Object Publishing

In Object Traversal , we learned about the nevow.inevow.IResource.renderHTTP method, which is the most basic way to send HTML to a browser when using Nevow. However, it is not very convenient (or clean) to generate HTML tags by concatenating strings in Python code. In the Deployment documentation, we saw that it was possible to render a Hello World page using a nevow.rend.Page subclass and providing a docFactory :

>>> from nevow import rend, loaders
>>> class HelloWorld(rend.Page):
...     docFactory = loaders.stan("Hello, world!")
>>> HelloWorld().renderSynchronously()
'Hello, world!'

This example does nothing interesting, but the concept of a loader is important in Nevow. The rend.Page.renderHTTP implementation always starts rendering HTML by loading a template from the docFactory .

The stan DOM

Nevow uses a DOM-based approach to rendering HTML. A tree of objects is first constructed in memory by the template loader. This tree is then processed one node at a time, applying functions which transform from various Python types to HTML strings.

Nevow uses a nonstandard DOM named “stan”. Unlike the W3C DOM, stan is made up of simple python lists, strings, and instances of the nevow.stan.Tag class. During the rendering process, “Flattener” functions convert from rich types to HTML strings. For example, we can load a template made up of some nested lists and Python types, render it, and see what happens:

>>> class PythonTypes(rend.Page):
...     docFactory = loaders.stan(["Hello", 1, 1.5, True, ["Goodbye", 3]])
>>> PythonTypes().renderSynchronously()

Tag instances

So far, we have only rendered simple strings as output. However, the main purpose of Nevow is HTML generation. In the stan DOM, HTML tags are represented by instances of the nevow.stan.Tag class. Tag is a very simple class, whose instances have an attributes dictionary and a children list. The Tag flattener knows how to recursively flatten attributes and children of the tag. To show you how Tag s really work before you layer Nevow’s convenience syntax on top, try this horrible example:

>>> from nevow import stan
>>> h = stan.Tag('html')
>>> d = stan.Tag('div')
>>> d.attributes['style'] = 'border: 1px solid black'
>>> h.children.append(d)
>>> class Tags(rend.Page):
...     docFactory = loaders.stan(h)
>>> Tags().renderSynchronously()
'<html><div style="border: 1px solid black"></div></html>'

So, we see how it is possible to programatically generate HTML by constructing and nesting stan Tag instances. However, it is far more convenient to use the overloaded operators Tag provides to manipulate them. Tag implements a __call__ method which takes any keyword arguments and values and updates the attributes dictionary; it also implements a __getitem__ method which takes whatever is between the square brackets and appends them to the children list. A simple example should clarify things:

>>> class Tags2(rend.Page):
...     docFactory = loaders.stan(stan.Tag('html')[stan.Tag('div')(style="border: 1px solid black")])
>>> Tags2().renderSynchronously()
'<html><div style="border: 1px solid black"></div></html>'

This isn’t very easy to read, but luckily we can simplify the example even further by using the nevow.tags module, which is full of “Tag prototypes” for every tag type described by the XHTML 1.0 specification:

>>> class Tags3(rend.Page):
...     docFactory = loaders.stan(tags.html[tags.div(style="border: 1px solid black")])
>>> Tags3().renderSynchronously()
'<html><div style="border: 1px solid black"></div></html>'

Using stan syntax is not the only way to construct template DOM for use by the Nevow rendering process. Nevow also includes loaders.xmlfile which implements a simple tag attribute language similar to the Zope Page Templates (ZPT) Tag Attribute Language (TAL). However, experience with the stan DOM should give you insight into how the Nevow rendering process really works. Rendering a template into HTML in Nevow is really nothing more than iterating a tree of objects and recursively applying “Flattener” functions to objects in this tree, until all HTML has been generated.

Functions in the DOM

So far, all of our examples have generated static HTML pages, which is not terribly interesting when discussing dynamic web applications. Nevow takes a very simple approach to dynamic HTML generation. If you put a Python function reference in the DOM, Nevow will call it when the page is rendered. The return value of the function replaces the function itself in the DOM, and the results are flattened further. This makes it easy to express looping and branching structures in Nevow, because normal Python looping and branching constructs are used to do the job:

>>> def repeat(ctx, data):
...     return [tags.div(style="color: %s" % (color, ))
...         for color in ['red', 'blue', 'green']]
>>> class Repeat(rend.Page):
...     docFactory = loaders.stan(tags.html[repeat])
>>> Repeat().renderSynchronously()
'<html><div style="color: red"></div><div style="color: blue"></div><div style="color: green"></div></html>'

However, in the example above, the repeat function isn’t even necessary, because we could have inlined the list comprehension right where we placed the function reference in the DOM. Things only really become interesting when we begin writing parameterized render functions which cause templates to render differently depending on the input to the web application.

