<script src="http://www.google-analytics.com/urchin.js" type="text/javascript"> </script> <script type="text/javascript"> _uacct = "UA-99018-3"; urchinTracker(); </script>
1 | Nevow Object Publishing |
---|---|
2 | ======================= |
3 | |
4 | In Nevow Object Traversal, we learned about the |
5 | nevow.inevow.IResource.renderHTTP method, which is the most basic way to send |
6 | HTML to a browser when using Nevow. However, it is not very convenient (or |
7 | clean) to generate HTML tags by concatenating strings in Python code. In the |
8 | Nevow Deployment documentation, we saw that it was possible to render a Hello |
9 | World page using a nevow.rend.Page subclass and providing a "docFactory":: |
10 | |
11 | >>> from nevow import rend, loaders |
12 | >>> class HelloWorld(rend.Page): |
13 | ... docFactory = loaders.stan("Hello, world!") |
14 | ... |
15 | >>> HelloWorld().renderSynchronously() |
16 | 'Hello, world!' |
17 | |
18 | This example does nothing interesting, but the concept of a loader is important |
19 | in Nevow. The rend.Page.renderHTTP implementation always starts rendering HTML |
20 | by loading a template from the docFactory. |
21 | |
22 | * `The stan DOM`_ |
23 | * `Tag instances`_ |
24 | * `Functions in the DOM`_ |
25 | * `Accessing query parameters and form post data`_ |
26 | * `Generators in the DOM`_ |
27 | * `Methods in the DOM`_ |
28 | * `Data specials`_ |
29 | * `Render specials`_ |
30 | * `Pattern specials`_ |
31 | * `Slot specials`_ |
32 | * `Data directives`_ |
33 | * `Render directives`_ |
34 | * `Flatteners`_ |
35 | |
36 | The stan DOM |
37 | ------------ |
38 | |
39 | Nevow uses a DOM-based approach to rendering HTML. A tree of objects is first |
40 | constructed in memory by the template loader. This tree is then processed one |
41 | node at a time, applying functions which transform from various Python types to |
42 | HTML strings. |
43 | |
44 | Nevow uses a nonstandard DOM named "stan". Unlike the W3C DOM, stan is made up |
45 | of simple python lists, strings, and instances of the nevow.stan.Tag class. |
46 | During the rendering process, "Flattener" functions convert from rich types to |
47 | HTML strings. For example, we can load a template made up of some nested lists |
48 | and Python types, render it, and see what happens:: |
49 | |
50 | >>> class PythonTypes(rend.Page): |
51 | ... docFactory = loaders.stan(["Hello", 1, 1.5, True, ["Goodbye", 3]]) |
52 | ... |
53 | >>> PythonTypes().renderSynchronously() |
54 | 'Hello11.5TrueGoodbye3' |
55 | |
56 | Tag instances |
57 | ------------- |
58 | |
59 | So far, we have only rendered simple strings as output. However, the main |
60 | purpose of Nevow is HTML generation. In the stan DOM, HTML tags are represented |
61 | by instances of the nevow.stan.Tag class. Tag is a very simple class, whose |
62 | instances have an "attributes" dictionary and a "children" list. The Tag |
63 | flattener knows how to recursively flatten attributes and children of the tag. |
64 | To show you how Tags really work before you layer Nevow's convenience syntax on |
65 | top, try this horrible example:: |
66 | |
67 | >>> from nevow import stan |
68 | >>> h = stan.Tag('html') |
69 | >>> d = stan.Tag('div') |
70 | >>> d.attributes['style'] = 'border: 1px solid black' |
71 | >>> h.children.append(d) |
72 | >>> class Tags(rend.Page): |
73 | ... docFactory = loaders.stan(h) |
74 | ... |
75 | >>> Tags().renderSynchronously() |
76 | '<html><div style="border: 1px solid black"></div></html>' |
77 | |
78 | So, we see how it is possible to programatically generate HTML by constructing |
79 | and nesting stan Tag instances. However, it is far more convenient to use the |
80 | overloaded operators Tag provides to manipulate them. Tag implements a __call__ |
81 | method which takes any keyword arguments and values and updates the attributes |
82 | dictionary; it also implements a __getitem__ method which takes whatever is |
83 | between the square brackets and appends them to the children list. A simple |
84 | example should clarify things:: |
85 | |
86 | >>> class Tags2(rend.Page): |
87 | ... docFactory = loaders.stan(stan.Tag('html')[stan.Tag('div')(style="border: 1px solid black")]) |
88 | ... |
