%PDF- %PDF-
Direktori : /opt/imunify360/venv/lib/python3.11/site-packages/docutils/writers/odf_odt/ |
Current File : //opt/imunify360/venv/lib/python3.11/site-packages/docutils/writers/odf_odt/__init__.py |
# $Id: __init__.py 9258 2022-11-21 14:51:43Z milde $ # Author: Dave Kuhlman <dkuhlman@davekuhlman.org> # Copyright: This module has been placed in the public domain. """ Open Document Format (ODF) Writer. This module is provisional: the API is not settled and may change with any minor Docutils version. """ __docformat__ = 'reStructuredText' from configparser import ConfigParser import copy from io import StringIO import itertools import locale import os import os.path import re import subprocess import tempfile import time from urllib.request import urlopen from urllib.error import HTTPError import weakref from xml.etree import ElementTree as etree from xml.dom import minidom import zipfile import docutils from docutils import frontend, nodes, utils, writers, languages from docutils.parsers.rst.directives.images import PIL # optional from docutils.readers import standalone from docutils.transforms import references # Import pygments and odtwriter pygments formatters if possible. try: import pygments import pygments.lexers from .pygmentsformatter import (OdtPygmentsProgFormatter, OdtPygmentsLaTeXFormatter) except (ImportError, SyntaxError): pygments = None # import warnings # warnings.warn('importing IPShellEmbed', UserWarning) # from IPython.Shell import IPShellEmbed # args = ['-pdb', '-pi1', 'In <\\#>: ', '-pi2', ' .\\D.: ', # '-po', 'Out<\\#>: ', '-nosep'] # ipshell = IPShellEmbed(args, # banner = 'Entering IPython. Press Ctrl-D to exit.', # exit_msg = 'Leaving Interpreter, back to program.') VERSION = '1.0a' IMAGE_NAME_COUNTER = itertools.count() # # ElementTree does not support getparent method (lxml does). # This wrapper class and the following support functions provide # that support for the ability to get the parent of an element. # _parents = weakref.WeakKeyDictionary() if isinstance(etree.Element, type): _ElementInterface = etree.Element else: _ElementInterface = etree._ElementInterface class _ElementInterfaceWrapper(_ElementInterface): def __init__(self, tag, attrib=None): _ElementInterface.__init__(self, tag, attrib) _parents[self] = None def setparent(self, parent): _parents[self] = parent def getparent(self): return _parents[self] # # Constants and globals SPACES_PATTERN = re.compile(r'( +)') TABS_PATTERN = re.compile(r'(\t+)') FILL_PAT1 = re.compile(r'^ +') FILL_PAT2 = re.compile(r' {2,}') TABLESTYLEPREFIX = 'rststyle-table-' TABLENAMEDEFAULT = '%s0' % TABLESTYLEPREFIX TABLEPROPERTYNAMES = ( 'border', 'border-top', 'border-left', 'border-right', 'border-bottom', ) GENERATOR_DESC = 'Docutils.org/odf_odt' NAME_SPACE_1 = 'urn:oasis:names:tc:opendocument:xmlns:office:1.0' CONTENT_NAMESPACE_DICT = CNSD = { # 'office:version': '1.0', 'chart': 'urn:oasis:names:tc:opendocument:xmlns:chart:1.0', 'dc': 'http://purl.org/dc/elements/1.1/', 'dom': 'http://www.w3.org/2001/xml-events', 'dr3d': 'urn:oasis:names:tc:opendocument:xmlns:dr3d:1.0', 'draw': 'urn:oasis:names:tc:opendocument:xmlns:drawing:1.0', 'fo': 'urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0', 'form': 'urn:oasis:names:tc:opendocument:xmlns:form:1.0', 'math': 'http://www.w3.org/1998/Math/MathML', 'meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'number': 'urn:oasis:names:tc:opendocument:xmlns:datastyle:1.0', 'office': NAME_SPACE_1, 'ooo': 'http://openoffice.org/2004/office', 'oooc': 'http://openoffice.org/2004/calc', 'ooow': 'http://openoffice.org/2004/writer', 'presentation': 'urn:oasis:names:tc:opendocument:xmlns:presentation:1.0', 'script': 'urn:oasis:names:tc:opendocument:xmlns:script:1.0', 'style': 'urn:oasis:names:tc:opendocument:xmlns:style:1.0', 'svg': 'urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0', 'table': 'urn:oasis:names:tc:opendocument:xmlns:table:1.0', 'text': 'urn:oasis:names:tc:opendocument:xmlns:text:1.0', 'xforms': 'http://www.w3.org/2002/xforms', 'xlink': 'http://www.w3.org/1999/xlink', 'xsd': 'http://www.w3.org/2001/XMLSchema', 'xsi': 'http://www.w3.org/2001/XMLSchema-instance', } STYLES_NAMESPACE_DICT = SNSD = { # 'office:version': '1.0', 'chart': 'urn:oasis:names:tc:opendocument:xmlns:chart:1.0', 'dc': 'http://purl.org/dc/elements/1.1/', 'dom': 'http://www.w3.org/2001/xml-events', 'dr3d': 'urn:oasis:names:tc:opendocument:xmlns:dr3d:1.0', 'draw': 'urn:oasis:names:tc:opendocument:xmlns:drawing:1.0', 'fo': 'urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0', 'form': 'urn:oasis:names:tc:opendocument:xmlns:form:1.0', 'math': 'http://www.w3.org/1998/Math/MathML', 'meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'number': 'urn:oasis:names:tc:opendocument:xmlns:datastyle:1.0', 'office': NAME_SPACE_1, 'presentation': 'urn:oasis:names:tc:opendocument:xmlns:presentation:1.0', 'ooo': 'http://openoffice.org/2004/office', 'oooc': 'http://openoffice.org/2004/calc', 'ooow': 'http://openoffice.org/2004/writer', 'script': 'urn:oasis:names:tc:opendocument:xmlns:script:1.0', 'style': 'urn:oasis:names:tc:opendocument:xmlns:style:1.0', 'svg': 'urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0', 'table': 'urn:oasis:names:tc:opendocument:xmlns:table:1.0', 'text': 'urn:oasis:names:tc:opendocument:xmlns:text:1.0', 'xlink': 'http://www.w3.org/1999/xlink', } MANIFEST_NAMESPACE_DICT = MANNSD = { 'manifest': 'urn:oasis:names:tc:opendocument:xmlns:manifest:1.0', } META_NAMESPACE_DICT = METNSD = { # 'office:version': '1.0', 'dc': 'http://purl.org/dc/elements/1.1/', 'meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'office': NAME_SPACE_1, 'ooo': 'http://openoffice.org/2004/office', 'xlink': 'http://www.w3.org/1999/xlink', } # Attribute dictionaries for use with ElementTree, which # does not support use of nsmap parameter on Element() and SubElement(). CONTENT_NAMESPACE_ATTRIB = { # 'office:version': '1.0', 'xmlns:chart': 'urn:oasis:names:tc:opendocument:xmlns:chart:1.0', 'xmlns:dc': 'http://purl.org/dc/elements/1.1/', 'xmlns:dom': 'http://www.w3.org/2001/xml-events', 'xmlns:dr3d': 'urn:oasis:names:tc:opendocument:xmlns:dr3d:1.0', 'xmlns:draw': 'urn:oasis:names:tc:opendocument:xmlns:drawing:1.0', 'xmlns:fo': 'urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0', 'xmlns:form': 'urn:oasis:names:tc:opendocument:xmlns:form:1.0', 'xmlns:math': 'http://www.w3.org/1998/Math/MathML', 'xmlns:meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'xmlns:number': 'urn:oasis:names:tc:opendocument:xmlns:datastyle:1.0', 'xmlns:office': NAME_SPACE_1, 'xmlns:presentation': 'urn:oasis:names:tc:opendocument:xmlns:presentation:1.0', 'xmlns:ooo': 'http://openoffice.org/2004/office', 'xmlns:oooc': 'http://openoffice.org/2004/calc', 'xmlns:ooow': 'http://openoffice.org/2004/writer', 'xmlns:script': 'urn:oasis:names:tc:opendocument:xmlns:script:1.0', 'xmlns:style': 'urn:oasis:names:tc:opendocument:xmlns:style:1.0', 'xmlns:svg': 'urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0', 'xmlns:table': 'urn:oasis:names:tc:opendocument:xmlns:table:1.0', 'xmlns:text': 'urn:oasis:names:tc:opendocument:xmlns:text:1.0', 'xmlns:xforms': 'http://www.w3.org/2002/xforms', 'xmlns:xlink': 'http://www.w3.org/1999/xlink', 'xmlns:xsd': 'http://www.w3.org/2001/XMLSchema', 'xmlns:xsi': 'http://www.w3.org/2001/XMLSchema-instance', } STYLES_NAMESPACE_ATTRIB = { # 'office:version': '1.0', 'xmlns:chart': 'urn:oasis:names:tc:opendocument:xmlns:chart:1.0', 'xmlns:dc': 'http://purl.org/dc/elements/1.1/', 'xmlns:dom': 'http://www.w3.org/2001/xml-events', 'xmlns:dr3d': 'urn:oasis:names:tc:opendocument:xmlns:dr3d:1.0', 'xmlns:draw': 'urn:oasis:names:tc:opendocument:xmlns:drawing:1.0', 'xmlns:fo': 'urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0', 'xmlns:form': 'urn:oasis:names:tc:opendocument:xmlns:form:1.0', 'xmlns:math': 'http://www.w3.org/1998/Math/MathML', 'xmlns:meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'xmlns:number': 'urn:oasis:names:tc:opendocument:xmlns:datastyle:1.0', 'xmlns:office': NAME_SPACE_1, 'xmlns:presentation': 'urn:oasis:names:tc:opendocument:xmlns:presentation:1.0', 'xmlns:ooo': 'http://openoffice.org/2004/office', 'xmlns:oooc': 'http://openoffice.org/2004/calc', 'xmlns:ooow': 'http://openoffice.org/2004/writer', 'xmlns:script': 'urn:oasis:names:tc:opendocument:xmlns:script:1.0', 'xmlns:style': 'urn:oasis:names:tc:opendocument:xmlns:style:1.0', 'xmlns:svg': 'urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0', 'xmlns:table': 'urn:oasis:names:tc:opendocument:xmlns:table:1.0', 'xmlns:text': 'urn:oasis:names:tc:opendocument:xmlns:text:1.0', 'xmlns:xlink': 'http://www.w3.org/1999/xlink', } MANIFEST_NAMESPACE_ATTRIB = { 'xmlns:manifest': 'urn:oasis:names:tc:opendocument:xmlns:manifest:1.0', } META_NAMESPACE_ATTRIB = { # 'office:version': '1.0', 'xmlns:dc': 'http://purl.org/dc/elements/1.1/', 'xmlns:meta': 'urn:oasis:names:tc:opendocument:xmlns:meta:1.0', 'xmlns:office': NAME_SPACE_1, 'xmlns:ooo': 'http://openoffice.org/2004/office', 'xmlns:xlink': 'http://www.w3.org/1999/xlink', } # # Functions # # # ElementTree support functions. # In order to be able to get the parent of elements, must use these # instead of the functions with same name provided by ElementTree. # def Element(tag, attrib=None, nsmap=None, nsdict=CNSD): if attrib is None: attrib = {} tag, attrib = fix_ns(tag, attrib, nsdict) return _ElementInterfaceWrapper(tag, attrib) def SubElement(parent, tag, attrib=None, nsmap=None, nsdict=CNSD): if attrib is None: attrib = {} tag, attrib = fix_ns(tag, attrib, nsdict) el = _ElementInterfaceWrapper(tag, attrib) parent.append(el) el.setparent(parent) return el def fix_ns(tag, attrib, nsdict): nstag = add_ns(tag, nsdict) nsattrib = {} for key, val in list(attrib.items()): nskey = add_ns(key, nsdict) nsattrib[nskey] = val return nstag, nsattrib def add_ns(tag, nsdict=CNSD): return tag def ToString(et): outstream = StringIO() et.write(outstream, encoding="unicode") s1 = outstream.getvalue() outstream.close() return s1 def escape_cdata(text): text = text.replace("&", "&") text = text.replace("<", "<") text = text.replace(">", ">") ascii = '' for char in text: if ord(char) >= ord("\x7f"): ascii += "&#x%X;" % (ord(char), ) else: ascii += char return ascii # # Classes # class TableStyle: def __init__(self, border=None, backgroundcolor=None): self.border = border self.backgroundcolor = backgroundcolor def get_border_(self): return self.border_ def set_border_(self, border): self.border_ = border border = property(get_border_, set_border_) def get_backgroundcolor_(self): return self.backgroundcolor_ def set_backgroundcolor_(self, backgroundcolor): self.backgroundcolor_ = backgroundcolor backgroundcolor = property(get_backgroundcolor_, set_backgroundcolor_) BUILTIN_DEFAULT_TABLE_STYLE = TableStyle( border='0.0007in solid #000000') # # Information about the indentation level for lists nested inside # other contexts, e.g. dictionary lists. class ListLevel: def __init__(self, level, sibling_level=True, nested_level=True): self.level = level self.sibling_level = sibling_level self.nested_level = nested_level def set_sibling(self, sibling_level): self.sibling_level = sibling_level def get_sibling(self): return self.sibling_level def set_nested(self, nested_level): self.nested_level = nested_level def get_nested(self): return self.nested_level def set_level(self, level): self.level = level def get_level(self): return self.level class Writer(writers.Writer): MIME_TYPE = 'application/vnd.oasis.opendocument.text' EXTENSION = '.odt' supported = ('odt', ) """Formats this writer supports.""" default_stylesheet = 'styles' + EXTENSION default_stylesheet_path = utils.relative_path( os.path.join(os.getcwd(), 'dummy'), os.path.join(os.path.dirname(__file__), default_stylesheet)) default_template = 'template.txt' default_template_path = utils.relative_path( os.path.join(os.getcwd(), 'dummy'), os.path.join(os.path.dirname(__file__), default_template)) settings_spec = ( 'ODF-Specific Options', None, ( ('Specify