%PDF- %PDF-
Direktori : /opt/alt/ruby27/share/gems/gems/rdoc-6.2.1.1/lib/rdoc/ |
Current File : //opt/alt/ruby27/share/gems/gems/rdoc-6.2.1.1/lib/rdoc/text.rb |
# frozen_string_literal: true ## # For RDoc::Text#to_html require 'strscan' ## # Methods for manipulating comment text module RDoc::Text attr_accessor :language ## # Maps markup formats to classes that can parse them. If the format is # unknown, "rdoc" format is used. MARKUP_FORMAT = { 'markdown' => RDoc::Markdown, 'rdoc' => RDoc::Markup, 'rd' => RDoc::RD, 'tomdoc' => RDoc::TomDoc, } MARKUP_FORMAT.default = RDoc::Markup ## # Maps an encoding to a Hash of characters properly transcoded for that # encoding. # # See also encode_fallback. TO_HTML_CHARACTERS = Hash.new do |h, encoding| h[encoding] = { :close_dquote => encode_fallback('”', encoding, '"'), :close_squote => encode_fallback('’', encoding, '\''), :copyright => encode_fallback('©', encoding, '(c)'), :ellipsis => encode_fallback('…', encoding, '...'), :em_dash => encode_fallback('—', encoding, '---'), :en_dash => encode_fallback('–', encoding, '--'), :open_dquote => encode_fallback('“', encoding, '"'), :open_squote => encode_fallback('‘', encoding, '\''), :trademark => encode_fallback('®', encoding, '(r)'), } end ## # Transcodes +character+ to +encoding+ with a +fallback+ character. def self.encode_fallback character, encoding, fallback character.encode(encoding, :fallback => { character => fallback }, :undef => :replace, :replace => fallback) end ## # Expands tab characters in +text+ to eight spaces def expand_tabs text expanded = [] text.each_line do |line| nil while line.gsub!(/(?:\G|\r)((?:.{8})*?)([^\t\r\n]{0,7})\t/) do r = "#{$1}#{$2}#{' ' * (8 - $2.size)}" r = RDoc::Encoding.change_encoding r, text.encoding r end expanded << line end expanded.join end ## # Flush +text+ left based on the shortest line def flush_left text indent = 9999 text.each_line do |line| line_indent = line =~ /\S/ || 9999 indent = line_indent if indent > line_indent end empty = '' empty = RDoc::Encoding.change_encoding empty, text.encoding text.gsub(/^ {0,#{indent}}/, empty) end ## # Convert a string in markup format into HTML. # # Requires the including class to implement #formatter def markup text if @store.rdoc.options locale = @store.rdoc.options.locale else locale = nil end if locale i18n_text = RDoc::I18n::Text.new(text) text = i18n_text.translate(locale) end parse(text).accept formatter end ## # Strips hashes, expands tabs then flushes +text+ to the left def normalize_comment text return text if text.empty? case language when :ruby text = strip_hashes text when :c text = strip_stars text end text = expand_tabs text text = flush_left text text = strip_newlines text text end ## # Normalizes +text+ then builds a RDoc::Markup::Document from it def parse text, format = 'rdoc' return text if RDoc::Markup::Document === text return text.parse if RDoc::Comment === text text = normalize_comment text # TODO remove, should not be necessary return RDoc::Markup::Document.new if text =~ /\A\n*\z/ MARKUP_FORMAT[format].parse text end ## # The first +limit+ characters of +text+ as HTML def snippet text, limit = 100 document = parse text RDoc::Markup::ToHtmlSnippet.new(options, limit).convert document end ## # Strips leading # characters from +text+ def strip_hashes text return text if text =~ /^(?>\s*)[^\#]/ empty = '' empty = RDoc::Encoding.change_encoding empty, text.encoding text.gsub(/^\s*(#+)/) { $1.tr '#', ' ' }.gsub(/^\s+$/, empty) end ## # Strips leading and trailing \n characters from +text+ def strip_newlines text text.gsub(/\A\n*(.*?)\n*\z/m) do $1 end # block preserves String encoding end ## # Strips /* */ style comments def strip_stars text return text unless text =~ %r%/\*.*\*/%m encoding = text.encoding text = text.gsub %r%Document-method:\s+[\w:.#=!?|^&<>~+\-/*\%@`\[\]]+%, '' space = ' ' space = RDoc::Encoding.change_encoding space, encoding if encoding text.sub! %r%/\*+% do space * $&.length end text.sub! %r%\*+/% do space * $&.length end text.gsub! %r%^[ \t]*\*%m do space * $&.length end empty = '' empty = RDoc::Encoding.change_encoding empty, encoding if encoding text.gsub(/^\s+$/, empty) end ## # Converts ampersand, dashes, ellipsis, quotes, copyright and registered # trademark symbols in +text+ to properly encoded characters. def to_html text html = (''.encode text.encoding).dup encoded = RDoc::Text::TO_HTML_CHARACTERS[text.encoding] s = StringScanner.new text insquotes = false indquotes = false after_word = nil until s.eos? do case when s.scan(/<(tt|code)>.*?<\/\1>/) then # skip contents of tt html << s.matched.gsub('\\\\', '\\') when s.scan(/<(tt|code)>.*?/) then warn "mismatched <#{s[1]}> tag" # TODO signal file/line html << s.matched when s.scan(/<[^>]+\/?s*>/) then # skip HTML tags html << s.matched when s.scan(/\\(\S)/) then # unhandled suppressed crossref html << s[1] after_word = nil when s.scan(/\.\.\.(\.?)/) then html << s[1] << encoded[:ellipsis] after_word = nil when s.scan(/\(c\)/) then html << encoded[:copyright] after_word = nil when s.scan(/\(r\)/) then html << encoded[:trademark] after_word = nil when s.scan(/---/) then html << encoded[:em_dash] after_word = nil when s.scan(/--/) then html << encoded[:en_dash] after_word = nil when s.scan(/"|"/) then html << encoded[indquotes ? :close_dquote : :open_dquote] indquotes = !indquotes after_word = nil when s.scan(/``/) then # backtick double quote html << encoded[:open_dquote] after_word = nil when s.scan(/''/) then # tick double quote html << encoded[:close_dquote] after_word = nil when s.scan(/'/) then # single quote if insquotes html << encoded[:close_squote] insquotes = false elsif after_word # Mary's dog, my parents' house: do not start paired quotes html << encoded[:close_squote] else html << encoded[:open_squote] insquotes = true end after_word = nil else # advance to the next potentially significant character match = s.scan(/.+?(?=[<\\.("'`&-])/) #" if match then html << match after_word = match =~ /\w$/ else html << s.rest break end end end html end ## # Wraps +txt+ to +line_len+ def wrap(txt, line_len = 76) res = [] sp = 0 ep = txt.length while sp < ep # scan back for a space p = sp + line_len - 1 if p >= ep p = ep else while p > sp and txt[p] != ?\s p -= 1 end if p <= sp p = sp + line_len while p < ep and txt[p] != ?\s p += 1 end end end res << txt[sp...p] << "\n" sp = p sp += 1 while sp < ep and txt[sp] == ?\s end res.join.strip end end