868 lines
29 KiB
Python
868 lines
29 KiB
Python
|
|
"""SCons.Tool.docbook
|
|
|
|
Tool-specific initialization for Docbook.
|
|
|
|
There normally shouldn't be any need to import this module directly.
|
|
It will usually be imported through the generic SCons.Tool.Tool()
|
|
selection method.
|
|
|
|
"""
|
|
|
|
#
|
|
# Copyright (c) 2001 - 2016 The SCons Foundation
|
|
#
|
|
# Permission is hereby granted, free of charge, to any person obtaining
|
|
# a copy of this software and associated documentation files (the
|
|
# "Software"), to deal in the Software without restriction, including
|
|
# without limitation the rights to use, copy, modify, merge, publish,
|
|
# distribute, sublicense, and/or sell copies of the Software, and to
|
|
# permit persons to whom the Software is furnished to do so, subject to
|
|
# the following conditions:
|
|
#
|
|
# The above copyright notice and this permission notice shall be included
|
|
# in all copies or substantial portions of the Software.
|
|
#
|
|
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY
|
|
# KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
|
|
# WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
|
|
# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
|
|
# LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
|
|
# OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
|
|
# WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
#
|
|
|
|
import os
|
|
import glob
|
|
import re
|
|
|
|
import SCons.Action
|
|
import SCons.Builder
|
|
import SCons.Defaults
|
|
import SCons.Script
|
|
import SCons.Tool
|
|
import SCons.Util
|
|
|
|
# Get full path to this script
|
|
scriptpath = os.path.dirname(os.path.realpath(__file__))
|
|
|
|
# Local folder for the collection of DocBook XSLs
|
|
db_xsl_folder = 'docbook-xsl-1.76.1'
|
|
|
|
# Do we have libxml2/libxslt/lxml?
|
|
has_libxml2 = True
|
|
has_lxml = True
|
|
try:
|
|
import libxml2
|
|
import libxslt
|
|
except:
|
|
has_libxml2 = False
|
|
try:
|
|
import lxml
|
|
except:
|
|
has_lxml = False
|
|
|
|
# Set this to True, to prefer xsltproc over libxml2 and lxml
|
|
prefer_xsltproc = False
|
|
|
|
# Regexs for parsing Docbook XML sources of MAN pages
|
|
re_manvolnum = re.compile("<manvolnum>([^<]*)</manvolnum>")
|
|
re_refname = re.compile("<refname>([^<]*)</refname>")
|
|
|
|
#
|
|
# Helper functions
|
|
#
|
|
def __extend_targets_sources(target, source):
|
|
""" Prepare the lists of target and source files. """
|
|
if not SCons.Util.is_List(target):
|
|
target = [target]
|
|
if not source:
|
|
source = target[:]
|
|
elif not SCons.Util.is_List(source):
|
|
source = [source]
|
|
if len(target) < len(source):
|
|
target.extend(source[len(target):])
|
|
|
|
return target, source
|
|
|
|
def __init_xsl_stylesheet(kw, env, user_xsl_var, default_path):
|
|
if kw.get('DOCBOOK_XSL','') == '':
|
|
xsl_style = kw.get('xsl', env.subst(user_xsl_var))
|
|
if xsl_style == '':
|
|
path_args = [scriptpath, db_xsl_folder] + default_path
|
|
xsl_style = os.path.join(*path_args)
|
|
kw['DOCBOOK_XSL'] = xsl_style
|
|
|
|
def __select_builder(lxml_builder, libxml2_builder, cmdline_builder):
|
|
""" Selects a builder, based on which Python modules are present. """
|
|
if prefer_xsltproc:
|
|
return cmdline_builder
|
|
|
|
if not has_libxml2:
|
|
# At the moment we prefer libxml2 over lxml, the latter can lead
|
|
# to conflicts when installed together with libxml2.
|
|
if has_lxml:
|
|
return lxml_builder
|
|
else:
|
|
return cmdline_builder
|
|
|
|
return libxml2_builder
|
|
|
|
def __ensure_suffix(t, suffix):
|
|
""" Ensure that the target t has the given suffix. """
|
|
tpath = str(t)
|
|
if not tpath.endswith(suffix):
|
|
return tpath+suffix
|
|
|
|
return t
|
|
|
|
def __ensure_suffix_stem(t, suffix):
|
|
""" Ensure that the target t has the given suffix, and return the file's stem. """
|
|
tpath = str(t)
|
|
if not tpath.endswith(suffix):
|
|
stem = tpath
|
|
tpath += suffix
|
|
|
|
return tpath, stem
|
|
else:
|
|
stem, ext = os.path.splitext(tpath)
|
|
|
|
return t, stem
|
|
|
|
def __get_xml_text(root):
|
|
""" Return the text for the given root node (xml.dom.minidom). """
|
|
txt = ""
|
|
for e in root.childNodes:
|
|
if (e.nodeType == e.TEXT_NODE):
|
|
txt += e.data
|
|
return txt
|
|
|
|
def __create_output_dir(base_dir):
|
|
""" Ensure that the output directory base_dir exists. """
|
|
root, tail = os.path.split(base_dir)
|
|
dir = None
|
|
if tail:
|
|
if base_dir.endswith('/'):
|
|
dir = base_dir
|
|
else:
|
|
dir = root
|
|
else:
|
|
if base_dir.endswith('/'):
|
|
dir = base_dir
|
|
|
|
if dir and not os.path.isdir(dir):
|
|
os.makedirs(dir)
|
|
|
|
|
|
#
|
|
# Supported command line tools and their call "signature"
|
|
#
|
|
xsltproc_com = {'xsltproc' : '$DOCBOOK_XSLTPROC $DOCBOOK_XSLTPROCFLAGS -o $TARGET $DOCBOOK_XSL $SOURCE',
|
|
'saxon' : '$DOCBOOK_XSLTPROC $DOCBOOK_XSLTPROCFLAGS -o $TARGET $DOCBOOK_XSL $SOURCE $DOCBOOK_XSLTPROCPARAMS',
|
|
'saxon-xslt' : '$DOCBOOK_XSLTPROC $DOCBOOK_XSLTPROCFLAGS -o $TARGET $DOCBOOK_XSL $SOURCE $DOCBOOK_XSLTPROCPARAMS',
|
|
'xalan' : '$DOCBOOK_XSLTPROC $DOCBOOK_XSLTPROCFLAGS -q -out $TARGET -xsl $DOCBOOK_XSL -in $SOURCE'}
|
|
xmllint_com = {'xmllint' : '$DOCBOOK_XMLLINT $DOCBOOK_XMLLINTFLAGS --xinclude $SOURCE > $TARGET'}
|
|
fop_com = {'fop' : '$DOCBOOK_FOP $DOCBOOK_FOPFLAGS -fo $SOURCE -pdf $TARGET',
|
|
'xep' : '$DOCBOOK_FOP $DOCBOOK_FOPFLAGS -valid -fo $SOURCE -pdf $TARGET',
|
|
'jw' : '$DOCBOOK_FOP $DOCBOOK_FOPFLAGS -f docbook -b pdf $SOURCE -o $TARGET'}
|
|
|
|
def __detect_cl_tool(env, chainkey, cdict):
|
|
"""
|
|
Helper function, picks a command line tool from the list
|
|
and initializes its environment variables.
|
|
"""
|
|
if env.get(chainkey,'') == '':
|
|
clpath = ''
|
|
for cltool in cdict:
|
|
clpath = env.WhereIs(cltool)
|
|
if clpath:
|
|
env[chainkey] = clpath
|
|
if not env[chainkey + 'COM']:
|
|
env[chainkey + 'COM'] = cdict[cltool]
|
|
|
|
def _detect(env):
|
|
"""
|
|
Detect all the command line tools that we might need for creating
|
|
the requested output formats.