a stylesheet. ' 'Default: "%s"' % default_stylesheet_path, ['--stylesheet'], { 'default': default_stylesheet_path, 'dest': 'stylesheet' }), ('Specify a configuration/mapping file relative to the ' 'current working ' 'directory for additional ODF options. ' 'In particular, this file may contain a section named ' '"Formats" that maps default style names to ' 'names to be used in the resulting output file allowing for ' 'adhering to external standards. ' 'For more info and the format of the ' 'configuration/mapping file, ' 'see the odtwriter doc.', ['--odf-config-file'], {'metavar': '<file>'}), ('Obfuscate email addresses to confuse harvesters while still ' 'keeping email links usable with ' 'standards-compliant browsers.', ['--cloak-email-addresses'], {'default': False, 'action': 'store_true', 'dest': 'cloak_email_addresses', 'validator': frontend.validate_boolean}), ('Do not obfuscate email addresses.', ['--no-cloak-email-addresses'], {'default': False, 'action': 'store_false', 'dest': 'cloak_email_addresses', 'validator': frontend.validate_boolean}), ('Specify the thickness of table borders in thousands of a cm. ' 'Default is 35.', ['--table-border-thickness'], {'default': None, 'validator': frontend.validate_nonnegative_int}), ('Add syntax highlighting in literal code blocks.', ['--add-syntax-highlighting'], {'default': False, 'action': 'store_true', 'dest': 'add_syntax_highlighting', 'validator': frontend.validate_boolean}), ('Do not add syntax highlighting in ' 'literal code blocks. (default)', ['--no-syntax-highlighting'], {'default': False, 'action': 'store_false', 'dest': 'add_syntax_highlighting', 'validator': frontend.validate_boolean}), ('Create sections for headers. (default)', ['--create-sections'], {'default': True, 'action': 'store_true', 'dest': 'create_sections', 'validator': frontend.validate_boolean}), ('Do not create sections for headers.', ['--no-sections'], {'default': True, 'action': 'store_false', 'dest': 'create_sections', 'validator': frontend.validate_boolean}), ('Create links.', ['--create-links'], {'default': False, 'action': 'store_true', 'dest': 'create_links', 'validator': frontend.validate_boolean}), ('Do not create links. (default)', ['--no-links'], {'default': False, 'action': 'store_false', 'dest': 'create_links', 'validator': frontend.validate_boolean}), ('Generate endnotes at end of document, not footnotes ' 'at bottom of page.', ['--endnotes-end-doc'], {'default': False, 'action': 'store_true', 'dest': 'endnotes_end_doc', 'validator': frontend.validate_boolean}), ('Generate footnotes at bottom of page, not endnotes ' 'at end of document. (default)', ['--no-endnotes-end-doc'], {'default': False, 'action': 'store_false', 'dest': 'endnotes_end_doc', 'validator': frontend.validate_boolean}), ('Generate a bullet list table of contents, not ' 'an ODF/oowriter table of contents.', ['--generate-list-toc'], {'default': True, 'action': 'store_false', 'dest': 'generate_oowriter_toc', 'validator': frontend.validate_boolean}), ('Generate an ODF/oowriter table of contents, not ' 'a bullet list. (default)', ['--generate-oowriter-toc'], {'default': True, 'action': 'store_true', 'dest': 'generate_oowriter_toc', 'validator': frontend.validate_boolean}), ('Specify the contents of an custom header line. ' 'See odf_odt writer documentation for details ' 'about special field character sequences.', ['--custom-odt-header'], {'default': '', 'dest': 'custom_header', }), ('Specify the contents of an custom footer line. ' 'See odf_odt writer documentation for details ' 'about special field character sequences.', ['--custom-odt-footer'], {'default': '', 'dest': 'custom_footer', }), ) ) settings_defaults = { 'output_encoding_error_handler': 'xmlcharrefreplace', } relative_path_settings = ( 'stylesheet_path', ) config_section = 'odf_odt writer' config_section_dependencies = ('writers',) def __init__(self): writers.Writer.__init__(self) self.translator_class = ODFTranslator def translate(self): self.settings = self.document.settings self.visitor = self.translator_class(self.document) self.visitor.retrieve_styles(self.EXTENSION) self.document.walkabout(self.visitor) self.visitor.add_doc_title() self.assemble_my_parts() self.output = self.parts['whole'] def assemble_my_parts(self): """Assemble the `self.parts` dictionary. Extend in subclasses. """ writers.Writer.assemble_parts(self) f = tempfile.NamedTemporaryFile() zfile = zipfile.ZipFile(f, 'w', zipfile.ZIP_DEFLATED) self.write_zip_str( zfile, 'mimetype', self.MIME_TYPE, compress_type=zipfile.ZIP_STORED) content = self.visitor.content_astext() self.write_zip_str(zfile, 'content.xml', content) s1 = self.create_manifest() self.write_zip_str(zfile, 'META-INF/manifest.xml', s1) s1 = self.create_meta() self.write_zip_str(zfile, 'meta.xml', s1) s1 = self.get_stylesheet() # Set default language in document to be generated. # Language is specified by the -l/--language command line option. # The format is described in BCP 47. If region is omitted, we use # local.normalize(ll) to obtain a region. language_code = None region_code = None if self.visitor.language_code: language_ids = self.visitor.language_code.replace('_', '-') language_ids = language_ids.split('-') # first tag is primary language tag language_code = language_ids[0].lower() # 2-letter region subtag may follow in 2nd or 3rd position for subtag in language_ids[1:]: if len(subtag) == 2 and subtag.isalpha(): region_code = subtag.upper() break elif len(subtag) == 1: break # 1-letter tag is never before valid region tag if region_code is None: try: rcode = locale.normalize(language_code) except NameError: rcode = language_code rcode = rcode.split('_') if len(rcode) > 1: rcode = rcode[1].split('.') region_code = rcode[0] if region_code is None: self.document.reporter.warning( 'invalid language-region.\n' ' Could not find region with locale.normalize().\n' ' Please specify both language and region (ll-RR).\n' ' Examples: es-MX (Spanish, Mexico),\n' ' en-AU (English, Australia).') # Update the style ElementTree with the language and region. # Note that we keep a reference to the modified node because # it is possible that ElementTree will throw away the Python # representation of the updated node if we do not. updated, new_dom_styles, updated_node = self.update_stylesheet( self.visitor.get_dom_stylesheet(), language_code, region_code) if updated: s1 = etree.tostring(new_dom_styles) self.write_zip_str(zfile, 'styles.xml', s1) self.store_embedded_files(zfile) self.copy_from_stylesheet(zfile) zfile.close() f.seek(0) whole = f.read() f.close() self.parts['whole'] = whole self.parts['encoding'] = self.document.settings.output_encoding self.parts['version'] = docutils.__version__ def update_stylesheet(self, stylesheet_root, language_code, region_code): """Update xml style sheet element with language and region/country.""" updated = False modified_nodes = set() if language_code is not None or region_code is not None: n1 = stylesheet_root.find( '{urn:oasis:names:tc:opendocument:xmlns:office:1.0}' 'styles') if n1 is None: raise RuntimeError( "Cannot find 'styles' element in styles.odt/styles.xml") n2_nodes = n1.findall( '{urn:oasis:names:tc:opendocument:xmlns:style:1.0}' 'default-style') if not n2_nodes: raise RuntimeError( "Cannot find 'default-style' " "element in styles.xml") for node in n2_nodes: family = node.attrib.get( '{urn:oasis:names:tc:opendocument:xmlns:style:1.0}' 'family') if family == 'paragraph' or family == 'graphic': n3 = node.find( '{urn:oasis:names:tc:opendocument:xmlns:style:1.0}' 'text-properties') if n3 is None: raise RuntimeError( "Cannot find 'text-properties' " "element in styles.xml") if language_code is not None: n3.attrib[ '{urn:oasis:names:tc:opendocument:xmlns:' 'xsl-fo-compatible:1.0}language'] = language_code n3.attrib[ '{urn:oasis:names:tc:opendocument:xmlns:' 'style:1.0}language-complex'] = language_code updated = True modified_nodes.add(n3) if region_code is not None: n3.attrib[ '{urn:oasis:names:tc:opendocument:xmlns:' 'xsl-fo-compatible:1.0}country'] = region_code n3.attrib[ '{urn:oasis:names:tc:opendocument:xmlns:' 'style:1.0}country-complex'] = region_code updated = True modified_nodes.add(n3) return updated, stylesheet_root, modified_nodes def write_zip_str( self, zfile, name, bytes, compress_type=zipfile.ZIP_DEFLATED): localtime = time.localtime(time.time()) zinfo = zipfile.ZipInfo(name, localtime) # Add some standard UNIX file access permissions (-rw-r--r--). zinfo.external_attr = (0x81a4 & 0xFFFF) << 16 zinfo.compress_type = compress_type zfile.writestr(zinfo, bytes) def store_embedded_files(self, zfile): embedded_files = self.visitor.get_embedded_file_list() for source, destination in embedded_files: if source is None: continue try: zfile.write(source, destination) except OSError: self.document.reporter.warning( "Can't open file %s." % (source, )) def get_settings(self): """ modeled after get_stylesheet """ stylespath = self.settings.stylesheet zfile = zipfile.ZipFile(stylespath, 'r') s1 = zfile.read('settings.xml') zfile.close() return s1 def get_stylesheet(self): """Get the stylesheet from the visitor. Ask the visitor to setup the page. """ return self.visitor.setup_page() def copy_from_stylesheet(self, outzipfile): """Copy images, settings, etc from the stylesheet doc into target doc. """ stylespath = self.settings.stylesheet inzipfile = zipfile.ZipFile(stylespath, 'r') # Copy the styles. s1 = inzipfile.read('settings.xml') self.write_zip_str(outzipfile, 'settings.xml', s1) # Copy the images. namelist = inzipfile.namelist() for name in namelist: if name.startswith('Pictures/'): imageobj = inzipfile.read(name) outzipfile.writestr(name, imageobj) inzipfile.close() def assemble_parts(self): pass def create_manifest(self): root = Element( 'manifest:manifest', attrib=MANIFEST_NAMESPACE_ATTRIB, nsdict=MANIFEST_NAMESPACE_DICT, ) doc = etree.ElementTree(root) SubElement(root, 'manifest:file-entry', attrib={ 'manifest:media-type': self.MIME_TYPE, 'manifest:full-path': '/', }, nsdict=MANNSD) SubElement(root, 'manifest:file-entry', attrib={ 'manifest:media-type': 'text/xml', 'manifest:full-path': 'content.xml', }, nsdict=MANNSD) SubElement(root, 'manifest:file-entry', attrib={ 'manifest:media-type': 'text/xml', 'manifest:full-path': 'styles.xml', }, nsdict=MANNSD) SubElement(root, 'manifest:file-entry', attrib={ 'manifest:media-type': 'text/xml', 'manifest:full-path': 'settings.xml', }, nsdict=MANNSD) SubElement(root, 'manifest:file-entry', attrib={ 'manifest:media-type': 'text/xml', 'manifest:full-path': 'meta.xml', }, nsdict=MANNSD) s1 = ToString(doc) doc = minidom.parseString(s1) return doc.toprettyxml(' ') def create_meta(self): root = Element( 'office:document-meta', attrib=META_NAMESPACE_ATTRIB, nsdict=META_NAMESPACE_DICT, ) doc = etree.ElementTree(root) root = SubElement(root, 'office:meta', nsdict=METNSD) el1 = SubElement(root, 'meta:generator', nsdict=METNSD) el1.text = 'Docutils/rst2odf.py/%s' % (VERSION, ) s1 = os.environ.get('USER', '') el1 = SubElement(root, 'meta:initial-creator', nsdict=METNSD) el1.text = s1 s2 = time.strftime('%Y-%m-%dT%H:%M:%S', time.localtime()) el1 = SubElement(root, 'meta:creation-date', nsdict=METNSD) el1.text = s2 el1 = SubElement(root, 'dc:creator', nsdict=METNSD) el1.text = s1 el1 = SubElement(root, 'dc:date', nsdict=METNSD) el1.text = s2 el1 = SubElement(root, 'dc:language', nsdict=METNSD) el1.text = 'en-US' el1 = SubElement(root, 'meta:editing-cycles', nsdict=METNSD) el1.text = '1' el1 = SubElement(root, 'meta:editing-duration', nsdict=METNSD) el1.text = 'PT00M01S' title = self.visitor.get_title() el1 = SubElement(root, 'dc:title', nsdict=METNSD) if title: el1.text = title else: el1.text = '[no title]' for prop, value in self.visitor.get_meta_dict().items(): # 'keywords', 'description', and 'subject' have their own fields: if prop == 'keywords': keywords = re.split(', *', value) for keyword in keywords: el1 = SubElement(root, 'meta:keyword', nsdict=METNSD) el1.text = keyword elif prop == 'description': el1 = SubElement(root, 'dc:description', nsdict=METNSD) el1.text = value elif prop == 'subject': el1 = SubElement(root, 'dc:subject', nsdict=METNSD) el1.text = value else: # Store remaining properties as custom/user-defined el1 = SubElement(root, 'meta:user-defined', attrib={'meta:name': prop}, nsdict=METNSD) el1.text = value s1 = ToString(doc) # doc = minidom.parseString(s1) # s1 = doc.toprettyxml(' ') return s1 # class ODFTranslator(nodes.SparseNodeVisitor): class ODFTranslator(nodes.GenericNodeVisitor): used_styles = ( 'attribution', 'blockindent', 'blockquote', 'blockquote-bulletitem', 'blockquote-bulletlist', 'blockquote-enumitem', 'blockquote-enumlist', 'bulletitem', 'bulletlist', 'caption', 'legend', 'centeredtextbody', 'codeblock', 'codeblock-indented', 'codeblock-classname', 'codeblock-comment', 'codeblock-functionname', 'codeblock-keyword', 'codeblock-name', 'codeblock-number', 'codeblock-operator', 'codeblock-string', 'emphasis', 'enumitem', 'enumlist', 'epigraph', 'epigraph-bulletitem', 'epigraph-bulletlist', 'epigraph-enumitem', 'epigraph-enumlist', 'footer', 'footnote', 'citation', 'header', 'highlights', 'highlights-bulletitem', 'highlights-bulletlist', 'highlights-enumitem', 'highlights-enumlist', 'horizontalline', 'inlineliteral', 'quotation', 'rubric', 'strong', 'table-title', 'textbody', 'tocbulletlist', 'tocenumlist', 'title', 'subtitle', 'heading1', 'heading2', 'heading3', 'heading4', 'heading5', 'heading6', 'heading7', 'admon-attention-hdr', 'admon-attention-body', 'admon-caution-hdr', 'admon-caution-body', 'admon-danger-hdr', 'admon-danger-body', 'admon-error-hdr', 'admon-error-body', 'admon-generic-hdr', 'admon-generic-body', 'admon-hint-hdr', 'admon-hint-body', 'admon-important-hdr', 'admon-important-body', 'admon-note-hdr', 'admon-note-body', 'admon-tip-hdr', 'admon-tip-body', 'admon-warning-hdr', 'admon-warning-body', 'tableoption', 'tableoption.%c', 'tableoption.%c%d', 'Table%d', 'Table%d.%c', 'Table%d.%c%d', 'lineblock1', 'lineblock2', 'lineblock3', 'lineblock4', 'lineblock5', 'lineblock6', 'image', 'figureframe', ) def __init__(self, document): # nodes.SparseNodeVisitor.__init__(self, document) nodes.GenericNodeVisitor.__init__(self, document) self.settings = document.settings self.language_code = self.settings.language_code self.language = languages.get_language( self.language_code, document.reporter) self.format_map = {} if self.settings.odf_config_file: parser = ConfigParser() parser.read(self.settings.odf_config_file) for rststyle, format in parser.items("Formats"): if rststyle not in self.used_styles: self.document.reporter.warning( 'Style "%s" is not a style used by odtwriter.' % ( rststyle, )) self.format_map[rststyle] = format self.section_level = 0 self.section_count = 0 # Create ElementTree content and styles documents. root = Element( 'office:document-content', attrib=CONTENT_NAMESPACE_ATTRIB, ) self.content_tree = etree.ElementTree(element=root) self.current_element = root SubElement(root, 'office:scripts') SubElement(root, 'office:font-face-decls') el = SubElement(root, 'office:automatic-styles') self.automatic_styles = el el = SubElement(root, 'office:body') el = self.generate_content_element(el) self.current_element = el self.body_text_element = el self.paragraph_style_stack = [self.rststyle('textbody'), ] self.list_style_stack = [] self.table_count = 0 self.column_count = ord('A') - 1 self.trace_level = -1 self.optiontablestyles_generated = False self.field_name = None self.field_element = None self.title = None self.image_count = 0 self.image_style_count = 0 self.image_dict = {} self.embedded_file_list = [] self.syntaxhighlighting = 1 self.syntaxhighlight_lexer = 'python' self.header_content = [] self.footer_content = [] self.in_header = False self.in_footer = False self.blockstyle = '' self.in_table_of_contents = False self.table_of_content_index_body = None self.list_level = 0 self.def_list_level = 0 self.footnote_ref_dict = {} self.footnote_list = [] self.footnote_chars_idx = 0 self.footnote_level = 0 self.pending_ids = [] self.in_paragraph = False self.found_doc_title = False self.bumped_list_level_stack = [] self.meta_dict = {} self.line_block_level = 0 self.line_indent_level = 0 self.citation_id = None self.style_index = 0 # use to form unique style names self.str_stylesheet = '' self.str_stylesheetcontent = '' self.dom_stylesheet = None self.table_styles = None self.in_citation = False # Keep track of nested styling classes self.inline_style_count_stack = [] def get_str_stylesheet(self): return self.str_stylesheet def retrieve_styles(self, extension): """Retrieve the stylesheet from either a .xml file or from a .odt (zip) file. Return the content as a string. """ s2 = None stylespath = self.settings.stylesheet ext = os.path.splitext(stylespath)[1] if ext == '.xml': with open(stylespath, 'r', encoding='utf-8') as stylesfile: s1 = stylesfile.read() elif ext == extension: zfile = zipfile.ZipFile(stylespath, 'r') s1 = zfile.read('styles.xml') s2 = zfile.read('content.xml') zfile.close() else: raise RuntimeError('stylesheet path (%s) must be %s or ' '.xml file' % (stylespath, extension)) self.str_stylesheet = s1 self.str_stylesheetcontent = s2 self.dom_stylesheet = etree.fromstring(self.str_stylesheet) self.dom_stylesheetcontent = etree.fromstring( self.str_stylesheetcontent) self.table_styles = self.extract_table_styles(s2) def extract_table_styles(self, styles_str): root = etree.fromstring(styles_str) table_styles = {} auto_styles = root.find( '{%s}automatic-styles' % (CNSD['office'], )) for stylenode in auto_styles: name = stylenode.get('{%s}name' % (CNSD['style'], )) tablename = name.split('.')[0] family = stylenode.get('{%s}family' % (CNSD['style'], )) if name.startswith(TABLESTYLEPREFIX): tablestyle = table_styles.get(tablename) if tablestyle is None: tablestyle = TableStyle() table_styles[tablename] = tablestyle if family == 'table': properties = stylenode.find( '{%s}table-properties' % (CNSD['style'], )) property = properties.get( '{%s}%s' % (CNSD['fo'], 'background-color', )) if property is not None and property != 'none': tablestyle.backgroundcolor = property elif family == 'table-cell': properties = stylenode.find( '{%s}table-cell-properties' % (CNSD['style'], )) if properties is not None: border = self.get_property(properties) if border is not None: tablestyle.border = border return table_styles def get_property(self, stylenode): border = None for propertyname in TABLEPROPERTYNAMES: border = stylenode.get('{%s}%s' % (CNSD['fo'], propertyname, )) if border is not None and border != 'none': return border return border def add_doc_title(self): text = self.settings.title if text: self.title = text if not self.found_doc_title: el = Element('text:p', attrib={ 'text:style-name': self.rststyle('title'), }) el.text = text self.body_text_element.insert(0, el) el = self.find_first_text_p(self.body_text_element) if el is not None: self.attach_page_style(el) def find_first_text_p(self, el): """Search the generated doc and return the first <text:p> element. """ if el.tag == 'text:p' or el.tag == 'text:h': return el else: for child in el: el1 = self.find_first_text_p(child) if el1 is not None: return el1 return None def attach_page_style(self, el): """Attach the default page style. Create an automatic-style that refers to the current style of this element and that refers to the default page style. """ current_style = el.get('text:style-name') style_name = 'P1003' el1 = SubElement( self.automatic_styles, 'style:style', attrib={ 'style:name': style_name, 'style:master-page-name': "rststyle-pagedefault", 'style:family': "paragraph", }, nsdict=SNSD) if current_style: el1.set('style:parent-style-name', current_style) el.set('text:style-name', style_name) def rststyle(self, name, parameters=()): """ Returns the style name to use for the given style. If `parameters` is given `name` must contain a matching number of ``%`` and is used as a format expression with `parameters` as the value. """ name1 = name % parameters return self.format_map.get(name1, 'rststyle-%s' % name1) def generate_content_element(self, root): return SubElement(root, 'office:text') def setup_page(self): self.setup_paper(self.dom_stylesheet) if (len(self.header_content) > 0 or len(self.footer_content) > 0 or self.settings.custom_header or self.settings.custom_footer): self.add_header_footer(self.dom_stylesheet) return etree.tostring(self.dom_stylesheet) def get_dom_stylesheet(self): return self.dom_stylesheet def setup_paper(self, root_el): try: dimensions = subprocess.check_output(('paperconf', '-s'), stderr=subprocess.STDOUT) w, h = (float(s) for s in dimensions.split()) except (subprocess.CalledProcessError, FileNotFoundError, ValueError): self.document.reporter.info( 'Cannot use `paperconf`, defaulting to Letter.') w, h = 612, 792 # default to Letter def walk(el): if el.tag == "{%s}page-layout-properties" % SNSD["style"] and \ "{%s}page-width" % SNSD["fo"] not in el.attrib: el.attrib["{%s}page-width" % SNSD["fo"]] = "%.3fpt" % w el.attrib["{%s}page-height" % SNSD["fo"]] = "%.3fpt" % h el.attrib["{%s}margin-left" % SNSD["fo"]] = \ el.attrib["{%s}margin-right" % SNSD["fo"]] = \ "%.3fpt" % (.1 * w) el.attrib["{%s}margin-top" % SNSD["fo"]] = \ el.attrib["{%s}margin-bottom" % SNSD["fo"]] = \ "%.3fpt" % (.1 * h) else: for subel in el: walk(subel) walk(root_el) def add_header_footer(self, root_el): automatic_styles = root_el.find( '{%s}automatic-styles' % SNSD['office']) path = '{%s}master-styles' % (NAME_SPACE_1, ) master_el = root_el.find(path) if master_el is None: return path = '{%s}master-page' % (SNSD['style'], ) master_el_container = master_el.findall(path) master_el = None target_attrib = '{%s}name' % (SNSD['style'], ) target_name = self.rststyle('pagedefault') for el in master_el_container: if el.get(target_attrib) == target_name: master_el = el break if master_el is None: return el1 = master_el if self.header_content or self.settings.custom_header: el2 = SubElement( el1, 'style:header', attrib=STYLES_NAMESPACE_ATTRIB, nsdict=STYLES_NAMESPACE_DICT, ) for el in self.header_content: attrkey = add_ns('text:style-name', nsdict=SNSD) el.attrib[attrkey] = self.rststyle('header') el2.append(el) if self.settings.custom_header: self.create_custom_headfoot( el2, self.settings.custom_header, 'header', automatic_styles) if self.footer_content or self.settings.custom_footer: el2 = SubElement( el1, 'style:footer', attrib=STYLES_NAMESPACE_ATTRIB, nsdict=STYLES_NAMESPACE_DICT, ) for el in self.footer_content: attrkey = add_ns('text:style-name', nsdict=SNSD) el.attrib[attrkey] = self.rststyle('footer') el2.append(el) if self.settings.custom_footer: self.create_custom_headfoot( el2, self.settings.custom_footer, 'footer', automatic_styles) code_none, code_field, code_text = list(range(3)) field_pat = re.compile(r'%(..?)