module ActiveSupport::Multibyte::Unicode

def self.codepoints_to_pattern(array_of_codepoints) #:nodoc:
  array_of_codepoints.collect{ |e| [e].pack 'U*'.freeze }.join('|'.freeze)
end

def apply_mapping(string, mapping) #:nodoc:
  database.codepoints
  string.each_codepoint.map do |codepoint|
    cp = database.codepoints[codepoint]
    if cp and (ncp = cp.send(mapping)) and ncp > 0
      ncp
    else
      codepoint
    end
  end.pack('U*')
end

def compose(codepoints)
  pos = 0
  eoa = codepoints.length - 1
  starter_pos = 0
  starter_char = codepoints[0]
  previous_combining_class = -1
  while pos < eoa
    pos += 1
    lindex = starter_char - HANGUL_LBASE
    # -- Hangul
    if 0 <= lindex and lindex < HANGUL_LCOUNT
      vindex = codepoints[starter_pos+1] - HANGUL_VBASE rescue vindex = -1
      if 0 <= vindex and vindex < HANGUL_VCOUNT
        tindex = codepoints[starter_pos+2] - HANGUL_TBASE rescue tindex = -1
        if 0 <= tindex and tindex < HANGUL_TCOUNT
          j = starter_pos + 2
          eoa -= 2
        else
          tindex = 0
          j = starter_pos + 1
          eoa -= 1
        end
        codepoints[starter_pos..j] = (lindex * HANGUL_VCOUNT + vindex) * HANGUL_TCOUNT + tindex + HANGUL_SBASE
      end
      starter_pos += 1
      starter_char = codepoints[starter_pos]
    # -- Other characters
    else
      current_char = codepoints[pos]
      current = database.codepoints[current_char]
      if current.combining_class > previous_combining_class
        if ref = database.composition_map[starter_char]
          composition = ref[current_char]
        else
          composition = nil
        end
        unless composition.nil?
          codepoints[starter_pos] = composition
          starter_char = composition
          codepoints.delete_at pos
          eoa -= 1
          pos -= 1
          previous_combining_class = -1
        else
          previous_combining_class = current.combining_class
        end
      else
        previous_combining_class = current.combining_class
      end
      if current.combining_class == 0
        starter_pos = pos
        starter_char = codepoints[pos]
      end
    end
  end
  codepoints
end

def database
  @database ||= UnicodeDatabase.new
end

def decompose(type, codepoints)
  codepoints.inject([]) do |decomposed, cp|
    # if it's a hangul syllable starter character
    if HANGUL_SBASE <= cp and cp < HANGUL_SLAST
      sindex = cp - HANGUL_SBASE
      ncp = [] # new codepoints
      ncp << HANGUL_LBASE + sindex / HANGUL_NCOUNT
      ncp << HANGUL_VBASE + (sindex % HANGUL_NCOUNT) / HANGUL_TCOUNT
      tindex = sindex % HANGUL_TCOUNT
      ncp << (HANGUL_TBASE + tindex) unless tindex == 0
      decomposed.concat ncp
    # if the codepoint is decomposable in with the current decomposition type
    elsif (ncp = database.codepoints[cp].decomp_mapping) and (!database.codepoints[cp].decomp_type || type == :compatibility)
      decomposed.concat decompose(type, ncp.dup)
    else
      decomposed << cp
    end
  end
end

def downcase(string)
  apply_mapping string, :lowercase_mapping
end

def in_char_class?(codepoint, classes)
  classes.detect { |c| database.boundary[c] === codepoint } ? true : false
end

def normalize(string, form=nil)
  form ||= @default_normalization_form
  # See http://www.unicode.org/reports/tr15, Table 1
  codepoints = string.codepoints.to_a
  case form
    when :d
      reorder_characters(decompose(:canonical, codepoints))
    when :c
      compose(reorder_characters(decompose(:canonical, codepoints)))
    when :kd
      reorder_characters(decompose(:compatibility, codepoints))
    when :kc
      compose(reorder_characters(decompose(:compatibility, codepoints)))
    else
      raise ArgumentError, "#{form} is not a valid normalization variant", caller
  end.pack('U*'.freeze)
end

