require 'rdoc/parser/ruby'
require 'rdoc/known_classes'

##
# RDoc::Parser::C attempts to parse C extension files.  It looks for
# the standard patterns that you find in extensions: <tt>rb_define_class,
# rb_define_method</tt> and so on.  It tries to find the corresponding
# C source for the methods and extract comments, but if we fail
# we don't worry too much.
#
# The comments associated with a Ruby method are extracted from the C
# comment block associated with the routine that _implements_ that
# method, that is to say the method whose name is given in the
# <tt>rb_define_method</tt> call. For example, you might write:
#
#   /*
#    * Returns a new array that is a one-dimensional flattening of this
#    * array (recursively). That is, for every element that is an array,
#    * extract its elements into the new array.
#    *
#    *    s = [ 1, 2, 3 ]           #=> [1, 2, 3]
#    *    t = [ 4, 5, 6, [7, 8] ]   #=> [4, 5, 6, [7, 8]]
#    *    a = [ s, t, 9, 10 ]       #=> [[1, 2, 3], [4, 5, 6, [7, 8]], 9, 10]
#    *    a.flatten                 #=> [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
#    */
#    static VALUE
#    rb_ary_flatten(ary)
#        VALUE ary;
#    {
#        ary = rb_obj_dup(ary);
#        rb_ary_flatten_bang(ary);
#        return ary;
#    }
#
#    ...
#
#    void
#    Init_Array()
#    {
#      ...
#      rb_define_method(rb_cArray, "flatten", rb_ary_flatten, 0);
#
# Here RDoc will determine from the rb_define_method line that there's a
# method called "flatten" in class Array, and will look for the implementation
# in the method rb_ary_flatten. It will then use the comment from that
# method in the HTML output. This method must be in the same source file
# as the rb_define_method.
#
# The comment blocks may include special directives:
#
# [Document-class: +name+]
#   Documentation for the named class.
#
# [Document-module: +name+]
#   Documentation for the named module.
#
# [Document-const: +name+]
#   Documentation for the named +rb_define_const+.
#
# [Document-global: +name+]
#   Documentation for the named +rb_define_global_const+
#
# [Document-variable: +name+]
#   Documentation for the named +rb_define_variable+
#
# [Document-method: +method_name+]
#   Documentation for the named method.  Use this when the method name is
#   unambiguous.
#
# [Document-method: <tt>ClassName::method_name<tt>]
#   Documentation for a singleton method in the given class.  Use this when
#   the method name alone is ambiguous.
#
# [Document-method: <tt>ClassName#method_name<tt>]
#   Documentation for a instance method in the given class.  Use this when the
#   method name alone is ambiguous.
#
# [Document-attr: +name+]
#   Documentation for the named attribute.
#
# [call-seq:  <i>text up to an empty line</i>]
#   Because C source doesn't give descriptive names to Ruby-level parameters,
#   you need to document the calling sequence explicitly
#
# In addition, RDoc assumes by default that the C method implementing a
# Ruby function is in the same source file as the rb_define_method call.
# If this isn't the case, add the comment:
#
#   rb_define_method(....);  // in filename
#
# As an example, we might have an extension that defines multiple classes
# in its Init_xxx method. We could document them using
#
#   /*
#    * Document-class:  MyClass
#    *
#    * Encapsulate the writing and reading of the configuration
#    * file. ...
#    */
#
#   /*
#    * Document-method: read_value
#    *
#    * call-seq:
#    *   cfg.read_value(key)            -> value
#    *   cfg.read_value(key} { |key| }  -> value
#    *
#    * Return the value corresponding to +key+ from the configuration.
#    * In the second form, if the key isn't found, invoke the
#    * block and return its value.
#    */

class RDoc::Parser::C < RDoc::Parser

  parse_files_matching(/\.(?:([CcHh])\1?|c([+xp])\2|y)\z/)