89 | >>> Tags2().renderSynchronously() |
90 | '<html><div style="border: 1px solid black"></div></html>' |
91 | |
92 | This isn't very easy to read, but luckily we can simplify the example even |
93 | further by using the nevow.tags module, which is full of "Tag prototypes" for |
94 | every tag type described by the XHTML 1.0 specification:: |
95 | |
96 | >>> class Tags3(rend.Page): |
97 | ... docFactory = loaders.stan(tags.html[tags.div(style="border: 1px solid black")]) |
98 | ... |
99 | >>> Tags3().renderSynchronously() |
100 | '<html><div style="border: 1px solid black"></div></html>' |
101 | |
102 | Using stan syntax is not the only way to construct template DOM for use by the |
103 | Nevow rendering process. Nevow also includes loaders.xmlfile which implements a |
104 | simple tag attribute language similar to the Zope Page Templates (ZPT) Tag |
105 | Attribute Language (TAL). However, experience with the stan DOM should give you |
106 | insight into how the Nevow rendering process really works. Rendering a template |
107 | into HTML in Nevow is really nothing more than iterating a tree of objects and |
108 | recursively applying "Flattener" functions to objects in this tree, until all |
109 | HTML has been generated. |
110 | |
111 | Functions in the DOM |
112 | -------------------- |
113 | |
114 | So far, all of our examples have generated static HTML pages, which is not |
115 | terribly interesting when discussing dynamic web applications. Nevow takes a |
116 | very simple approach to dynamic HTML generation. If you put a Python function |
117 | reference in the DOM, Nevow will call it when the page is rendered. The return |
118 | value of the function replaces the function itself in the DOM, and the results |
119 | are flattened further. This makes it easy to express looping and branching |
120 | structures in Nevow, because normal Python looping and branching constructs are |
121 | used to do the job:: |
122 | |
123 | >>> def repeat(ctx, data): |
124 | ... return [tags.div(style="color: %s" % (color, )) |
125 | ... for color in ['red', 'blue', 'green']] |
126 | ... |
127 | >>> class Repeat(rend.Page): |
128 | ... docFactory = loaders.stan(tags.html[repeat]) |
129 | ... |
130 | >>> Repeat().renderSynchronously() |
131 | '<html><div style="color: red"></div><div style="color: blue"></div><div style="color: green"></div></html>' |
132 | |
133 | However, in the example above, the repeat function isn't even necessary, because |
134 | we could have inlined the list comprehension right where we placed the function |
135 | reference in the DOM. Things only really become interesting when we begin |
136 | writing parameterized render functions which cause templates to render |
137 | differently depending on the input to the web application. |
138 | |
139 | The required signature of functions which we can place in the DOM is (ctx, |
140 | data). The "context" object is essentially opaque for now, and we will learn how |
141 | to extract useful information out of it later. The "data" object is anything we |
142 | want it to be, and can change during the rendering of the page. By default, the |
143 | data object is whatever we pass as the first argument to the Page constructor, |
144 | **or** the Page instance itself if nothing is passed. Armed with this knowledge, |
145 | we can create a Page which renders differently depending on the data we pass to |
146 | the Page constructor:: |
147 | |
148 | class Root(rend.Page): |
149 | docFactory = loaders.stan(tags.html[ |
150 | tags.h1["Welcome."], |
151 | tags.a(href="foo")["Foo"], |
152 | tags.a(href="bar")["Bar"], |
153 | tags.a(href="baz")["Baz"]]) |
154 | |
155 | def childFactory(self, ctx, name): |
156 | return Leaf(name) |
157 | |
158 | |
159 | def greet(ctx, name): |
160 | return "Hello. You are visiting the ", name, " page." |
161 | |
162 | class Leaf(rend.Page): |
163 | docFactory = loaders.stan(tags.html[greet]) |
164 | |
165 | Armed with this knowledge and the information in the Object Traversal |
166 | documentation, we now have enough information to create dynamic websites with |
167 | arbitrary URL hierarchies whose pages render dynamically depending on which URL |
168 | was used to access them. |
169 | |
170 | Accessing query parameters and form post data |
