Nevow Object Traversal
======================

*Object traversal* is the process Nevow uses to determine what object to use to
render HTML for a particular URL. When an HTTP request comes in to the web
server, the object publisher splits the URL into segments, and repeatedly calls
methods which consume path segments and return objects which represent that
path, until all segments have been consumed. At the core, the Nevow traversal
API is very simple. However, it provides some higher level functionality layered
on top of this to satisfy common use cases.

* `Object Traversal Basics`_
* `locateChild in depth`_
* `childFactory method`_
* `child_* methods and attributes`_
* `Dots in child names`_
* `children dictionary`_
* `The default trailing slash handler`_
* `ICurrentSegments and IRemainingSegments`_

Object Traversal Basics
-----------------------

The *root resource* is the top-level object in the URL space; it conceptually
represents the URI "/". The Nevow *object traversal* and *object publishing*
machinery uses only two methods to locate an object suitable for publishing and
to generate the HTML from it; these methods are described in the interface
``nevow.inevow.IResource``::


  class IResource(compy.Interface):
      def locateChild(self, ctx, segments):
          """Locate another object which can be adapted to IResource
          Return a tuple of resource, path segments
          """

      def renderHTTP(self, ctx):
          """Render a request
          """

``renderHTTP`` can be as simple as a method which simply returns a string of HTML.
Let's examine what happens when object traversal occurs over a very simple root
resource::

  from zope.interface import implements

  class SimpleRoot(object):
      implements(inevow.IResource)

      def locateChild(self, ctx, segments):
          return self, ()

      def renderHTTP(self, ctx):
          return "Hello, world!"

This resource, when passed as the root resource to ``appserver.NevowSite`` or
``wsgi.createWSGIApplication``, will immediately return itself, consuming all path
segments. This means that for every URI a user visits on a web server which is
serving this root resource, the text "Hello, world!" will be rendered. Let's
examine the value of ``segments`` for various values of URI:

/foo/bar
  ('foo', 'bar')

/
  ('', )

/foo/bar/baz.html
  ('foo', 'bar', 'baz.html')

/foo/bar/directory/
  ('foo', 'bar', 'directory', '')

So we see that Nevow does nothing more than split the URI on the string '/' and
pass these path segments to our application for consumption. Armed with these
two methods alone, we already have enough information to write applications
which service any form of URL imaginable in any way we wish. However, there are
some common URL handling patterns which Nevow provides higher level support for.

``locateChild`` in depth
------------------------

One common URL handling pattern involves parents which only know about their
direct children. For example, a ``Directory`` object may only know about the
contents of a single directory, but if it contains other directories, it does
not know about the contents of them. Let's examine a simple ``Directory`` object
which can provide directory listings and serves up objects for child directories
and files::

  from zope.interface import implements            

  class Directory(object):
      implements(inevow.IResource)

      def __init__(self, directory):
          self.directory = directory

      def renderHTTP(self, ctx):
          html = ['<ul>']
          for child in os.listdir(self.directory):
              fullpath = os.path.join(self.directory, child)
              if os.path.isdir(fullpath):
                  child += '/'
              html.extend(['<li><a href="', child, '">', child, '</a></li>'])
          html.append('</ul>')
          return ''.join(html)

      def locateChild(self, ctx, segments):
          name = segments[0]
          fullpath = os.path.join(self.directory, name)
          if not os.path.exists(fullpath):
              return None, () # 404

          if os.path.isdir(fullpath):
              return Directory(fullpath), segments[1:]
          if os.path.isfile(fullpath):
              return static.File(fullpath), segments[1:]

Because this implementation of ``locateChild`` only consumed one segment and
returned the rest of them (``segments[1:]``), the object traversal process will
continue by calling ``locateChild`` on the returned resource and passing the
partially-consumed segments. In this way, a directory structure of any depth can
be traversed, and directory listings or file contents can be rendered for any
existing directories and files.

So, let us examine what happens when the URI "/foo/bar/baz.html" is traversed,
where "foo" and "bar" are directories, and "baz.html" is a file.

Directory('/').locateChild(ctx, ('foo', 'bar', 'baz.html'))
    Returns Directory('/foo'), ('bar', 'baz.html')

Directory('/foo').locateChild(ctx, ('bar', 'baz.html'))
    Returns Directory('/foo/bar'), ('baz.html, )

Directory('/foo/bar').locateChild(ctx, ('baz.html'))
    Returns File('/foo/bar/baz.html'), ()

No more segments to be consumed; ``File('/foo/bar/baz.html').renderHTTP(ctx)`` is
called, and the result is sent to the browser.
                        
``childFactory`` method
-----------------------

Consuming one URI segment at a time by checking to see if a requested resource
exists and returning a new object is a very common pattern. Nevow's default
implementation of ``IResource``, ``nevow.rend.Page``, contains an implementation of
``locateChild`` which provides more convenient hooks for implementing object
traversal. One of these hooks is ``childFactory``. Let us imagine for the sake of
example that we wished to render a tree of dictionaries. Our data structure
might look something like this::

    tree = dict(
        one=dict(
            foo=None,
            bar=None),
        two=dict(
            baz=dict(
                quux=None)))

Given this data structure, the valid URIs would be:

* /
* /one
* /one/foo
* /one/bar
* /two
* /two/baz
* /two/baz/quux

Let us construct a ``rend.Page`` subclass which uses the default ``locateChild``
implementation and overrides the ``childFactory`` hook instead::

  class DictTree(rend.Page):
      def __init__(self, dataDict):
          self.dataDict = dataDict

      def renderHTTP(self, ctx):
          if self.dataDict is None:
              return "Leaf"
          html = ['<ul>']
          for key in self.dataDict.keys():
              html.extend(['<li><a href="', key, '">', key, '</a></li>'])
          html.append('</ul>')
          return ''.join(html)

      def childFactory(self, ctx, name):
          if name not in self.dataDict:
              return rend.NotFound # 404
          return DictTree(self.dataDict[name])

As you can see, the ``childFactory`` implementation is considerably shorter than the
equivalent ``locateChild`` implementation would have been.

