native_client_sdk/src/doc/_sphinxext/devsite_builder.py - chromium/src - Git at Google

 # Copyright (c) 2013 The Chromium Authors. All rights reserved.
 # Use of this source code is governed by a BSD-style license that can be
 # found in the LICENSE file.

 #
 # This is a Sphinx extension.
 #

 from __future__ import print_function
 import codecs
 from collections import namedtuple, OrderedDict
 import os
 import string
 from docutils import nodes
 from docutils.parsers.rst import Directive, directives
 from sphinx.util.osutil import ensuredir
 from sphinx.builders.html import StandaloneHTMLBuilder
 from sphinx.writers.html import HTMLWriter
 from sphinx.writers.html import SmartyPantsHTMLTranslator as HTMLTranslator
 from sphinx.util.console import bold

 PEPPER_VERSION = "31"

 # TODO(eliben): it may be interesting to use an actual Sphinx template here at
 # some point.
 PAGE_TEMPLATE = string.Template(r'''
 ${devsite_prefix}
 <html devsite>
   <head>
     ${nonprod_meta_head}
     <title>${doc_title}</title>
     <meta name="project_path" value="/native-client${folder}/_project.yaml" />
     <meta name="book_path" value="/native-client${folder}/_book.yaml" />
     <link href="/native-client/css/local_extensions.css" rel="stylesheet" type="text/css"/>
     ${nonprod_css}
     <style type="text/css">
       .nested-def {list-style-type: none; margin-bottom: 0.3em;}
       .maxheight {height: 200px;}
     </style>
   </head>
   <body>
     ${devsite_butterbar}

 ${doc_body}

   </body>
 </html>
 '''.lstrip())

 DEVSITE_PREFIX = r'''
 {% setvar pepperversion %}pepper''' + PEPPER_VERSION + '''{% endsetvar %}
 {% include "native-client/_local_variables.html" %}'''

 DEVSITE_BUTTERBAR = '{{butterbar}}'

 # We want the non-production-mode HTML to resemble the real one, so this points
 # to a copied version of the devsite CSS that we'll keep locally. It's for
 # testing purposes only.
 NONPROD_CSS = '<link href="/_static/css/local_extensions.css"'\
               'rel="stylesheet" type="text/css"/>'
 NONPROD_META_HEAD = '<meta charset="utf-8" />'

 # Path to the top-level YAML table-of-contents file for the devsite
 BOOK_TOC_TEMPLATE = '_book_template.yaml'


 class DevsiteHTMLTranslator(HTMLTranslator):
   """ Custom HTML translator for devsite output.

       Hooked into the HTML builder by setting the html_translator_class
       option in conf.py

       HTMLTranslator is provided by Sphinx. We're actually using
       SmartyPantsHTMLTranslator to use its quote and dash-formatting
       capabilities. It's a subclass of the HTMLTranslator provided by docutils,
       with Sphinx-specific features added. Here we provide devsite-specific
       behavior by overriding some of the visiting methods.
   """
   def __init__(self, builder, *args, **kwds):
     # HTMLTranslator is an old-style Python class, so 'super' doesn't work: use
     # direct parent invocation.
     HTMLTranslator.__init__(self, builder, *args, **kwds)

     self.within_ignored_h1 = False
     self.within_toc = False

   def visit_bullet_list(self, node):
     if self.within_toc:
       # Within a TOC, devsite wants <ol>
       self.body.append(self.starttag(node, 'ol'))
     else:
       # Use our own class attribute for <ul>. Don't care about compacted lists.
       self.body.append(self.starttag(node, 'ul', **{'class': 'small-gap'}))

   def depart_bullet_list(self, node):
     if self.within_toc:
       self.body.append('</ol>\n')
     else:
       # Override to not pop anything from context
       self.body.append('</ul>\n')

   def visit_literal(self, node):
     # Don't insert "smart" quotes here
     self.no_smarty += 1
     # Sphinx emits <tt></tt> for literals (``like this``), with <span> per word
     # to protect against wrapping, etc. We're required to emit plain <code>
     # tags for them.
     # Emit a simple <code> tag without enabling "protect_literal_text" mode,
     # so Sphinx's visit_Text doesn't mess with the contents.
     self.body.append(self.starttag(node, 'code', suffix=''))

   def depart_literal(self, node):
     self.no_smarty -= 1
     self.body.append('</code>')