171 | --------------------------------------------- |
172 | |
173 | Before we move on to more advanced rendering techniques, let us first examine |
174 | how one could further customize the rendering of a Page based on the URL query |
175 | parameters and form post information provided to us by a browser. Recall that |
176 | URL parameters are expressed in the form:: |
177 | |
178 | http://example.com/foo/bar?baz=1&quux=2 |
179 | |
180 | And form post data can be generated by providing a form to a browser:: |
181 | |
182 | <form action="" method="POST"> |
183 | <input type="text" name="baz" /> |
184 | <input type="text" name="quux" /> |
185 | <input type="submit" /> |
186 | </form> |
187 | |
188 | Accessing this information is such a common procedure that Nevow provides a |
189 | convenience method on the context to do it. Let's examine a simple page whose |
190 | output can be influenced by the query parameters in the URL used to access it:: |
191 | |
192 | def showChoice(ctx, data): |
193 | choice = ctx.arg('choice') |
194 | if choice is None: |
195 | return '' |
196 | return "You chose ", choice, "." |
197 | |
198 | class Custom(rend.Page): |
199 | docFactory = loaders.stan(tags.html[ |
200 | tags.a(href="?choice=baz")["Baz"], |
201 | tags.a(href="?choice=quux")["Quux"], |
202 | tags.p[showChoice]]) |
203 | |
204 | The procedure is exactly the same for simple form post information:: |
205 | |
206 | def greet(ctx, data): |
207 | name = ctx.arg('name') |
208 | if name is None: |
209 | return '' |
210 | return "Greetings, ", name, "!" |
211 | |
212 | class Form(rend.Page): |
213 | docFactory = loaders.stan(tags.html[ |
214 | tags.form(action="", method="POST")[ |
215 | tags.input(name="name"), |
216 | tags.input(type="submit")], |
217 | greet]) |
218 | |
219 | Note that ctx.arg returns only the first argument with the given name. For |
220 | complex cases where multiple arguments and lists of argument values are |
221 | required, you can access the request argument dictionary directly using the |
222 | syntax:: |
223 | |
224 | def arguments(ctx, data): |
225 | args = inevow.IRequest(ctx).args |
226 | return "Request arguments are: ", str(args) |
227 | |
228 | Generators in the DOM |
229 | --------------------- |
230 | |
231 | One common operation when building dynamic pages is iterating a list of data and |
232 | emitting some HTML for each item. Python generators are well suited for |
233 | expressing this sort of logic, and code which is written as a python generator |
234 | can perform tests (if) and loops of various kinds (while, for) and emit a row of |
235 | html whenever it has enough data to do so. Nevow can handle generators in the |
236 | DOM just as gracefully as it can handle anything else:: |
237 | |
238 | >>> from nevow import rend, loaders, tags |
239 | >>> def generate(ctx, items): |
240 | ... for item in items: |
241 | ... yield tags.div[ item ] |
242 | ... |
243 | >>> class List(rend.Page): |
244 | ... docFactory = loaders.stan(tags.html[ generate ]) |
245 | ... |
246 | >>> List(['one', 'two', 'three']).renderSynchronously() |
247 | '<html><div>one</div><div>two</div><div>three</div></html>' |
248 | |
249 | As you can see, generating HTML inside of functions or generators can be very |
250 | convenient, and can lead to very rapid application development. However, it is |
251 | also what I would call a "template abstraction violation", and we will learn how |
252 | we can keep knowledge of HTML out of our python code when we learn about |
253 | patterns and slots. |
254 | |
255 | Methods in the DOM |
256 | ------------------ |
257 | |
258 | Up until now, we have been placing our template manipulation logic inside of |
259 | simple Python functions and generators. However, it is often appropriate to use |
260 | a method instead of a function. Nevow makes it just as easy to use a method to |
261 | render HTML:: |
262 | |
263 | class MethodRender(rend.Page): |
264 | def __init__(self, foo): |
265 | self.foo = foo |
266 | |
267 | def render_foo(self, ctx, data): |
268 | return self.foo |
269 | |
270 | docFactory = loaders.stan(tags.html[ render_foo ]) |
271 | |
272 | Using render methods makes it possible to parameterize your Page class with more |