The required signature of functions which we can place in the DOM is (ctx, data). The “context” object is essentially opaque for now, and we will learn how to extract useful information out of it later. The “data” object is anything we want it to be, and can change during the rendering of the page. By default, the data object is whatever we pass as the first argument to the Page constructor, or the Page instance itself if nothing is passed. Armed with this knowledge, we can create a Page which renders differently depending on the data we pass to the Page constructor:

class Root(rend.Page):
    docFactory = loaders.stan(tags.html[

    def childFactory(self, ctx, name):
        return Leaf(name)

def greet(ctx, name):
    return "Hello. You are visiting the ", name, " page."

class Leaf(rend.Page):
    docFactory = loaders.stan(tags.html[greet])

Armed with this knowledge and the information in the Object Traversal documentation, we now have enough information to create dynamic websites with arbitrary URL hierarchies whose pages render dynamically depending on which URL was used to access them.

Accessing query parameters and form post data

Before we move on to more advanced rendering techniques, let us first examine how one could further customize the rendering of a Page based on the URL query parameters and form post information provided to us by a browser. Recall that URL parameters are expressed in the form:


And form post data can be generated by providing a form to a browser:

<form action="" method="POST">
  <input type="text" name="baz" />
  <input type="text" name="quux" />
  <input type="submit" />

Accessing this information is such a common procedure that Nevow provides a convenience method on the context to do it. Let’s examine a simple page whose output can be influenced by the query parameters in the URL used to access it:

def showChoice(ctx, data):
    choice = ctx.arg('choice')
    if choice is None:
        return ''
    return "You chose ", choice, "."

class Custom(rend.Page):
    docFactory = loaders.stan(tags.html[

The procedure is exactly the same for simple form post information:

def greet(ctx, data):
    name = ctx.arg('name')
    if name is None:
        return ''
    return "Greetings, ", name, "!"

class Form(rend.Page):
    docFactory = loaders.stan(tags.html[
        tags.form(action="", method="POST")[

Note that ctx.arg returns only the first argument with the given name. For complex cases where multiple arguments and lists of argument values are required, you can access the request argument dictionary directly using the syntax:

def arguments(ctx, data):
    args = inevow.IRequest(ctx).args
    return "Request arguments are: ", str(args)

Generators in the DOM

One common operation when building dynamic pages is iterating a list of data and emitting some HTML for each item. Python generators are well suited for expressing this sort of logic, and code which is written as a python generator can perform tests (if ) and loops of various kinds (while , for ) and emit a row of html whenever it has enough data to do so. Nevow can handle generators in the DOM just as gracefully as it can handle anything else:

>>> from nevow import rend, loaders, tags
>>> def generate(ctx, items):
...     for item in items:
...         yield tags.div[ item ]
>>> class List(rend.Page):
...     docFactory = loaders.stan(tags.html[ generate ])
>>> List(['one', 'two', 'three']).renderSynchronously()

As you can see, generating HTML inside of functions or generators can be very convenient, and can lead to very rapid application development. However, it is also what I would call a “template abstraction violation”, and we will learn how we can keep knowledge of HTML out of our python code when we learn about patterns and slots.

Methods in the DOM

Up until now, we have been placing our template manipulation logic inside of simple Python functions and generators. However, it is often appropriate to use a method instead of a function. Nevow makes it just as easy to use a method to render HTML:

class MethodRender(rend.Page):
    def __init__(self, foo):
        self.foo = foo

    def render_foo(self, ctx, data):
        return self.foo

    docFactory = loaders.stan(tags.html[ render_foo ])

Using render methods makes it possible to parameterize your Page class with more parameters. With render methods, you can also use the Page instance as a state machine to keep track of the state of the render. While Nevow is designed to allow you to render the same Page instance repeatedly, it can also be convenient to know that a Page instance will only be used one time, and that the Page instance can be used as a scratch pad to manage information about the render.

