====================================
Using custom Element classes in lxml
====================================

lxml has very sophisticated support for custom Element classes.  You
can provide your own classes for Elements and have lxml use them by
default for all elements generated by a specific parser, only for a
specific tag name in a specific namespace or even for an exact element
at a specific position in the tree.

Custom Elements must inherit from the ``lxml.etree.ElementBase`` class, which
provides the Element interface for subclasses:

.. sourcecode:: pycon

  >>> from lxml import etree

  >>> class honk(etree.ElementBase):
  ...    @property
  ...    def honking(self):
  ...       return self.get('honking') == 'true'

This defines a new Element class ``honk`` with a property ``honking``.

The following document describes how you can make lxml.etree use these
custom Element classes.

.. contents::
..
   1  Background on Element proxies
   2  Element initialization
   3  Setting up a class lookup scheme
     3.1  Default class lookup
     3.2  Namespace class lookup
     3.3  Attribute based lookup
     3.4  Custom element class lookup
     3.5  Tree based element class lookup in Python
   4  Generating XML with custom classes
   5  Implementing namespaces


Background on Element proxies
=============================

Being based on libxml2, lxml.etree holds the entire XML tree in a C
structure.  To communicate with Python code, it creates Python proxy
objects for the XML elements on demand.

   .. image:: proxies.png

The mapping between C elements and Python Element classes is
completely configurable.  When you ask lxml.etree for an Element by
using its API, it will instantiate your classes for you.  All you have
to do is tell lxml which class to use for which kind of Element.  This
is done through a class lookup scheme, as described in the sections
below.


Element initialization
======================

There is one thing to know up front.  Element classes *must not* have
an ``__init___`` or ``__new__`` method.  There should not be any
internal state either, except for the data stored in the underlying
XML tree.  Element instances are created and garbage collected at
need, so there is normally no way to predict when and how often a
proxy is created for them.  Even worse, when the ``__init__`` method
is called, the object is not even initialized yet to represent the XML
tag, so there is not much use in providing an ``__init__`` method in
subclasses.

Most use cases will not require any class initialisation or proxy
state, so you can content yourself with skipping to the next section
for now.  However, if you really need to set up your element class on
instantiation, or need a way to persistently store state in the proxy
instances instead of the XML tree, here is a way to do so.

There is one important guarantee regarding Element proxies.  Once a
proxy has been instantiated, it will keep alive as long as there is a
Python reference to it, and any access to the XML element in the tree
will return this very instance.  Therefore, if you need to store local
state in a custom Element class (which is generally discouraged), you
can do so by keeping the Elements in a tree alive.  If the tree
doesn't change, you can simply do this:

.. sourcecode:: python

  proxy_cache = list(root.iter())

or

.. sourcecode:: python

  proxy_cache = set(root.iter())

or use any other suitable container.  Note that you have to keep this
cache manually up to date if the tree changes, which can get tricky in
cases.

For proxy initialisation, ElementBase classes have an ``_init()``
method that can be overridden, as oppose to the normal ``__init__()``
method.  It can be used to modify the XML tree, e.g. to construct
special children or verify and update attributes.

The semantics of ``_init()`` are as follows:

* It is called once on Element class instantiation time.  That is,
  when a Python representation of the element is created by lxml.  At
  that time, the element object is completely initialized to represent
  a specific XML element within the tree.

* The method has complete access to the XML tree.  Modifications can be done
  in exactly the same way as anywhere else in the program.

* Python representations of elements may be created multiple times during the
  lifetime of an XML element in the underlying C tree.  The ``_init()`` code
  provided by subclasses must take special care by itself that multiple
  executions either are harmless or that they are prevented by some kind of
  flag in the XML tree.  The latter can be achieved by modifying an attribute
  value or by removing or adding a specific child node and then verifying this
  before running through the init process.

* Any exceptions raised in ``_init()`` will be propagated through the API
  call that lead to the creation of the Element.  So be careful with the code
  you write here as its exceptions may turn up in various unexpected places.


Setting up a class lookup scheme
================================

The first thing to do when deploying custom element classes is to register a
class lookup scheme on a parser.  lxml.etree provides quite a number of
different schemes that also support class lookup based on namespaces or
attribute values.  Most lookups support fallback chaining, which allows the
next lookup mechanism to take over when the previous one fails to find a
class.