   def visit_literal_block(self, node):
     # Don't insert "smart" quotes here
     self.no_smarty += 1
     # We don't use Sphinx's buildin pygments integration for code highlighting,
     # because the devsite requires special <pre> tags for that and handles the
     # highlighting on its own.
     attrs = {'class': 'prettyprint'} if node.get('prettyprint', 1) else {}
     self.body.append(self.starttag(node, 'pre', **attrs))

   def depart_literal_block(self, node):
     self.no_smarty -= 1
     self.body.append('\n</pre>\n')

   def visit_paragraph(self, node):
     # Don't generate <p>s within the table of contents
     if not self.within_toc:
       HTMLTranslator.visit_paragraph(self, node)

   def depart_paragraph(self, node):
     if not self.within_toc:
       HTMLTranslator.depart_paragraph(self, node)

   def visit_section(self, node):
     # devsite needs <section> instead of <div class='section'>
     self.section_level += 1
     self.body.append(self.starttag(node, 'section'))

   def depart_section(self, node):
     self.section_level -= 1
     self.body.append('</section>')

   def visit_image(self, node):
     # Paths to images in .rst sources should be absolute. This visitor does the
     # required transformation for the path to be correct in the final HTML.
     if self.builder.devsite_production_mode:
       node['uri'] = self.builder.get_production_url(node['uri'])
     HTMLTranslator.visit_image(self, node)

   def visit_reference(self, node):
     # In "kill_internal_links" mode, we don't emit the actual links for internal
     # nodes.
     if self.builder.kill_internal_links and node.get('internal'):
       pass
     else:
       HTMLTranslator.visit_reference(self, node)

   def depart_reference(self, node):
     if self.builder.kill_internal_links and node.get('internal'):
       pass
     else:
       HTMLTranslator.depart_reference(self, node)

   def visit_title(self, node):
     # Why this?
     #
     # Sphinx insists on inserting a <h1>Page Title</h1> into the page, but this
     # is not how the devsite wants it. The devsite inserts the page title on
     # its own, the the extra h1 is duplication.
     #
     # Killing the doctree title node in a custom transform doesn't work, because
     # Sphinx explicitly looks for it when writing a document. So instead we rig
     # the HTML produced.
     #
     # When a title node is visited, and this is the h1-to-be, we ignore it and
     # also set a flag that tells visit_Text not to print the actual text of the
     # header.

     # The h1 node is in the section whose parent is the document itself. Other
     # sections have this top-section as their parent.
     if (node.parent and node.parent.parent and
         isinstance(node.parent.parent, nodes.document)):
       # Internal flag. Also, nothing is pushed to the context. Our depart_title
       # doesn't pop anything when this flag is set.
       self.within_ignored_h1 = True
     else:
       HTMLTranslator.visit_title(self, node)

   def depart_title(self, node):
     if not self.within_ignored_h1:
       HTMLTranslator.depart_title(self, node)
     self.within_ignored_h1 = False

   def visit_Text(self, node):
     if not self.within_ignored_h1:
       HTMLTranslator.visit_Text(self, node)

   def visit_topic(self, node):
     if 'contents' in node['classes']:
       # Detect a TOC: this requires special treatment for devsite.
       self.within_toc = True
       # Emit <nav> manually and not through starttage because we don't want
       # additional class components to be added
       self.body.append('\n<nav class="inline-toc">')

   def depart_topic(self, node):
     if self.within_toc:
       self.within_toc = False
       self.body.append('</nav>\n')

   def write_colspecs(self):
     # Override this method from docutils to do nothing. We don't need those
     # pesky <col width=NN /> tags in our markup.
     pass

   def visit_admonition(self, node, name=''):
     self.body.append(self.starttag(node, 'aside', CLASS=node.get('class', '')))

   def depart_admonition(self, node=''):
     self.body.append('\n</aside>\n')

   def unknown_visit(self, node):
     raise NotImplementedError('Unknown node: ' + node.__class__.__name__)


 class DevsiteBuilder(StandaloneHTMLBuilder):
   """ Builder for the NaCl devsite HTML output.

