The FunctionNamespace

In order to use a function in XPath/XSLT, it needs to have a (namespaced) name by which it can be called during evaluation. This is done using the FunctionNamespace class. For simplicity, we choose the empty namespace (None):

>>> from lxml import etree
>>> ns = etree.FunctionNamespace(None)
>>> ns['hello'] = hello
>>> ns['countargs'] = loadsofargs

This registers the function hello with the name hello in the default namespace (None), and the function loadsofargs with the name countargs. Now we're going to create a document that we can run XPath expressions against:

>>> root = etree.XML('<a><b>Haegar</b></a>')
>>> doc = etree.ElementTree(root)

Done. Now we can have XPath expressions call our new function:

>>> print root.xpath("hello('world')")
Hello world
>>> print root.xpath('hello(local-name(*))')
Hello b
>>> print root.xpath('hello(string(b))')
Hello Haegar
>>> print root.xpath('countargs(., b, ./*)')
Got 3 arguments.

Note how we call both a Python function (hello) and an XPath built-in function (string) in exactly the same way. Normally, however, you would want to separate the two in different namespaces. The FunctionNamespace class allows you to do this:

>>> ns = etree.FunctionNamespace('http://mydomain.org/myfunctions')
>>> ns['hello'] = hello
>>> prefixmap = {'f' : 'http://mydomain.org/myfunctions'}
>>> print root.xpath('f:hello(local-name(*))', namespaces=prefixmap)
Hello b

Global prefix assignment

In the last example, you had to specify a prefix for the function namespace. If you always use the same prefix for a function namespace, you can also register it with the namespace:

>>> ns = etree.FunctionNamespace('http://mydomain.org/myother/functions')
>>> ns.prefix = 'es'
>>> ns['hello'] = ola
>>> print root.xpath('es:hello(local-name(*))')
Ola b

This is a global assignment, so take care not to assign the same prefix to more than one namespace. The resulting behaviour in that case is completely undefined. It is always a good idea to consistently use the same meaningful prefix for each namespace throughout your application.

The prefix assignment only works with functions and FunctionNamespace objects, not with the general Namespace object that registers element classes. The reasoning is that elements in lxml do not care about prefixes anyway, so it would rather complicate things than be of any help.

The XPath context

Functions get a context object as first parameter. In lxml 1.x, this value was None, but since lxml 2.0 it provides two properties: eval_context and context_node. The context node is the Element where the current function is called:

>>> def print_tag(context, nodes):
...     print context.context_node.tag, [ n.tag for n in nodes ]

>>> ns = etree.FunctionNamespace('http://mydomain.org/printtag')
>>> ns.prefix = "pt"
>>> ns["print_tag"] = print_tag

>>> ignore = root.xpath("//*[pt:print_tag(.//*)]")
a ['b']
b []

The eval_context is a dictionary that is local to the evaluation. It allows functions to keep state:

>>> def print_context(context):
...     context.eval_context[context.context_node.tag] = "done"
...     entries = context.eval_context.items()
...     entries.sort()
...     print entries
>>> ns["print_context"] = print_context

>>> ignore = root.xpath("//*[pt:print_context()]")
[('a', 'done')]
[('a', 'done'), ('b', 'done')]

Evaluators and XSLT

Extension functions work for all ways of evaluating XPath expressions and for XSL transformations:

>>> e = etree.XPathEvaluator(doc)
>>> print e.evaluate('es:hello(local-name(/a))')
Ola a

>>> namespaces = {'f' : 'http://mydomain.org/myfunctions'}
>>> e = etree.XPathEvaluator(doc, namespaces=namespaces)
>>> print e.evaluate('f:hello(local-name(/a))')
Hello a

>>> xslt = etree.XSLT(etree.XML('''
...   <stylesheet version="1.0"
...          xmlns="http://www.w3.org/1999/XSL/Transform"
...          xmlns:es="http://mydomain.org/myother/functions">
...     <output method="text" encoding="ASCII"/>
...     <template match="/">
...       <value-of select="es:hello(string(//b))"/>
...     </template>
...   </stylesheet>
... '''))
>>> print xslt(doc)
Ola Haegar

It is also possible to register namespaces with a single evaluator after its creation. While the following example involves no functions, the idea should still be clear:

>>> from StringIO import StringIO
>>> f = StringIO('<a xmlns="http://mydomain.org/myfunctions" />')
>>> ns_doc = etree.parse(f)
>>> e = etree.XPathEvaluator(ns_doc)
>>> e.evaluate('/a')
[]

This returns nothing, as we did not ask for the right namespace. When we register the namespace with the evaluator, however, we can access it via a prefix:

>>> e.registerNamespace('foo', 'http://mydomain.org/myfunctions')
>>> e.evaluate('/foo:a')[0].tag
'{http://mydomain.org/myfunctions}a'

Note that this prefix mapping is only known to this evaluator, as opposed to the global mapping of the FunctionNamespace objects:

>>> e2 = etree.XPathEvaluator(ns_doc)
>>> e2.evaluate('/foo:a')
Traceback (most recent call last):
...
XPathEvalError: Undefined namespace prefix

Evaluator-local extensions

Apart from the global registration of extension functions, there is also a way of making extensions known to a single Evaluator or XSLT. All evaluators and the XSLT object accept a keyword argument extensions in their constructor. The value is a dictionary mapping (namespace, name) tuples to functions:

>>> extensions = {('local-ns', 'local-hello') : hello}
>>> namespaces = {'l' : 'local-ns'}