273 | parameters. With render methods, you can also use the Page instance as a state |
274 | machine to keep track of the state of the render. While Nevow is designed to |
275 | allow you to render the same Page instance repeatedly, it can also be convenient |
276 | to know that a Page instance will only be used one time, and that the Page |
277 | instance can be used as a scratch pad to manage information about the render. |
278 | |
279 | Data specials |
280 | ------------- |
281 | |
282 | Previously we saw how passing a parameter to the default Page constructor makes |
283 | it available as the "data" parameter to all of our render methods. This "data" |
284 | parameter can change as the page render proceeds, and is a useful way to ensure |
285 | that render functions are isolated and only act upon the data which is available |
286 | to them. Render functions which do not pull information from sources other than |
287 | the "data" parameter are more easily reusable and can be composed into larger |
288 | parts more easily. |
289 | |
290 | Deciding which data gets passed as the data parameter is as simple as changing |
291 | the "Data special" for a Tag. See the Glossary under "Tag Specials" for more |
292 | information about specials. Assigning to the data special is as simple as |
293 | assigning to a tag attribute:: |
294 | |
295 | >>> def hello(ctx, name): |
296 | ... return "Hello, ", name |
297 | ... |
298 | >>> class DataSpecial(rend.Page): |
299 | ... docFactory = loaders.stan(tags.html[ |
300 | ... tags.div(data="foo")[ hello ], |
301 | ... tags.div(data="bar")[ hello ]]) |
302 | ... |
303 | >>> DataSpecial().renderSynchronously() |
304 | '<html><div>Hello, foo</div><div>Hello, bar</div></html>' |
305 | |
306 | Data specials may be assigned any python value. Data specials are only in scope |
307 | during the rendering of the tag they are assigned to, so if the "hello" renderer |
308 | were placed in the DOM inside the html node directly, "Hello, None" would be |
309 | output. |
310 | |
311 | Before data is passed to a render function, Nevow first checks to see if there |
312 | is an IGettable adapter for it. If there is, it calls IGettable.get(), and |
313 | passes the result of this as the data parameter instead. Nevow includes an |
314 | IGettable adapter for python functions, which means you can set a Tag data |
315 | special to a function reference and Nevow will call it to obtain the data when |
316 | the Tag is rendered. The signature for data methods is similar to that of render |
317 | methods, (ctx, data). For example:: |
318 | |
319 | def getName(ctx, data): |
320 | return ctx.arg('name') |
321 | |
322 | def greet(ctx, name): |
323 | return "Greetings, ", name |
324 | |
325 | class GreetName(rend.Page): |
326 | docFactory = loaders.stan(tags.html[ |
327 | tags.form(action="")[ |
328 | tags.input(name="name"), |
329 | tags.input(type="submit")], |
330 | tags.div(data=getName)[ greet ]]) |
331 | |
332 | Data specials exist mainly to allow you to construct and enforce a |
333 | Model-View-Controller style separation of the Model code from the View. Here we |
334 | see that the greet function is capable of rendering a greeting view for a name |
335 | model, and that the implementation of getName may change without the view code |
336 | changing. |
337 | |
338 | Render specials |
339 | --------------- |
340 | |
341 | Previously, we have seen how render functions can be placed directly in the DOM, |
342 | and the return value replaces the render function in the DOM. However, these |
343 | free functions and methods are devoid of any contextual information about the |
344 | template they are living in. The render special is a way to associate a render |
345 | function or method with a particular Tag instance, which the render function can |
346 | then examine to decide how to render:: |
347 | |
348 | >>> def alignment(ctx, data): |
349 | ... align = ctx.tag.attributes.get('align') |
350 | ... if align == 'right': |
351 | ... return ctx.tag["Aligned right"] |
352 | ... elif align == 'center': |
353 | ... return ctx.tag["Aligned center"] |
354 | ... else: |
355 | ... return ctx.tag["Aligned left"] |
356 | ... |
357 | >>> class AlignmentPage(rend.Page): |