       Loosely based on the code of Sphinx's standard SerializingHTMLBuilder.
   """
   name = 'devsite'
   out_suffix = '.html'
   link_suffix = '.html'

   # Disable the addition of "pi"-permalinks to each section header
   add_permalinks = False

   def init(self):
     self.devsite_production_mode = int(self.config.devsite_production_mode) == 1
     self.kill_internal_links = int(self.config.kill_internal_links) == 1
     self.info("----> Devsite builder with production mode = %d" % (
         self.devsite_production_mode,))
     self.config_hash = ''
     self.tags_hash = ''
     self.theme = None       # no theme necessary
     self.templates = None   # no template bridge necessary
     self.init_translator_class()
     self.init_highlighter()

   def finish(self):
     super(DevsiteBuilder, self).finish()
     if self.devsite_production_mode:
       # We decided to keep the manual _book.yaml for now;
       # The code for auto-generating YAML TOCs from index.rst was removed in
       # https://codereview.chromium.org/57923006/
       self.info(bold('generating YAML table-of-contents... '))
       subs = {
         'version': PEPPER_VERSION,
         'folder': self.config.devsite_foldername or ''}
       with open(os.path.join(self.env.srcdir, '_book.yaml')) as in_f:
         with open(os.path.join(self.outdir, '_book.yaml'), 'w') as out_f:
           out_f.write(string.Template(in_f.read()).substitute(subs))
     self.info()

   def dump_inventory(self):
     # We don't want an inventory file when building for devsite
     if not self.devsite_production_mode:
       super(DevsiteBuilder, self).dump_inventory()

   def get_production_url(self, url):
     if not self.devsite_production_mode:
       return url

     if self.config.devsite_foldername:
       return '/native-client/%s/%s' % (self.config.devsite_foldername, url)

     return '/native-client/%s' % url

   def get_target_uri(self, docname, typ=None):
     if self.devsite_production_mode:
       return self.get_production_url(docname)
     else:
       return docname + self.link_suffix

   def handle_page(self, pagename, ctx, templatename='page.html',
                   outfilename=None, event_arg=None):
     ctx['current_page_name'] = pagename

     if not outfilename:
       outfilename = os.path.join(self.outdir,
                                  pagename + self.out_suffix)

     # Emit an event to Sphinx
     self.app.emit('html-page-context', pagename, templatename,
                   ctx, event_arg)

     ensuredir(os.path.dirname(outfilename))
     self._dump_context(ctx, outfilename)

   def _dump_context(self, context, filename):
     """ Do the actual dumping of the page to the file. context is a dict. Some
         important fields:
           body - document contents
           title
           current_page_name
         Some special pages (genindex, etc.) may not have some of the fields, so
         fetch them conservatively.
     """
     if not 'body' in context:
       return

     folder = ''
     if self.devsite_production_mode and self.config.devsite_foldername:
       folder = "/" + self.config.devsite_foldername

     # codecs.open is the fast Python 2.x way of emulating the encoding= argument
     # in Python 3's builtin open.
     with codecs.open(filename, 'w', encoding='utf-8') as f:
       f.write(PAGE_TEMPLATE.substitute(
         doc_title=context.get('title', ''),
         doc_body=context.get('body'),
         folder=folder,
         nonprod_css=self._conditional_nonprod(NONPROD_CSS),
         nonprod_meta_head=self._conditional_nonprod(NONPROD_META_HEAD),
         devsite_prefix=self._conditional_devsite(DEVSITE_PREFIX),
         devsite_butterbar=self._conditional_devsite(DEVSITE_BUTTERBAR)))

   def _conditional_devsite(self, s):
     return s if self.devsite_production_mode else ''

   def _conditional_nonprod(self, s):
     return s if not self.devsite_production_mode else ''


 class NaclCodeDirective(Directive):
   """ Custom "naclcode" directive for code snippets. To keep it under our
       control.
   """
   has_content = True
   required_arguments = 0
   optional_arguments = 1
   option_spec = {
       'prettyprint': int,
   }

   def run(self):
     code = u'\n'.join(self.content)
     literal = nodes.literal_block(code, code)
     literal['prettyprint'] = self.options.get('prettyprint', 1)
     return [literal]

 def setup(app):
   """ Extension registration hook.
   """
   # linkcheck issues HEAD requests to save time, but some Google properties
   # reject them and we get spurious 405 responses. Monkey-patch sphinx to
   # just use normal GET requests.
   # See: https://bitbucket.org/birkenfeld/sphinx/issue/1292/
   from sphinx.builders import linkcheck
   import urllib2
   linkcheck.HeadRequest = urllib2.Request

   app.add_directive('naclcode', NaclCodeDirective)
   app.add_builder(DevsiteBuilder)

   # "Production mode" for local testing vs. on-server documentation.
   app.add_config_value('devsite_production_mode', default='1', rebuild='html')
   app.add_config_value('kill_internal_links', default='0', rebuild='html')
   app.add_config_value('devsite_foldername', default=None, rebuild='html')
	# Copyright (c) 2013 The Chromium Authors. All rights reserved.
	# Use of this source code is governed by a BSD-style license that can be
	# found in the LICENSE file.