  include RDoc::Text

  ##
  # C file the parser is parsing

  attr_accessor :content


  ##
  # Maps C variable names to names of ruby classes (andsingleton classes)

  attr_reader :known_classes

  ##
  # Maps C variable names to names of ruby singleton classes

  attr_reader :singleton_classes

  ##
  # Resets cross-file state.  Call when parsing different projects that need
  # separate documentation.

  def self.reset
    @@enclosure_classes = {}
    @@known_bodies = {}
  end

  reset

  ##
  # Prepare to parse a C file

  def initialize(top_level, file_name, content, options, stats)
    super

    @known_classes = RDoc::KNOWN_CLASSES.dup
    @content = handle_tab_width handle_ifdefs_in(@content)
    @classes = {}
    @singleton_classes = {}
    @file_dir = File.dirname(@file_name)
  end

  ##
  # Scans #content for rb_define_alias

  def do_aliases
    @content.scan(/rb_define_alias\s*\(
                   \s*(\w+),
                   \s*"(.+?)",
                   \s*"(.+?)"
                   \s*\)/xm) do |var_name, new_name, old_name|
      class_name = @known_classes[var_name]

      unless class_name then
        warn "Enclosing class/module %p for alias %s %s not known" % [
          var_name, new_name, old_name]
        next
      end

      class_obj = find_class var_name, class_name

      al = RDoc::Alias.new '', old_name, new_name, ''
      al.singleton = @singleton_classes.key? var_name

      comment = find_alias_comment var_name, new_name, old_name
      comment = strip_stars comment
      al.comment = comment

      al.record_location @top_level

      class_obj.add_alias al
      @stats.add_alias al
    end
  end

  ##
  # Scans #content for rb_attr and rb_define_attr

  def do_attrs
    @content.scan(/rb_attr\s*\(
                   \s*(\w+),
                   \s*([\w"()]+),
                   \s*([01]),
                   \s*([01]),
                   \s*\w+\);/xm) do |var_name, attr_name, read, write|
      handle_attr var_name, attr_name, read, write
    end

    @content.scan(%r%rb_define_attr\(
                             \s*([\w\.]+),
                             \s*"([^"]+)",
                             \s*(\d+),
                             \s*(\d+)\s*\);
                %xm) do |var_name, attr_name, read, write|
      handle_attr var_name, attr_name, read, write
    end
  end

  ##
  # Scans #content for rb_define_module, rb_define_class, boot_defclass,
  # rb_define_module_under, rb_define_class_under and rb_singleton_class

  def do_classes
    @content.scan(/(\w+)\s* = \s*rb_define_module\s*\(\s*"(\w+)"\s*\)/mx) do
      |var_name, class_name|
      handle_class_module(var_name, "module", class_name, nil, nil)
    end

    # The '.' lets us handle SWIG-generated files
    @content.scan(/([\w\.]+)\s* = \s*rb_define_class\s*
              \(
                 \s*"(\w+)",
                 \s*(\w+)\s*
              \)/mx) do |var_name, class_name, parent|
      handle_class_module(var_name, "class", class_name, parent, nil)
    end

    @content.scan(/(\w+)\s*=\s*boot_defclass\s*\(\s*"(\w+?)",\s*(\w+?)\s*\)/) do
      |var_name, class_name, parent|
      parent = nil if parent == "0"
      handle_class_module(var_name, "class", class_name, parent, nil)
    end

    @content.scan(/(\w+)\s* = \s*rb_define_module_under\s*
              \(
                 \s*(\w+),
                 \s*"(\w+)"
              \s*\)/mx) do |var_name, in_module, class_name|
      handle_class_module(var_name, "module", class_name, nil, in_module)
    end

    @content.scan(/([\w\.]+)\s* =                  # var_name
                   \s*rb_define_class_under\s*
                   \(
                     \s* (\w+),                    # under
                     \s* "(\w+)",                  # class_name
                     \s*
                     (?:
                       ([\w\*\s\(\)\.\->]+) |      # parent_name
                       rb_path2class\("([\w:]+)"\) # path
                     )
                     \s*
                   \)
                  /mx) do |var_name, under, class_name, parent_name, path|
      parent = path || parent_name

      handle_class_module var_name, 'class', class_name, parent, under
    end

    @content.scan(/([\w\.]+)\s* = \s*rb_singleton_class\s*
                  \(
                    \s*(\w+)
                  \s*\)/mx) do |sclass_var, class_var|
      handle_singleton sclass_var, class_var
    end
  end