For example, setting the ``honk`` Element as a default element class
for a parser works as follows:

.. sourcecode:: pycon

  >>> parser_lookup = etree.ElementDefaultClassLookup(element=honk)
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(parser_lookup)

There is one drawback of the parser based scheme: the ``Element()`` factory
does not know about your specialised parser and creates a new document that
deploys the default parser:

.. sourcecode:: pycon

  >>> el = etree.Element("root")
  >>> print(isinstance(el, honk))
  False

You should therefore avoid using this factory function in code that
uses custom classes.  The ``makeelement()`` method of parsers provides
a simple replacement:

.. sourcecode:: pycon

  >>> el = parser.makeelement("root")
  >>> print(isinstance(el, honk))
  True

If you use a parser at the module level, you can easily redirect a module
level ``Element()`` factory to the parser method by adding code like this:

.. sourcecode:: pycon

  >>> module_level_parser = etree.XMLParser()
  >>> Element = module_level_parser.makeelement

While the ``XML()`` and ``HTML()`` factories also depend on the default
parser, you can pass them a different parser as second argument:

.. sourcecode:: pycon

  >>> element = etree.XML("<test/>")
  >>> print(isinstance(element, honk))
  False

  >>> element = etree.XML("<test/>", parser)
  >>> print(isinstance(element, honk))
  True

Whenever you create a document with a parser, it will inherit the lookup
scheme and all subsequent element instantiations for this document will use
it:

.. sourcecode:: pycon

  >>> element = etree.fromstring("<test/>", parser)
  >>> print(isinstance(element, honk))
  True
  >>> el = etree.SubElement(element, "subel")
  >>> print(isinstance(el, honk))
  True

For testing code in the Python interpreter and for small projects, you
may also consider setting a lookup scheme on the default parser.  To
avoid interfering with other modules, however, it is usually a better
idea to use a dedicated parser for each module (or a parser pool when
using threads) and then register the required lookup scheme only for
this parser.


Default class lookup
--------------------

This is the most simple lookup mechanism.  It always returns the default
element class.  Consequently, no further fallbacks are supported, but this
scheme is a nice fallback for other custom lookup mechanisms.  Specifically,
it also handles comments and processing instructions, which are easy to
forget about when mapping proxies to classes.

Usage:

.. sourcecode:: pycon

  >>> lookup = etree.ElementDefaultClassLookup()
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(lookup)

Note that the default for new parsers is to use the global fallback, which is
also the default lookup (if not configured otherwise).

To change the default element implementation, you can pass your new class to
the constructor.  While it accepts classes for ``element``, ``comment`` and
``pi`` nodes, most use cases will only override the element class:

.. sourcecode:: pycon

  >>> el = parser.makeelement("myelement")
  >>> print(isinstance(el, honk))
  False

  >>> lookup = etree.ElementDefaultClassLookup(element=honk)
  >>> parser.set_element_class_lookup(lookup)

  >>> el = parser.makeelement("myelement")
  >>> print(isinstance(el, honk))
  True
  >>> el.honking
  False
  >>> el = parser.makeelement("myelement", honking='true')
  >>> etree.tostring(el)
  b'<myelement honking="true"/>'
  >>> el.honking
  True

  >>> root = etree.fromstring(
  ...     '<root honking="true"><!--comment--></root>', parser)
  >>> root.honking
  True
  >>> print(root[0].text)
  comment


Namespace class lookup
----------------------

This is an advanced lookup mechanism that supports namespace/tag-name specific
element classes.  You can select it by calling:

.. sourcecode:: pycon

  >>> lookup = etree.ElementNamespaceClassLookup()
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(lookup)

See the separate section on `implementing namespaces`_ below to learn how to
make use of it.

.. _`implementing namespaces`: #implementing-namespaces

This scheme supports a fallback mechanism that is used in the case where the
namespace is not found or no class was registered for the element name.
Normally, the default class lookup is used here.  To change it, pass the
desired fallback lookup scheme to the constructor:

.. sourcecode:: pycon

  >>> fallback = etree.ElementDefaultClassLookup(element=honk)
  >>> lookup = etree.ElementNamespaceClassLookup(fallback)
  >>> parser.set_element_class_lookup(lookup)

  >>> root = etree.fromstring(
  ...     '<root honking="true"><!--comment--></root>', parser)
  >>> root.honking
  True
  >>> print(root[0].text)
  comment


Attribute based lookup
----------------------

This scheme uses a mapping from attribute values to classes.  An attribute
name is set at initialisation time and is then used to find the corresponding
value in a dictionary.  It is set up as follows:

.. sourcecode:: pycon

  >>> id_class_mapping = {'1234' : honk} # maps attribute values to classes

  >>> lookup = etree.AttributeBasedElementClassLookup(
  ...                                      'id', id_class_mapping)
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(lookup)

And here is how to use it:

.. sourcecode:: pycon

  >>> xml = '<a id="123"><b id="1234"/><b id="1234" honking="true"/></a>'
  >>> a = etree.fromstring(xml, parser)