%') def create_custom_headfoot( self, parent, text, style_name, automatic_styles): parent = SubElement(parent, 'text:p', attrib={ 'text:style-name': self.rststyle(style_name), }) current_element = None field_iter = self.split_field_specifiers_iter(text) for item in field_iter: if item[0] == ODFTranslator.code_field: if item[1] not in ( 'p', 'P', 't1', 't2', 't3', 't4', 'd1', 'd2', 'd3', 'd4', 'd5', 's', 't', 'a'): msg = 'bad field spec: %%%s%%' % (item[1], ) raise RuntimeError(msg) el1 = self.make_field_element( parent, item[1], style_name, automatic_styles) if el1 is None: msg = 'bad field spec: %%%s%%' % (item[1], ) raise RuntimeError(msg) else: current_element = el1 else: if current_element is None: parent.text = item[1] else: current_element.tail = item[1] def make_field_element(self, parent, text, style_name, automatic_styles): if text == 'p': el1 = SubElement(parent, 'text:page-number', attrib={ # 'text:style-name': self.rststyle(style_name), 'text:select-page': 'current', }) elif text == 'P': el1 = SubElement(parent, 'text:page-count', attrib={ # 'text:style-name': self.rststyle(style_name), }) elif text == 't1': self.style_index += 1 el1 = SubElement(parent, 'text:time', attrib={ 'text:style-name': self.rststyle(style_name), 'text:fixed': 'true', 'style:data-style-name': 'rst-time-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:time-style', attrib={ 'style:name': 'rst-time-style-%d' % self.style_index, 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:hours', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:minutes', attrib={ 'number:style': 'long', }) elif text == 't2': self.style_index += 1 el1 = SubElement(parent, 'text:time', attrib={ 'text:style-name': self.rststyle(style_name), 'text:fixed': 'true', 'style:data-style-name': 'rst-time-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:time-style', attrib={ 'style:name': 'rst-time-style-%d' % self.style_index, 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:hours', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:minutes', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:seconds', attrib={ 'number:style': 'long', }) elif text == 't3': self.style_index += 1 el1 = SubElement(parent, 'text:time', attrib={ 'text:style-name': self.rststyle(style_name), 'text:fixed': 'true', 'style:data-style-name': 'rst-time-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:time-style', attrib={ 'style:name': 'rst-time-style-%d' % self.style_index, 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:hours', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:minutes', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ' ' el3 = SubElement(el2, 'number:am-pm') elif text == 't4': self.style_index += 1 el1 = SubElement(parent, 'text:time', attrib={ 'text:style-name': self.rststyle(style_name), 'text:fixed': 'true', 'style:data-style-name': 'rst-time-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:time-style', attrib={ 'style:name': 'rst-time-style-%d' % self.style_index, 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:hours', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:minutes', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ':' el3 = SubElement(el2, 'number:seconds', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ' ' el3 = SubElement(el2, 'number:am-pm') elif text == 'd1': self.style_index += 1 el1 = SubElement(parent, 'text:date', attrib={ 'text:style-name': self.rststyle(style_name), 'style:data-style-name': 'rst-date-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:date-style', attrib={ 'style:name': 'rst-date-style-%d' % self.style_index, 'number:automatic-order': 'true', 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:month', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '/' el3 = SubElement(el2, 'number:day', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '/' el3 = SubElement(el2, 'number:year') elif text == 'd2': self.style_index += 1 el1 = SubElement(parent, 'text:date', attrib={ 'text:style-name': self.rststyle(style_name), 'style:data-style-name': 'rst-date-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:date-style', attrib={ 'style:name': 'rst-date-style-%d' % self.style_index, 'number:automatic-order': 'true', 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:month', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '/' el3 = SubElement(el2, 'number:day', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '/' el3 = SubElement(el2, 'number:year', attrib={ 'number:style': 'long', }) elif text == 'd3': self.style_index += 1 el1 = SubElement(parent, 'text:date', attrib={ 'text:style-name': self.rststyle(style_name), 'style:data-style-name': 'rst-date-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:date-style', attrib={ 'style:name': 'rst-date-style-%d' % self.style_index, 'number:automatic-order': 'true', 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:month', attrib={ 'number:textual': 'true', }) el3 = SubElement(el2, 'number:text') el3.text = ' ' el3 = SubElement(el2, 'number:day', attrib={}) el3 = SubElement(el2, 'number:text') el3.text = ', ' el3 = SubElement(el2, 'number:year', attrib={ 'number:style': 'long', }) elif text == 'd4': self.style_index += 1 el1 = SubElement(parent, 'text:date', attrib={ 'text:style-name': self.rststyle(style_name), 'style:data-style-name': 'rst-date-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:date-style', attrib={ 'style:name': 'rst-date-style-%d' % self.style_index, 'number:automatic-order': 'true', 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:month', attrib={ 'number:textual': 'true', 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = ' ' el3 = SubElement(el2, 'number:day', attrib={}) el3 = SubElement(el2, 'number:text') el3.text = ', ' el3 = SubElement(el2, 'number:year', attrib={ 'number:style': 'long', }) elif text == 'd5': self.style_index += 1 el1 = SubElement(parent, 'text:date', attrib={ 'text:style-name': self.rststyle(style_name), 'style:data-style-name': 'rst-date-style-%d' % self.style_index, }) el2 = SubElement(automatic_styles, 'number:date-style', attrib={ 'style:name': 'rst-date-style-%d' % self.style_index, 'xmlns:number': SNSD['number'], 'xmlns:style': SNSD['style'], }) el3 = SubElement(el2, 'number:year', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '-' el3 = SubElement(el2, 'number:month', attrib={ 'number:style': 'long', }) el3 = SubElement(el2, 'number:text') el3.text = '-' el3 = SubElement(el2, 'number:day', attrib={ 'number:style': 'long', }) elif text == 's': el1 = SubElement(parent, 'text:subject', attrib={ 'text:style-name': self.rststyle(style_name), }) elif text == 't': el1 = SubElement(parent, 'text:title', attrib={ 'text:style-name': self.rststyle(style_name), }) elif text == 'a': el1 = SubElement(parent, 'text:author-name', attrib={ 'text:fixed': 'false', }) else: el1 = None return el1 def split_field_specifiers_iter(self, text): pos1 = 0 while True: mo = ODFTranslator.field_pat.search(text, pos1) if mo: pos2 = mo.start() if pos2 > pos1: yield ODFTranslator.code_text, text[pos1:pos2] yield ODFTranslator.code_field, mo.group(1) pos1 = mo.end() else: break trailing = text[pos1:] if trailing: yield ODFTranslator.code_text, trailing def astext(self): root = self.content_tree.getroot() et = etree.ElementTree(root) return ToString(et) def content_astext(self): return self.astext() def set_title(self, title): self.title = title def get_title(self): return self.title def set_embedded_file_list(self, embedded_file_list): self.embedded_file_list = embedded_file_list def get_embedded_file_list(self): return self.embedded_file_list def get_meta_dict(self): return self.meta_dict def process_footnotes(self): for node, el1 in self.footnote_list: backrefs = node.attributes.get('backrefs', []) first = True for ref in backrefs: el2 = self.footnote_ref_dict.get(ref) if el2 is not None: if first: first = False el3 = copy.deepcopy(el1) el2.append(el3) else: if len(el2) > 0: # and 'id' in el2.attrib: child = el2[0] ref1 = child.text attribkey = add_ns('text:id', nsdict=SNSD) id1 = el2.get(attribkey, 'footnote-error') if id1 is None: id1 = '' tag = add_ns('text:note-ref', nsdict=SNSD) el2.tag = tag if self.settings.endnotes_end_doc: note_class = 'endnote' else: note_class = 'footnote' el2.attrib.clear() attribkey = add_ns('text:note-class', nsdict=SNSD) el2.attrib[attribkey] = note_class attribkey = add_ns('text:ref-name', nsdict=SNSD) el2.attrib[attribkey] = id1 attribkey = add_ns( 'text:reference-format', nsdict=SNSD) el2.attrib[attribkey] = 'page' el2.text = ref1 # # Utility methods def append_child(self, tag, attrib=None, parent=None): if parent is None: parent = self.current_element return SubElement(parent, tag, attrib) def append_p(self, style, text=None): result = self.append_child('text:p', attrib={ 'text:style-name': self.rststyle(style)}) self.append_pending_ids(result) if text is not None: result.text = text return result def append_pending_ids(self, el): if self.settings.create_links: for id in self.pending_ids: SubElement(el, 'text:reference-mark', attrib={ 'text:name': id}) self.pending_ids = [] def set_current_element(self, el): self.current_element = el def set_to_parent(self): self.current_element = self.current_element.getparent() def generate_labeled_block(self, node, label): label = '%s:' % (self.language.labels[label], ) el = self.append_p('textbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) el1.text = label return self.append_p('blockindent') def generate_labeled_line(self, node, label): label = '%s:' % (self.language.labels[label], ) el = self.append_p('textbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) el1.text = label el1.tail = node.astext() return el def encode(self, text): return text.replace('\n', " ") # # Visitor functions # # In alphabetic order, more or less. # See docutils.docutils.nodes.node_class_names. # def dispatch_visit(self, node): """Override to catch basic attributes which many nodes have.""" self.handle_basic_atts(node) nodes.GenericNodeVisitor.dispatch_visit(self, node) def handle_basic_atts(self, node): if isinstance(node, nodes.Element) and node['ids']: self.pending_ids += node['ids'] def default_visit(self, node): self.document.reporter.warning('missing visit_%s' % (node.tagname, )) def default_departure(self, node): self.document.reporter.warning('missing depart_%s' % (node.tagname, )) def visit_Text(self, node): # Skip nodes whose text has been processed in parent nodes. if isinstance(node.parent, docutils.nodes.literal_block): return text = node.astext() # Are we in mixed content? If so, add the text to the # etree tail of the previous sibling element. if len(self.current_element) > 0: if self.current_element[-1].tail: self.current_element[-1].tail += text else: self.current_element[-1].tail = text else: if self.current_element.text: self.current_element.text += text else: self.current_element.text = text def depart_Text(self, node): pass # # Pre-defined fields # def visit_address(self, node): el = self.generate_labeled_block(node, 'address') self.set_current_element(el) def depart_address(self, node): self.set_to_parent() def visit_author(self, node): if isinstance(node.parent, nodes.authors): el = self.append_p('blockindent') else: el = self.generate_labeled_block(node, 'author') self.set_current_element(el) def depart_author(self, node): self.set_to_parent() def visit_authors(self, node): label = '%s:' % (self.language.labels['authors'], ) el = self.append_p('textbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) el1.text = label def depart_authors(self, node): pass def visit_contact(self, node): el = self.generate_labeled_block(node, 'contact') self.set_current_element(el) def depart_contact(self, node): self.set_to_parent() def visit_copyright(self, node): el = self.generate_labeled_block(node, 'copyright') self.set_current_element(el) def depart_copyright(self, node): self.set_to_parent() def visit_date(self, node): self.generate_labeled_line(node, 'date') def depart_date(self, node): pass def visit_organization(self, node): el = self.generate_labeled_block(node, 'organization') self.set_current_element(el) def depart_organization(self, node): self.set_to_parent() def visit_status(self, node): el = self.generate_labeled_block(node, 'status') self.set_current_element(el) def depart_status(self, node): self.set_to_parent() def visit_revision(self, node): self.generate_labeled_line(node, 'revision') def depart_revision(self, node): pass def visit_version(self, node): self.generate_labeled_line(node, 'version') # self.set_current_element(el) def depart_version(self, node): # self.set_to_parent() pass def visit_attribution(self, node): self.append_p('attribution', node.astext()) def depart_attribution(self, node): pass def visit_block_quote(self, node): if 'epigraph' in node.attributes['classes']: self.paragraph_style_stack.append(self.rststyle('epigraph')) self.blockstyle = self.rststyle('epigraph') elif 'highlights' in node.attributes['classes']: self.paragraph_style_stack.append(self.rststyle('highlights')) self.blockstyle = self.rststyle('highlights') else: self.paragraph_style_stack.append(self.rststyle('blockquote')) self.blockstyle = self.rststyle('blockquote') self.line_indent_level += 1 def depart_block_quote(self, node): self.paragraph_style_stack.pop() self.blockstyle = '' self.line_indent_level -= 1 def visit_bullet_list(self, node): self.list_level += 1 if self.in_table_of_contents: if self.settings.generate_oowriter_toc: pass else: if 'classes' in node and \ 'auto-toc' in node.attributes['classes']: el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('tocenumlist'), }) self.list_style_stack.append(self.rststyle('enumitem')) else: el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('tocbulletlist'), }) self.list_style_stack.append(self.rststyle('bulletitem')) self.set_current_element(el) else: if self.blockstyle == self.rststyle('blockquote'): el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('blockquote-bulletlist'), }) self.list_style_stack.append( self.rststyle('blockquote-bulletitem')) elif self.blockstyle == self.rststyle('highlights'): el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('highlights-bulletlist'), }) self.list_style_stack.append( self.rststyle('highlights-bulletitem')) elif self.blockstyle == self.rststyle('epigraph'): el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('epigraph-bulletlist'), }) self.list_style_stack.append( self.rststyle('epigraph-bulletitem')) else: el = SubElement(self.current_element, 'text:list', attrib={ 'text:style-name': self.rststyle('bulletlist'), }) self.list_style_stack.append(self.rststyle('bulletitem')) self.set_current_element(el) def depart_bullet_list(self, node): if self.in_table_of_contents: if self.settings.generate_oowriter_toc: pass else: self.set_to_parent() self.list_style_stack.pop() else: self.set_to_parent() self.list_style_stack.pop() self.list_level -= 1 def visit_caption(self, node): raise nodes.SkipChildren() def depart_caption(self, node): pass def visit_comment(self, node): el = self.append_p('textbody') el1 = SubElement(el, 'office:annotation', attrib={}) el2 = SubElement(el1, 'dc:creator', attrib={}) s1 = os.environ.get('USER', '') el2.text = s1 el2 = SubElement(el1, 'text:p', attrib={}) el2.text = node.astext() def depart_comment(self, node): pass def visit_compound(self, node): # The compound directive currently receives no special treatment. pass def depart_compound(self, node): pass def visit_container(self, node): styles = node.attributes.get('classes', ()) if len(styles) > 0: self.paragraph_style_stack.append(self.rststyle(styles[0])) def depart_container(self, node): styles = node.attributes.get('classes', ()) if len(styles) > 0: self.paragraph_style_stack.pop() def visit_decoration(self, node): pass def depart_decoration(self, node): pass def visit_definition_list(self, node): self.def_list_level += 1 if self.list_level > 5: raise RuntimeError( 'max definition list nesting level exceeded') def depart_definition_list(self, node): self.def_list_level -= 1 def visit_definition_list_item(self, node): pass def depart_definition_list_item(self, node): pass def visit_term(self, node): el = self.append_p('deflist-term-%d' % self.def_list_level) el.text = node.astext() self.set_current_element(el) raise nodes.SkipChildren() def depart_term(self, node): self.set_to_parent() def visit_definition(self, node): self.paragraph_style_stack.append( self.rststyle('deflist-def-%d' % self.def_list_level)) self.bumped_list_level_stack.append(ListLevel(1)) def depart_definition(self, node): self.paragraph_style_stack.pop() self.bumped_list_level_stack.pop() def visit_classifier(self, node): if len(self.current_element) > 0: el = self.current_element[-1] el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('emphasis')}) el1.text = ' (%s)' % (node.astext(), ) def depart_classifier(self, node): pass def visit_document(self, node): pass def depart_document(self, node): self.process_footnotes() def visit_docinfo(self, node): self.section_level += 1 self.section_count += 1 if self.settings.create_sections: el = self.append_child( 'text:section', attrib={ 'text:name': 'Section%d' % self.section_count, 'text:style-name': 'Sect%d' % self.section_level, } ) self.set_current_element(el) def depart_docinfo(self, node): self.section_level -= 1 if self.settings.create_sections: self.set_to_parent() def visit_emphasis(self, node): el = SubElement( self.current_element, 'text:span', attrib={'text:style-name': self.rststyle('emphasis')}) self.set_current_element(el) def depart_emphasis(self, node): self.set_to_parent() def visit_enumerated_list(self, node): el1 = self.current_element if self.blockstyle == self.rststyle('blockquote'): el2 = SubElement(el1, 'text:list', attrib={ 'text:style-name': self.rststyle('blockquote-enumlist'), }) self.list_style_stack.append(self.rststyle('blockquote-enumitem')) elif self.blockstyle == self.rststyle('highlights'): el2 = SubElement(el1, 'text:list', attrib={ 'text:style-name': self.rststyle('highlights-enumlist'), }) self.list_style_stack.append(self.rststyle('highlights-enumitem')) elif self.blockstyle == self.rststyle('epigraph'): el2 = SubElement(el1, 'text:list', attrib={ 'text:style-name': self.rststyle('epigraph-enumlist'), }) self.list_style_stack.append(self.rststyle('epigraph-enumitem')) else: liststylename = 'enumlist-%s' % (node.get('enumtype', 'arabic'), ) el2 = SubElement(el1, 'text:list', attrib={ 'text:style-name': self.rststyle(liststylename), }) self.list_style_stack.append(self.rststyle('enumitem')) self.set_current_element(el2) def depart_enumerated_list(self, node): self.set_to_parent() self.list_style_stack.pop() def visit_list_item(self, node): # If we are in a "bumped" list level, then wrap this # list in an outer lists in order to increase the # indentation level. if self.in_table_of_contents: if self.settings.generate_oowriter_toc: self.paragraph_style_stack.append( self.rststyle('contents-%d' % (self.list_level, ))) else: el1 = self.append_child('text:list-item') self.set_current_element(el1) else: el1 = self.append_child('text:list-item') el3 = el1 if len(self.bumped_list_level_stack) > 0: level_obj = self.bumped_list_level_stack[-1] if level_obj.get_sibling(): level_obj.set_nested(False) for level_obj1 in self.bumped_list_level_stack: for idx in range(level_obj1.get_level()): el2 = self.append_child('text:list', parent=el3) el3 = self.append_child( 'text:list-item', parent=el2) self.paragraph_style_stack.append(self.list_style_stack[-1]) self.set_current_element(el3) def depart_list_item(self, node): if self.in_table_of_contents: if self.settings.generate_oowriter_toc: self.paragraph_style_stack.pop() else: self.set_to_parent() else: if len(self.bumped_list_level_stack) > 0: level_obj = self.bumped_list_level_stack[-1] if level_obj.get_sibling(): level_obj.set_nested(True) for level_obj1 in self.bumped_list_level_stack: for idx in range(level_obj1.get_level()): self.set_to_parent() self.set_to_parent() self.paragraph_style_stack.pop() self.set_to_parent() def visit_header(self, node): self.in_header = True def depart_header(self, node): self.in_header = False def visit_footer(self, node): self.in_footer = True def depart_footer(self, node): self.in_footer = False def visit_field(self, node): pass def depart_field(self, node): pass def visit_field_list(self, node): pass def depart_field_list(self, node): pass def visit_field_name(self, node): el = self.append_p('textbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) el1.text = node.astext() def depart_field_name(self, node): pass def visit_field_body(self, node): self.paragraph_style_stack.append(self.rststyle('blockindent')) def depart_field_body(self, node): self.paragraph_style_stack.pop() def visit_figure(self, node): pass def depart_figure(self, node): pass def visit_footnote(self, node): self.footnote_level += 1 self.save_footnote_current = self.current_element el1 = Element('text:note-body') self.current_element = el1 self.footnote_list.append((node, el1)) if isinstance(node, docutils.nodes.citation): self.paragraph_style_stack.append(self.rststyle('citation')) else: self.paragraph_style_stack.append(self.rststyle('footnote')) def depart_footnote(self, node): self.paragraph_style_stack.pop() self.current_element = self.save_footnote_current self.footnote_level -= 1 footnote_chars = [ '*', '**', '***', '++', '+++', '##', '###', '@@', '@@@', ] def visit_footnote_reference(self, node): if self.footnote_level <= 0: id = node.attributes['ids'][0] refid = node.attributes.get('refid') if refid is None: refid = '' if self.settings.endnotes_end_doc: note_class = 'endnote' else: note_class = 'footnote' el1 = self.append_child('text:note', attrib={ 'text:id': '%s' % (refid, ), 'text:note-class': note_class, }) note_auto = str(node.attributes.get('auto', 1)) if isinstance(node, docutils.nodes.citation_reference): citation = '[%s]' % node.astext() el2 = SubElement(el1, 'text:note-citation', attrib={ 'text:label': citation, }) el2.text = citation elif note_auto == '1': el2 = SubElement(el1, 'text:note-citation', attrib={ 'text:label': node.astext(), }) el2.text = node.astext() elif note_auto == '*': if self.footnote_chars_idx >= len( ODFTranslator.footnote_chars): self.footnote_chars_idx = 0 footnote_char = ODFTranslator.footnote_chars[ self.footnote_chars_idx] self.footnote_chars_idx += 1 el2 = SubElement(el1, 'text:note-citation', attrib={ 'text:label': footnote_char, }) el2.text = footnote_char self.footnote_ref_dict[id] = el1 raise nodes.SkipChildren() def depart_footnote_reference(self, node): pass def visit_citation(self, node): self.in_citation = True for id in node.attributes['ids']: self.citation_id = id break self.paragraph_style_stack.append(self.rststyle('blockindent')) self.bumped_list_level_stack.append(ListLevel(1)) def depart_citation(self, node): self.citation_id = None self.paragraph_style_stack.pop() self.bumped_list_level_stack.pop() self.in_citation = False def visit_citation_reference(self, node): if self.settings.create_links: id = node.attributes['refid'] el = self.append_child('text:reference-ref', attrib={ 'text:ref-name': '%s' % (id, ), 'text:reference-format': 'text', }) el.text = '[' self.set_current_element(el) elif self.current_element.text is None: self.current_element.text = '[' else: self.current_element.text += '[' def depart_citation_reference(self, node): self.current_element.text += ']' if self.settings.create_links: self.set_to_parent() def visit_label(self, node): if isinstance(node.parent, docutils.nodes.footnote): raise nodes.SkipChildren() elif self.citation_id is not None: el = self.append_p('textbody') self.set_current_element(el) if self.settings.create_links: el0 = SubElement(el, 'text:span') el0.text = '[' self.append_child('text:reference-mark-start', attrib={ 'text:name': '%s' % (self.citation_id, ), }) else: el.text = '[' def depart_label(self, node): if isinstance(node.parent, docutils.nodes.footnote): pass elif self.citation_id is not None: if self.settings.create_links: self.append_child('text:reference-mark-end', attrib={ 'text:name': '%s' % (self.citation_id, ), }) el0 = SubElement(self.current_element, 'text:span') el0.text = ']' else: self.current_element.text += ']' self.set_to_parent() def visit_generated(self, node): pass def depart_generated(self, node): pass def check_file_exists(self, path): if os.path.exists(path): return 1 else: return 0 def visit_image(self, node): # Capture the image file. if 'uri' in node.attributes: source = node.attributes['uri'] if not (source.startswith('http:') or source.startswith('https:')): if not source.startswith(os.sep): docsource, line = utils.get_source_line(node) if docsource: dirname = os.path.dirname(docsource) if dirname: source = '%s%s%s' % (dirname, os.sep, source, ) if not self.check_file_exists(source): self.document.reporter.warning( 'Cannot find image file %s.' % (source, )) return else: return if source in self.image_dict: filename, destination = self.image_dict[source] else: self.image_count += 1 filename = os.path.split(source)[1] destination = 'Pictures/1%08x%s' % (self.image_count, filename, ) if source.startswith('http:') or