# -*- coding: utf-8 -*-
"""
sphinx.builders.htmlhelp
~~~~~~~~~~~~~~~~~~~~~~~~
Build HTML help support files.
Parts adapted from Python's Doc/tools/prechm.py.
:copyright: Copyright 2007-2010 by the Sphinx team, see AUTHORS.
:license: BSD, see LICENSE for details.
"""
import os
import cgi
import codecs
from os import path
from docutils import nodes
from sphinx import addnodes
from sphinx.builders.html import StandaloneHTMLBuilder
# Project file (*.hhp) template. 'outname' is the file basename (like
# the pythlp in pythlp.hhp); 'version' is the doc version number (like
# the 2.2 in Python 2.2).
# The magical numbers in the long line under [WINDOWS] set most of the
# user-visible features (visible buttons, tabs, etc).
# About 0x10384e: This defines the buttons in the help viewer. The
# following defns are taken from htmlhelp.h. Not all possibilities
# actually work, and not all those that work are available from the Help
# Workshop GUI. In particular, the Zoom/Font button works and is not
# available from the GUI. The ones we're using are marked with 'x':
#
# 0x000002 Hide/Show x
# 0x000004 Back x
# 0x000008 Forward x
# 0x000010 Stop
# 0x000020 Refresh
# 0x000040 Home x
# 0x000080 Forward
# 0x000100 Back
# 0x000200 Notes
# 0x000400 Contents
# 0x000800 Locate x
# 0x001000 Options x
# 0x002000 Print x
# 0x004000 Index
# 0x008000 Search
# 0x010000 History
# 0x020000 Favorites
# 0x040000 Jump 1
# 0x080000 Jump 2
# 0x100000 Zoom/Font x
# 0x200000 TOC Next
# 0x400000 TOC Prev
project_template = '''\
[OPTIONS]
Binary TOC=Yes
Binary Index=No
Compiled file=%(outname)s.chm
Contents file=%(outname)s.hhc
Default Window=%(outname)s
Default topic=index.html
Display compile progress=No
Full text search stop list file=%(outname)s.stp
Full-text search=Yes
Index file=%(outname)s.hhk
Language=%(lcid)#x
Title=%(title)s
[WINDOWS]
%(outname)s="%(title)s","%(outname)s.hhc","%(outname)s.hhk",\
"index.html","index.html",,,,,0x63520,220,0x10384e,[0,0,1024,768],,,,,,,0
[FILES]
'''
contents_header = '''\
'''
contents_footer = '''\
'''
object_sitemap = '''\
'''
# List of words the full text search facility shouldn't index. This
# becomes file outname.stp. Note that this list must be pretty small!
# Different versions of the MS docs claim the file has a maximum size of
# 256 or 512 bytes (including \r\n at the end of each line).
# Note that "and", "or", "not" and "near" are operators in the search
# language, so no point indexing them even if we wanted to.
stopwords = """
a and are as at
be but by
for
if in into is it
near no not
of on or
such
that the their then there these they this to
was will with
""".split()
# The following list includes only languages supported by Sphinx.
# See http://msdn.microsoft.com/en-us/library/ms930130.aspx for more.
chm_locales = {
# lang: LCID, encoding
'cs': (0x405, 'iso8859_2'),
'de': (0x407, 'iso8859_1'),
'en': (0x409, 'iso8859_1'),
'es': (0x40a, 'iso8859_1'),
'fi': (0x40b, 'iso8859_1'),
'fr': (0x40c, 'iso8859_1'),
'it': (0x410, 'iso8859_1'),
'ja': (0x411, 'cp932'),
'nl': (0x413, 'iso8859_1'),
'pl': (0x415, 'iso8859_2'),
'pt_BR': (0x416, 'iso8859_1'),
'ru': (0x419, 'cp1251'),
'sl': (0x424, 'iso8859_2'),
'uk_UA': (0x422, 'cp1251'),
'zh_CN': (0x804, 'cp936'),
'zh_TW': (0x404, 'cp950'),
}
class HTMLHelpBuilder(StandaloneHTMLBuilder):
"""
Builder that also outputs Windows HTML help project, contents and
index files. Adapted from the original Doc/tools/prechm.py.
"""
name = 'htmlhelp'
# don't copy the reST source
copysource = False
supported_image_types = ['image/png', 'image/gif', 'image/jpeg']
# don't add links
add_permalinks = False
# don't add sidebar etc.
embedded = True
lcid = 0x409
encoding = 'iso8859_1'
def init(self):
StandaloneHTMLBuilder.init(self)
# the output files for HTML help must be .html only
self.out_suffix = '.html'
# determine the correct locale setting
locale = chm_locales.get(self.config.language)
if locale is not None:
self.lcid, self.encoding = locale
def open_file(self, outdir, basename, mode='w'):
# open a file with the correct encoding for the selected language
return codecs.open(path.join(outdir, basename), mode,
self.encoding, 'xmlcharrefreplace')
def handle_finish(self):
self.build_hhx(self.outdir, self.config.htmlhelp_basename)
def build_hhx(self, outdir, outname):
self.info('dumping stopword list...')
f = self.open_file(outdir, outname+'.stp')
try:
for word in sorted(stopwords):
print >>f, word
finally:
f.close()
self.info('writing project file...')
f = self.open_file(outdir, outname+'.hhp')
try:
f.write(project_template % {'outname': outname,
'title': self.config.html_title,
'version': self.config.version,
'project': self.config.project,
'lcid': self.lcid})
if not outdir.endswith(os.sep):
outdir += os.sep
olen = len(outdir)
for root, dirs, files in os.walk(outdir):
staticdir = (root == path.join(outdir, '_static'))
for fn in files:
if (staticdir and not fn.endswith('.js')) or \
fn.endswith('.html'):
print >>f, path.join(root, fn)[olen:].replace(os.sep,
'\\')
finally:
f.close()
self.info('writing TOC file...')
f = self.open_file(outdir, outname+'.hhc')
try:
f.write(contents_header)
# special books
f.write('
' + object_sitemap % (self.config.html_short_title,
'index.html'))
if self.config.html_use_modindex:
f.write('
')
for subnode in node:
write_toc(subnode, ullevel)
elif isinstance(node, nodes.reference):
link = node['refuri']
title = cgi.escape(node.astext()).replace('"','"')
f.write(object_sitemap % (title, link))
elif isinstance(node, nodes.bullet_list):
if ullevel != 0:
f.write('
\n')
for subnode in node:
write_toc(subnode, ullevel+1)
if ullevel != 0:
f.write('
\n')
elif isinstance(node, addnodes.compact_paragraph):
for subnode in node:
write_toc(subnode, ullevel)
def istoctree(node):
return isinstance(node, addnodes.compact_paragraph) and \
node.has_key('toctree')
for node in tocdoc.traverse(istoctree):
write_toc(node)
f.write(contents_footer)
finally:
f.close()
self.info('writing index file...')
index = self.env.create_index(self)
f = self.open_file(outdir, outname+'.hhk')
try:
f.write('