|
|
"""
|
|
global prefer_xsltproc
|
|
|
|
if env.get('DOCBOOK_PREFER_XSLTPROC',''):
|
|
prefer_xsltproc = True
|
|
|
|
if ((not has_libxml2 and not has_lxml) or (prefer_xsltproc)):
|
|
# Try to find the XSLT processors
|
|
__detect_cl_tool(env, 'DOCBOOK_XSLTPROC', xsltproc_com)
|
|
__detect_cl_tool(env, 'DOCBOOK_XMLLINT', xmllint_com)
|
|
|
|
__detect_cl_tool(env, 'DOCBOOK_FOP', fop_com)
|
|
|
|
#
|
|
# Scanners
|
|
#
|
|
include_re = re.compile('fileref\\s*=\\s*["|\']([^\\n]*)["|\']')
|
|
sentity_re = re.compile('<!ENTITY\\s+%*\\s*[^\\s]+\\s+SYSTEM\\s+["|\']([^\\n]*)["|\']>')
|
|
|
|
def __xml_scan(node, env, path, arg):
|
|
""" Simple XML file scanner, detecting local images and XIncludes as implicit dependencies. """
|
|
# Does the node exist yet?
|
|
if not os.path.isfile(str(node)):
|
|
return []
|
|
|
|
if env.get('DOCBOOK_SCANENT',''):
|
|
# Use simple pattern matching for system entities..., no support
|
|
# for recursion yet.
|
|
contents = node.get_text_contents()
|
|
return sentity_re.findall(contents)
|
|
|
|
xsl_file = os.path.join(scriptpath,'utils','xmldepend.xsl')
|
|
if not has_libxml2 or prefer_xsltproc:
|
|
if has_lxml and not prefer_xsltproc:
|
|
|
|
from lxml import etree
|
|
|
|
xsl_tree = etree.parse(xsl_file)
|
|
doc = etree.parse(str(node))
|
|
result = doc.xslt(xsl_tree)
|
|
|
|
depfiles = [x.strip() for x in str(result).splitlines() if x.strip() != "" and not x.startswith("<?xml ")]
|
|
return depfiles
|
|
else:
|
|
# Try to call xsltproc
|
|
xsltproc = env.subst("$DOCBOOK_XSLTPROC")
|
|
if xsltproc and xsltproc.endswith('xsltproc'):
|
|
result = env.backtick(' '.join([xsltproc, xsl_file, str(node)]))
|
|
depfiles = [x.strip() for x in str(result).splitlines() if x.strip() != "" and not x.startswith("<?xml ")]
|
|
return depfiles
|
|
else:
|
|
# Use simple pattern matching, there is currently no support
|
|
# for xi:includes...
|
|
contents = node.get_text_contents()
|
|
return include_re.findall(contents)
|
|
|
|
styledoc = libxml2.parseFile(xsl_file)
|
|
style = libxslt.parseStylesheetDoc(styledoc)
|
|
doc = libxml2.readFile(str(node), None, libxml2.XML_PARSE_NOENT)
|
|
result = style.applyStylesheet(doc, None)
|
|
|
|
depfiles = []
|
|
for x in str(result).splitlines():
|
|
if x.strip() != "" and not x.startswith("<?xml "):
|
|
depfiles.extend(x.strip().split())
|
|
|
|
style.freeStylesheet()
|
|
doc.freeDoc()
|
|
result.freeDoc()
|
|
|
|
return depfiles
|
|
|
|
# Creating the instance of our XML dependency scanner
|
|
docbook_xml_scanner = SCons.Script.Scanner(function = __xml_scan,
|
|
argument = None)
|
|
|
|
|
|
#
|
|
# Action generators
|
|
#
|
|
def __generate_xsltproc_action(source, target, env, for_signature):
|
|
cmd = env['DOCBOOK_XSLTPROCCOM']
|
|
# Does the environment have a base_dir defined?
|
|
base_dir = env.subst('$base_dir')
|
|
if base_dir:
|
|
# Yes, so replace target path by its filename
|
|
return cmd.replace('$TARGET','${TARGET.file}')
|
|
return cmd
|
|
|
|
|
|
#
|
|
# Emitters
|
|
#
|
|
def __emit_xsl_basedir(target, source, env):
|
|
# Does the environment have a base_dir defined?
|
|
base_dir = env.subst('$base_dir')
|
|
if base_dir:
|
|
# Yes, so prepend it to each target
|
|
return [os.path.join(base_dir, str(t)) for t in target], source
|
|
|
|
# No, so simply pass target and source names through
|
|
return target, source
|
|
|
|
|
|
#
|
|
# Builders
|
|
#
|
|
def __build_libxml2(target, source, env):
|
|
"""
|
|
General XSLT builder (HTML/FO), using the libxml2 module.
