YATL helpers¶
Consider the following code in a view:
[[=DIV('this', 'is', 'a', 'test', _id='123', _class='myclass')]]
it is rendered as:
<div id="123" class="myclass">thisisatest</div>
DIV
is a helper class, i.e., something that can be used to build
HTML programmatically. It corresponds to the HTML <div>
tag.
Positional arguments are interpreted as objects contained between the open and close tags. Named arguments that start with an underscore are interpreted as HTML tag attributes (without the underscore). Some helpers also have named arguments that do not start with underscore; these arguments are tag-specific.
Instead of a set of unnamed arguments, a helper can also take a single
list or tuple as its set of components using the *
notation and it
can take a single dictionary as its set of attributes using the **
,
for example:
[[
contents = ['this', 'is', 'a', 'test']
attributes = {'_id':'123', '_class':'myclass'}
=DIV(*contents, **attributes)
]]
(produces the same output as before).
The following set of helpers:
A
, BEAUTIFY
, BODY
, CAT
, CODE
, DIV
, EM
,
FORM
, H1
, H2
, H3
, H4
, H5
, H6
, HEAD
,
HTML
, IMG
, INPUT
, LABEL
, LI
, METATAG
,
OL
, OPTION
, P
, PRE
, SELECT
, SPAN
, STRONG
,
TABLE
, TAG
, TAGGER
, THEAD
, TBODY
, TD
,
TEXTAREA
, TH
, TT
, TR
, UL
, XML
, xmlescape
,
I
, META
, LINK
, TITLE
, STYLE
, SCRIPT
can be used to build complex expressions that can then be serialized to XML. For example:
[[=DIV(STRONG(I("hello ", "<world>")), _class="myclass")]]
is rendered:
<div class="myclass"><strong><i>hello <world></i></strong></div>
Helpers can also be serialized into strings, equivalently, with the
__str__
and the xml
methods:
>>> str(DIV("hello world"))
'<div>hello world</div>'
>>> DIV("hello world").xml()
'<div>hello world</div>'
The helpers mechanism in py4web is more than a system to generate HTML without concatenating strings. It provides a server-side representation of the document object model (DOM).
Components of helpers can be referenced via their position, and helpers act as lists with respect to their components:
>>> a = DIV(SPAN('a', 'b'), 'c')
>>> print(a)
<div><span>ab</span>c</div>
>>> del a[1]
>>> a.append(STRONG('x'))
>>> a[0][0] = 'y'
>>> print(a)
<div><span>yb</span><strong>x</strong></div>
Attributes of helpers can be referenced by name, and helpers act as dictionaries with respect to their attributes:
>>> a = DIV(SPAN('a', 'b'), 'c')
>>> a['_class'] = 's'
>>> a[0]['_class'] = 't'
>>> print(a)
<div class="s"><span class="t">ab</span>c</div>
Note, the complete set of components can be accessed via a list called
a.children
, and the complete set of attributes can be accessed via
a dictionary called a.attributes
. So, a[i]
is equivalent to
a.children[i]
when i
is an integer, and a[s]
is equivalent
to a.attributes[s]
when s
is a string.
Notice that helper attributes are passed as keyword arguments to the helper. In some cases, however, attribute names include special characters that are not allowed in Python identifiers (e.g., hyphens) and therefore cannot be used as keyword argument names. For example:
DIV('text', _data-role='collapsible')
will not work because “_data-role” includes a hyphen, which will produce a Python syntax error.
In such cases you can pass the attributes as a dictionary and make use
of Python’s **
function arguments notation, which maps a dictionary
of (key:value) pairs into a set of keyword arguments:
>>> print(DIV('text', **{'_data-role': 'collapsible'}))
<div data-role="collapsible">text</div>
You can also dynamically create special TAGs:
>>> print(TAG['soap:Body']('whatever', **{'_xmlns:m':'http://www.example.org'}))
<soap:Body xmlns:m="http://www.example.org">whatever</soap:Body>
XML
¶
XML
is an object used to encapsulate text that should not be
escaped. The text may or may not contain valid XML. For example, it
could contain JavaScript.
