class RDoc::Markup::ToRdoc

def accept_blank_line blank_line

def accept_blank_line blank_line
  @res << "\n"
end

def accept_block_quote block_quote

def accept_block_quote block_quote
  @indent += 2
  block_quote.parts.each do |part|
    @prefix = '> '
    part.accept self
  end
  @indent -= 2
end

def accept_heading heading

def accept_heading heading
  use_prefix or @res << ' ' * @indent
  @res << @headings[heading.level][0]
  @res << attributes(heading.text)
  @res << @headings[heading.level][1]
  @res << "\n"
end

def accept_indented_paragraph paragraph

def accept_indented_paragraph paragraph
  @indent += paragraph.indent
  text = paragraph.text @hard_break
  wrap attributes text
  @indent -= paragraph.indent
end

def accept_list_end list

def accept_list_end list
  @list_index.pop
  @list_type.pop
  @list_width.pop
end

def accept_list_item_end list_item

def accept_list_item_end list_item
  width = case @list_type.last
          when :BULLET then
            2
          when :NOTE, :LABEL then
            if @prefix then
              @res << @prefix.strip
              @prefix = nil
            end
            @res << "\n"
            2
          else
            bullet = @list_index.last.to_s
            @list_index[-1] = @list_index.last.succ
            bullet.length + 2
          end
  @indent -= width
end

def accept_list_item_start list_item

def accept_list_item_start list_item
  type = @list_type.last
  case type
  when :NOTE, :LABEL then
    stripped_labels = Array(list_item.label).map do |label|
      attributes(label).strip
    end
    bullets = case type
    when :NOTE
      stripped_labels.map { |b| "#{b}::" }
    when :LABEL
      stripped_labels.map { |b| "[#{b}]" }
    end
    bullets = bullets.join("\n")
    bullets << "\n" unless stripped_labels.empty?
    @prefix = ' ' * @indent
    @indent += 2
    @prefix << bullets + (' ' * @indent)
  else
    bullet = type == :BULLET ? '*' :  @list_index.last.to_s + '.'
    @prefix = (' ' * @indent) + bullet.ljust(bullet.length + 1)
    width = bullet.length + 1
    @indent += width
  end
end

def accept_list_start list

def accept_list_start list
  case list.type
  when :BULLET then
    @list_index << nil
    @list_width << 1
  when :LABEL, :NOTE then
    @list_index << nil
    @list_width << 2
  when :LALPHA then
    @list_index << 'a'
    @list_width << list.items.length.to_s.length
  when :NUMBER then
    @list_index << 1
    @list_width << list.items.length.to_s.length
  when :UALPHA then
    @list_index << 'A'
    @list_width << list.items.length.to_s.length
  else
    raise RDoc::Error, "invalid list type #{list.type}"
  end
  @list_type << list.type
end

def accept_paragraph paragraph

def accept_paragraph paragraph
  text = paragraph.text @hard_break
  wrap attributes text
end

def accept_raw raw

def accept_raw raw
  @res << raw.parts.join("\n")
end

def accept_rule rule

def accept_rule rule
  use_prefix or @res << ' ' * @indent
  @res << '-' * (@width - @indent)
  @res << "\n"
end

def accept_table header, body, aligns

def accept_table header, body, aligns
  widths = header.zip(*body).map do |cols|
    cols.map(&:size).max
  end
  aligns = aligns.map do |a|
    case a
    when nil, :center
      :center
    when :left
      :ljust
    when :right
      :rjust
    end
  end
  @res << header.zip(widths, aligns).map do |h, w, a|
    h.__send__(a, w)
  end.join("|").rstrip << "\n"
  @res << widths.map {|w| "-" * w }.join("|") << "\n"
  body.each do |row|
    @res << row.zip(widths, aligns).map do |t, w, a|
      t.__send__(a, w)
    end.join("|").rstrip << "\n"
  end
end

def accept_verbatim verbatim

def accept_verbatim verbatim
  indent = ' ' * (@indent + 2)
  verbatim.parts.each do |part|
    @res << indent unless part == "\n"
    @res << part
  end
  @res << "\n"
end

def attributes text

def attributes text
  flow = @am.flow text.dup
  convert_flow flow
end

def end_accepting

def end_accepting
  @res.join
end

def handle_regexp_HARD_BREAK target

def handle_regexp_HARD_BREAK target
  "\n"
end

def handle_regexp_SUPPRESSED_CROSSREF target

def handle_regexp_SUPPRESSED_CROSSREF target
  text = target.text
  text = text.sub('\\', '') unless in_tt?
  text
end

def init_tags

def init_tags
  add_tag :BOLD, "<b>", "</b>"
  add_tag :TT,   "<tt>", "</tt>"
  add_tag :EM,   "<em>", "</em>"
end

def initialize markup = nil

def initialize markup = nil
  super nil, markup
  @markup.add_regexp_handling(/\\\S/, :SUPPRESSED_CROSSREF)
  @width = 78
  init_tags
  @headings = {}
  @headings.default = []
  @headings[1] = ['= ',      '']
  @headings[2] = ['== ',     '']
  @headings[3] = ['=== ',    '']
  @headings[4] = ['==== ',   '']
  @headings[5] = ['===== ',  '']
  @headings[6] = ['====== ', '']
  @hard_break = "\n"
end

def start_accepting

def start_accepting
  @res = [""]
  @indent = 0
  @prefix = nil
  @list_index = []
  @list_type  = []
  @list_width = []
end

def use_prefix

def use_prefix
  prefix, @prefix = @prefix, nil
  @res << prefix if prefix
  prefix
end

def wrap text

def wrap text
  return unless text && !text.empty?
  text_len = @width - @indent
  text_len = 20 if text_len < 20
  next_prefix = ' ' * @indent
  prefix = @prefix || next_prefix
  @prefix = nil
  text.scan(/\G(?:([^ \n]{#{text_len}})(?=[^ \n])|(.{1,#{text_len}})(?:[ \n]|\z))/) do
    @res << prefix << ($1 || $2) << "\n"
    prefix = next_prefix
  end
end