358 | ... docFactory = loaders.stan(tags.html[ |
359 | ... tags.p(render=alignment), |
360 | ... tags.p(render=alignment, align="center"), |
361 | ... tags.p(render=alignment, align="right")]) |
362 | ... |
363 | >>> AlignmentPage().renderSynchronously() |
364 | '<html><p>Aligned left</p><p align="center">Aligned center</p><p align="right">Aligned right</p></html>' |
365 | |
366 | Note how the alignment renderer has access to the template node as "ctx.tag". It |
367 | can examine and change this node, and the return value of the render function |
368 | replaces the original node in the DOM. Note that here we are returning the |
369 | template node after changing it. We will see later how we can instead mutate the |
370 | context and use slots so that the knowledge the renderer requires about the |
371 | structure of the template is reduced even more. |
372 | |
373 | Pattern specials |
374 | ---------------- |
375 | |
376 | When writing render methods, it is easy to inline the construction of Tag |
377 | instances to generate HTML programatically. However, this creates a template |
378 | abstraction violation, where part of the HTML which will show up in the final |
379 | page output is hidden away inside of render methods instead of inside the |
380 | template. Pattern specials are designed to avoid this problem. A node which has |
381 | been tagged with a pattern special can then be located and copied by a render |
382 | method. The render method does not need to know anything about the structure or |
383 | location of the pattern, only it's name. |
384 | |
385 | We can rewrite our previous generator example so that the generator does not |
386 | have to know what type of tag the template designer would like repeated for each |
387 | item in the list:: |
388 | |
389 | >>> from nevow import rend, loaders, tags, inevow |
390 | >>> def generate(ctx, items): |
391 | ... pat = inevow.IQ(ctx).patternGenerator('item') |
392 | ... for item in items: |
393 | ... ctx.tag[ pat(data=item) ] |
394 | ... return ctx.tag |
395 | ... |
396 | >>> def string(ctx, item): |
397 | ... return ctx.tag[ str(item) ] |
398 | ... |
399 | >>> class List(rend.Page): |
400 | ... docFactory = loaders.stan(tags.html[ |
401 | ... tags.ul(render=generate)[ |
402 | ... tags.li(pattern="item", render=string)]]) |
403 | ... |
404 | >>> List([1, 2, 3]).renderSynchronously() |
405 | '<html><ol><li>1</li><li>2</li><li>3</li></ol></html>' |
406 | |
407 | Note that we have to mutate the tag in place and repeatedly copy the item |
408 | pattern, applying the item as the data special to the resulting Tag. It turns |
409 | out that this is such a common operation that nevow comes out of the box with |
410 | these two render functions:: |
411 | |
412 | >>> class List(rend.Page): |
413 | ... docFactory = loaders.stan(tags.html[ |
414 | ... tags.ul(render=rend.sequence)[ |
415 | ... tags.li(pattern="item", render=rend.data)]]) |
416 | ... |
417 | >>> List([1, 2, 3]).renderSynchronously() |
418 | '<html><ul><li>1</li><li>2</li><li>3</li></ul></html>' |
419 | |
420 | Slot specials |
421 | ------------- |
422 | |
423 | The problem with render methods is that they are only capable of making changes |
424 | to their direct children. Because of the architecture of Nevow, they should not |
425 | attempt to change grandchildren or parent nodes. It is possible to write one |
426 | render method for every node you wish to change, but there is a better way. A |
427 | node with a slot special can be "filled" with content by any renderer above the |
428 | slot. Creating a slot special is such a frequent task that there is a prototype |
429 | in nevow.tags which is usually used. |
430 | |
431 | Let us examine a renderer which fills a template with information about a |
432 | person: |
433 | |
434 | >>> from nevow import loaders, rend, tags |
435 | ... |
436 | >>> person = ('Donovan', 'Preston', 'Male', 'California') |
437 | ... |
438 | >>> def render_person(ctx, person): |
439 | ... firstName, lastName, sex, location = person |
440 | ... ctx.fillSlots('firstName', firstName) |
441 | ... ctx.fillSlots('lastName', lastName) |
442 | ... ctx.fillSlots('sex', sex) |
443 | ... ctx.fillSlots('location', location) |