source.startswith('https:'): try: imgfile = urlopen(source) content = imgfile.read() imgfile.close() imgfile2 = tempfile.NamedTemporaryFile('wb', delete=False) imgfile2.write(content) imgfile2.close() imgfilename = imgfile2.name source = imgfilename except HTTPError: self.document.reporter.warning( "Can't open image url %s." % (source, )) spec = (source, destination,) else: spec = (os.path.abspath(source), destination,) self.embedded_file_list.append(spec) self.image_dict[source] = (source, destination,) # Is this a figure (containing an image) or just a plain image? if self.in_paragraph: el1 = self.current_element else: el1 = SubElement( self.current_element, 'text:p', attrib={'text:style-name': self.rststyle('textbody')}) el2 = el1 if isinstance(node.parent, docutils.nodes.figure): el3, el4, el5, caption = self.generate_figure( node, source, destination, el2) attrib = {} el6, width = self.generate_image( node, source, destination, el5, attrib) if caption is not None: el6.tail = caption else: # if isinstance(node.parent, docutils.nodes.image): self.generate_image(node, source, destination, el2) def depart_image(self, node): pass def get_image_width_height(self, node, attr): size = None unit = None if attr in node.attributes: size = node.attributes[attr] size = size.strip() # For conversion factors, see: # http://www.unitconversion.org/unit_converter/typography-ex.html try: if size.endswith('%'): if attr == 'height': # Percentage allowed for width but not height. raise ValueError('percentage not allowed for height') size = size.rstrip(' %') size = float(size) / 100.0 unit = '%' else: size, unit = self.convert_to_cm(size) except ValueError as exp: self.document.reporter.warning( 'Invalid %s for image: "%s". ' 'Error: "%s".' % ( attr, node.attributes[attr], exp)) return size, unit def convert_to_cm(self, size): """Convert various units to centimeters. Note that a call to this method should be wrapped in: try: except ValueError: """ size = size.strip() if size.endswith('px'): size = float(size[:-2]) * 0.026 # convert px to cm elif size.endswith('in'): size = float(size[:-2]) * 2.54 # convert in to cm elif size.endswith('pt'): size = float(size[:-2]) * 0.035 # convert pt to cm elif size.endswith('pc'): size = float(size[:-2]) * 2.371 # convert pc to cm elif size.endswith('mm'): size = float(size[:-2]) * 0.1 # convert mm to cm elif size.endswith('cm'): size = float(size[:-2]) else: raise ValueError('unknown unit type') unit = 'cm' return size, unit def get_image_scale(self, node): if 'scale' in node.attributes: scale = node.attributes['scale'] try: scale = int(scale) except ValueError: self.document.reporter.warning( 'Invalid scale for image: "%s"' % ( node.attributes['scale'], )) if scale < 1: # or scale > 100: self.document.reporter.warning( 'scale out of range (%s), using 1.' % (scale, )) scale = 1 scale = scale * 0.01 else: scale = 1.0 return scale def get_image_scaled_width_height(self, node, source): """Return the image size in centimeters adjusted by image attrs.""" scale = self.get_image_scale(node) width, width_unit = self.get_image_width_height(node, 'width') height, _ = self.get_image_width_height(node, 'height') dpi = (72, 72) if PIL is not None and source in self.image_dict: filename, destination = self.image_dict[source] with PIL.Image.open(filename, 'r') as img: img_size = img.size dpi = img.info.get('dpi', dpi) # dpi information can be (xdpi, ydpi) or xydpi try: iter(dpi) except TypeError: dpi = (dpi, dpi) else: img_size = None if width is None or height is None: if img_size is None: raise RuntimeError( 'image size not fully specified and PIL not installed') if width is None: width = img_size[0] width = float(width) * 0.026 # convert px to cm if height is None: height = img_size[1] height = float(height) * 0.026 # convert px to cm if width_unit == '%': factor = width image_width = img_size[0] image_width = float(image_width) * 0.026 # convert px to cm image_height = img_size[1] image_height = float(image_height) * 0.026 # convert px to cm line_width = self.get_page_width() width = factor * line_width factor = (factor * line_width) / image_width height = factor * image_height width *= scale height *= scale width = '%.2fcm' % width height = '%.2fcm' % height return width, height def get_page_width(self): """Return the document's page width in centimeters.""" root = self.get_dom_stylesheet() nodes = root.iterfind( './/{urn:oasis:names:tc:opendocument:xmlns:style:1.0}' 'page-layout/' '{urn:oasis:names:tc:opendocument:xmlns:style:1.0}' 'page-layout-properties') width = None for node in nodes: page_width = node.get( '{urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0}' 'page-width') margin_left = node.get( '{urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0}' 'margin-left') margin_right = node.get( '{urn:oasis:names:tc:opendocument:xmlns:xsl-fo-compatible:1.0}' 'margin-right') if (page_width is None or margin_left is None or margin_right is None): continue try: page_width, _ = self.convert_to_cm(page_width) margin_left, _ = self.convert_to_cm(margin_left) margin_right, _ = self.convert_to_cm(margin_right) except ValueError: self.document.reporter.warning( 'Stylesheet file contains invalid page width ' 'or margin size.') width = page_width - margin_left - margin_right if width is None: # We can't find the width in styles, so we make a guess. # Use a width of 6 in = 15.24 cm. width = 15.24 return width def generate_figure(self, node, source, destination, current_element): caption = None width, height = self.get_image_scaled_width_height(node, source) for node1 in node.parent.children: if node1.tagname == 'caption': caption = node1.astext() self.image_style_count += 1 # # Add the style for the caption. if caption is not None: attrib = { 'style:class': 'extra', 'style:family': 'paragraph', 'style:name': 'Caption', 'style:parent-style-name': 'Standard', } el1 = SubElement(self.automatic_styles, 'style:style', attrib=attrib, nsdict=SNSD) attrib = { 'fo:margin-bottom': '0.0835in', 'fo:margin-top': '0.0835in', 'text:line-number': '0', 'text:number-lines': 'false', } SubElement(el1, 'style:paragraph-properties', attrib=attrib, nsdict=SNSD) attrib = { 'fo:font-size': '12pt', 'fo:font-style': 'italic', 'style:font-name': 'Times', 'style:font-name-complex': 'Lucidasans1', 'style:font-size-asian': '12pt', 'style:font-size-complex': '12pt', 'style:font-style-asian': 'italic', 'style:font-style-complex': 'italic', } SubElement(el1, 'style:text-properties', attrib=attrib, nsdict=SNSD) style_name = 'rstframestyle%d' % self.image_style_count draw_name = 'graphics%d' % next(IMAGE_NAME_COUNTER) # Add the styles attrib = { 'style:name': style_name, 'style:family': 'graphic', 'style:parent-style-name': self.rststyle('figureframe'), } el1 = SubElement(self.automatic_styles, 'style:style', attrib=attrib, nsdict=SNSD) attrib = {} wrap = False classes = node.parent.attributes.get('classes') if classes and 'wrap' in classes: wrap = True if wrap: attrib['style:wrap'] = 'dynamic' else: attrib['style:wrap'] = 'none' SubElement(el1, 'style:graphic-properties', attrib=attrib, nsdict=SNSD) attrib = { 'draw:style-name': style_name, 'draw:name': draw_name, 'text:anchor-type': 'paragraph', 'draw:z-index': '0', } attrib['svg:width'] = width el3 = SubElement(current_element, 'draw:frame', attrib=attrib) attrib = {} el4 = SubElement(el3, 'draw:text-box', attrib=attrib) attrib = { 'text:style-name': self.rststyle('caption'), } el5 = SubElement(el4, 'text:p', attrib=attrib) return el3, el4, el5, caption def generate_image(self, node, source, destination, current_element, frame_attrs=None): width, height = self.get_image_scaled_width_height(node, source) self.image_style_count += 1 style_name = 'rstframestyle%d' % self.image_style_count # Add the style. attrib = { 'style:name': style_name, 'style:family': 'graphic', 'style:parent-style-name': self.rststyle('image'), } el1 = SubElement(self.automatic_styles, 'style:style', attrib=attrib, nsdict=SNSD) halign = None valign = None if 'align' in node.attributes: align = node.attributes['align'].split() for val in align: if val in ('left', 'center', 'right'): halign = val elif val in ('top', 'middle', 'bottom'): valign = val if frame_attrs is None: attrib = { 'style:vertical-pos': 'top', 'style:vertical-rel': 'paragraph', 'style:horizontal-rel': 'paragraph', 'style:mirror': 'none', 'fo:clip': 'rect(0cm 0cm 0cm 0cm)', 'draw:luminance': '0%', 'draw:contrast': '0%', 'draw:red': '0%', 'draw:green': '0%', 'draw:blue': '0%', 'draw:gamma': '100%', 'draw:color-inversion': 'false', 'draw:image-opacity': '100%', 'draw:color-mode': 'standard', } else: attrib = frame_attrs if halign is not None: attrib['style:horizontal-pos'] = halign if valign is not None: attrib['style:vertical-pos'] = valign # If there is a classes/wrap directive or we are # inside a table, add a no-wrap style. wrap = False classes = node.attributes.get('classes') if classes and 'wrap' in classes: wrap = True if wrap: attrib['style:wrap'] = 'dynamic' else: attrib['style:wrap'] = 'none' # If we are inside a table, add a no-wrap style. if self.is_in_table(node): attrib['style:wrap'] = 'none' SubElement(el1, 'style:graphic-properties', attrib=attrib, nsdict=SNSD) draw_name = 'graphics%d' % next(IMAGE_NAME_COUNTER) # Add the content. # el = SubElement(current_element, 'text:p', # attrib={'text:style-name': self.rststyle('textbody')}) attrib = { 'draw:style-name': style_name, 'draw:name': draw_name, 'draw:z-index': '1', } if isinstance(node.parent, nodes.TextElement): attrib['text:anchor-type'] = 'as-char' # vds else: attrib['text:anchor-type'] = 'paragraph' attrib['svg:width'] = width attrib['svg:height'] = height el1 = SubElement(current_element, 'draw:frame', attrib=attrib) SubElement(el1, 'draw:image', attrib={ 'xlink:href': '%s' % (destination, ), 'xlink:type': 'simple', 'xlink:show': 'embed', 'xlink:actuate': 'onLoad', }) return el1, width def is_in_table(self, node): node1 = node.parent while node1: if isinstance(node1, docutils.nodes.entry): return True node1 = node1.parent return False def visit_legend(self, node): if isinstance(node.parent, docutils.nodes.figure): el1 = self.current_element[-1] el1 = el1[0][0] self.current_element = el1 self.paragraph_style_stack.append(self.rststyle('legend')) def depart_legend(self, node): if isinstance(node.parent, docutils.nodes.figure): self.paragraph_style_stack.pop() self.set_to_parent() self.set_to_parent() self.set_to_parent() def visit_line_block(self, node): self.line_indent_level += 1 self.line_block_level += 1 def depart_line_block(self, node): self.line_indent_level -= 1 self.line_block_level -= 1 def visit_line(self, node): style = 'lineblock%d' % self.line_indent_level el1 = SubElement(self.current_element, 'text:p', attrib={'text:style-name': self.rststyle(style), }) self.current_element = el1 def depart_line(self, node): self.set_to_parent() def visit_literal(self, node): el = SubElement( self.current_element, 'text:span', attrib={'text:style-name': self.rststyle('inlineliteral')}) self.set_current_element(el) def depart_literal(self, node): self.set_to_parent() def visit_inline(self, node): styles = node.attributes.get('classes', ()) if styles: el = self.current_element for inline_style in styles: el = SubElement(el, 'text:span', attrib={'text:style-name': self.rststyle(inline_style)}) count = len(styles) else: # No style was specified so use a default style (old code # crashed if no style was given) el = SubElement(self.current_element, 'text:span') count = 1 self.set_current_element(el) self.inline_style_count_stack.append(count) def depart_inline(self, node): count = self.inline_style_count_stack.pop() for x in range(count): self.set_to_parent() def _calculate_code_block_padding(self, line): count = 0 matchobj = SPACES_PATTERN.match(line) if matchobj: pad = matchobj.group() count = len(pad) else: matchobj = TABS_PATTERN.match(line) if matchobj: pad = matchobj.group() count = len(pad) * 8 return count def _add_syntax_highlighting(self, insource, language): lexer = pygments.lexers.get_lexer_by_name(language, stripall=True) if language in ('latex', 'tex'): fmtr = OdtPygmentsLaTeXFormatter( lambda name, parameters=(): self.rststyle(name, parameters), escape_function=escape_cdata) else: fmtr = OdtPygmentsProgFormatter( lambda name, parameters=(): self.rststyle(name, parameters), escape_function=escape_cdata) return pygments.highlight(insource, lexer, fmtr) def fill_line(self, line): line = FILL_PAT1.sub(self.fill_func1, line) return FILL_PAT2.sub(self.fill_func2, line) def fill_func1(self, matchobj): spaces = matchobj.group(0) return '<text:s text:c="%d"/>' % (len(spaces), ) def fill_func2(self, matchobj): spaces = matchobj.group(0) return ' <text:s text:c="%d"/>' % (len(spaces) - 1, ) def visit_literal_block(self, node): if len(self.paragraph_style_stack) > 1: wrapper1 = '<text:p text:style-name="%s">%%s</text:p>' % ( self.rststyle('codeblock-indented'), ) else: wrapper1 = '<text:p text:style-name="%s">%%s</text:p>' % ( self.rststyle('codeblock'), ) source = node.astext() if (pygments and self.settings.add_syntax_highlighting): language = node.get('language', 'python') source = self._add_syntax_highlighting(source, language) else: source = escape_cdata(source) lines = source.split('\n') # If there is an empty last line, remove it. if lines[-1] == '': del lines[-1] lines1 = ['<wrappertag1 xmlns:text="urn:oasis:names:tc:' 'opendocument:xmlns:text:1.0">'] my_lines = [] for my_line in lines: my_line = self.fill_line(my_line) my_line = my_line.replace(" ", "\n") my_lines.append(my_line) my_lines_str = '<text:line-break/>'.join(my_lines) my_lines_str2 = wrapper1 % (my_lines_str, ) lines1.append(my_lines_str2) lines1.append('</wrappertag1>') s1 = ''.join(lines1) s1 = s1.encode("utf-8") el1 = etree.fromstring(s1) for child in el1: self.current_element.append(child) def depart_literal_block(self, node): pass visit_doctest_block = visit_literal_block depart_doctest_block = depart_literal_block # placeholder for math (see docs/dev/todo.txt) def visit_math(self, node): self.document.reporter.warning('"math" role not supported', base_node=node) self.visit_literal(node) def depart_math(self, node): self.depart_literal(node) def visit_math_block(self, node): self.document.reporter.warning('"math" directive not supported', base_node=node) self.visit_literal_block(node) def depart_math_block(self, node): self.depart_literal_block(node) def visit_meta(self, node): name = node.attributes.get('name') content = node.attributes.get('content') if name is not None and content is not None: self.meta_dict[name] = content def depart_meta(self, node): pass def visit_option_list(self, node): table_name = 'tableoption' # # Generate automatic styles if not self.optiontablestyles_generated: self.optiontablestyles_generated = True el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle(table_name), 'style:family': 'table'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-properties', attrib={ 'style:width': '17.59cm', 'table:align': 'left', 'style:shadow': 'none'}, nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle('%s.%%c' % table_name, ('A', )), 'style:family': 'table-column'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-column-properties', attrib={ 'style:column-width': '4.999cm'}, nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle('%s.%%c' % table_name, ('B', )), 'style:family': 'table-column'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-column-properties', attrib={ 'style:column-width': '12.587cm'}, nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s.%%c%%d' % table_name, ('A', 1, )), 'style:family': 'table-cell'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-cell-properties', attrib={ 'fo:background-color': 'transparent', 'fo:padding': '0.097cm', 'fo:border-left': '0.035cm solid #000000', 'fo:border-right': 'none', 'fo:border-top': '0.035cm solid #000000', 'fo:border-bottom': '0.035cm solid #000000'}, nsdict=SNSD) el2 = SubElement(el1, 'style:background-image', nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s.%%c%%d' % table_name, ('B', 1, )), 'style:family': 'table-cell'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-cell-properties', attrib={ 'fo:padding': '0.097cm', 'fo:border': '0.035cm solid #000000'}, nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s.%%c%%d' % table_name, ('A', 2, )), 'style:family': 'table-cell'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-cell-properties', attrib={ 'fo:padding': '0.097cm', 'fo:border-left': '0.035cm solid #000000', 'fo:border-right': 'none', 'fo:border-top': 'none', 'fo:border-bottom': '0.035cm solid #000000'}, nsdict=SNSD) el = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s.%%c%%d' % table_name, ('B', 2, )), 'style:family': 'table-cell'}, nsdict=SNSD) el1 = SubElement(el, 'style:table-cell-properties', attrib={ 'fo:padding': '0.097cm', 'fo:border-left': '0.035cm solid #000000', 'fo:border-right': '0.035cm solid #000000', 'fo:border-top': 'none', 'fo:border-bottom': '0.035cm solid #000000'}, nsdict=SNSD) # # Generate table data el = self.append_child('table:table', attrib={ 'table:name': self.rststyle(table_name), 'table:style-name': self.rststyle(table_name), }) el1 = SubElement(el, 'table:table-column', attrib={ 'table:style-name': self.rststyle( '%s.%%c' % table_name, ('A', ))}) el1 = SubElement(el, 'table:table-column', attrib={ 'table:style-name': self.rststyle( '%s.%%c' % table_name, ('B', ))}) el1 = SubElement(el, 'table:table-header-rows') el2 = SubElement(el1, 'table:table-row') el3 = SubElement(el2, 'table:table-cell', attrib={ 'table:style-name': self.rststyle( '%s.%%c%%d' % table_name, ('A', 1, )), 'office:value-type': 'string'}) el4 = SubElement(el3, 'text:p', attrib={ 'text:style-name': 'Table_20_Heading'}) el4.text = 'Option' el3 = SubElement(el2, 'table:table-cell', attrib={ 'table:style-name': self.rststyle( '%s.%%c%%d' % table_name, ('B', 1, )), 'office:value-type': 'string'}) el4 = SubElement(el3, 'text:p', attrib={ 'text:style-name': 'Table_20_Heading'}) el4.text = 'Description' self.set_current_element(el) def depart_option_list(self, node): self.set_to_parent() def visit_option_list_item(self, node): el = self.append_child('table:table-row') self.set_current_element(el) def depart_option_list_item(self, node): self.set_to_parent() def visit_option_group(self, node): el = self.append_child('table:table-cell', attrib={ 'table:style-name': 'Table%d.A2' % self.table_count, 'office:value-type': 'string', }) self.set_current_element(el) def depart_option_group(self, node): self.set_to_parent() def visit_option(self, node): el = self.append_child('text:p', attrib={ 'text:style-name': 'Table_20_Contents'}) el.text = node.astext() def depart_option(self, node): pass def visit_option_string(self, node): pass def depart_option_string(self, node): pass def visit_option_argument(self, node): pass def depart_option_argument(self, node): pass def visit_description(self, node): el = self.append_child('table:table-cell', attrib={ 'table:style-name': 'Table%d.B2' % self.table_count, 'office:value-type': 'string', }) el1 = SubElement(el, 'text:p', attrib={ 'text:style-name': 'Table_20_Contents'}) el1.text = node.astext() raise nodes.SkipChildren() def depart_description(self, node): pass def visit_paragraph(self, node): self.in_paragraph = True if self.in_header: el = self.append_p('header') elif self.in_footer: el = self.append_p('footer') else: style_name = self.paragraph_style_stack[-1] el = self.append_child( 'text:p', attrib={'text:style-name': style_name}) self.append_pending_ids(el) self.set_current_element(el) def depart_paragraph(self, node): self.in_paragraph = False self.set_to_parent() if self.in_header: self.header_content.append(self.current_element[-1]) self.current_element.remove(self.current_element[-1]) elif self.in_footer: self.footer_content.append(self.current_element[-1]) self.current_element.remove(self.current_element[-1]) def visit_problematic(self, node): pass def depart_problematic(self, node): pass def visit_raw(self, node): if 'format' in node.attributes: formats = node.attributes['format'] formatlist = formats.split() if 'odt' in formatlist: rawstr = node.astext() attrstr = ' '.join( '%s="%s"' % (k, v, ) for k, v in list(CONTENT_NAMESPACE_ATTRIB.items())) contentstr = '<stuff %s>%s</stuff>' % (attrstr, rawstr, ) contentstr = contentstr.encode("utf-8") content = etree.fromstring(contentstr) if len(content) > 0: el1 = content[0] if self.in_header: pass elif self.in_footer: pass else: self.current_element.append(el1) raise nodes.SkipChildren() def depart_raw(self, node): if self.in_header: pass elif self.in_footer: pass else: pass def visit_reference(self, node): # text = node.astext() if self.settings.create_links: if 'refuri' in node: href = node['refuri'] if (self.settings.cloak_email_addresses and href.startswith('mailto:')): href = self.cloak_mailto(href) el = self.append_child('text:a', attrib={ 'xlink:href': '%s' % href, 'xlink:type': 'simple', }) self.set_current_element(el) elif 'refid' in node: if self.settings.create_links: href = node['refid'] el = self.append_child('text:reference-ref', attrib={ 'text:ref-name': '%s' % href, 'text:reference-format': 'text', }) else: self.document.reporter.warning( 'References must have "refuri" or "refid" attribute.') if (self.in_table_of_contents and len(node.children) >= 1 and isinstance(node.children[0], docutils.nodes.generated)): node.remove(node.children[0]) def depart_reference(self, node): if self.settings.create_links: if 'refuri' in node: self.set_to_parent() def visit_rubric(self, node): style_name = self.rststyle('rubric') classes = node.get('classes') if classes: class1 = classes[0] if class1: style_name = class1 el = SubElement(self.current_element, 'text:h', attrib={ # 'text:outline-level': '%d' % section_level, # 'text:style-name': 'Heading_20_%d' % section_level, 'text:style-name': style_name, }) text = node.astext() el.text = self.encode(text) def depart_rubric(self, node): pass def visit_section(self, node, move_ids=1): self.section_level += 1 self.section_count += 1 if self.settings.create_sections: el = self.append_child('text:section', attrib={ 'text:name': 'Section%d' % self.section_count, 'text:style-name': 'Sect%d' % self.section_level, }) self.set_current_element(el) def depart_section(self, node): self.section_level -= 1 if self.settings.create_sections: self.set_to_parent() def visit_strong(self, node): el = SubElement(self.current_element, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) self.set_current_element(el) def depart_strong(self, node): self.set_to_parent() def visit_substitution_definition(self, node): raise nodes.SkipChildren() def depart_substitution_definition(self, node): pass def visit_system_message(self, node): pass def depart_system_message(self, node): pass def get_table_style(self, node): table_style = None table_name = None str_classes = node.get('classes') if str_classes is not None: for str_class in str_classes: if str_class.startswith(TABLESTYLEPREFIX): table_name = str_class break if table_name is not None: table_style = self.table_styles.get(table_name) if table_style is None: # If we can't find the table style, issue warning # and use the default table style. self.document.reporter.warning( 'Can\'t find table style "%s". Using default.' % ( table_name, )) table_name = TABLENAMEDEFAULT table_style = self.table_styles.get(table_name) if table_style is None: # If we can't find the default table style, issue a warning # and use a built-in default style. self.document.reporter.warning( 'Can\'t find default table style "%s". ' 'Using built-in default.' % ( table_name, )) table_style = BUILTIN_DEFAULT_TABLE_STYLE else: table_name = TABLENAMEDEFAULT table_style = self.table_styles.get(table_name) if table_style is None: # If we can't find the default table style, issue a warning # and use a built-in default style. self.document.reporter.warning( 'Can\'t find default table style "%s". ' 'Using built-in default.' % ( table_name, )) table_style = BUILTIN_DEFAULT_TABLE_STYLE return table_style def visit_table(self, node): self.table_count += 1 table_style = self.get_table_style(node) table_name = '%s%%d' % TABLESTYLEPREFIX el1 = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s' % table_name, (self.table_count, )), 'style:family': 'table', }, nsdict=SNSD) if table_style.backgroundcolor is None: SubElement(el1, 'style:table-properties', attrib={ # 'style:width': '17.59cm', # 'table:align': 'margins', 'table:align': 'left', 'fo:margin-top': '0in', 'fo:margin-bottom': '0.10in', }, nsdict=SNSD) else: SubElement(el1, 'style:table-properties', attrib={ # 'style:width': '17.59cm', 'table:align': 'margins', 'fo:margin-top': '0in', 'fo:margin-bottom': '0.10in', 'fo:background-color': table_style.backgroundcolor, }, nsdict=SNSD) # We use a single cell style for all cells in this table. # That's probably not correct, but seems to work. el2 = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': self.rststyle( '%s.