  >>> a.honking       # id does not match !
  Traceback (most recent call last):
  AttributeError: 'lxml.etree._Element' object has no attribute 'honking'

  >>> a[0].honking
  False
  >>> a[1].honking
  True

This lookup scheme uses its fallback if the attribute is not found or
its value is not in the mapping.  Normally, the default class lookup
is used here.  If you want to use the namespace lookup, for example,
you can use this code:

.. sourcecode:: pycon

  >>> fallback = etree.ElementNamespaceClassLookup()
  >>> lookup = etree.AttributeBasedElementClassLookup(
  ...                       'id', id_class_mapping, fallback)
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(lookup)


Custom element class lookup
---------------------------

This is the most customisable way of finding element classes on a per-element
basis.  It allows you to implement a custom lookup scheme in a subclass:

.. sourcecode:: pycon

  >>> class MyLookup(etree.CustomElementClassLookup):
  ...     def lookup(self, node_type, document, namespace, name):
  ...         if node_type == 'element':
  ...             return honk  # be a bit more selective here ...
  ...         else:
  ...             return None  # pass on to (default) fallback

  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(MyLookup())

  >>> root = etree.fromstring(
  ...     '<root honking="true"><!--comment--></root>', parser)
  >>> root.honking
  True
  >>> print(root[0].text)
  comment

The ``.lookup()`` method must return either None (which triggers the
fallback mechanism) or a subclass of ``lxml.etree.ElementBase``.  It
can take any decision it wants based on the node type (one of
"element", "comment", "PI", "entity"), the XML document of the
element, or its namespace or tag name.


Tree based element class lookup in Python
-----------------------------------------

Taking more elaborate decisions than allowed by the custom scheme is
difficult to achieve in pure Python, as it results in a
chicken-and-egg problem.  It would require access to the tree - before
the elements in the tree have been instantiated as Python Element
proxies.

Luckily, there is a way to do this.  The ``PythonElementClassLookup``
works similar to the custom lookup scheme:

.. sourcecode:: pycon

  >>> class MyLookup(etree.PythonElementClassLookup):
  ...     def lookup(self, document, element):
  ...         return MyElementClass # defined elsewhere

  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(MyLookup())

As before, the first argument to the ``lookup()`` method is the opaque
document instance that contains the Element.  The second arguments is a
lightweight Element proxy implementation that is only valid during the lookup.
Do not try to keep a reference to it.  Once the lookup is finished, the proxy
will become invalid.  You will get an ``AssertionError`` if you access any of
the properties or methods outside the scope of the lookup call where they were
instantiated.

During the lookup, the element object behaves mostly like a normal Element
instance.  It provides the properties ``tag``, ``text``, ``tail`` etc. and
supports indexing, slicing and the ``getchildren()``, ``getparent()``
etc. methods.  It does *not* support iteration, nor does it support any kind
of modification.  All of its properties are read-only and it cannot be removed
or inserted into other trees.  You can use it as a starting point to freely
traverse the tree and collect any kind of information that its elements
provide.  Once you have taken the decision which class to use for this
element, you can simply return it and have lxml take care of cleaning up the
instantiated proxy classes.

Sidenote: this lookup scheme originally lived in a separate module called
``lxml.pyclasslookup``.


Generating XML with custom classes
==================================

Up to lxml 2.1, you could not instantiate proxy classes yourself.
Only lxml.etree could do that when creating an object representation
of an existing XML element.  Since lxml 2.2, however, instantiating
this class will simply create a new Element:

.. sourcecode:: pycon

  >>> el = honk(honking='true')
  >>> el.tag
  'honk'
  >>> el.honking
  True

Note, however, that the proxy you create here will be garbage
collected just like any other proxy.  You can therefore not count on
lxml.etree using the same class that you instantiated when you access
this Element a second time after letting its reference go.  You should
therefore always use a corresponding class lookup scheme that returns
your Element proxy classes for the elements that they create.  The
``ElementNamespaceClassLookup`` is generally a good match.

You can use custom Element classes to quickly create XML fragments:

.. sourcecode:: pycon

  >>> class hale(etree.ElementBase): pass
  >>> class bopp(etree.ElementBase): pass

  >>> el = hale( "some ", honk(honking = 'true'), bopp, " text" )

  >>> print(etree.tostring(el, encoding='unicode'))
  <hale>some <honk honking="true"/><bopp/> text</hale>


Implementing namespaces
=======================

lxml allows you to implement namespaces, in a rather literal sense.  After
setting up the namespace class lookup mechanism as described above, you can
build a new element namespace (or retrieve an existing one) by calling the
``get_namespace(uri)`` method of the lookup:

.. sourcecode:: pycon

  >>> lookup = etree.ElementNamespaceClassLookup()
  >>> parser = etree.XMLParser()
  >>> parser.set_element_class_lookup(lookup)

  >>> namespace = lookup.get_namespace('http://hui.de/honk')

and then register the new element type with that namespace, say, under the tag
name ``honk``:

.. sourcecode:: pycon

  >>> namespace['honk'] = honk

If you have many Element classes declared in one module, and they are
all named like the elements they create, you can simply use
``namespace.update(globals())`` at the end of your module to declare them
automatically.  The implementation is smart enough to ignore
everything that is not an Element class.