Data specials

Previously we saw how passing a parameter to the default Page constructor makes it available as the “data” parameter to all of our render methods. This “data” parameter can change as the page render proceeds, and is a useful way to ensure that render functions are isolated and only act upon the data which is available to them. Render functions which do not pull information from sources other than the “data” parameter are more easily reusable and can be composed into larger parts more easily.

Deciding which data gets passed as the data parameter is as simple as changing the “Data special” for a Tag. See the Glossary under “Tag Specials” for more information about specials. Assigning to the data special is as simple as assigning to a tag attribute:

>>> def hello(ctx, name):
...     return "Hello, ", name
>>> class DataSpecial(rend.Page):
...     docFactory = loaders.stan(tags.html[
...     tags.div(data="foo")[ hello ],
...     tags.div(data="bar")[ hello ]])
>>> DataSpecial().renderSynchronously()
'<html><div>Hello, foo</div><div>Hello, bar</div></html>'

Data specials may be assigned any python value. Data specials are only in scope during the rendering of the tag they are assigned to, so if the “hello” renderer were placed in the DOM inside the html node directly, “Hello, None” would be output.

Before data is passed to a render function, Nevow first checks to see if there is an IGettable adapter for it. If there is, it calls IGettable.get() , and passes the result of this as the data parameter instead. Nevow includes an IGettable adapter for python functions, which means you can set a Tag data special to a function reference and Nevow will call it to obtain the data when the Tag is rendered. The signature for data methods is similar to that of render methods, (ctx, data). For example:

def getName(ctx, data):
    return ctx.arg('name')

def greet(ctx, name):
    return "Greetings, ", name

class GreetName(rend.Page):
    docFactory = loaders.stan(tags.html[
            tags.div(data=getName)[ greet ]])

Data specials exist mainly to allow you to construct and enforce a Model-View-Controller style separation of the Model code from the View. Here we see that the greet function is capable of rendering a greeting view for a name model, and that the implementation of getName may change without the view code changing.

Render specials

Previously, we have seen how render functions can be placed directly in the DOM, and the return value replaces the render function in the DOM. However, these free functions and methods are devoid of any contextual information about the template they are living in. The render special is a way to associate a render function or method with a particular Tag instance, which the render function can then examine to decide how to render:

>>> def alignment(ctx, data):
...     align = ctx.tag.attributes.get('align')
...     if align == 'right':
...         return ctx.tag["Aligned right"]
...     elif align == 'center':
...         return ctx.tag["Aligned center"]
...     else:
...         return ctx.tag["Aligned left"]
>>> class AlignmentPage(rend.Page):
...     docFactory = loaders.stan(tags.html[
...     tags.p(render=alignment),
...     tags.p(render=alignment, align="center"),
...     tags.p(render=alignment, align="right")])
>>> AlignmentPage().renderSynchronously()
'<html><p>Aligned left</p><p align="center">Aligned center</p><p align="right">Aligned right</p></html>'

Note how the alignment renderer has access to the template node as ctx.tag . It can examine and change this node, and the return value of the render function replaces the original node in the DOM. Note that here we are returning the template node after changing it. We will see later how we can instead mutate the context and use slots so that the knowledge the renderer requires about the structure of the template is reduced even more.

Pattern specials

When writing render methods, it is easy to inline the construction of Tag instances to generate HTML programatically. However, this creates a template abstraction violation, where part of the HTML which will show up in the final page output is hidden away inside of render methods instead of inside the template. Pattern specials are designed to avoid this problem. A node which has been tagged with a pattern special can then be located and copied by a render method. The render method does not need to know anything about the structure or location of the pattern, only it’s name.