%%c%%d' % table_name, (self.table_count, 'A', 1, )), 'style:family': 'table-cell', }, nsdict=SNSD) thickness = self.settings.table_border_thickness if thickness is None: line_style1 = table_style.border else: line_style1 = '0.%03dcm solid #000000' % (thickness, ) SubElement(el2, 'style:table-cell-properties', attrib={ 'fo:padding': '0.049cm', 'fo:border-left': line_style1, 'fo:border-right': line_style1, 'fo:border-top': line_style1, 'fo:border-bottom': line_style1, }, nsdict=SNSD) title = None for child in node.children: if child.tagname == 'title': title = child.astext() break if title is not None: self.append_p('table-title', title) else: pass el4 = SubElement(self.current_element, 'table:table', attrib={ 'table:name': self.rststyle( '%s' % table_name, (self.table_count, )), 'table:style-name': self.rststyle( '%s' % table_name, (self.table_count, )), }) self.set_current_element(el4) self.current_table_style = el1 self.table_width = 0.0 def depart_table(self, node): attribkey = add_ns('style:width', nsdict=SNSD) attribval = '%.4fin' % (self.table_width, ) el1 = self.current_table_style el2 = el1[0] el2.attrib[attribkey] = attribval self.set_to_parent() def visit_tgroup(self, node): self.column_count = ord('A') - 1 def depart_tgroup(self, node): pass def visit_colspec(self, node): self.column_count += 1 colspec_name = self.rststyle( '%s%%d.%%s' % TABLESTYLEPREFIX, (self.table_count, chr(self.column_count), ) ) colwidth = node['colwidth'] / 12.0 el1 = SubElement(self.automatic_styles, 'style:style', attrib={ 'style:name': colspec_name, 'style:family': 'table-column', }, nsdict=SNSD) SubElement(el1, 'style:table-column-properties', attrib={'style:column-width': '%.4fin' % colwidth}, nsdict=SNSD) self.append_child('table:table-column', attrib={'table:style-name': colspec_name, }) self.table_width += colwidth def depart_colspec(self, node): pass def visit_thead(self, node): el = self.append_child('table:table-header-rows') self.set_current_element(el) self.in_thead = True self.paragraph_style_stack.append('Table_20_Heading') def depart_thead(self, node): self.set_to_parent() self.in_thead = False self.paragraph_style_stack.pop() def visit_row(self, node): self.column_count = ord('A') - 1 el = self.append_child('table:table-row') self.set_current_element(el) def depart_row(self, node): self.set_to_parent() def visit_entry(self, node): self.column_count += 1 cellspec_name = self.rststyle( '%s%%d.%%c%%d' % TABLESTYLEPREFIX, (self.table_count, 'A', 1, ) ) attrib = { 'table:style-name': cellspec_name, 'office:value-type': 'string', } morecols = node.get('morecols', 0) if morecols > 0: attrib['table:number-columns-spanned'] = '%d' % (morecols + 1,) self.column_count += morecols morerows = node.get('morerows', 0) if morerows > 0: attrib['table:number-rows-spanned'] = '%d' % (morerows + 1,) el1 = self.append_child('table:table-cell', attrib=attrib) self.set_current_element(el1) def depart_entry(self, node): self.set_to_parent() def visit_tbody(self, node): pass def depart_tbody(self, node): pass def visit_target(self, node): # # I don't know how to implement targets in ODF. # How do we create a target in oowriter? A cross-reference? if ('refuri' not in node and 'refid' not in node and 'refname' not in node): pass else: pass def depart_target(self, node): pass def visit_title(self, node, move_ids=1, title_type='title'): if isinstance(node.parent, docutils.nodes.section): section_level = self.section_level if section_level > 7: self.document.reporter.warning( 'Heading/section levels greater than 7 not supported.') self.document.reporter.warning( ' Reducing to heading level 7 for heading: "%s"' % ( node.astext(), )) section_level = 7 el1 = self.append_child( 'text:h', attrib={ 'text:outline-level': '%d' % section_level, # 'text:style-name': 'Heading_20_%d' % section_level, 'text:style-name': self.rststyle( 'heading%d', (section_level, )), }) self.append_pending_ids(el1) self.set_current_element(el1) elif isinstance(node.parent, docutils.nodes.document): # text = self.settings.title # else: # text = node.astext() el1 = SubElement(self.current_element, 'text:p', attrib={ 'text:style-name': self.rststyle(title_type), }) self.append_pending_ids(el1) text = node.astext() self.title = text self.found_doc_title = True self.set_current_element(el1) def depart_title(self, node): if (isinstance(node.parent, docutils.nodes.section) or isinstance(node.parent, docutils.nodes.document)): self.set_to_parent() def visit_subtitle(self, node, move_ids=1): self.visit_title(node, move_ids, title_type='subtitle') def depart_subtitle(self, node): self.depart_title(node) def visit_title_reference(self, node): el = self.append_child('text:span', attrib={ 'text:style-name': self.rststyle('quotation')}) el.text = self.encode(node.astext()) raise nodes.SkipChildren() def depart_title_reference(self, node): pass def generate_table_of_content_entry_template(self, el1): for idx in range(1, 11): el2 = SubElement( el1, 'text:table-of-content-entry-template', attrib={ 'text:outline-level': "%d" % (idx, ), 'text:style-name': self.rststyle('contents-%d' % (idx, )), }) SubElement(el2, 'text:index-entry-chapter') SubElement(el2, 'text:index-entry-text') SubElement(el2, 'text:index-entry-tab-stop', attrib={ 'style:leader-char': ".", 'style:type': "right", }) SubElement(el2, 'text:index-entry-page-number') def find_title_label(self, node, class_type, label_key): label = '' title_node = None for child in node.children: if isinstance(child, class_type): title_node = child break if title_node is not None: label = title_node.astext() else: label = self.language.labels[label_key] return label def visit_topic(self, node): if 'classes' in node.attributes: if 'contents' in node.attributes['classes']: label = self.find_title_label( node, docutils.nodes.title, 'contents') if self.settings.generate_oowriter_toc: el1 = self.append_child('text:table-of-content', attrib={ 'text:name': 'Table of Contents1', 'text:protected': 'true', 'text:style-name': 'Sect1', }) el2 = SubElement( el1, 'text:table-of-content-source', attrib={ 'text:outline-level': '10', }) el3 = SubElement(el2, 'text:index-title-template', attrib={ 'text:style-name': 'Contents_20_Heading', }) el3.text = label self.generate_table_of_content_entry_template(el2) el4 = SubElement(el1, 'text:index-body') el5 = SubElement(el4, 'text:index-title') el6 = SubElement(el5, 'text:p', attrib={ 'text:style-name': self.rststyle('contents-heading'), }) el6.text = label self.save_current_element = self.current_element self.table_of_content_index_body = el4 self.set_current_element(el4) else: el = self.append_p('horizontalline') el = self.append_p('centeredtextbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) el1.text = label self.in_table_of_contents = True elif 'abstract' in node.attributes['classes']: el = self.append_p('horizontalline') el = self.append_p('centeredtextbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) label = self.find_title_label( node, docutils.nodes.title, 'abstract') el1.text = label elif 'dedication' in node.attributes['classes']: el = self.append_p('horizontalline') el = self.append_p('centeredtextbody') el1 = SubElement( el, 'text:span', attrib={'text:style-name': self.rststyle('strong')}) label = self.find_title_label( node, docutils.nodes.title, 'dedication') el1.text = label def depart_topic(self, node): if 'classes' in node.attributes: if 'contents' in node.attributes['classes']: if self.settings.generate_oowriter_toc: self.update_toc_page_numbers( self.table_of_content_index_body) self.set_current_element(self.save_current_element) else: self.append_p('horizontalline') self.in_table_of_contents = False def update_toc_page_numbers(self, el): collection = [] self.update_toc_collect(el, 0, collection) self.update_toc_add_numbers(collection) def update_toc_collect(self, el, level, collection): collection.append((level, el)) level += 1 for child_el in el: if child_el.tag != 'text:index-body': self.update_toc_collect(child_el, level, collection) def update_toc_add_numbers(self, collection): for level, el1 in collection: if (el1.tag == 'text:p' and el1.text != 'Table of Contents'): el2 = SubElement(el1, 'text:tab') el2.tail = '9999' def visit_transition(self, node): self.append_p('horizontalline') def depart_transition(self, node): pass # # Admonitions # def visit_warning(self, node): self.generate_admonition(node, 'warning') def depart_warning(self, node): self.paragraph_style_stack.pop() def visit_attention(self, node): self.generate_admonition(node, 'attention') depart_attention = depart_warning def visit_caution(self, node): self.generate_admonition(node, 'caution') depart_caution = depart_warning def visit_danger(self, node): self.generate_admonition(node, 'danger') depart_danger = depart_warning def visit_error(self, node): self.generate_admonition(node, 'error') depart_error = depart_warning def visit_hint(self, node): self.generate_admonition(node, 'hint') depart_hint = depart_warning def visit_important(self, node): self.generate_admonition(node, 'important') depart_important = depart_warning def visit_note(self, node): self.generate_admonition(node, 'note') depart_note = depart_warning def visit_tip(self, node): self.generate_admonition(node, 'tip') depart_tip = depart_warning def visit_admonition(self, node): title = None for child in node.children: if child.tagname == 'title': title = child.astext() if title is None: classes1 = node.get('classes') if classes1: title = classes1[0] self.generate_admonition(node, 'generic', title) depart_admonition = depart_warning def generate_admonition(self, node, label, title=None): if hasattr(self.language, 'labels'): translated_label = self.language.labels.get(label, label) else: translated_label = label el1 = SubElement(self.current_element, 'text:p', attrib={ 'text:style-name': self.rststyle( 'admon-%s-hdr', (label, )), }) if title: el1.text = title else: el1.text = '%s!' % (translated_label.capitalize(), ) s1 = self.rststyle('admon-%s-body', (label, )) self.paragraph_style_stack.append(s1) # # Roles (e.g. subscript, superscript, strong, ... # def visit_subscript(self, node): el = self.append_child('text:span', attrib={ 'text:style-name': 'rststyle-subscript', }) self.set_current_element(el) def depart_subscript(self, node): self.set_to_parent() def visit_superscript(self, node): el = self.append_child('text:span', attrib={ 'text:style-name': 'rststyle-superscript', }) self.set_current_element(el) def depart_superscript(self, node): self.set_to_parent() def visit_abbreviation(self, node): pass def depart_abbreviation(self, node): pass def visit_acronym(self, node): pass def depart_acronym(self, node): pass def visit_sidebar(self, node): pass def depart_sidebar(self, node): pass # Use an own reader to modify transformations done. class Reader(standalone.Reader): def get_transforms(self): transforms = super().get_transforms() if not self.settings.create_links: transforms.remove(references.DanglingReferences) return transforms