  ##
  # Scans #content for rb_define_variable, rb_define_readonly_variable,
  # rb_define_const and rb_define_global_const

  def do_constants
    @content.scan(%r%\Wrb_define_
                   ( variable          |
                     readonly_variable |
                     const             |
                     global_const        )
               \s*\(
                 (?:\s*(\w+),)?
                 \s*"(\w+)",
                 \s*(.*?)\s*\)\s*;
                 %xm) do |type, var_name, const_name, definition|
      var_name = "rb_cObject" if !var_name or var_name == "rb_mKernel"
      handle_constants type, var_name, const_name, definition
    end

    @content.scan(%r%
                  \Wrb_curses_define_const
                  \s*\(
                    \s*
                    (\w+)
                    \s*
                  \)
                  \s*;%xm) do |consts|
      const = consts.first
      handle_constants 'const', 'mCurses', const, "UINT2NUM(#{const})"
    end
  end

  ##
  # Scans #content for rb_include_module

  def do_includes
    @content.scan(/rb_include_module\s*\(\s*(\w+?),\s*(\w+?)\s*\)/) do |c,m|
      if cls = @classes[c]
        m = @known_classes[m] || m
        incl = cls.add_include RDoc::Include.new(m, "")
        incl.record_location @top_level
      end
    end
  end

  ##
  # Scans #content for rb_define_method, rb_define_singleton_method,
  # rb_define_module_function, rb_define_private_method,
  # rb_define_global_function and define_filetest_function

  def do_methods
    @content.scan(%r%rb_define_
                   (
                      singleton_method |
                      method           |
                      module_function  |
                      private_method
                   )
                   \s*\(\s*([\w\.]+),
                     \s*"([^"]+)",
                     \s*(?:RUBY_METHOD_FUNC\(|VALUEFUNC\()?(\w+)\)?,
                     \s*(-?\w+)\s*\)
                   (?:;\s*/[*/]\s+in\s+(\w+?\.(?:cpp|c|y)))?
                 %xm) do |type, var_name, meth_name, function, param_count, source_file|

      # Ignore top-object and weird struct.c dynamic stuff
      next if var_name == "ruby_top_self"
      next if var_name == "nstr"
      next if var_name == "envtbl"
      next if var_name == "argf"   # it'd be nice to handle this one

      var_name = "rb_cObject" if var_name == "rb_mKernel"
      handle_method(type, var_name, meth_name, function, param_count,
                    source_file)
    end

    @content.scan(%r%rb_define_global_function\s*\(
                             \s*"([^"]+)",
                             \s*(?:RUBY_METHOD_FUNC\(|VALUEFUNC\()?(\w+)\)?,
                             \s*(-?\w+)\s*\)
                (?:;\s*/[*/]\s+in\s+(\w+?\.[cy]))?
                %xm) do |meth_name, function, param_count, source_file|
      handle_method("method", "rb_mKernel", meth_name, function, param_count,
                    source_file)
    end

    @content.scan(/define_filetest_function\s*\(
                     \s*"([^"]+)",
                     \s*(?:RUBY_METHOD_FUNC\(|VALUEFUNC\()?(\w+)\)?,
                     \s*(-?\w+)\s*\)/xm) do |meth_name, function, param_count|

      handle_method("method", "rb_mFileTest", meth_name, function, param_count)
      handle_method("singleton_method", "rb_cFile", meth_name, function,
                    param_count)
    end
  end