The text in this example is escaped:
>>> print(DIV("<strong>hello</strong>"))
<div><strong>hello</strong></div>
by using XML
you can prevent escaping:
>>> print(DIV(XML("<strong>hello</strong>")))
<div><strong>hello</strong></div>
Sometimes you want to render HTML stored in a variable, but the HTML may contain unsafe tags such as scripts:
>>> print(XML('<script>alert("unsafe!")</script>'))
<script>alert("unsafe!")</script>
Un-escaped executable input such as this (for example, entered in the body of a comment in a blog) is unsafe, because it can be used to generate cross site scripting (XSS) attacks against other visitors to the page.
The py4web XML
helper can sanitize our text to prevent injections
and escape all tags except those that you explicitly allow. Here is an
example:
>>> print(XML('<script>alert("unsafe!")</script>', sanitize=True))
<script>alert("unsafe!")</script>
The XML
constructors, by default, consider the content of some tags
and some of their attributes safe. You can override the defaults using
the optional permitted_tags
and allowed_attributes
arguments.
Here are the default values of the optional arguments of the XML
helper.
XML(text, sanitize=False,
permitted_tags=['a', 'b', 'blockquote', 'br/', 'i', 'li',
'ol', 'ul', 'p', 'cite', 'code', 'pre', 'img/',
'h1', 'h2', 'h3', 'h4', 'h5', 'h6', 'table', 'tr', 'td',
'div', 'strong', 'span'],
allowed_attributes={'a': ['href', 'title', 'target'],
'img': ['src', 'alt'], 'blockquote': ['type'], 'td': ['colspan']})
Built-in helpers¶
A
¶
This helper is used to build links.
>>> print(A('<click>', XML('<strong>me</strong>'),
_href='http://www.py4web.com'))
<a href="http://www.py4web.com"><click><strong>me</strong></a>
BODY
¶
This helper makes the body of a page.
>>> print(BODY('<hello>', XML('<strong>world</strong>'), _bgcolor='red'))
<body bgcolor="red"><hello><strong>world</strong></body>
CAT
¶
This helper concatenates other helpers.
>>> print(CAT('Here is a ', A('link', _href='target'), ', and here is some ', STRONG('bold text'), '.'))
Here is a <a href="target">link</a>, and here is some <strong>bold text</strong>.
DIV
¶
This is the content division element.
>>> print(DIV('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<div id="0" class="test"><hello><strong>world</strong></div>
EM
¶
Emphasizes its content.
>>> print(EM('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<em id="0" class="test"><hello><strong>world</strong></em>
FORM
¶
Use this helper to make a FORM for user input. Forms will be discussed in detail in Chapter 12.
>>> print(FORM(INPUT(_type='submit'), _action='', _method='post'))
<form action="" method="post"><input type="submit"/></form>
H1
, H2
, H3
, H4
, H5
, H6
¶
These helpers are for paragraph headings and subheadings.
>>> print(H1('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<h1 id="0" class="test"><hello><strong>world</strong></h1>
HEAD
¶
For tagging the HEAD of an HTML page.
>>> print(HEAD(TITLE('<hello>', XML('<strong>world</strong>'))))
<head><title><hello><strong>world</strong></title></head>
HTML
¶
For tagging an HTML page.
>>> print(HTML(BODY('<hello>', XML('<strong>world</strong>'))))
<html><body><hello><strong>world</strong></body></html>
I
¶
This helper makes its contents italic.
>>> print(I('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<i id="0" class="test"><hello><strong>world</strong></i>
IMG
¶
It can be used to embed images into HTML.
>>> print(IMG(_src='http://example.com/image.png', _alt='test'))
<img alt="test" src="http://example.com/image.png"/>
Here is a combination of A, IMG, and URL helpers for including a static image with a link:
>>> print(A(IMG(_src=URL('static', 'logo.png'), _alt="My Logo"),
... _href=URL('default', 'index')))
<a href="/default/index"><img alt="My Logo" src="/static/logo.png"/></a>
INPUT
¶
Creates an <input.../>
tag. An input tag may not contain other tags,
and is closed by />
instead of >
. The input tag has an optional
attribute _type
that can be set to “text” (the default), “submit”,
“checkbox”, or “radio”.
>>> print(INPUT(_name='test', _value='a'))
<input name="test" value="a"/>
For radio buttons use the _checked
attribute:
>>> for v in ['a', 'b', 'c']:
... print(INPUT(_type='radio', _name='test', _value=v, _checked=v=='b'), v)
...