def pack_graphemes(unpacked)
  unpacked.flatten.pack('U*')
end

def recode_windows1252_chars(string)
  string.encode(Encoding::UTF_8, Encoding::Windows_1252, invalid: :replace, undef: :replace)
end

def reorder_characters(codepoints)
  length = codepoints.length- 1
  pos = 0
  while pos < length do
    cp1, cp2 = database.codepoints[codepoints[pos]], database.codepoints[codepoints[pos+1]]
    if (cp1.combining_class > cp2.combining_class) && (cp2.combining_class > 0)
      codepoints[pos..pos+1] = cp2.code, cp1.code
      pos += (pos > 0 ? -1 : 1)
    else
      pos += 1
    end
  end
  codepoints
end

def swapcase(string)
  apply_mapping string, :swapcase_mapping
end

def tidy_bytes(string, force = false)
  return string if string.empty?
  return recode_windows1252_chars(string) if force
  string.scrub { |bad| recode_windows1252_chars(bad) }
end

def tidy_bytes(string, force = false)
  return string if string.empty?
  return recode_windows1252_chars(string) if force
  # We can't transcode to the same format, so we choose a nearly-identical encoding.
  # We're going to 'transcode' bytes from UTF-8 when possible, then fall back to
  # CP1252 when we get errors. The final string will be 'converted' back to UTF-8
  # before returning.
  reader = Encoding::Converter.new(Encoding::UTF_8, Encoding::UTF_16LE)
  source = string.dup
  out = ''.force_encoding(Encoding::UTF_16LE)
  loop do
    reader.primitive_convert(source, out)
    _, _, _, error_bytes, _ = reader.primitive_errinfo
    break if error_bytes.nil?
    out << error_bytes.encode(Encoding::UTF_16LE, Encoding::Windows_1252, invalid: :replace, undef: :replace)
  end
  reader.finish
  out.encode!(Encoding::UTF_8)
end

def unpack_graphemes(string)
  codepoints = string.codepoints.to_a
  unpacked = []
  pos = 0
  marker = 0
  eoc = codepoints.length
  while(pos < eoc)
    pos += 1
    previous = codepoints[pos-1]
    current = codepoints[pos]
    should_break =
      # GB3. CR X LF
      if previous == database.boundary[:cr] and current == database.boundary[:lf]
        false
      # GB4. (Control|CR|LF) ÷
      elsif previous and in_char_class?(previous, [:control,:cr,:lf])
        true
      # GB5. ÷ (Control|CR|LF)
      elsif in_char_class?(current, [:control,:cr,:lf])
        true
      # GB6. L X (L|V|LV|LVT)
      elsif database.boundary[:l] === previous and in_char_class?(current, [:l,:v,:lv,:lvt])
        false
      # GB7. (LV|V) X (V|T)
      elsif in_char_class?(previous, [:lv,:v]) and in_char_class?(current, [:v,:t])
        false
      # GB8. (LVT|T) X (T)
      elsif in_char_class?(previous, [:lvt,:t]) and database.boundary[:t] === current
        false
      # GB8a. Regional_Indicator X Regional_Indicator
      elsif database.boundary[:regional_indicator] === previous and database.boundary[:regional_indicator] === current
        false
      # GB9. X Extend
      elsif database.boundary[:extend] === current
        false
      # GB9a. X SpacingMark
      elsif database.boundary[:spacingmark] === current
        false
      # GB9b. Prepend X
      elsif database.boundary[:prepend] === previous
        false
      # GB10. Any ÷ Any
      else
        true
      end
    if should_break
      unpacked << codepoints[marker..pos-1]
      marker = pos
    end
  end
  unpacked
end

def upcase(string)
  apply_mapping string, :uppercase_mapping
end