After this, you create and use your XML elements through the normal API of
lxml:

.. sourcecode:: pycon

  >>> xml = '<honk xmlns="http://hui.de/honk" honking="true"/>'
  >>> honk_element = etree.XML(xml, parser)
  >>> print(honk_element.honking)
  True

The same works when creating elements by hand:

.. sourcecode:: pycon

  >>> honk_element = parser.makeelement('{http://hui.de/honk}honk',
  ...                                   honking='true')
  >>> print(honk_element.honking)
  True

Essentially, what this allows you to do, is to give Elements a custom API
based on their namespace and tag name.

A somewhat related topic are `extension functions`_ which use a similar
mechanism for registering Python functions for use in XPath and XSLT.

.. _`extension functions`: extensions.html

In the setup example above, we associated the ``honk`` Element class
only with the 'honk' element.  If an XML tree contains different
elements in the same namespace, they do not pick up the same
implementation:

.. sourcecode:: pycon

  >>> xml = ('<honk xmlns="http://hui.de/honk" honking="true">'
  ...        '<bla/><!--comment-->'
  ...        '</honk>')
  >>> honk_element = etree.XML(xml, parser)
  >>> print(honk_element.honking)
  True
  >>> print(honk_element[0].honking)
  Traceback (most recent call last):
    ...
  AttributeError: 'lxml.etree._Element' object has no attribute 'honking'
  >>> print(honk_element[1].text)
  comment

You can therefore provide one implementation per element name in each
namespace and have lxml select the right one on the fly.  If you want one
element implementation per namespace (ignoring the element name) or prefer
having a common class for most elements except a few, you can specify a
default implementation for an entire namespace by registering that class with
the empty element name (``None``).

You may consider following an object oriented approach here.  If you build a
class hierarchy of element classes, you can also implement a base class for a
namespace that is used if no specific element class is provided.  Again, you
can just pass None as an element name:

.. sourcecode:: pycon

  >>> class HonkNSElement(etree.ElementBase):
  ...    def honk(self):
  ...       return "HONK"
  >>> namespace[None] = HonkNSElement  # default Element for namespace

  >>> class HonkElement(HonkNSElement):
  ...    @property
  ...    def honking(self):
  ...       return self.get('honking') == 'true'
  >>> namespace['honk'] = HonkElement  # Element for specific tag

Now you can rely on lxml to always return objects of type HonkNSElement or its
subclasses for elements of this namespace:

.. sourcecode:: pycon

  >>> xml = ('<honk xmlns="http://hui.de/honk" honking="true">'
  ...        '<bla/><!--comment-->'
  ...        '</honk>')
  >>> honk_element = etree.fromstring(xml, parser)

  >>> print(type(honk_element))
  <class 'HonkElement'>
  >>> print(type(honk_element[0]))
  <class 'HonkNSElement'>

  >>> print(honk_element.honking)
  True
  >>> print(honk_element.honk())
  HONK

  >>> print(honk_element[0].honk())
  HONK
  >>> print(honk_element[0].honking)
  Traceback (most recent call last):
  ...
  AttributeError: 'HonkNSElement' object has no attribute 'honking'

  >>> print(honk_element[1].text)  # uses fallback for non-elements
  comment

Since lxml 4.1, the registration is more conveniently done with
class decorators.  The namespace registry object is callable with
a name (or ``None``) as argument and can then be used as decorator.

.. sourcecode:: pycon

  >>> honk_elements = lookup.get_namespace('http://hui.de/honk')

  >>> @honk_elements(None)
  ... class HonkNSElement(etree.ElementBase):
  ...    def honk(self):
  ...       return "HONK"

If the class has the same name as the tag, you can also leave out the call
and use the blank decorator instead:

  >>> @honk_elements
  ... class honkel(HonkNSElement):
  ...    @property
  ...    def honking(self):
  ...       return self.get('honking') == 'true'

  >>> xml = '<honkel xmlns="http://hui.de/honk" honking="true"><bla/><!--comment--></honkel>'
  >>> honk_element = etree.fromstring(xml, parser)

  >>> print(type(honk_element))
  <class 'honkel'>
  >>> print(type(honk_element[0]))
  <class 'HonkNSElement'>