	#
	# This is a Sphinx extension.
	#

	from __future__ import print_function
	import codecs
	from collections import namedtuple, OrderedDict
	import os
	import string
	from docutils import nodes
	from docutils.parsers.rst import Directive, directives
	from sphinx.util.osutil import ensuredir
	from sphinx.builders.html import StandaloneHTMLBuilder
	from sphinx.writers.html import HTMLWriter
	from sphinx.writers.html import SmartyPantsHTMLTranslator as HTMLTranslator
	from sphinx.util.console import bold

	PEPPER_VERSION = "31"

	# TODO(eliben): it may be interesting to use an actual Sphinx template here at
	# some point.
	PAGE_TEMPLATE = string.Template(r'''
	${devsite_prefix}
	<html devsite>
	<head>
	${nonprod_meta_head}
	<title>${doc_title}</title>
	<meta name="project_path" value="/native-client${folder}/_project.yaml" />
	<meta name="book_path" value="/native-client${folder}/_book.yaml" />
	<link href="/native-client/css/local_extensions.css" rel="stylesheet" type="text/css"/>
	${nonprod_css}
	<style type="text/css">
	.nested-def {list-style-type: none; margin-bottom: 0.3em;}
	.maxheight {height: 200px;}
	</style>
	</head>
	<body>
	${devsite_butterbar}

	${doc_body}

	</body>
	</html>
	'''.lstrip())

	DEVSITE_PREFIX = r'''
	{% setvar pepperversion %}pepper''' + PEPPER_VERSION + '''{% endsetvar %}
	{% include "native-client/_local_variables.html" %}'''

	DEVSITE_BUTTERBAR = '{{butterbar}}'

	# We want the non-production-mode HTML to resemble the real one, so this points
	# to a copied version of the devsite CSS that we'll keep locally. It's for
	# testing purposes only.
	NONPROD_CSS = '<link href="/_static/css/local_extensions.css"'\
	'rel="stylesheet" type="text/css"/>'
	NONPROD_META_HEAD = '<meta charset="utf-8" />'

	# Path to the top-level YAML table-of-contents file for the devsite
	BOOK_TOC_TEMPLATE = '_book_template.yaml'


	class DevsiteHTMLTranslator(HTMLTranslator):
	""" Custom HTML translator for devsite output.

	Hooked into the HTML builder by setting the html_translator_class
	option in conf.py

	HTMLTranslator is provided by Sphinx. We're actually using
	SmartyPantsHTMLTranslator to use its quote and dash-formatting
	capabilities. It's a subclass of the HTMLTranslator provided by docutils,
	with Sphinx-specific features added. Here we provide devsite-specific
	behavior by overriding some of the visiting methods.
	"""
	def __init__(self, builder, args, *kwds):
	# HTMLTranslator is an old-style Python class, so 'super' doesn't work: use
	# direct parent invocation.
	HTMLTranslator.__init__(self, builder, args, *kwds)

	self.within_ignored_h1 = False
	self.within_toc = False

	def visit_bullet_list(self, node):
	if self.within_toc:
	# Within a TOC, devsite wants <ol>
	self.body.append(self.starttag(node, 'ol'))
	else:
	# Use our own class attribute for <ul>. Don't care about compacted lists.
	self.body.append(self.starttag(node, 'ul', **{'class': 'small-gap'}))

	def depart_bullet_list(self, node):
	if self.within_toc:
	self.body.append('</ol>\n')
	else:
	# Override to not pop anything from context
	self.body.append('</ul>\n')

	def visit_literal(self, node):
	# Don't insert "smart" quotes here
	self.no_smarty += 1
	# Sphinx emits <tt></tt> for literals (``like this``), with <span> per word
	# to protect against wrapping, etc. We're required to emit plain <code>
	# tags for them.
	# Emit a simple <code> tag without enabling "protect_literal_text" mode,
	# so Sphinx's visit_Text doesn't mess with the contents.
	self.body.append(self.starttag(node, 'code', suffix=''))

	def depart_literal(self, node):
	self.no_smarty -= 1
	self.body.append('</code>')