``child_*`` methods and attributes
----------------------------------

Often we may wish to have some hardcoded URLs which are not dynamically
generated based on some data structure. For example, we might have an
application which uses an external CSS stylesheet, an external JavaScript file,
and a folder full of images. The ``rend.Page`` ``locateChild`` implementation provides a
convenient way for us to express these relationships by using ``child``-prefixed
methods::

  class Linker(rend.Page):
      def renderHTTP(self, ctx):
          return """<html>
    <head>
      <link href="css" rel="stylesheet" />
      <script type="text/javascript" src="scripts" />
    <body>
      <img src="images/logo.png" />
    </body>
  </html>"""

      def child_css(self, ctx):
          return static.File('/Users/dp/styles.css')

      def child_scripts(self, ctx):
          return static.File('/Users/dp/scripts.js')

      def child_images(self, ctx):
          return static.File('/Users/dp/images/')

One thing you may have noticed is that all of the examples so far have returned
new object instances whenever they were implementing a traversal API. However,
there is no reason these instances cannot be shared. One could for example
return a global resource instance, an instance which was previously inserted in
a dict, or lazily create and cache dynamic resource instances on the fly. The
``rend.Page`` ``locateChild`` implementation also provides a convenient way to express
that one global resource instance should always be used for a particular url,
the ``child``-prefixed attribute::

  class FasterLinker(Linker):
      child_css = static.File('/Users/dp/styles.css')
      child_scripts = static.File('/Users/dp/scripts.js')
      child_images = static.File('/Users/dp/images/')

Dots in child names
-------------------

When a URL contains dots, which is quite common in normal URLs, it is simple
enough to handle these URL segments in ``locateChild`` or ``childFactory`` -- one of the
passed segments will simply be a string containing a dot. However, it is not
immediately obvious how one would express a URL segment with a dot in it when
using ``child``-prefixed methods. The solution is really quite simple::

  class DotChildren(rend.Page):
      return '<html><head><script type="text/javascript" src="scripts.js" /></head></html>'

  setattr(DotChildren, 'child_scripts.js', static.File('/Users/dp/scripts.js'))

The same technique could be used to install a child method with a dot in the
name.

children dictionary
-------------------

The final hook supported by the default implementation of locateChild is the
``rend.Page.children`` dictionary::

  class Main(rend.Page):
      children = {
          'people': People(),
          'jobs': Jobs(),
          'events': Events()}

      def renderHTTP(self, ctx):
          return """\
<html>
    <head>
        <title>Our Site</title>
    </head>
    <body>
        <p>bla bla bla</p>
    </body>
</html>"""


Hooks are checked in the following order:

  1. ``self.dictionary``
  2. ``self.child_*``
  3. ``self.childFactory``

The default trailing slash handler
----------------------------------

When a URI which is being handled ends in a slash, such as when the '/' URI is
being rendered or when a directory-like URI is being rendered, the string ''
appears in the path segments which will be traversed. Again, handling this case
is trivial inside either ``locateChild`` or ``childFactory``, but it may not be
immediately obvious what ``child``-prefixed method or attribute will be looked up.
The method or attribute name which will be used is simply ``child`` with a single
trailing underscore.

The ``rend.Page`` class provides an implementation of this method which can work in
two different ways. If the attribute ``addSlash`` is True, the default trailing
slash handler will return ``self``. In the case when ``addSlash`` is True, the default
``rend.Page.renderHTTP`` implementation will simply perform a redirect which adds
the missing slash to the URL.

The default trailing slash handler also returns self if ``addSlash`` is false, but
emits a warning as it does so. This warning may become an exception at some
point in the future.

``ICurrentSegments`` and ``IRemainingSegments``
-----------------------------------------------

During the object traversal process, it may be useful to discover which segments
have already been handled and which segments are remaining to be handled. This
information may be obtained from the ``context`` object which is passed to all the
traversal APIs. The interfaces ``nevow.inevow.ICurrentSegments`` and
``nevow.inevow.IRemainingSegments`` are used to retrieve this information. To
retrieve a tuple of segments which have previously been consumed during object
traversal, use this syntax::

  segs = ICurrentSegments(ctx)

The same is true of ``IRemainingSegments``. ``IRemainingSegments`` is the same value
which is passed as ``segments`` to ``locateChild``, but may also be useful in the
implementations of ``childFactory`` or a ``child``-prefixed method, where this
information would not otherwise be available.
 
Conclusion
==========

Nevow makes it easy to handle complex URL hierarchies. The most basic object
traversal interface, ``nevow.inevow.IResource.locateChild``, provides powerful and
flexible control over the entire object traversal process. Nevow's canonical
``IResource`` implementation, ``rend.Page``, also includes the convenience hooks
``childFactory`` along with ``child``-prefixed method and attribute semantics to
simplify common use cases.