<input name="test" type="radio" value="a"/> a
<input checked="checked" name="test" type="radio" value="b"/> b
<input name="test" type="radio" value="c"/> c
and similarly for checkboxes:
>>> print(INPUT(_type='checkbox', _name='test', _value='a', _checked=True))
<input checked="checked" name="test" type="checkbox" value="a"/>
>>> print(INPUT(_type='checkbox', _name='test', _value='a', _checked=False))
<input name="test" type="checkbox" value="a"/>
LABEL
¶
It is used to create a LABEL tag for an INPUT field.
>>> print(LABEL('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<label id="0" class="test"><hello><strong>world</strong></label>
LI
¶
It makes a list item and should be contained in a UL
or OL
tag.
>>> print(LI('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<li id="0" class="test"><hello><strong>world</strong></li>
OL
¶
It stands for ordered list. The list should contain LI tags.
>>> print(OL(LI('<hello>'), LI(XML('<strong>world</strong>')), _class='test', _id=0))
<ol class="test" id="0"><li><hello></li><li><strong>world</strong></li></ol>
OPTION
¶
This should only be used as argument of a SELECT
.
>>> print(OPTION('<hello>', XML('<strong>world</strong>'), _value='a'))
<option value="a"><hello><strong>world</strong></option>
For selected options use the _selected
attribute:
>>> print(OPTION('Thank You', _value='ok', _selected=True))
<option selected="selected" value="ok">Thank You</option>
P
¶
This is for tagging a paragraph.
>>> print(P('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<p id="0" class="test"><hello><strong>world</strong></p>
PRE
¶
Generates a <pre>...</pre>
tag for displaying pre-formatted text.
The CODE
helper is generally preferable for code listings.
>>> print(SELECT(OPTION('first', _value='1'), OPTION('second', _value='2'), _class='test', _id=0))
<pre id="0" class="test"><hello><strong>world</strong></pre>
SCRIPT
¶
This is for include or link a script, such as JavaScript.
>>> print(SCRIPT('console.log("hello world");', _type='text/javascript'))
<script type="text/javascript">console.log("hello world");</script>
SELECT
¶
Makes a <select>...</select>
tag. This is used with the OPTION
helper.
>>> print(SELECT(OPTION('first', _value='1'), OPTION('second', _value='2'),
... _class='test', _id=0))
<select class="test" id="0"><option value="1">first</option><option value="2">second</option></select>
SPAN
¶
Similar to DIV
but used to tag inline (rather than block) content.
>>> print(SPAN('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<span id="0" class="test"><hello><strong>world</strong></span>
STYLE
¶
Similar to script, but used to either include or link CSS code. Here the CSS is included:
>>> print(STYLE(XML('body {color: white}')))
<style>body {color: white}</style>
and here it is linked:
>>> print(STYLE(_src='style.css'))
<style src="style.css"></style>
TABLE
, TR
, TD
¶
These tags (along with the optional THEAD
and TBODY
helpers) are
used to build HTML tables.
>>> print(TABLE(TR(TD('a'), TD('b')), TR(TD('c'), TD('d'))))
<table><tr><td>a</td><td>b</td></tr><tr><td>c</td><td>d</td></tr></table>
TR
expects TD
content.
It is easy to convert a Python array into an HTML table using Python’s
*
function arguments notation, which maps list elements to
positional function arguments.
Here, we will do it line by line:
>>> table = [['a', 'b'], ['c', 'd']]
>>> print(TABLE(TR(*map(TD, table[0])), TR(*map(TD, table[1]))))
<table><tr><td>a</td><td>b</td></tr><tr><td>c</td><td>d</td></tr></table>
Here we do all lines at once:
>>> table = [['a', 'b'], ['c', 'd']]
>>> print(TABLE(*[TR(*map(TD, rows)) for rows in table]))
<table><tr><td>a</td><td>b</td></tr><tr><td>c</td><td>d</td></tr></table>
TBODY
¶
This is used to tag rows contained in the table body, as opposed to header or footer rows. It is optional.
>>> print(TBODY(TR(TD('<hello>')), _class='test', _id=0))
<tbody id="0" class="test"><tr><td><hello></td></tr></tbody>
TEXTAREA
¶
This helper makes a <textarea>...</textarea>
tag.
>>> print(TEXTAREA('<hello>', XML('<strong>world</strong>'), _class='test',
... _cols="40", _rows="10"))
<textarea class="test" cols="40" rows="10"><hello><strong>world</strong></textarea>
TH
¶
This is used instead of TD
in table headers.