We can rewrite our previous generator example so that the generator does not have to know what type of tag the template designer would like repeated for each item in the list:

>>> from nevow import rend, loaders, tags, inevow
>>> def generate(ctx, items):
...     pat = inevow.IQ(ctx).patternGenerator('item')
...     for item in items:
...         ctx.tag[ pat(data=item) ]
...     return ctx.tag
>>> def string(ctx, item):
...     return ctx.tag[ str(item) ]
>>> class List(rend.Page):
...     docFactory = loaders.stan(tags.html[
...     tags.ul(render=generate)[
...         tags.li(pattern="item", render=string)]])
>>> List([1, 2, 3]).renderSynchronously()

Note that we have to mutate the tag in place and repeatedly copy the item pattern, applying the item as the data special to the resulting Tag. It turns out that this is such a common operation that nevow comes out of the box with these two render functions:

>>> class List(rend.Page):
...     docFactory = loaders.stan(tags.html[
...     tags.ul(render=rend.sequence)[
...         tags.li(pattern="item", render=rend.data)]])
>>> List([1, 2, 3]).renderSynchronously()

Slot specials

The problem with render methods is that they are only capable of making changes to their direct children. Because of the architecture of Nevow, they should not attempt to change grandchildren or parent nodes. It is possible to write one render method for every node you wish to change, but there is a better way. A node with a slot special can be “filled” with content by any renderer above the slot. Creating a slot special is such a frequent task that there is a prototype in nevow.tags which is usually used.

Let us examine a renderer which fills a template with information about a person:

>>> from nevow import loaders, rend, tags
>>> person = ('Donovan', 'Preston', 'Male', 'California')
>>> def render_person(ctx, person):
...     firstName, lastName, sex, location = person
...     ctx.fillSlots('firstName', firstName)
...     ctx.fillSlots('lastName', lastName)
...     ctx.fillSlots('sex', sex)
...     ctx.fillSlots('location', location)
...     return ctx.tag
>>> class PersonPage(rend.Page):
...     docFactory = loaders.stan(tags.html(render=render_person)[
...     tags.table[
...         tags.tr[
...             tags.td[tags.slot('firstName')],
...             tags.td[tags.slot('lastName')],
...             tags.td[tags.slot('sex')],
...             tags.td[tags.slot('location')]]]])
>>> PersonPage(person).renderSynchronously()

Using patterns in combination with slots can lead to very powerful template abstraction. Nevow also includes another standard renderer called “mapping” which takes any data which responds to the “items()” message and inserts the items into appropriate slots:

>>> class DictPage(rend.Page):
...     docFactory = loaders.stan(tags.html(render=rend.mapping)[
...         tags.span[ tags.slot('foo') ], tags.span[ tags.slot('bar') ]])
>>> DictPage(dict(foo=1, bar=2)).renderSynchronously()

Data directives

So far, we have always placed data functions directly in the Data special attribute of a Tag. Sometimes, it is preferable to look up a data method from the Page class as the Page has being rendered. For example, a base class may define a template and a subclass may provide the implementation of the data method. We can accomplish this effect by using a data directive as a Tag’s data special:

class Base(rend.Page):
    docFactory = loaders.stan(tags.html[
        tags.div(data=tags.directive('name'), render=rend.data)])

class Subclass(Base):
    def data_name(self, ctx, data):
        return "Your name"

The data directive is resolved by searching for the IContainer implementation in the context. rend.Page implements IContainer.get by performing an attribute lookup on the Page with the prefix ‘data_*’. You can provide your own IContainer implementation if you wish, and also you should know that IContainer implementations for list and dict are included in the nevow.accessors module.

A common gotcha is that the closest IContainer is used to resolve data directives. This means that if a list is being used as the data during the rendering process, data directives below this will be resolved against the IContainer implementation in nevow.accessors.ListAccessor . If you are expecting a data directive to invoke a Page’s data_* method but instead get a KeyError , this is why.

Render directives

Render directives are almost exactly the same, except they are resolved using the closest IRendererFactory implementation in the context. Render directives can be used to allow subclasses to override certain render methods, and also can be used to allow Fragments to locate their own prefixed render methods.


TODO This section isn’t done yet.

Nevow’s flatteners use a type/function registry to determine how to render objects which Nevow encounters in the DOM during the rendering process. “Explicit is better than implicit”, so in most cases, explicitly applying render methods to data will be better than registering a flattener, but in some cases it can be useful:

class Person(object):
    def __init__(self, firstName, lastName):
        self.firstName = firstName
        self.lastName = lastName

def flattenPerson(person, ctx):
    return flat.partialflatten(ctx, (person.firstName, " ", person.lastName))

from nevow import flat
flat.registerFlattener(flattenPerson, Person)

def insertData(ctx, data):
    return data

class PersonPage(rend.Page):
    docFactory = loaders.stan(tags.html[insertData])