Axes¶

Every location step in an XPath expression has three parts:

axis::node-test[predicate]

The axis sets the direction the step travels from the context node (down to children, up to ancestors, sideways to siblings, …); the node-test says which kind of node to keep along that axis; the optional predicate filters what survives.

The paths you have written so far used the abbreviated syntax, where the most common axes are implied. Every abbreviation expands to a full axis::node-test form:

So catalog/cd/@genre is just a friendlier spelling of child::catalog/child::cd/attribute::genre. The abbreviations cover the steps you reach for daily; the explicit axes below are what you fall back to when they do not.

The running example¶

<?xml version="1.0" encoding="UTF-8"?>
<catalog>
  <cd genre="rock">
    <title>Empire Burlesque</title>
    <artist>Bob Dylan</artist>
    <price>10.90</price>
    <year>1985</year>
  </cd>
  <cd genre="pop">
    <title>Hide your heart</title>
    <artist>Bonnie Tyler</artist>
    <price>9.90</price>
  </cd>
  <cd genre="country">
    <title>Greatest Hits</title>
    <artist>Dolly Parton</artist>
    <price>9.90</price>
  </cd>
</catalog>

Forward axes¶

Forward axes select nodes that come at or after the context node in document order.

child — the immediate children (the default axis).

child::cd          <!-- same as: cd -->

descendant — children, grandchildren, and so on, but not the context node itself.

descendant::title

descendant-or-self — the descendants plus the context node. This is the axis hiding inside //.

descendant-or-self::node()

following-sibling — siblings that come after the context node, sharing the same parent.

following-sibling::cd      <!-- from the first cd: the 2nd and 3rd -->

following — every node after the context node in document order, except its own descendants (and excluding attribute and namespace nodes).

following::price           <!-- from the first title: all later prices -->

attribute — the attributes of the context element. This is what @ abbreviates.

attribute::genre           <!-- same as: @genre -->

self — the context node itself. This is what . abbreviates; it is most useful for adding a node-test, as in self::cd.

self::node()               <!-- same as: . -->

namespace — the namespace nodes in scope on an element. Rarely used directly, and dropped entirely in XPath 2.0+.

Reverse axes¶

Reverse axes select nodes that come at or before the context node.

parent — the single parent node. This is what .. abbreviates.

parent::node()             <!-- same as: .. -->

ancestor — the parent, grandparent, and so on up to the document root.

ancestor::catalog          <!-- from any title: the enclosing catalog -->

ancestor-or-self — the ancestors plus the context node.

ancestor-or-self::*

preceding-sibling — siblings that come before the context node, sharing the same parent.

preceding-sibling::cd      <!-- from the third cd: the 1st and 2nd -->

preceding — every node before the context node in document order, except its own ancestors (and excluding attribute and namespace nodes).

preceding::title           <!-- from the last price: all earlier titles -->

descendant vs following, child vs descendant¶

Two distinctions trip people up.

child reaches one level down; descendant reaches all levels down. For the flat catalog child::title from a cd finds its title, while descendant::title from catalog finds every title in the document.

descendant stays inside the context node — only nodes it contains. following looks outside and after it — everything later in the document that is not one of its descendants. Standing on the first <cd>:

• descendant::* → its own title, artist, price, year.
• following::* → the second and third <cd> elements and all their children, but nothing inside the first <cd>.

The two axes never overlap: together with ancestor, preceding, and self they partition the whole document.

When you actually need explicit axes¶

The abbreviations handle "down" and "up one level". You write an explicit axis when you need to move sideways, reach far up, or look backward.

following-sibling::cd[1]    <!-- the very next cd after this one -->
ancestor::catalog          <!-- the catalog this node lives in -->
preceding-sibling::*       <!-- every earlier sibling element -->

following-sibling::cd[1] is the idiomatic "next record" navigation: from one <cd> it lands on the one immediately after. There is no abbreviation for any of these — .. only climbs one step, and // only descends.

Summary¶

Next¶

A step is more than its axis: the node-test decides which nodes survive, and the predicate filters them — see Node tests and predicates.