>>> print(TH('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<th id="0" class="test"><hello><strong>world</strong></th>
THEAD
¶
This is used to tag table header rows.
>>> print(THEAD(TR(TH('<hello>')), _class='test', _id=0))
<thead id="0" class="test"><tr><th><hello></th></tr></thead>
TITLE
¶
This is used to tag the title of a page in an HTML header.
>>> print(TITLE('<hello>', XML('<strong>world</strong>')))
<title><hello><strong>world</strong></title>
TT
¶
Tags text as typewriter (monospaced) text.
>>> print(TT('<hello>', XML('<strong>world</strong>'), _class='test', _id=0))
<tt id="0" class="test"><hello><strong>world</strong></tt>
UL
¶
It stands for unordered list. The list should contain LI tags.
>>> print(UL(LI('<hello>'), LI(XML('<strong>world</strong>')), _class='test', _id=0))
<ul class="test" id="0"><li><hello></li><li><strong>world</strong></li></ul>
URL
¶
The URL helper is not part of yatl package, instead it is provided by py4web.
Custom helpers¶
TAG
¶
Sometimes you need to generate custom XML tags. py4web provides TAG
,
a universal tag generator.
[[=TAG.name('a', 'b', _c='d')]]
generates the following XML:
<name c="d">ab</name>
Arguments “a”, “b”, and “d” are automatically escaped; use the XML
helper to suppress this behavior. Using TAG
you can generate
HTML/XML tags not already provided by the API. TAGs can be nested, and
are serialized with str().
An equivalent syntax is:
[[=TAG['name']('a', 'b', _c='d')]]
Self-closing tags can be generated with the TAG helper. The tag name must end with a “/”.
[[=TAG['link/'](_href='http://py4web.com')]]
generates the following XML:
<link ref="http://py4web.com"/>
Notice that TAG
is an object, and TAG.name
or TAG['name']
is
a function that returns an helper instance.
BEAUTIFY
¶
BEAUTIFY
is used to build HTML representations of compound objects,
including lists, tuples and dictionaries:
[[=BEAUTIFY({"a": ["hello", STRONG("world")], "b": (1, 2)})]]
BEAUTIFY
returns an XML-like object serializable to XML, with a nice
looking representation of its constructor argument. In this case, the
XML representation of:
{"a": ["hello", STRONG("world")], "b": (1, 2)}
will render as:
<table><tbody>
<tr><th>a</th><td><ul><li>hello</li><li><strong>world</strong></li></ul></td></tr>
<tr><th>b</th><td>(1, 2)</td></tr>
</tbody></table>
Server-side DOM¶
children
¶
Each helper object keep the list of its components into the children
attribute.
>>> CAT('hello', STRONG('world')).children
['hello', <yatl.helpers.TAGGER object at 0x7fa533ff7640>]
find
¶
To help searching into the DOM, all helpers have a find
method with
the following signature:
def find(self, query=None, **kargs)
that returns all the components matching supplied arguments.
A very simple query
can be a tag name:
>>> a = DIV(DIV(SPAN('x'), 3, DIV(SPAN('y'))))
>>> for c in a.find('span', first_only=True): c[0]='z'
>>> print(a) # We should .xml() here instead of print
<div><div><span>z</span>3<div><span>y</span></div></div></div>
>>> for c in a.find('span'): c[0]='z'
>>> print(a)
<div><div><span>z</span>3<div><span>z</span></div></div></div>
It also supports a syntax compatible with jQuery, accepting the following expressions:
jQuery Multiple Selector, e.g. “selector1, selector2, selectorN”,
jQuery Descendant Selector, e.g. “ancestor descendant”,
jQuery ID Selector, e.g. “#id”,
jQuery Class Selector, e.g. “.class”, and
jQuery Attribute Equals Selector, e.g. “[name=value]”, notice that here the value must be unquoted.