  ##
  # Finds the comment for an alias on +class_name+ from +new_name+ to
  # +old_name+

  def find_alias_comment class_name, new_name, old_name
    content =~ %r%((?>/\*.*?\*/\s+))
                  rb_define_alias\(\s*#{Regexp.escape class_name}\s*,
                                   \s*"#{Regexp.escape new_name}"\s*,
                                   \s*"#{Regexp.escape old_name}"\s*\);%xm

    $1 || ''
  end

  ##
  # Finds a comment for rb_define_attr, rb_attr or Document-attr.
  #
  # +var_name+ is the C class variable the attribute is defined on.
  # +attr_name+ is the attribute's name.
  #
  # +read+ and +write+ are the read/write flags ('1' or '0').  Either both or
  # neither must be provided.

  def find_attr_comment var_name, attr_name, read = nil, write = nil
    attr_name = Regexp.escape attr_name

    rw = if read and write then
           /\s*#{read}\s*,\s*#{write}\s*/xm
         else
           /.*?/m
         end

    if @content =~ %r%((?>/\*.*?\*/\s+))
                      rb_define_attr\((?:\s*#{var_name},)?\s*
                                      "#{attr_name}"\s*,
                                      #{rw}\)\s*;%xm then
      $1
    elsif @content =~ %r%((?>/\*.*?\*/\s+))
                         rb_attr\(\s*#{var_name}\s*,
                                  \s*#{attr_name}\s*,
                                  #{rw},.*?\)\s*;%xm then
      $1
    elsif @content =~ %r%Document-attr:\s#{attr_name}\s*?\n
                         ((?>.*?\*/))%xm then
      $1
    else
      ''
    end
  end

  ##
  # Find the C code corresponding to a Ruby method

  def find_body class_name, meth_name, meth_obj, file_content, quiet = false
    case file_content
    when %r%((?>/\*.*?\*/\s*)?)
            ((?:(?:static|SWIGINTERN)\s+)?
             (?:intern\s+)?VALUE\s+#{meth_name}
             \s*(\([^)]*\))([^;]|$))%xm then
      comment = $1
      body = $2
      offset = $~.offset(2).first

      remove_private_comments comment if comment

      # try to find the whole body
      body = $& if /#{Regexp.escape body}[^(]*?\{.*?^\}/m =~ file_content

      # The comment block may have been overridden with a 'Document-method'
      # block. This happens in the interpreter when multiple methods are
      # vectored through to the same C method but those methods are logically
      # distinct (for example Kernel.hash and Kernel.object_id share the same
      # implementation

      override_comment = find_override_comment class_name, meth_obj
      comment = override_comment if override_comment

      find_modifiers comment, meth_obj if comment

      #meth_obj.params = params
      meth_obj.start_collecting_tokens
      tk = RDoc::RubyToken::Token.new nil, 1, 1
      tk.set_text body
      meth_obj.add_token tk
      meth_obj.comment = strip_stars comment
      meth_obj.offset  = offset
      meth_obj.line    = file_content[0, offset].count("\n") + 1

      body
    when %r%((?>/\*.*?\*/\s*))^\s*(\#\s*define\s+#{meth_name}\s+(\w+))%m then
      comment = $1
      body = $2
      offset = $~.offset(2).first

      find_body class_name, $3, meth_obj, file_content, true
      find_modifiers comment, meth_obj

      meth_obj.start_collecting_tokens
      tk = RDoc::RubyToken::Token.new nil, 1, 1
      tk.set_text body
      meth_obj.add_token tk
      meth_obj.comment = strip_stars(comment) + meth_obj.comment.to_s
      meth_obj.offset  = offset
      meth_obj.line    = file_content[0, offset].count("\n") + 1

      body
    when %r%^\s*\#\s*define\s+#{meth_name}\s+(\w+)%m then
      # with no comment we hope the aliased definition has it and use it's
      # definition

      body = find_body(class_name, $1, meth_obj, file_content, true)

      return body if body

      warn "No definition for #{meth_name}" if @options.verbosity > 1
      false
    else # No body, but might still have an override comment
      comment = find_override_comment class_name, meth_obj

      if comment then
        find_modifiers comment, meth_obj
        meth_obj.comment = strip_stars comment

