%PDF- %PDF-
Direktori : /opt/imunify360/venv/lib64/python3.11/site-packages/docutils/transforms/ |
Current File : //opt/imunify360/venv/lib64/python3.11/site-packages/docutils/transforms/universal.py |
# $Id: universal.py 9320 2023-01-17 15:40:15Z milde $ # Authors: David Goodger <goodger@python.org>; Ueli Schlaepfer; Günter Milde # Maintainer: docutils-develop@lists.sourceforge.net # Copyright: This module has been placed in the public domain. """ Transforms needed by most or all documents: - `Decorations`: Generate a document's header & footer. - `ExposeInternals`: Expose internal attributes. - `Messages`: Placement of system messages generated after parsing. - `FilterMessages`: Remove system messages below verbosity threshold. - `TestMessages`: Like `Messages`, used on test runs. - `StripComments`: Remove comment elements from the document tree. - `StripClassesAndElements`: Remove elements with classes in `self.document.settings.strip_elements_with_classes` and class values in `self.document.settings.strip_classes`. - `SmartQuotes`: Replace ASCII quotation marks with typographic form. """ __docformat__ = 'reStructuredText' import re import time from docutils import nodes, utils from docutils.transforms import Transform from docutils.utils import smartquotes class Decorations(Transform): """ Populate a document's decoration element (header, footer). """ default_priority = 820 def apply(self): header_nodes = self.generate_header() if header_nodes: decoration = self.document.get_decoration() header = decoration.get_header() header.extend(header_nodes) footer_nodes = self.generate_footer() if footer_nodes: decoration = self.document.get_decoration() footer = decoration.get_footer() footer.extend(footer_nodes) def generate_header(self): return None def generate_footer(self): # @@@ Text is hard-coded for now. # Should be made dynamic (language-dependent). # @@@ Use timestamp from the `SOURCE_DATE_EPOCH`_ environment variable # for the datestamp? # See https://sourceforge.net/p/docutils/patches/132/ # and https://reproducible-builds.org/specs/source-date-epoch/ settings = self.document.settings if (settings.generator or settings.datestamp or settings.source_link or settings.source_url): text = [] if (settings.source_link and settings._source or settings.source_url): if settings.source_url: source = settings.source_url else: source = utils.relative_path(settings._destination, settings._source) text.extend([ nodes.reference('', 'View document source', refuri=source), nodes.Text('.\n')]) if settings.datestamp: datestamp = time.strftime(settings.datestamp, time.gmtime()) text.append(nodes.Text('Generated on: ' + datestamp + '.\n')) if settings.generator: text.extend([ nodes.Text('Generated by '), nodes.reference('', 'Docutils', refuri='https://docutils.sourceforge.io/'), nodes.Text(' from '), nodes.reference('', 'reStructuredText', refuri='https://docutils.sourceforge.io/' 'rst.html'), nodes.Text(' source.\n')]) return [nodes.paragraph('', '', *text)] else: return None class ExposeInternals(Transform): """ Expose internal attributes if ``expose_internals`` setting is set. """ default_priority = 840 def not_Text(self, node): return not isinstance(node, nodes.Text) def apply(self): if self.document.settings.expose_internals: for node in self.document.findall(self.not_Text): for att in self.document.settings.expose_internals: value = getattr(node, att, None) if value is not None: node['internal:' + att] = value class Messages(Transform): """ Place any system messages generated after parsing into a dedicated section of the document. """ default_priority = 860 def apply(self): messages = self.document.transform_messages loose_messages = [msg for msg in messages if not msg.parent] if loose_messages: section = nodes.section(classes=['system-messages']) # @@@ get this from the language module? section += nodes.title('', 'Docutils System Messages') section += loose_messages self.document.transform_messages[:] = [] self.document += section # TODO: fix bug #435: # Messages are filtered at a very late stage # This breaks the link from inline error messages to the corresponding # system message at the end of document. class FilterMessages(Transform): """ Remove system messages below verbosity threshold. Convert <problematic> nodes referencing removed messages to <Text>. Remove "System Messages" section if empty. """ default_priority = 870 def apply(self): for node in tuple(self.document.findall(nodes.system_message)): if node['level'] < self.document.reporter.report_level: node.parent.remove(node) try: # also remove id-entry del self.document.ids[node['ids'][0]] except (IndexError): pass for node in tuple(self.document.findall(nodes.problematic)): if node['refid'] not in self.document.ids: node.parent.replace(node, nodes.Text(node.astext())) for node in self.document.findall(nodes.section): if "system-messages" in node['classes'] and len(node) == 1: node.parent.remove(node) class