|
|
"""
|
|
xsl_style = env.subst('$DOCBOOK_XSL')
|
|
styledoc = libxml2.parseFile(xsl_style)
|
|
style = libxslt.parseStylesheetDoc(styledoc)
|
|
doc = libxml2.readFile(str(source[0]),None,libxml2.XML_PARSE_NOENT)
|
|
# Support for additional parameters
|
|
parampass = {}
|
|
if parampass:
|
|
result = style.applyStylesheet(doc, parampass)
|
|
else:
|
|
result = style.applyStylesheet(doc, None)
|
|
style.saveResultToFilename(str(target[0]), result, 0)
|
|
style.freeStylesheet()
|
|
doc.freeDoc()
|
|
result.freeDoc()
|
|
|
|
return None
|
|
|
|
def __build_lxml(target, source, env):
|
|
"""
|
|
General XSLT builder (HTML/FO), using the lxml module.
|
|
"""
|
|
from lxml import etree
|
|
|
|
xslt_ac = etree.XSLTAccessControl(read_file=True,
|
|
write_file=True,
|
|
create_dir=True,
|
|
read_network=False,
|
|
write_network=False)
|
|
xsl_style = env.subst('$DOCBOOK_XSL')
|
|
xsl_tree = etree.parse(xsl_style)
|
|
transform = etree.XSLT(xsl_tree, access_control=xslt_ac)
|
|
doc = etree.parse(str(source[0]))
|
|
# Support for additional parameters
|
|
parampass = {}
|
|
if parampass:
|
|
result = transform(doc, **parampass)
|
|
else:
|
|
result = transform(doc)
|
|
|
|
try:
|
|
of = open(str(target[0]), "w")
|
|
of.write(of.write(etree.tostring(result, pretty_print=True)))
|
|
of.close()
|
|
except:
|
|
pass
|
|
|
|
return None
|
|
|
|
def __xinclude_libxml2(target, source, env):
|
|
"""
|
|
Resolving XIncludes, using the libxml2 module.
|
|
"""
|
|
doc = libxml2.readFile(str(source[0]), None, libxml2.XML_PARSE_NOENT)
|
|
doc.xincludeProcessFlags(libxml2.XML_PARSE_NOENT)
|
|
doc.saveFile(str(target[0]))
|
|
doc.freeDoc()
|
|
|
|
return None
|
|
|
|
def __xinclude_lxml(target, source, env):
|
|
"""
|
|
Resolving XIncludes, using the lxml module.
|
|
"""
|
|
from lxml import etree
|
|
|
|
doc = etree.parse(str(source[0]))
|
|
doc.xinclude()
|
|
try:
|
|
doc.write(str(target[0]), xml_declaration=True,
|
|
encoding="UTF-8", pretty_print=True)
|
|
except:
|
|
pass
|
|
|
|
return None
|
|
|
|
__libxml2_builder = SCons.Builder.Builder(
|
|
action = __build_libxml2,
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner,
|
|
emitter = __emit_xsl_basedir)
|
|
__lxml_builder = SCons.Builder.Builder(
|
|
action = __build_lxml,
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner,
|
|
emitter = __emit_xsl_basedir)
|
|
|
|
__xinclude_libxml2_builder = SCons.Builder.Builder(
|
|
action = __xinclude_libxml2,
|
|
suffix = '.xml',
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner)
|
|
__xinclude_lxml_builder = SCons.Builder.Builder(
|
|
action = __xinclude_lxml,
|
|
suffix = '.xml',
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner)
|
|
|
|
__xsltproc_builder = SCons.Builder.Builder(
|
|
action = SCons.Action.CommandGeneratorAction(__generate_xsltproc_action,
|
|
{'cmdstr' : '$DOCBOOK_XSLTPROCCOMSTR'}),
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner,
|
|
emitter = __emit_xsl_basedir)
|
|
__xmllint_builder = SCons.Builder.Builder(
|
|
action = SCons.Action.Action('$DOCBOOK_XMLLINTCOM','$DOCBOOK_XMLLINTCOMSTR'),
|
|
suffix = '.xml',
|
|
src_suffix = '.xml',
|
|
source_scanner = docbook_xml_scanner)
|
|
__fop_builder = SCons.Builder.Builder(
|
|
action = SCons.Action.Action('$DOCBOOK_FOPCOM','$DOCBOOK_FOPCOMSTR'),
|
|
suffix = '.pdf',
|
|
src_suffix = '.fo',
|
|
ensure_suffix=1)
|
|
|
|
def DocbookEpub(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for ePub output.
