointer unusable for any application not supporting the application-specific extension. It is useful to note that the XPointer specification states that interpretation proceeds from left to right and stops once a scheme part that can be successfully interpreted as a locator into the resource has been identified. Consequently, it is possible to concatenate multiple scheme parts, and if the interpretation of a scheme part fails (a so-called subresource error19), then the interpretation of the XPointer continues with the next scheme part. A simple example for this is the following URI reference:
http://www.w3.org/TR/#xpointer(id('xptr'))
xpointer(/*[1]/*[2]/*[17]/*[15]/*[1]/*[1]/*[1])
In this case, the XPointer consists of two scheme parts, both being of the xpointer scheme. If an element with the ID xptr is present in the resource, then the XPointer will locate it.20 Otherwise, the first scheme part results in a subresource error, and the second part then successfully locates the resource by simply counting element children.21 Given our example used so far, this XPointer would continue to work even if the ID xptr was removed from the document. However, the XPointer would still break if the ID was removed and the resource's structure was changed in such a way that the "child sequence" no longer identified the element we want to locate.22
Formally, as already mentioned, the XPointer specification states that evaluation of XPointer scheme parts must be left to right but also that evaluation of scheme parts must continue if one of the following conditions is met while evaluating a scheme part:
The scheme is unknown to the application evaluating the XPointer.
The scheme is not applicable to the media type of the resource.
The scheme does not locate any subresource present in the resource.23 This applies to XPointer expressions evaluating to an empty location set (a subresource error).
If the scheme being interpreted is xpointer, the following applies:
The string argument in a string-range function is not found in the string-value of the location, or the third or fourth argument of the function indicates a string that is beyond the beginning or end of the document.
The point returned by the start-point function is of type attribute or namespace.
This set of rules for the evaluation of XPointer scheme parts makes it possible to create XPointers with built-in "fault tolerance" (through providing fall-back scheme parts to be used if the original scheme part does not work anymore). In section 6.4.3 we will discuss some interesting examples.