>>> e = etree.XPathEvaluator(doc, namespaces=namespaces, extensions=extensions)
>>> print e.evaluate('l:local-hello(string(b))')
Hello Haegar

For larger numbers of extension functions, you can define classes or modules and use the Extension helper:

>>> class MyExt:
...     def function1(self, _, arg):
...         return '1'+arg
...     def function2(self, _, arg):
...         return '2'+arg
...     def function3(self, _, arg):
...         return '3'+arg

>>> ext_module = MyExt()
>>> functions = ('function1', 'function2')
>>> extensions = etree.Extension( ext_module, functions, ns='local-ns' )

>>> e = etree.XPathEvaluator(doc, namespaces=namespaces, extensions=extensions)
>>> print e.evaluate('l:function1(string(b))')
1Haegar

The optional second argument to Extension can either be be a sequence of names to select from the module, a dictionary that explicitly maps function names to their XPath alter-ego or None (explicitly passed) to take all available functions under their original name (if their name does not start with '_').

The additional ns keyword argument takes a namespace URI or None (also if left out) for the default namespace. The following examples will therefore all do the same thing:

>>> functions = ('function1', 'function2', 'function3')
>>> extensions = etree.Extension( ext_module, functions )
>>> e = etree.XPathEvaluator(doc, extensions=extensions)
>>> print e.evaluate('function1(function2(function3(string(b))))')
123Haegar

>>> extensions = etree.Extension( ext_module, functions, ns=None )
>>> e = etree.XPathEvaluator(doc, extensions=extensions)
>>> print e.evaluate('function1(function2(function3(string(b))))')
123Haegar

>>> extensions = etree.Extension(ext_module)
>>> e = etree.XPathEvaluator(doc, extensions=extensions)
>>> print e.evaluate('function1(function2(function3(string(b))))')
123Haegar

>>> functions = {
...     'function1' : 'function1',
...     'function2' : 'function2',
...     'function3' : 'function3'
...     }
>>> extensions = etree.Extension(ext_module, functions)
>>> e = etree.XPathEvaluator(doc, extensions=extensions)
>>> print e.evaluate('function1(function2(function3(string(b))))')
123Haegar

For convenience, you can also pass a sequence of extensions:

>>> extensions1 = etree.Extension(ext_module)
>>> extensions2 = etree.Extension(ext_module, ns='local-ns')
>>> e = etree.XPathEvaluator(doc, extensions=[extensions1, extensions2],
...                          namespaces=namespaces)
>>> print e.evaluate('function1(l:function2(function3(string(b))))')
123Haegar

What to return from a function

Extension functions can return any data type for which there is an XPath equivalent (see the documentation on XPath return values). This includes numbers, boolean values, elements and lists of elements. Note that integers will also be returned as floats:

>>> def returnsFloat(_):
...    return 1.7
>>> def returnsInteger(_):
...    return 1
>>> def returnsBool(_):
...    return True
>>> def returnFirstNode(_, nodes):
...    return nodes[0]

>>> ns = etree.FunctionNamespace(None)
>>> ns['float'] = returnsFloat
>>> ns['int']   = returnsInteger
>>> ns['bool']  = returnsBool
>>> ns['first'] = returnFirstNode

>>> e = etree.XPathEvaluator(doc)
>>> e.evaluate("float()")
1.7
>>> e.evaluate("int()")
1.0
>>> int( e.evaluate("int()") )
1
>>> e.evaluate("bool()")
True
>>> e.evaluate("count(first(//b))")
1.0

As the last example shows, you can pass the results of functions back into the XPath expression. Elements and sequences of elements are treated as XPath node-sets:

>>> def returnsNodeSet(_):
...     results1 = etree.Element('results1')
...     etree.SubElement(results1, 'result').text = "Alpha"
...     etree.SubElement(results1, 'result').text = "Beta"
...
...     results2 = etree.Element('results2')
...     etree.SubElement(results2, 'result').text = "Gamma"
...     etree.SubElement(results2, 'result').text = "Delta"
...
...     results3 = etree.SubElement(results2, 'subresult')
...     return [results1, results2, results3]

>>> ns['new-node-set'] = returnsNodeSet

>>> e = etree.XPathEvaluator(doc)

>>> r = e.evaluate("new-node-set()/result")
>>> print [ t.text for t in r ]
['Alpha', 'Beta', 'Gamma', 'Delta']

>>> r = e.evaluate("new-node-set()")
>>> print [ t.tag for t in r ]
['results1', 'results2', 'subresult']
>>> print [ len(t) for t in r ]
[2, 3, 0]
>>> r[0][0].text
'Alpha'

>>> print etree.tostring(r[0])
<results1><result>Alpha</result><result>Beta</result></results1>

>>> print etree.tostring(r[1])
<results2><result>Gamma</result><result>Delta</result><subresult/></results2>

>>> print etree.tostring(r[2])
<subresult/>

The current implementation deep-copies newly created elements in node-sets. Only the elements and their children are passed on, no outlying parents or tail texts will be available in the result. This also means that in the above example, the subresult elements in results2 and results3 are no longer identical within the node-set, they belong to independent trees:

>>> print r[1][-1].tag, r[2].tag
subresult subresult
>>> print r[1][-1] == r[2]
False
>>> print r[1][-1].getparent().tag
results2
>>> print r[2].getparent()
None

This is an implementation detail that you should be aware of, but you should avoid relying on it in your code. Note that elements taken from the source document (the most common case) do not suffer from this restriction. They will always be passed unchanged.