|
|
"""
|
|
import zipfile
|
|
import shutil
|
|
|
|
def build_open_container(target, source, env):
|
|
"""Generate the *.epub file from intermediate outputs
|
|
|
|
Constructs the epub file according to the Open Container Format. This
|
|
function could be replaced by a call to the SCons Zip builder if support
|
|
was added for different compression formats for separate source nodes.
|
|
"""
|
|
zf = zipfile.ZipFile(str(target[0]), 'w')
|
|
mime_file = open('mimetype', 'w')
|
|
mime_file.write('application/epub+zip')
|
|
mime_file.close()
|
|
zf.write(mime_file.name, compress_type = zipfile.ZIP_STORED)
|
|
for s in source:
|
|
if os.path.isfile(str(s)):
|
|
head, tail = os.path.split(str(s))
|
|
if not head:
|
|
continue
|
|
s = head
|
|
for dirpath, dirnames, filenames in os.walk(str(s)):
|
|
for fname in filenames:
|
|
path = os.path.join(dirpath, fname)
|
|
if os.path.isfile(path):
|
|
zf.write(path, os.path.relpath(path, str(env.get('ZIPROOT', ''))),
|
|
zipfile.ZIP_DEFLATED)
|
|
zf.close()
|
|
|
|
def add_resources(target, source, env):
|
|
"""Add missing resources to the OEBPS directory
|
|
|
|
Ensure all the resources in the manifest are present in the OEBPS directory.
|
|
"""
|
|
hrefs = []
|
|
content_file = os.path.join(source[0].get_abspath(), 'content.opf')
|
|
if not os.path.isfile(content_file):
|
|
return
|
|
|
|
hrefs = []
|
|
if has_libxml2:
|
|
nsmap = {'opf' : 'http://www.idpf.org/2007/opf'}
|
|
# Read file and resolve entities
|
|
doc = libxml2.readFile(content_file, None, 0)
|
|
opf = doc.getRootElement()
|
|
# Create xpath context
|
|
xpath_context = doc.xpathNewContext()
|
|
# Register namespaces
|
|
for key, val in nsmap.iteritems():
|
|
xpath_context.xpathRegisterNs(key, val)
|
|
|
|
if hasattr(opf, 'xpathEval') and xpath_context:
|
|
# Use the xpath context
|
|
xpath_context.setContextNode(opf)
|
|
items = xpath_context.xpathEval(".//opf:item")
|
|
else:
|
|
items = opf.findall(".//{'http://www.idpf.org/2007/opf'}item")
|
|
|
|
for item in items:
|
|
if hasattr(item, 'prop'):
|
|
hrefs.append(item.prop('href'))
|
|
else:
|
|
hrefs.append(item.attrib['href'])
|
|
|
|
doc.freeDoc()
|
|
xpath_context.xpathFreeContext()
|
|
elif has_lxml:
|
|
from lxml import etree
|
|
|
|
opf = etree.parse(content_file)
|
|
# All the opf:item elements are resources
|
|
for item in opf.xpath('//opf:item',
|
|
namespaces= { 'opf': 'http://www.idpf.org/2007/opf' }):
|
|
hrefs.append(item.attrib['href'])
|
|
|
|
for href in hrefs:
|
|
# If the resource was not already created by DocBook XSL itself,
|
|
# copy it into the OEBPS folder
|
|
referenced_file = os.path.join(source[0].get_abspath(), href)
|
|
if not os.path.exists(referenced_file):
|
|
shutil.copy(href, os.path.join(source[0].get_abspath(), href))
|
|
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_EPUB', ['epub','docbook.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
if not env.GetOption('clean'):
|
|
# Ensure that the folders OEBPS and META-INF exist
|
|
__create_output_dir('OEBPS/')
|
|
__create_output_dir('META-INF/')
|
|
dirs = env.Dir(['OEBPS', 'META-INF'])
|
|
|
|
# Set the fixed base_dir
|
|
kw['base_dir'] = 'OEBPS/'
|
|
tocncx = __builder.__call__(env, 'toc.ncx', source[0], **kw)
|
|
cxml = env.File('META-INF/container.xml')
|
|
env.SideEffect(cxml, tocncx)
|
|
|
|
env.Depends(tocncx, kw['DOCBOOK_XSL'])
|
|
result.extend(tocncx+[cxml])