Here are some examples:
>>> a = DIV(SPAN(A('hello', **{'_id': '1-1', '_u:v': '$'})), P('world', _class='this is a test'))
>>> for e in a.find('div a#1-1, p.is'): print(e)
<a id="1-1" u:v="$">hello</a>
<p class="this is a test">world</p>
>>> for e in a.find('#1-1'): print(e)
<a id="1-1" u:v="$">hello</a>
>>> a.find('a[u:v=$]')[0].xml()
'<a id="1-1" u:v="$">hello</a>'
>>> a = FORM(INPUT(_type='text'), SELECT(OPTION(0)), TEXTAREA())
>>> for c in a.find('input, select, textarea'): c['_disabled'] = True
>>> a.xml()
'<form><input disabled="disabled" type="text"/><select disabled="disabled"><option>0</option></select><textarea disabled="disabled"></textarea></form>'
>>> for c in a.find('input, select, textarea'): c['_disabled'] = False
>>> a.xml()
'<form><input type="text"/><select><option>0</option></select><textarea></textarea></form>'
Elements that are matched can also be replaced or removed by specifying
a replace
argument (note, a list of the original matching elements
is still returned as usual).
>>> a = DIV(DIV(SPAN('x', _class='abc'), DIV(SPAN('y', _class='abc'), SPAN('z', _class='abc'))))
>>> b = a.find('span.abc', replace=P('x', _class='xyz'))
>>> print(a)
<div><div><p class="xyz">x</p><div><p class="xyz">x</p><p class="xyz">x</p></div></div></div>
replace
can be a callable, which will be passed the original element and
should return a new element to replace it.
>>> a = DIV(DIV(SPAN('x', _class='abc'), DIV(SPAN('y', _class='abc'), SPAN('z', _class='abc'))))
>>> b = a.find('span.abc', replace=lambda el: P(el[0], _class='xyz'))
>>> print(a)
<div><div><p class="xyz">x</p><div><p class="xyz">y</p><p class="xyz">z</p></div></div></div>
If replace=None
, matching elements will be removed completely.
>>> a = DIV(DIV(SPAN('x', _class='abc'), DIV(SPAN('y', _class='abc'), SPAN('z', _class='abc'))))
>>> b = a.find('span', text='y', replace=None)
>>> print(a)
<div><div><span class="abc">x</span><div><span class="abc"></span><span class="abc">z</span></div></div></div>
If a text
argument is specified, elements will be searched for text
components that match text, and any matching text components will be
replaced (text
is ignored if replace
is not also specified, use
a find
argument when you only need searching for textual elements).
Like the find
argument, text
can be a string or a compiled regex.
>>> a = DIV(DIV(SPAN('x', _class='abc'), DIV(SPAN('y', _class='abc'), SPAN('z', _class='abc'))))
>>> b = a.find(text=re.compile('x|y|z'), replace='hello')
>>> print(a)
<div><div><span class="abc">hello</span><div><span class="abc">hello</span><span class="abc">hello</span></div></div></div>
If other attributes are specified along with text
, then only components
that match the specified attributes will be searched for text.
>>> a = DIV(DIV(SPAN('x', _class='abc'), DIV(SPAN('y', _class='efg'), SPAN('z', _class='abc'))))
>>> b = a.find('span.efg', text=re.compile('x|y|z'), replace='hello')
>>> print(a)
<div><div><span class="abc">x</span><div><span class="efg">hello</span><span class="abc">z</span></div></div></div>
Page layout¶
Views can extend and include other views in a tree-like structure.
For example, we can think of a view “index.html” that extends “layout.html” and includes “body.html”. At the same time, “layout.html” may include “header.html” and “footer.html”.
The root of the tree is what we call a layout view. Just like any other HTML template file, you can edit it using the py4web administrative interface. The file name “layout.html” is just a convention.
Here is a minimalist page that extends the “layout.html” view and includes the “page.html” view:
[[extend 'layout.html']]
<h1>Hello World</h1>
[[include 'page.html']]
The extended layout file must contain an [[include]]
directive,
something like:
<html>
<head>
<title>Page Title</title>
</head>
<body>
[[include]]
</body>
</html>
When the view is called, the extended (layout) view is loaded, and the
calling view replaces the [[include]]
directive inside the layout.
Processing continues recursively until all extend
and include
directives have been processed. The resulting template is then
translated into Python code. Note, when an application is bytecode
compiled, it is this Python code that is compiled, not the original view
files themselves. So, the bytecode compiled version of a given view is a
single .pyc file that includes the Python code not just for the original
view file, but for its entire tree of extended and included views.
extend
,include
,block
andsuper
are special template directives, not Python commands.