TestMessages(Transform): """ Append all post-parse system messages to the end of the document. Used for testing purposes. """ # marker for pytest to ignore this class during test discovery __test__ = False default_priority = 880 def apply(self): for msg in self.document.transform_messages: if not msg.parent: self.document += msg class StripComments(Transform): """ Remove comment elements from the document tree (only if the ``strip_comments`` setting is enabled). """ default_priority = 740 def apply(self): if self.document.settings.strip_comments: for node in tuple(self.document.findall(nodes.comment)): node.parent.remove(node) class StripClassesAndElements(Transform): """ Remove from the document tree all elements with classes in `self.document.settings.strip_elements_with_classes` and all "classes" attribute values in `self.document.settings.strip_classes`. """ default_priority = 420 def apply(self): if self.document.settings.strip_elements_with_classes: self.strip_elements = {*self.document.settings .strip_elements_with_classes} # Iterate over a tuple as removing the current node # corrupts the iterator returned by `iter`: for node in tuple(self.document.findall(self.check_classes)): node.parent.remove(node) if not self.document.settings.strip_classes: return strip_classes = self.document.settings.strip_classes for node in self.document.findall(nodes.Element): for class_value in strip_classes: try: node['classes'].remove(class_value) except ValueError: pass def check_classes(self, node): if not isinstance(node, nodes.Element): return False for class_value in node['classes'][:]: if class_value in self.strip_elements: return True return False class SmartQuotes(Transform): """ Replace ASCII quotation marks with typographic form. Also replace multiple dashes with em-dash/en-dash characters. """ default_priority = 855 nodes_to_skip = (nodes.FixedTextElement, nodes.Special) """Do not apply "smartquotes" to instances of these block-level nodes.""" literal_nodes = (nodes.FixedTextElement, nodes.Special, nodes.image, nodes.literal, nodes.math, nodes.raw, nodes.problematic) """Do not apply smartquotes to instances of these inline nodes.""" smartquotes_action = 'qDe' """Setting to select smartquote transformations. The default 'qDe' educates normal quote characters: (", '), em- and en-dashes (---, --) and ellipses (...). """ def __init__(self, document, startnode): Transform.__init__(self, document, startnode=startnode) self.unsupported_languages = set() def get_tokens(self, txtnodes): # A generator that yields ``(texttype, nodetext)`` tuples for a list # of "Text" nodes (interface to ``smartquotes.educate_tokens()``). for node in txtnodes: if (isinstance(node.parent, self.literal_nodes) or isinstance(node.parent.parent, self.literal_nodes)): yield 'literal', str(node) else: # SmartQuotes uses backslash escapes instead of null-escapes # Insert backslashes before escaped "active" characters. txt = re.sub('(?<=\x00)([-\\\'".`])', r'\\\1', str(node)) yield 'plain', txt def apply(self): smart_quotes = self.document.settings.setdefault('smart_quotes', False) if not smart_quotes: return try: alternative = smart_quotes.startswith('alt') except AttributeError: alternative = False document_language = self.document.settings.language_code lc_smartquotes = self.document.settings.smartquotes_locales if lc_smartquotes: smartquotes.smartchars.quotes.update(dict(lc_smartquotes)) # "Educate" quotes in normal text. Handle each block of text # (TextElement node) as a unit to keep context around inline nodes: for node in self.document.findall(nodes.TextElement): # skip preformatted text blocks and special elements: if isinstance(node, self.nodes_to_skip): continue # nested TextElements are not "block-level" elements: if isinstance(node.parent, nodes.TextElement): continue # list of text nodes in the "text block": txtnodes = [txtnode for txtnode in node.findall(nodes.Text) if not isinstance(txtnode.parent, nodes.option_string)] # language: use typographical quotes for language "lang" lang = node.get_language_code(document_language) # use alternative form if `smart-quotes` setting starts with "alt": if alternative: if '-x-altquot' in lang: lang = lang.replace('-x-altquot', '') else: lang += '-x-altquot' # drop unsupported subtags: for tag in utils.normalize_language_tag(lang): if tag in smartquotes.smartchars.quotes: lang = tag break else: # language not supported -- keep ASCII quotes if lang not in self.unsupported_languages: self.document.reporter.warning( 'No smart quotes defined for language "%s".' % lang, base_node=node) self.unsupported_languages.add(lang) lang = '' # Iterator educating quotes in plain text: # (see "utils/smartquotes.py" for the attribute setting) teacher = smartquotes.educate_tokens( self.get_tokens(txtnodes), attr=self.smartquotes_action, language=lang) for txtnode, newtext in zip(txtnodes, teacher): txtnode.parent.replace(txtnode, nodes.Text(newtext)) self.unsupported_languages.clear()