	def visit_literal_block(self, node):
	# Don't insert "smart" quotes here
	self.no_smarty += 1
	# We don't use Sphinx's buildin pygments integration for code highlighting,
	# because the devsite requires special <pre> tags for that and handles the
	# highlighting on its own.
	attrs = {'class': 'prettyprint'} if node.get('prettyprint', 1) else {}
	self.body.append(self.starttag(node, 'pre', **attrs))

	def depart_literal_block(self, node):
	self.no_smarty -= 1
	self.body.append('\n</pre>\n')

	def visit_paragraph(self, node):
	# Don't generate <p>s within the table of contents
	if not self.within_toc:
	HTMLTranslator.visit_paragraph(self, node)

	def depart_paragraph(self, node):
	if not self.within_toc:
	HTMLTranslator.depart_paragraph(self, node)

	def visit_section(self, node):
	# devsite needs <section> instead of <div class='section'>
	self.section_level += 1
	self.body.append(self.starttag(node, 'section'))

	def depart_section(self, node):
	self.section_level -= 1
	self.body.append('</section>')

	def visit_image(self, node):
	# Paths to images in .rst sources should be absolute. This visitor does the
	# required transformation for the path to be correct in the final HTML.
	if self.builder.devsite_production_mode:
	node['uri'] = self.builder.get_production_url(node['uri'])
	HTMLTranslator.visit_image(self, node)

	def visit_reference(self, node):
	# In "kill_internal_links" mode, we don't emit the actual links for internal
	# nodes.
	if self.builder.kill_internal_links and node.get('internal'):
	pass
	else:
	HTMLTranslator.visit_reference(self, node)

	def depart_reference(self, node):
	if self.builder.kill_internal_links and node.get('internal'):
	pass
	else:
	HTMLTranslator.depart_reference(self, node)

	def visit_title(self, node):
	# Why this?
	#
	# Sphinx insists on inserting a <h1>Page Title</h1> into the page, but this
	# is not how the devsite wants it. The devsite inserts the page title on
	# its own, the the extra h1 is duplication.
	#
	# Killing the doctree title node in a custom transform doesn't work, because
	# Sphinx explicitly looks for it when writing a document. So instead we rig
	# the HTML produced.
	#
	# When a title node is visited, and this is the h1-to-be, we ignore it and
	# also set a flag that tells visit_Text not to print the actual text of the
	# header.

	# The h1 node is in the section whose parent is the document itself. Other
	# sections have this top-section as their parent.
	if (node.parent and node.parent.parent and
	isinstance(node.parent.parent, nodes.document)):
	# Internal flag. Also, nothing is pushed to the context. Our depart_title
	# doesn't pop anything when this flag is set.
	self.within_ignored_h1 = True
	else:
	HTMLTranslator.visit_title(self, node)

	def depart_title(self, node):
	if not self.within_ignored_h1:
	HTMLTranslator.depart_title(self, node)
	self.within_ignored_h1 = False

	def visit_Text(self, node):
	if not self.within_ignored_h1:
	HTMLTranslator.visit_Text(self, node)

	def visit_topic(self, node):
	if 'contents' in node['classes']:
	# Detect a TOC: this requires special treatment for devsite.
	self.within_toc = True
	# Emit <nav> manually and not through starttage because we don't want
	# additional class components to be added
	self.body.append('\n<nav class="inline-toc">')

	def depart_topic(self, node):
	if self.within_toc:
	self.within_toc = False
	self.body.append('</nav>\n')

	def write_colspecs(self):
	# Override this method from docutils to do nothing. We don't need those
	# pesky <col width=NN /> tags in our markup.
	pass

	def visit_admonition(self, node, name=''):
	self.body.append(self.starttag(node, 'aside', CLASS=node.get('class', '')))

	def depart_admonition(self, node=''):
	self.body.append('\n</aside>\n')

	def unknown_visit(self, node):
	raise NotImplementedError('Unknown node: ' + node.__class__.__name__)


	class DevsiteBuilder(StandaloneHTMLBuilder):
	""" Builder for the NaCl devsite HTML output.