Any content or code that precedes the [[extend ...]]
directive will
be inserted (and therefore executed) before the beginning of the
extended view’s content/code. Although this is not typically used to
insert actual HTML content before the extended view’s content, it can be
useful as a means to define variables or functions that you want to make
available to the extended view. For example, consider a view
“index.html”:
[[sidebar_enabled=True]]
[[extend 'layout.html']]
<h1>Home Page</h1>
and an excerpt from “layout.html”:
[[if sidebar_enabled:]]
<div id="sidebar">
Sidebar Content
</div>
[[pass]]
Because the sidebar_enabled
assignment in “index.html” comes before
the extend
, that line gets inserted before the beginning of
“layout.html”, making sidebar_enabled
available anywhere within the
“layout.html” code (a somewhat more sophisticated version of this is
used in the welcome app).
It is also worth pointing out that the variables returned by the controller function are available not only in the function’s main view, but in all of its extended and included views as well.
The argument of an extend
or include
(i.e., the extended or
included view name) can be a Python variable (though not a Python
expression). However, this imposes a limitation – views that use
variables in extend
or include
statements cannot be bytecode
compiled. As noted above, bytecode-compiled views include the entire
tree of extended and included views, so the specific extended and
included views must be known at compile time, which is not possible if
the view names are variables (whose values are not determined until run
time). Because bytecode compiling views can provide a significant speed
boost, using variables in extend
and include
should generally be
avoided if possible.
In some cases, an alternative to using a variable in an include
is
simply to place regular [[include ...]]
directives inside an
if...else
block.
[[if some_condition:]]
[[include 'this_view.html']]
[[else:]]
[[include 'that_view.html']]
[[pass]]
The above code does not present any problem for bytecode compilation
because no variables are involved. Note, however, that the bytecode
compiled view will actually include the Python code for both
“this_view.html” and “that_view.html”, though only the code for one of
those views will be executed, depending on the value of
some_condition
.
Keep in mind, this only works for include
– you cannot place
[[extend ...]]
directives inside if...else
blocks.
Layouts are used to encapsulate page commonality (headers, footers, menus), and though they are not mandatory, they will make your application easier to write and maintain. In particular, we suggest writing layouts that take advantage of the following variables that can be set in the controller. Using these well known variables will help make your layouts interchangeable:
response.title
response.subtitle
response.meta.author
response.meta.keywords
response.meta.description
response.flash
response.menu
response.files
Except for menu
and files
, these are all strings and their
meaning should be obvious.
response.menu
menu is a list of 3-tuples or 4-tuples. The three
elements are: the link name, a boolean representing whether the link is
active (is the current link), and the URL of the linked page. For
example:
response.menu = [('Google', False, 'http://www.google.com', []),
('Index', True, URL('index'), [])]
The fourth tuple element is an optional sub-menu.
response.files
is a list of CSS and JS files that are needed by your
page.
We also recommend that you use:
[[include 'py4web_ajax.html']]
in the HTML head, since this will include the jQuery libraries and
define some backward-compatible JavaScript functions for special effects
and Ajax. “py4web_ajax.html” includes the response.meta
tags in the
view, jQuery base, the calendar datepicker, and all required CSS and JS
response.files
.
Default page layout¶
The “views/layout.html” that ships with the py4web scaffolding application welcome (stripped down of some optional parts) is quite complex but it has the following structure:
<!DOCTYPE html>
<head>
<meta charset="utf-8" />
<title>[[=response.title or request.application]]</title>
...
