%PDF- %PDF-
Direktori : /opt/imunify360/venv/lib/python3.11/site-packages/docutils/parsers/rst/directives/ |
Current File : //opt/imunify360/venv/lib/python3.11/site-packages/docutils/parsers/rst/directives/body.py |
# $Id: body.py 9030 2022-03-05 23:28:32Z milde $ # Author: David Goodger <goodger@python.org> # Copyright: This module has been placed in the public domain. """ Directives for additional body elements. See `docutils.parsers.rst.directives` for API details. """ __docformat__ = 'reStructuredText' from docutils import nodes from docutils.parsers.rst import Directive from docutils.parsers.rst import directives from docutils.parsers.rst.roles import set_classes from docutils.utils.code_analyzer import Lexer, LexerError, NumberLines class BasePseudoSection(Directive): required_arguments = 1 optional_arguments = 0 final_argument_whitespace = True option_spec = {'class': directives.class_option, 'name': directives.unchanged} has_content = True node_class = None """Node class to be used (must be set in subclasses).""" def run(self): if not (self.state_machine.match_titles or isinstance(self.state_machine.node, nodes.sidebar)): raise self.error('The "%s" directive may not be used within ' 'topics or body elements.' % self.name) self.assert_has_content() if self.arguments: # title (in sidebars optional) title_text = self.arguments[0] textnodes, messages = self.state.inline_text( title_text, self.lineno) titles = [nodes.title(title_text, '', *textnodes)] # Sidebar uses this code. if 'subtitle' in self.options: textnodes, more_messages = self.state.inline_text( self.options['subtitle'], self.lineno) titles.append(nodes.subtitle(self.options['subtitle'], '', *textnodes)) messages.extend(more_messages) else: titles = [] messages = [] text = '\n'.join(self.content) node = self.node_class(text, *(titles + messages)) node['classes'] += self.options.get('class', []) self.add_name(node) if text: self.state.nested_parse(self.content, self.content_offset, node) return [node] class Topic(BasePseudoSection): node_class = nodes.topic class Sidebar(BasePseudoSection): node_class = nodes.sidebar required_arguments = 0 optional_arguments = 1 option_spec = BasePseudoSection.option_spec.copy() option_spec['subtitle'] = directives.unchanged_required def run(self): if isinstance(self.state_machine.node, nodes.sidebar): raise self.error('The "%s" directive may not be used within a ' 'sidebar element.' % self.name) if 'subtitle' in self.options and not self.arguments: raise self.error('The "subtitle" option may not be used ' 'without a title.') return BasePseudoSection.run(self) class LineBlock(Directive): option_spec = {'class': directives.class_option, 'name': directives.unchanged} has_content = True def run(self): self.assert_has_content() block = nodes.line_block(classes=self.options.get('class', [])) self.add_name(block) node_list = [block] for line_text in self.content: text_nodes, messages = self.state.inline_text( line_text.strip(), self.lineno + self.content_offset) line = nodes.line(line_text, '', *text_nodes) if line_text.strip(): line.indent = len(line_text) - len(line_text.lstrip()) block += line node_list.extend(messages) self.content_offset += 1 self.state.nest_line_block_lines(block) return node_list class ParsedLiteral(Directive): option_spec = {'class': directives.class_option, 'name': directives.unchanged} has_content = True def run(self): set_classes(self.options) self.assert_has_content() text = '\n'.join(self.content) text_nodes, messages = self.state.inline_text(text, self.lineno) node = nodes.literal_block(text, '', *text_nodes, **self.options) node.line = self.content_offset + 1 self.add_name(node) return [node] + messages class CodeBlock(Directive): """Parse and mark up content of a code block. Configuration setting: syntax_highlight Highlight Code content with Pygments? Possible values: ('long', 'short', 'none') """ optional_arguments = 1 option_spec = {'class': directives.class_option, 'name': directives.unchanged, 'number-lines': directives.unchanged # integer or None } has_content = True def run(self): self.assert_has_content() if self.arguments: language = self.arguments[0] else: language = '' set_classes(self.options) classes = ['code'] if language: classes.append(language) if 'classes' in self.options: classes.extend(self.options['classes']) # set up lexical analyzer try: tokens = Lexer('\n'.join(self.content), language, self.state.document.settings.syntax_highlight) except LexerError as error: if self.state.document.settings.report_level > 2: # don't report warnings -> insert without syntax highlight tokens = Lexer('\n'.join(self.content), language, 'none') else: raise self.warning(error) if 'number-lines' in self.options: # optional argument `startline`, defaults to 1 try: startline = int(self.options['number-lines'] or 1) except ValueError: raise self.error(':number-lines: with non-integer start value') endline = startline + len(self.content) # add linenumber filter: tokens = NumberLines(tokens, startline, endline) node = nodes.literal_block('\n'.join(self.content), classes=classes) self.add_name(node) # if called from "include", set the source if 'source' in self.options: node.attributes['source'] = self.options['source'] # analyze content and add nodes for every token for classes, value in tokens: if classes: node += nodes.inline(value, value, classes=classes) else: # insert as Text to decrease the verbosity of the output node += nodes.Text(value) return [node] class MathBlock(Directive): option_spec = {'class': directives.class_option, 'name': directives.unchanged, # TODO: Add Sphinx' ``mathbase.py`` option 'nowrap'? # 'nowrap': directives.flag, } has_content = True def run(self): set_classes(self.options) self.assert_has_content() # join lines, separate blocks content = '\n'.join(self.content).split('\n\n') _nodes = [] for block in content: if not block: continue node = nodes.math_block(self.block_text, block, **self.options) node.line = self.content_offset + 1 self.add_name(node) _nodes.append(node) return _nodes class Rubric(Directive): required_arguments = 1 optional_arguments = 0 final_argument_whitespace = True option_spec = {'class': directives.class_option, 'name': directives.unchanged} def run(self): set_classes(self.options) rubric_text = self.arguments[0] textnodes, messages = self.state.inline_text(rubric_text, self.lineno) rubric = nodes.rubric(rubric_text, '', *textnodes, **self.options) self.add_name(rubric) return [rubric] + messages class BlockQuote(Directive): has_content = True classes = [] def run(self): self.assert_has_content() elements = self.state.block_quote(self.content, self.content_offset) for element in elements: if isinstance(element, nodes.block_quote): element['classes'] += self.classes return elements class Epigraph(BlockQuote): classes = ['epigraph'] class Highlights(BlockQuote): classes = ['highlights'] class PullQuote(BlockQuote): classes = ['pull-quote'] class Compound(Directive): option_spec = {'class': directives.class_option, 'name': directives.unchanged} has_content = True def run(self): self.assert_has_content() text = '\n'.join(self.content) node = nodes.compound(text) node['classes'] += self.options.get('class', []) self.add_name(node) self.state.nested_parse(self.content, self.content_offset, node) return [node] class Container(Directive): optional_arguments = 1 final_argument_whitespace = True option_spec = {'name': directives.unchanged} has_content = True def run(self): self.assert_has_content() text = '\n'.join(self.content) try: if self.arguments: classes = directives.class_option(self.arguments[0]) else: classes = [] except ValueError: raise self.error( 'Invalid class attribute value for "%s" directive: "%s".' % (self.name, self.arguments[0])) node = nodes.container(text) node['classes'].extend(classes) self.add_name(node) self.state.nested_parse(self.content, self.content_offset, node) return [node]