444 | ... return ctx.tag |
445 | ... |
446 | >>> class PersonPage(rend.Page): |
447 | ... docFactory = loaders.stan(tags.html(render=render_person)[ |
448 | ... tags.table[ |
449 | ... tags.tr[ |
450 | ... tags.td[tags.slot('firstName')], |
451 | ... tags.td[tags.slot('lastName')], |
452 | ... tags.td[tags.slot('sex')], |
453 | ... tags.td[tags.slot('location')]]]]) |
454 | ... |
455 | >>> PersonPage(person).renderSynchronously() |
456 | '<html><table><tr><td>Donovan</td><td>Preston</td><td>Male</td><td>California</td></tr></table></html>' |
457 | |
458 | Using patterns in combination with slots can lead to very powerful template |
459 | abstraction. Nevow also includes another standard renderer called "mapping" |
460 | which takes any data which responds to the "items()" message and inserts the |
461 | items into appropriate slots:: |
462 | |
463 | >>> class DictPage(rend.Page): |
464 | ... docFactory = loaders.stan(tags.html(render=rend.mapping)[ |
465 | ... tags.span[ tags.slot('foo') ], tags.span[ tags.slot('bar') ]]) |
466 | ... |
467 | >>> DictPage(dict(foo=1, bar=2)).renderSynchronously() |
468 | '<html><span>1</span><span>2</span></html>' |
469 | |
470 | Data directives |
471 | --------------- |
472 | |
473 | So far, we have always placed data functions directly in the Data special |
474 | attribute of a Tag. Sometimes, it is preferable to look up a data method from |
475 | the Page class as the Page has being rendered. For example, a base class may |
476 | define a template and a subclass may provide the implementation of the data |
477 | method. We can accomplish this effect by using a data directive as a Tag's data |
478 | special:: |
479 | |
480 | class Base(rend.Page): |
481 | docFactory = loaders.stan(tags.html[ |
482 | tags.div(data=tags.directive('name'), render=rend.data)]) |
483 | |
484 | class Subclass(Base): |
485 | def data_name(self, ctx, data): |
486 | return "Your name" |
487 | |
488 | The data directive is resolved by searching for the IContainer implementation in |
489 | the context. rend.Page implements IContainer.get by performing an attribute |
490 | lookup on the Page with the prefix 'data_*'. You can provide your own IContainer |
491 | implementation if you wish, and also you should know that IContainer |
492 | implementations for list and dict are included in the nevow.accessors module. |
493 | |
494 | A common gotcha is that the closest IContainer is used to resolve data |
495 | directives. This means that if a list is being used as the data during the |
496 | rendering process, data directives below this will be resolved against the |
497 | IContainer implementation in nevow.accessors.ListAccessor. If you are expecting |
498 | a data directive to invoke a Page's data_* method but instead get a KeyError, |
499 | this is why. |
500 | |
501 | Render directives |
502 | ----------------- |
503 | |
504 | Render directives are almost exactly the same, except they are resolved using |
505 | the closest IRendererFactory implementation in the context. Render directives |
506 | can be used to allow subclasses to override certain render methods, and also can |
507 | be used to allow Fragments to locate their own prefixed render methods. |
508 | |
509 | Flatteners |
510 | ---------- |
511 | |
512 | TODO This section isn't done yet. |
513 | |
514 | Nevow's flatteners use a type/function registry to determine how to render |
515 | objects which Nevow encounters in the DOM during the rendering process. |
516 | "Explicit is better than implicit", so in most cases, explicitly applying render |
517 | methods to data will be better than registering a flattener, but in some cases |
518 | it can be useful:: |
519 | |
520 | class Person(object): |
521 | def __init__(self, firstName, lastName): |
522 | self.firstName = firstName |
523 | self.lastName = lastName |
524 | |
525 | def flattenPerson(person, ctx): |
526 | return flat.partialflatten( |
527 | ctx, |
528 | (person.firstName, " ", person.lastName)) |
529 | |
530 | from nevow import flat |
531 | flat.registerFlattener(flattenPerson, Person) |
532 | |
533 | def insertData(ctx, data): |
534 | return data |
535 | |
536 | class PersonPage(rend.Page): |
537 | docFactory = loaders.stan(tags.html[ insertData ]) |
538 |