|
|
|
|
container = env.Command(__ensure_suffix(str(target[0]), '.epub'),
|
|
tocncx+[cxml], [add_resources, build_open_container])
|
|
mimetype = env.File('mimetype')
|
|
env.SideEffect(mimetype, container)
|
|
|
|
result.extend(container)
|
|
# Add supporting files for cleanup
|
|
env.Clean(tocncx, dirs)
|
|
|
|
return result
|
|
|
|
def DocbookHtml(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for HTML output.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_HTML', ['html','docbook.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
r = __builder.__call__(env, __ensure_suffix(t,'.html'), s, **kw)
|
|
env.Depends(r, kw['DOCBOOK_XSL'])
|
|
result.extend(r)
|
|
|
|
return result
|
|
|
|
def DocbookHtmlChunked(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for chunked HTML output.
|
|
"""
|
|
# Init target/source
|
|
if not SCons.Util.is_List(target):
|
|
target = [target]
|
|
if not source:
|
|
source = target
|
|
target = ['index.html']
|
|
elif not SCons.Util.is_List(source):
|
|
source = [source]
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_HTMLCHUNKED', ['html','chunkfast.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Detect base dir
|
|
base_dir = kw.get('base_dir', '')
|
|
if base_dir:
|
|
__create_output_dir(base_dir)
|
|
|
|
# Create targets
|
|
result = []
|
|
r = __builder.__call__(env, __ensure_suffix(str(target[0]), '.html'), source[0], **kw)
|
|
env.Depends(r, kw['DOCBOOK_XSL'])
|
|
result.extend(r)
|
|
# Add supporting files for cleanup
|
|
env.Clean(r, glob.glob(os.path.join(base_dir, '*.html')))
|
|
|
|
return result
|
|
|
|
|
|
def DocbookHtmlhelp(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for HTMLHELP output.
|
|
"""
|
|
# Init target/source
|
|
if not SCons.Util.is_List(target):
|
|
target = [target]
|
|
if not source:
|
|
source = target
|
|
target = ['index.html']
|
|
elif not SCons.Util.is_List(source):
|
|
source = [source]
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_HTMLHELP', ['htmlhelp','htmlhelp.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Detect base dir
|
|
base_dir = kw.get('base_dir', '')
|
|
if base_dir:
|
|
__create_output_dir(base_dir)
|
|
|
|
# Create targets
|
|
result = []
|
|
r = __builder.__call__(env, __ensure_suffix(str(target[0]), '.html'), source[0], **kw)
|
|
env.Depends(r, kw['DOCBOOK_XSL'])
|
|
result.extend(r)
|
|
# Add supporting files for cleanup
|
|
env.Clean(r, ['toc.hhc', 'htmlhelp.hhp', 'index.hhk'] +
|
|
glob.glob(os.path.join(base_dir, '[ar|bk|ch]*.html')))
|
|
|
|
return result
|
|
|
|
def DocbookPdf(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for PDF output.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_PDF', ['fo','docbook.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
t, stem = __ensure_suffix_stem(t, '.pdf')
|
|
xsl = __builder.__call__(env, stem+'.fo', s, **kw)
|
|
result.extend(xsl)
|
|
env.Depends(xsl, kw['DOCBOOK_XSL'])
|
|
result.extend(__fop_builder.__call__(env, t, xsl, **kw))
|
|
|
|
return result
|
|
|
|
def DocbookMan(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for Man page output.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_MAN', ['manpages','docbook.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
volnum = "1"
|
|
outfiles = []
|
|
srcfile = __ensure_suffix(str(s),'.xml')
|
|
if os.path.isfile(srcfile):
|
|
try:
|
|