<script src="[[=URL('static', 'js/modernizr.custom.js')]]"></script>
[[
response.files.append(URL('static', 'css/py4web.css'))
response.files.append(URL('static', 'css/bootstrap.min.css'))
response.files.append(URL('static', 'css/bootstrap-responsive.min.css'))
response.files.append(URL('static', 'css/py4web_bootstrap.css'))
]]
[[include 'py4web_ajax.html']]
[[
# using sidebars need to know what sidebar you want to use
left_sidebar_enabled = globals().get('left_sidebar_enabled', False)
right_sidebar_enabled = globals().get('right_sidebar_enabled', False)
middle_columns = {0:'span12', 1:'span9', 2:'span6'}[
(left_sidebar_enabled and 1 or 0)+(right_sidebar_enabled and 1 or 0)]
]]
[[block head]][[end]]
</head>
<body>
<!-- Navbar ================================================== -->
<div class="navbar navbar-inverse navbar-fixed-top">
<div class="flash">[[=response.flash or '']]</div>
<div class="navbar-inner">
<div class="container">
[[=response.logo or '']]
<ul id="navbar" class="nav pull-right">
[[='auth' in globals() and auth.navbar(mode="dropdown") or '']]
</ul>
<div class="nav-collapse">
[[if response.menu:]]
[[=MENU(response.menu)]]
[[pass]]
</div><!--/.nav-collapse -->
</div>
</div>
</div><!--/top navbar -->
<div class="container">
<!-- Masthead ================================================== -->
<header class="mastheader row" id="header">
<div class="span12">
<div class="page-header">
<h1>
[[=response.title or request.application]]
<small>[[=response.subtitle or '']]</small>
</h1>
</div>
</div>
</header>
<section id="main" class="main row">
[[if left_sidebar_enabled:]]
<div class="span3 left-sidebar">
[[block left_sidebar]]
<h3>Left Sidebar</h3>
<p></p>
[[end]]
</div>
[[pass]]
<div class="[[=middle_columns]]">
[[block center]]
[[include]]
[[end]]
</div>
[[if right_sidebar_enabled:]]
<div class="span3">
[[block right_sidebar]]
<h3>Right Sidebar</h3>
<p></p>
[[end]]
</div>
[[pass]]
</section><!--/main-->
<!-- Footer ================================================== -->
<div class="row">
<footer class="footer span12" id="footer">
<div class="footer-content">
[[block footer]] <!-- this is default footer -->
...
[[end]]
</div>
</footer>
</div>
</div> <!-- /container -->
<!-- The javascript =============================================
(Placed at the end of the document so the pages load faster) -->
<script src="[[=URL('static', 'js/bootstrap.min.js')]]"></script>
<script src="[[=URL('static', 'js/py4web_bootstrap.js')]]"></script>
[[if response.google_analytics_id:]]
<script src="[[=URL('static', 'js/analytics.js')]]"></script>
<script type="text/javascript">
analytics.initialize({
'Google Analytics':{trackingId:'[[=response.google_analytics_id]]'}
});</script>
[[pass]]
</body>
</html>
There are a few features of this default layout that make it very easy to use and customize:
It is written in HTML5 and uses the “modernizr” library for backward compatibility. The actual layout includes some extra conditional statements required by IE and they are omitted for brevity.
It displays both
response.title
andresponse.subtitle
which can be set in a model or a controller. If they are not set, it adopts the application name as title.It includes the
py4web_ajax.html
file in the header which generated all the link and script import statements.It uses a modified version of Twitter Bootstrap for flexible layouts which works on mobile devices and re-arranges columns to fit small screens.
It uses “analytics.js” to connect to Google Analytics.
The
[[=auth.navbar(...)]]
displays a welcome to the current user and links to the auth functions like login, logout, register, change password, etc. depending on context.auth.navbar
is a helper factory and its output can be manipulated as any other helper. It is placed in an expression to check for auth definition, the expression evaluates to ’’ in case auth is undefined.The
[[=MENU(response.menu)]]
displays the menu structure as<ul>...</ul>
.[[include]]
is replaced by the content of the extending view when the page is rendered.By default it uses a conditional three column (the left and right sidebars can be turned off by the extending views)
It uses the following classes: page-header, main, footer.
It contains the following blocks: head, left_sidebar, center, right_sidebar, footer.
In views, you can turn on and customize sidebars as follows:
[[left_sidebar_enabled=True]]
[[extend 'layout.html']]
This text goes in center
[[block left_sidebar]]
This text goes in sidebar
[[end]]
Customizing the default layout¶
Customizing the default layout without editing is easy because the welcome application is based on Twitter Bootstrap which is well documented and supports themes. In py4web four static files which are relevant to style:
“css/py4web.css” contains py4web specific styles
“css/bootstrap.min.css” contains the Twitter Bootstrap CSS style
“css/py4web_bootstrap.css” which overrides some Bootstrap styles to conform to py4web needs.
“js/bootstrap.min.js” which includes the libraries for menu effects, modals, panels.