        ''
      else
        warn "No definition for #{meth_name}" if @options.verbosity > 1
        false
      end
    end
  end

  ##
  # Finds a RDoc::NormalClass or RDoc::NormalModule for +raw_name+

  def find_class(raw_name, name)
    unless @classes[raw_name]
      if raw_name =~ /^rb_m/
        container = @top_level.add_module RDoc::NormalModule, name
      else
        container = @top_level.add_class RDoc::NormalClass, name
      end

      container.record_location @top_level
      @classes[raw_name] = container
    end
    @classes[raw_name]
  end

  ##
  # Look for class or module documentation above Init_+class_name+(void),
  # in a Document-class +class_name+ (or module) comment or above an
  # rb_define_class (or module).  If a comment is supplied above a matching
  # Init_ and a rb_define_class the Init_ comment is used.
  #
  #   /*
  #    * This is a comment for Foo
  #    */
  #   Init_Foo(void) {
  #       VALUE cFoo = rb_define_class("Foo", rb_cObject);
  #   }
  #
  #   /*
  #    * Document-class: Foo
  #    * This is a comment for Foo
  #    */
  #   Init_foo(void) {
  #       VALUE cFoo = rb_define_class("Foo", rb_cObject);
  #   }
  #
  #   /*
  #    * This is a comment for Foo
  #    */
  #   VALUE cFoo = rb_define_class("Foo", rb_cObject);

  def find_class_comment(class_name, class_mod)
    comment = nil

    if @content =~ %r%
        ((?>/\*.*?\*/\s+))
        (static\s+)?
        void\s+
        Init_#{class_name}\s*(?:_\(\s*)?\(\s*(?:void\s*)?\)%xmi then
      comment = $1.sub(%r%Document-(?:class|module):\s+#{class_name}%, '')
    elsif @content =~ %r%Document-(?:class|module):\s+#{class_name}\s*?
                         (?:<\s+[:,\w]+)?\n((?>.*?\*/))%xm then
      comment = $1
    elsif @content =~ %r%((?>/\*.*?\*/\s+))
                         ([\w\.\s]+\s* = \s+)?rb_define_(class|module).*?"(#{class_name})"%xm then
      comment = $1
    end

    return unless comment

    comment = strip_stars comment

    comment = look_for_directives_in class_mod, comment

    class_mod.add_comment comment, @top_level
  end

  ##
  # Finds a comment matching +type+ and +const_name+ either above the
  # comment or in the matching Document- section.

  def find_const_comment(type, const_name, class_name = nil)
    if @content =~ %r%((?>^\s*/\*.*?\*/\s+))
                   rb_define_#{type}\((?:\s*(\w+),)?\s*
                                      "#{const_name}"\s*,
                                      .*?\)\s*;%xmi then
      $1
    elsif class_name and
          @content =~ %r%Document-(?:const|global|variable):\s
                         #{class_name}::#{const_name}
                         \s*?\n((?>.*?\*/))%xm then
      $1
    elsif @content =~ %r%Document-(?:const|global|variable):\s#{const_name}
                         \s*?\n((?>.*?\*/))%xm then
      $1
    else
      ''
    end
  end

  ##
  # Handles modifiers in +comment+ and updates +meth_obj+ as appropriate.
  #
  # If <tt>:nodoc:</tt> is found, documentation on +meth_obj+ is suppressed.
  #
  # If <tt>:yields:</tt> is followed by an argument list it is used for the
  # #block_params of +meth_obj+.
  #
  # If the comment block contains a <tt>call-seq:</tt> section like:
  #
  #   call-seq:
  #      ARGF.readlines(sep=$/)     -> array
  #      ARGF.readlines(limit)      -> array
  #      ARGF.readlines(sep, limit) -> array
  #
  #      ARGF.to_a(sep=$/)     -> array
  #      ARGF.to_a(limit)      -> array
  #      ARGF.to_a(sep, limit) -> array
  #
  # it is used for the parameters of +meth_obj+.