import xml.dom.minidom
|
|
|
|
dom = xml.dom.minidom.parse(__ensure_suffix(str(s),'.xml'))
|
|
# Extract volume number, default is 1
|
|
for node in dom.getElementsByTagName('refmeta'):
|
|
for vol in node.getElementsByTagName('manvolnum'):
|
|
volnum = __get_xml_text(vol)
|
|
|
|
# Extract output filenames
|
|
for node in dom.getElementsByTagName('refnamediv'):
|
|
for ref in node.getElementsByTagName('refname'):
|
|
outfiles.append(__get_xml_text(ref)+'.'+volnum)
|
|
|
|
except:
|
|
# Use simple regex parsing
|
|
f = open(__ensure_suffix(str(s),'.xml'), 'r')
|
|
content = f.read()
|
|
f.close()
|
|
|
|
for m in re_manvolnum.finditer(content):
|
|
volnum = m.group(1)
|
|
|
|
for m in re_refname.finditer(content):
|
|
outfiles.append(m.group(1)+'.'+volnum)
|
|
|
|
if not outfiles:
|
|
# Use stem of the source file
|
|
spath = str(s)
|
|
if not spath.endswith('.xml'):
|
|
outfiles.append(spath+'.'+volnum)
|
|
else:
|
|
stem, ext = os.path.splitext(spath)
|
|
outfiles.append(stem+'.'+volnum)
|
|
else:
|
|
# We have to completely rely on the given target name
|
|
outfiles.append(t)
|
|
|
|
__builder.__call__(env, outfiles[0], s, **kw)
|
|
env.Depends(outfiles[0], kw['DOCBOOK_XSL'])
|
|
result.append(outfiles[0])
|
|
if len(outfiles) > 1:
|
|
env.Clean(outfiles[0], outfiles[1:])
|
|
|
|
|
|
return result
|
|
|
|
def DocbookSlidesPdf(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for PDF slides output.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_SLIDESPDF', ['slides','fo','plain.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
t, stem = __ensure_suffix_stem(t, '.pdf')
|
|
xsl = __builder.__call__(env, stem+'.fo', s, **kw)
|
|
env.Depends(xsl, kw['DOCBOOK_XSL'])
|
|
result.extend(xsl)
|
|
result.extend(__fop_builder.__call__(env, t, xsl, **kw))
|
|
|
|
return result
|
|
|
|
def DocbookSlidesHtml(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, providing a Docbook toolchain for HTML slides output.
|
|
"""
|
|
# Init list of targets/sources
|
|
if not SCons.Util.is_List(target):
|
|
target = [target]
|
|
if not source:
|
|
source = target
|
|
target = ['index.html']
|
|
elif not SCons.Util.is_List(source):
|
|
source = [source]
|
|
|
|
# Init XSL stylesheet
|
|
__init_xsl_stylesheet(kw, env, '$DOCBOOK_DEFAULT_XSL_SLIDESHTML', ['slides','html','plain.xsl'])
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Detect base dir
|
|
base_dir = kw.get('base_dir', '')
|
|
if base_dir:
|
|
__create_output_dir(base_dir)
|
|
|
|
# Create targets
|
|
result = []
|
|
r = __builder.__call__(env, __ensure_suffix(str(target[0]), '.html'), source[0], **kw)
|
|
env.Depends(r, kw['DOCBOOK_XSL'])
|
|
result.extend(r)
|
|
# Add supporting files for cleanup
|
|
env.Clean(r, [os.path.join(base_dir, 'toc.html')] +
|
|
glob.glob(os.path.join(base_dir, 'foil*.html')))
|
|
|
|
return result
|
|
|
|
def DocbookXInclude(env, target, source, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, for resolving XIncludes in a separate processing step.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__xinclude_lxml_builder,__xinclude_libxml2_builder,__xmllint_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
result.extend(__builder.__call__(env, t, s, **kw))
|
|
|
|
return result
|
|
|
|
def DocbookXslt(env, target, source=None, *args, **kw):
|
|
"""
|
|
A pseudo-Builder, applying a simple XSL transformation to the input file.