To change colors and background images, try append the following code to layout.html header:
<style>
body { background: url('images/background.png') repeat-x #3A3A3A; }
a { color: #349C01; }
.page-header h1 { color: #349C01; }
.page-header h2 { color: white; font-style: italic; font-size: 14px;}
.statusbar { background: #333333; border-bottom: 5px #349C01 solid; }
.statusbar a { color: white; }
.footer { border-top: 5px #349C01 solid; }
</style>
Of course you can also completely replace the “layout.html” and “py4web.css” files with your own.
Mobile development¶
Although the default layout.html is designed to be mobile-friendly, one may sometimes need to use different views when a page is visited by a mobile device.
To make developing for desktop and mobile devices easier, py4web
includes the @mobilize
decorator. This decorator is applied to
actions that should have a normal view and a mobile view. This is
demonstrated here:
from gluon.contrib.user_agent_parser import mobilize
@mobilize
def index():
return dict()
Notice that the decorator must be imported before using it in a controller. When the “index” function is called from a regular browser (desktop computer), py4web will render the returned dictionary using the view “[controller]/index.html”. However, when it is called by a mobile device, the dictionary will be rendered by “[controller]/index.mobile.html”. Notice that mobile views have the “mobile.html” extension.
Alternatively you can apply the following logic to make all views mobile friendly:
if request.user_agent().is_mobile:
response.view.replace('.html', '.mobile.html')
The task of creating the “*.mobile.html” views is left to the developer but we strongly suggest using the “jQuery Mobile” plugin which makes the task very easy.
Functions in views¶
Consider this “layout.html”:
<html>
<body>
[[include]]
<div class="sidebar">
[[if 'mysidebar' in globals():]][[mysidebar()]][[else:]]
my default sidebar
[[pass]]
</div>
</body>
</html>
and this extending view
[[def mysidebar():]]
my new sidebar!!!
[[return]]
[[extend 'layout.html']]
Hello World!!!
Notice the function is defined before the [[extend...]]
statement –
this results in the function being created before the “layout.html” code
is executed, so the function can be called anywhere within
“layout.html”, even before the [[include]]
. Also notice the function
is included in the extended view without the =
prefix.
The code generates the following output:
<html>
<body>
Hello World!!!
<div class="sidebar">
my new sidebar!!!
</div>
</body>
</html>
Notice that the function is defined in HTML (although it could also
contain Python code) so that response.write
is used to write its
content (the function does not return the content). This is why the
layout calls the view function using [[mysidebar()]]
rather than
[[=mysidebar()]]
. Functions defined in this way can take arguments.
Blocks in views¶
The main way to make a view more modular is by using
[[block ...]]
s and this mechanism is an alternative to the
mechanism discussed in the previous section.
To understand how this works, consider apps based on the scaffolding app
welcome, which has a view layout.html. This view is extended by the view
default/index.html
via [[extend 'layout.html']]
. The contents of
layout.html predefine certain blocks with certain default content, and
these are therefore included into default/index.html.
You can override these default content blocks by enclosing your new content inside the same block name. The location of the block in the layout.html is not changed, but the contents is.
Here is a simplifed version. Imagine this is “layout.html”:
<html>
<body>
[[include]]
<div class="sidebar">
[[block mysidebar]]
my default sidebar (this content to be replaced)
[[end]]
</div>
</body>
</html>
and this is a simple extending view default/index.html
:
[[extend 'layout.html']]
Hello World!!!
[[block mysidebar]]
my new sidebar!!!
[[end]]
It generates the following output, where the content is provided by the over-riding block in the extending view, yet the enclosing DIV and class comes from layout.html. This allows consistency across views:
<html>
<body>
Hello World!!!
<div class="sidebar">
my new sidebar!!!
</div>
</body>
</html>
The real layout.html defines a number of useful blocks, and you can easily add more to match the layout your desire.
You can have many blocks, and if a block is present in the extended view
but not in the extending view, the content of the extended view is used.
Also, notice that unlike with functions, it is not necessary to define
blocks before the [[extend ...]]
– even if defined after the
extend
, they can be used to make substitutions anywhere in the
extended view.
Inside a block, you can use the expression [[super]]
to include the
content of the parent. For example, if we replace the above extending
view with:
[[extend 'layout.html']]
Hello World!!!
[[block mysidebar]]
[[super]]
my new sidebar!!!
[[end]]
we get:
<html>
<body>
Hello World!!!
<div class="sidebar">
my default sidebar
my new sidebar!
</div>
</body>
</html>