  def find_modifiers comment, meth_obj
    # we must handle situations like the above followed by an unindented first
    # comment.  The difficulty is to make sure not to match lines starting
    # with ARGF at the same indent, but that are after the first description
    # paragraph.

    if comment =~ /call-seq:(.*?(?:\S|\*\/?).*?)^\s*(?:\*\/?)?\s*$/m then
      all_start, all_stop = $~.offset(0)
      seq_start, seq_stop = $~.offset(1)

      # we get the following lines that start with the leading word at the
      # same indent, even if they have blank lines before
      if $1 =~ /(^\s*\*?\s*\n)+^(\s*\*?\s*\w+)/m then
        leading = $2 # ' *    ARGF' in the example above
        re = %r%
          \A(
             (^\s*\*?\s*\n)+
             (^#{Regexp.escape leading}.*?\n)+
            )+
          ^\s*\*?\s*$
        %xm
        if comment[seq_stop..-1] =~ re then
          all_stop = seq_stop + $~.offset(0).last
          seq_stop = seq_stop + $~.offset(1).last
        end
      end

      seq = comment[seq_start..seq_stop]
      seq.gsub!(/^(\s*\*?\s*?)(\S|\n)/m, '\2')
      comment.slice! all_start...all_stop
      meth_obj.call_seq = seq
    elsif comment.sub!(/\A\/\*\s*call-seq:(.*?)\*\/\Z/, '') then
      meth_obj.call_seq = $1.strip
    end

    if comment.sub!(/\s*:(nodoc|doc|yields?|args?):\s*(.*)/, '') then
      RDoc::Parser.process_directive meth_obj, $1, $2
    end
  end

  ##
  # Finds a <tt>Document-method</tt> override for +meth_obj+ on +class_name+

  def find_override_comment class_name, meth_obj
    name = Regexp.escape meth_obj.name
    prefix = Regexp.escape meth_obj.name_prefix

    if @content =~ %r%Document-method:\s+#{class_name}#{prefix}#{name}\s*?\n((?>.*?\*/))%m then
      $1
    elsif @content =~ %r%Document-method:\s#{name}\s*?\n((?>.*?\*/))%m then
      $1
    end
  end

  ##
  # Creates a new RDoc::Attr +attr_name+ on class +var_name+ that is either
  # +read+, +write+ or both

  def handle_attr(var_name, attr_name, read, write)
    rw = ''
    rw << 'R' if '1' == read
    rw << 'W' if '1' == write

    class_name = @known_classes[var_name]

    return unless class_name

    class_obj = find_class var_name, class_name

    return unless class_obj

    comment = find_attr_comment var_name, attr_name
    comment = strip_stars comment

    name = attr_name.gsub(/rb_intern\("([^"]+)"\)/, '\1')

    attr = RDoc::Attr.new '', name, rw, comment

    attr.record_location @top_level
    class_obj.add_attribute attr
    @stats.add_attribute attr
  end

  ##
  # Creates a new RDoc::NormalClass or RDoc::NormalModule based on +type+
  # named +class_name+ in +parent+ which was assigned to the C +var_name+.

  def handle_class_module(var_name, type, class_name, parent, in_module)
    parent_name = @known_classes[parent] || parent

    if in_module then
      enclosure = @classes[in_module] || @@enclosure_classes[in_module]

      if enclosure.nil? and enclosure = @known_classes[in_module] then
        enc_type = /^rb_m/ =~ in_module ? "module" : "class"
        handle_class_module in_module, enc_type, enclosure, nil, nil
        enclosure = @classes[in_module]
      end

      unless enclosure then
        warn "Enclosing class/module '#{in_module}' for #{type} #{class_name} not known"
        return
      end
    else
      enclosure = @top_level
    end

    if type == "class" then
      full_name = if RDoc::ClassModule === enclosure then
                    enclosure.full_name + "::#{class_name}"
                  else
                    class_name
                  end

      if @content =~ %r%Document-class:\s+#{full_name}\s*<\s+([:,\w]+)% then
        parent_name = $1
      end

      cm = enclosure.add_class RDoc::NormalClass, class_name, parent_name
    else
      cm = enclosure.add_module RDoc::NormalModule, class_name
    end

    cm.record_location enclosure.top_level

    find_class_comment cm.full_name, cm

    case cm
    when RDoc::NormalClass
      @stats.add_class cm
    when RDoc::NormalModule
      @stats.add_module cm
    end