	Loosely based on the code of Sphinx's standard SerializingHTMLBuilder.
	"""
	name = 'devsite'
	out_suffix = '.html'
	link_suffix = '.html'

	# Disable the addition of "pi"-permalinks to each section header
	add_permalinks = False

	def init(self):
	self.devsite_production_mode = int(self.config.devsite_production_mode) == 1
	self.kill_internal_links = int(self.config.kill_internal_links) == 1
	self.info("----> Devsite builder with production mode = %d" % (
	self.devsite_production_mode,))
	self.config_hash = ''
	self.tags_hash = ''
	self.theme = None # no theme necessary
	self.templates = None # no template bridge necessary
	self.init_translator_class()
	self.init_highlighter()

	def finish(self):
	super(DevsiteBuilder, self).finish()
	if self.devsite_production_mode:
	# We decided to keep the manual _book.yaml for now;
	# The code for auto-generating YAML TOCs from index.rst was removed in
	# https://codereview.chromium.org/57923006/
	self.info(bold('generating YAML table-of-contents... '))
	subs = {
	'version': PEPPER_VERSION,
	'folder': self.config.devsite_foldername or ''}
	with open(os.path.join(self.env.srcdir, '_book.yaml')) as in_f:
	with open(os.path.join(self.outdir, '_book.yaml'), 'w') as out_f:
	out_f.write(string.Template(in_f.read()).substitute(subs))
	self.info()

	def dump_inventory(self):
	# We don't want an inventory file when building for devsite
	if not self.devsite_production_mode:
	super(DevsiteBuilder, self).dump_inventory()

	def get_production_url(self, url):
	if not self.devsite_production_mode:
	return url

	if self.config.devsite_foldername:
	return '/native-client/%s/%s' % (self.config.devsite_foldername, url)

	return '/native-client/%s' % url

	def get_target_uri(self, docname, typ=None):
	if self.devsite_production_mode:
	return self.get_production_url(docname)
	else:
	return docname + self.link_suffix

	def handle_page(self, pagename, ctx, templatename='page.html',
	outfilename=None, event_arg=None):
	ctx['current_page_name'] = pagename

	if not outfilename:
	outfilename = os.path.join(self.outdir,
	pagename + self.out_suffix)

	# Emit an event to Sphinx
	self.app.emit('html-page-context', pagename, templatename,
	ctx, event_arg)

	ensuredir(os.path.dirname(outfilename))
	self._dump_context(ctx, outfilename)

	def _dump_context(self, context, filename):
	""" Do the actual dumping of the page to the file. context is a dict. Some
	important fields:
	body - document contents
	title
	current_page_name
	Some special pages (genindex, etc.) may not have some of the fields, so
	fetch them conservatively.
	"""
	if not 'body' in context:
	return

	folder = ''
	if self.devsite_production_mode and self.config.devsite_foldername:
	folder = "/" + self.config.devsite_foldername

	# codecs.open is the fast Python 2.x way of emulating the encoding= argument
	# in Python 3's builtin open.
	with codecs.open(filename, 'w', encoding='utf-8') as f:
	f.write(PAGE_TEMPLATE.substitute(
	doc_title=context.get('title', ''),
	doc_body=context.get('body'),
	folder=folder,
	nonprod_css=self._conditional_nonprod(NONPROD_CSS),
	nonprod_meta_head=self._conditional_nonprod(NONPROD_META_HEAD),
	devsite_prefix=self._conditional_devsite(DEVSITE_PREFIX),
	devsite_butterbar=self._conditional_devsite(DEVSITE_BUTTERBAR)))

	def _conditional_devsite(self, s):
	return s if self.devsite_production_mode else ''

	def _conditional_nonprod(self, s):
	return s if not self.devsite_production_mode else ''


	class NaclCodeDirective(Directive):
	""" Custom "naclcode" directive for code snippets. To keep it under our
	control.
	"""
	has_content = True
	required_arguments = 0
	optional_arguments = 1
	option_spec = {
	'prettyprint': int,
	}

	def run(self):
	code = u'\n'.join(self.content)
	literal = nodes.literal_block(code, code)
	literal['prettyprint'] = self.options.get('prettyprint', 1)
	return [literal]

	def setup(app):
	""" Extension registration hook.
	"""
	# linkcheck issues HEAD requests to save time, but some Google properties
	# reject them and we get spurious 405 responses. Monkey-patch sphinx to
	# just use normal GET requests.
	# See: https://bitbucket.org/birkenfeld/sphinx/issue/1292/
	from sphinx.builders import linkcheck
	import urllib2
	linkcheck.HeadRequest = urllib2.Request

	app.add_directive('naclcode', NaclCodeDirective)
	app.add_builder(DevsiteBuilder)

	# "Production mode" for local testing vs. on-server documentation.
	app.add_config_value('devsite_production_mode', default='1', rebuild='html')
	app.add_config_value('kill_internal_links', default='0', rebuild='html')
	app.add_config_value('devsite_foldername', default=None, rebuild='html')