|
|
"""
|
|
# Init list of targets/sources
|
|
target, source = __extend_targets_sources(target, source)
|
|
|
|
# Init XSL stylesheet
|
|
kw['DOCBOOK_XSL'] = kw.get('xsl', 'transform.xsl')
|
|
|
|
# Setup builder
|
|
__builder = __select_builder(__lxml_builder, __libxml2_builder, __xsltproc_builder)
|
|
|
|
# Create targets
|
|
result = []
|
|
for t,s in zip(target,source):
|
|
r = __builder.__call__(env, t, s, **kw)
|
|
env.Depends(r, kw['DOCBOOK_XSL'])
|
|
result.extend(r)
|
|
|
|
return result
|
|
|
|
|
|
def generate(env):
|
|
"""Add Builders and construction variables for docbook to an Environment."""
|
|
|
|
env.SetDefault(
|
|
# Default names for customized XSL stylesheets
|
|
DOCBOOK_DEFAULT_XSL_EPUB = '',
|
|
DOCBOOK_DEFAULT_XSL_HTML = '',
|
|
DOCBOOK_DEFAULT_XSL_HTMLCHUNKED = '',
|
|
DOCBOOK_DEFAULT_XSL_HTMLHELP = '',
|
|
DOCBOOK_DEFAULT_XSL_PDF = '',
|
|
DOCBOOK_DEFAULT_XSL_MAN = '',
|
|
DOCBOOK_DEFAULT_XSL_SLIDESPDF = '',
|
|
DOCBOOK_DEFAULT_XSL_SLIDESHTML = '',
|
|
|
|
# Paths to the detected executables
|
|
DOCBOOK_XSLTPROC = '',
|
|
DOCBOOK_XMLLINT = '',
|
|
DOCBOOK_FOP = '',
|
|
|
|
# Additional flags for the text processors
|
|
DOCBOOK_XSLTPROCFLAGS = SCons.Util.CLVar(''),
|
|
DOCBOOK_XMLLINTFLAGS = SCons.Util.CLVar(''),
|
|
DOCBOOK_FOPFLAGS = SCons.Util.CLVar(''),
|
|
DOCBOOK_XSLTPROCPARAMS = SCons.Util.CLVar(''),
|
|
|
|
# Default command lines for the detected executables
|
|
DOCBOOK_XSLTPROCCOM = xsltproc_com['xsltproc'],
|
|
DOCBOOK_XMLLINTCOM = xmllint_com['xmllint'],
|
|
DOCBOOK_FOPCOM = fop_com['fop'],
|
|
|
|
# Screen output for the text processors
|
|
DOCBOOK_XSLTPROCCOMSTR = None,
|
|
DOCBOOK_XMLLINTCOMSTR = None,
|
|
DOCBOOK_FOPCOMSTR = None,
|
|
|
|
)
|
|
_detect(env)
|
|
|
|
env.AddMethod(DocbookEpub, "DocbookEpub")
|
|
env.AddMethod(DocbookHtml, "DocbookHtml")
|
|
env.AddMethod(DocbookHtmlChunked, "DocbookHtmlChunked")
|
|
env.AddMethod(DocbookHtmlhelp, "DocbookHtmlhelp")
|
|
env.AddMethod(DocbookPdf, "DocbookPdf")
|
|
env.AddMethod(DocbookMan, "DocbookMan")
|
|
env.AddMethod(DocbookSlidesPdf, "DocbookSlidesPdf")
|
|
env.AddMethod(DocbookSlidesHtml, "DocbookSlidesHtml")
|
|
env.AddMethod(DocbookXInclude, "DocbookXInclude")
|
|
env.AddMethod(DocbookXslt, "DocbookXslt")
|
|
|
|
|
|
def exists(env):
|
|
return 1
|