    @classes[var_name] = cm
    @@enclosure_classes[var_name] = cm
    @known_classes[var_name] = cm.full_name
  end

  ##
  # Adds constants.  By providing some_value: at the start of the comment you
  # can override the C value of the comment to give a friendly definition.
  #
  #   /* 300: The perfect score in bowling */
  #   rb_define_const(cFoo, "PERFECT", INT2FIX(300);
  #
  # Will override <tt>INT2FIX(300)</tt> with the value +300+ in the output
  # RDoc.  Values may include quotes and escaped colons (\:).

  def handle_constants(type, var_name, const_name, definition)
    class_name = @known_classes[var_name]

    return unless class_name

    class_obj = find_class var_name, class_name

    unless class_obj then
      warn "Enclosing class/module #{const_name.inspect} not known"
      return
    end

    comment = find_const_comment type, const_name, class_name
    comment = strip_stars comment
    comment = normalize_comment comment

    # In the case of rb_define_const, the definition and comment are in
    # "/* definition: comment */" form.  The literal ':' and '\' characters
    # can be escaped with a backslash.
    if type.downcase == 'const' then
      elements = comment.split ':'

      if elements.nil? or elements.empty? then
        con = RDoc::Constant.new const_name, definition, comment
      else
        new_definition = elements[0..-2].join(':')

        if new_definition.empty? then # Default to literal C definition
          new_definition = definition
        else
          new_definition.gsub!("\:", ":")
          new_definition.gsub!("\\", '\\')
        end

        new_definition.sub!(/\A(\s+)/, '')

        new_comment = if $1.nil? then
                        elements.last.lstrip
                      else
                        "#{$1}#{elements.last.lstrip}"
                      end

        con = RDoc::Constant.new const_name, new_definition, new_comment
      end
    else
      con = RDoc::Constant.new const_name, definition, comment
    end

    con.record_location @top_level
    @stats.add_constant con
    class_obj.add_constant con
  end

  ##
  # Removes #ifdefs that would otherwise confuse us

  def handle_ifdefs_in(body)
    body.gsub(/^#ifdef HAVE_PROTOTYPES.*?#else.*?\n(.*?)#endif.*?\n/m, '\1')
  end

  ##
  # Adds an RDoc::AnyMethod +meth_name+ defined on a class or module assigned
  # to +var_name+.  +type+ is the type of method definition function used.
  # +singleton_method+ and +module_function+ create a singleton method.

  def handle_method(type, var_name, meth_name, function, param_count,
                    source_file = nil)
    class_name = @known_classes[var_name]
    singleton  = @singleton_classes.key? var_name

    return unless class_name

    class_obj = find_class var_name, class_name

    if class_obj then
      if meth_name == 'initialize' then
        meth_name = 'new'
        singleton = true
        type = 'method' # force public
      end

      meth_obj = RDoc::AnyMethod.new '', meth_name
      meth_obj.c_function = function
      meth_obj.singleton =
        singleton || %w[singleton_method module_function].include?(type)

      p_count = Integer(param_count) rescue -1

      if source_file then
        file_name = File.join @file_dir, source_file

        if File.exist? file_name then
          file_content = (@@known_bodies[file_name] ||= File.read(file_name))
        else
          warn "unknown source #{source_file} for #{meth_name} in #{@file_name}"
        end
      else
        file_content = @content
      end

      body = find_body class_name, function, meth_obj, file_content

      if body and meth_obj.document_self then
        meth_obj.params = if p_count < -1 then # -2 is Array
                            '(*args)'
                          elsif p_count == -1 then # argc, argv
                            rb_scan_args body
                          else
                            "(#{(1..p_count).map { |i| "p#{i}" }.join ', '})"
                          end


        meth_obj.record_location @top_level
        class_obj.add_method meth_obj
        @stats.add_method meth_obj
        meth_obj.visibility = :private if 'private_method' == type
      end
    end
  end

  ##
  # Registers a singleton class +sclass_var+ as a singleton of +class_var+

  def handle_singleton sclass_var, class_var
    class_name = @known_classes[class_var]

    @known_classes[sclass_var]     = class_name
    @singleton_classes[sclass_var] = class_name
  end

  ##
  # Normalizes tabs in +body+

  def handle_tab_width(body)
    if /\t/ =~ body
      tab_width = @options.tab_width
      body.split(/\n/).map do |line|
        1 while line.gsub!(/\t+/) do
          ' ' * (tab_width * $&.length - $`.length % tab_width)
        end && $~
        line
      end.join "\n"
    else
      body
    end
  end

  ##
  # Look for directives in a normal comment block:
  #
  #   /*
  #    * :title: My Awesome Project
  #    */
  #
  # This routine modifies its parameter

  def look_for_directives_in(context, comment)
    preprocess = RDoc::Markup::PreProcess.new @file_name, @options.rdoc_include

    preprocess.handle comment, context do |directive, param|
      case directive
      when 'main' then
        @options.main_page = param
        ''
      when 'title' then
        @options.default_title = param if @options.respond_to? :default_title=
        ''
      end
    end

    comment
  end

  ##
  # Extracts parameters from the +method_body+ and returns a method
  # parameter string.  Follows 1.9.3dev's scan-arg-spec, see README.EXT

  def rb_scan_args method_body
    method_body =~ /rb_scan_args\((.*?)\)/m
    return '(*args)' unless $1

    $1.split(/,/)[2] =~ /"(.*?)"/ # format argument
    format = $1.split(//)

    lead = opt = trail = 0

    if format.first =~ /\d/ then
      lead = $&.to_i
      format.shift
      if format.first =~ /\d/ then
        opt = $&.to_i
        format.shift
        if format.first =~ /\d/ then
          trail = $&.to_i
          format.shift
          block_arg = true
        end
      end
    end

    if format.first == '*' and not block_arg then
      var = true
      format.shift
      if format.first =~ /\d/ then
        trail = $&.to_i
        format.shift
      end
    end

    if format.first == ':' then
      hash = true
      format.shift
    end

    if format.first == '&' then
      block = true
      format.shift
    end

    # if the format string is not empty there's a bug in the C code, ignore it

    args = []
    position = 1

    (1...(position + lead)).each do |index|
      args << "p#{index}"
    end

    position += lead

    (position...(position + opt)).each do |index|
      args << "p#{index} = v#{index}"
    end

    position += opt

    if var then
      args << '*args'
      position += 1
    end

    (position...(position + trail)).each do |index|
      args << "p#{index}"
    end

    position += trail

    if hash then
      args << "p#{position} = {}"
      position += 1
    end

    args << '&block' if block

    "(#{args.join ', '})"
  end

  ##
  # Removes lines that are commented out that might otherwise get picked up
  # when scanning for classes and methods

  def remove_commented_out_lines
    @content.gsub!(%r%//.*rb_define_%, '//')
  end

  ##
  # Removes private comments from +comment+

  def remove_private_comments(comment)
    comment.gsub!(/\/?\*--\n(.*?)\/?\*\+\+/m, '')
    comment.sub!(/\/?\*--\n.*/m, '')
  end

  ##
  # Extracts the classes, modules, methods, attributes, constants and aliases
  # from a C file and returns an RDoc::TopLevel for this file

  def scan
    remove_commented_out_lines
    do_classes
    do_constants
    do_methods
    do_includes
    do_aliases
    do_attrs
    @top_level
  end

end