#
# kconv.rb - Kanji Converter.
#
# $Id: kconv.rb 11708 2007-02-12 23:01:19Z shyouhei $
#
# ----
#
# kconv.rb implements the Kconv class for Kanji Converter. Additionally,
# some methods in String classes are added to allow easy conversion.
#
require 'nkf'
#
# Kanji Converter for Ruby.
#
module Kconv
#
# Public Constants
#
#Constant of Encoding
# Auto-Detect
AUTO = NKF::AUTO
# ISO-2022-JP
JIS = NKF::JIS
# EUC-JP
EUC = NKF::EUC
# Shift_JIS
SJIS = NKF::SJIS
# BINARY
BINARY = NKF::BINARY
# NOCONV
NOCONV = NKF::NOCONV
# ASCII
ASCII = NKF::ASCII
# UTF-8
UTF8 = NKF::UTF8
# UTF-16
UTF16 = NKF::UTF16
# UTF-32
UTF32 = NKF::UTF32
# UNKNOWN
UNKNOWN = NKF::UNKNOWN
#
# Private Constants
#
# Revision of kconv.rb
REVISION = %q$Revision: 11708 $
#Regexp of Encoding
# Regexp of Shift_JIS string (private constant)
RegexpShiftjis = /\A(?:
[\x00-\x7f\xa1-\xdf] |
[\x81-\x9f\xe0-\xfc][\x40-\x7e\x80-\xfc]
)*\z/nx
# Regexp of EUC-JP string (private constant)
RegexpEucjp = /\A(?:
[\x00-\x7f] |
\x8e [\xa1-\xdf] |
\x8f [\xa1-\xfe] [\xa1-\xfe] |
[\xa1-\xfe] [\xa1-\xfe]
)*\z/nx
# Regexp of UTF-8 string (private constant)
RegexpUtf8 = /\A(?:
[\x00-\x7f] |
[\xc2-\xdf] [\x80-\xbf] |
\xe0 [\xa0-\xbf] [\x80-\xbf] |
[\xe1-\xef] [\x80-\xbf] [\x80-\xbf] |
\xf0 [\x90-\xbf] [\x80-\xbf] [\x80-\xbf] |
[\xf1-\xf3] [\x80-\xbf] [\x80-\xbf] [\x80-\xbf] |
\xf4 [\x80-\x8f] [\x80-\xbf] [\x80-\xbf]
)*\z/nx
#
# Public Methods
#
# call-seq:
# Kconv.kconv(str, out_code, in_code = Kconv::AUTO)
#
# Convert str
to out_code.
# out_code
and in_code
are given as constants of Kconv.
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want to decode them, use NKF.nkf.
def kconv(str, out_code, in_code = AUTO)
opt = '-'
case in_code
when ::NKF::JIS
opt << 'J'
when ::NKF::EUC
opt << 'E'
when ::NKF::SJIS
opt << 'S'
when ::NKF::UTF8
opt << 'W'
when ::NKF::UTF16
opt << 'W16'
end
case out_code
when ::NKF::JIS
opt << 'j'
when ::NKF::EUC
opt << 'e'
when ::NKF::SJIS
opt << 's'
when ::NKF::UTF8
opt << 'w'
when ::NKF::UTF16
opt << 'w16'
when ::NKF::NOCONV
return str
end
opt = '' if opt == '-'
::NKF::nkf(opt, str)
end
module_function :kconv
#
# Encode to
#
# call-seq:
# Kconv.tojis(str) -> string
#
# Convert str
to ISO-2022-JP
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-jxm0', str).
def tojis(str)
::NKF::nkf('-jm', str)
end
module_function :tojis
# call-seq:
# Kconv.toeuc(str) -> string
#
# Convert str
to EUC-JP
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-exm0', str).
def toeuc(str)
::NKF::nkf('-em', str)
end
module_function :toeuc
# call-seq:
# Kconv.tosjis(str) -> string
#
# Convert str
to Shift_JIS
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-sxm0', str).
def tosjis(str)
::NKF::nkf('-sm', str)
end
module_function :tosjis
# call-seq:
# Kconv.toutf8(str) -> string
#
# Convert str
to UTF-8
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-wxm0', str).
def toutf8(str)
::NKF::nkf('-wm', str)
end
module_function :toutf8
# call-seq:
# Kconv.toutf16(str) -> string
#
# Convert str
to UTF-16
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-w16xm0', str).
def toutf16(str)
::NKF::nkf('-w16m', str)
end
module_function :toutf16
#
# guess
#
# call-seq:
# Kconv.guess(str) -> integer
#
# Guess input encoding by NKF.guess2
def guess(str)
::NKF::guess(str)
end
module_function :guess
# call-seq:
# Kconv.guess_old(str) -> integer
#
# Guess input encoding by NKF.guess1
def guess_old(str)
::NKF::guess1(str)
end
module_function :guess_old
#
# isEncoding
#
# call-seq:
# Kconv.iseuc(str) -> obj or nil
#
# Returns whether input encoding is EUC-JP or not.
#
# *Note* don't expect this return value is MatchData.
def iseuc(str)
RegexpEucjp.match( str )
end
module_function :iseuc
# call-seq:
# Kconv.issjis(str) -> obj or nil
#
# Returns whether input encoding is Shift_JIS or not.
#
# *Note* don't expect this return value is MatchData.
def issjis(str)
RegexpShiftjis.match( str )
end
module_function :issjis
# call-seq:
# Kconv.isutf8(str) -> obj or nil
#
# Returns whether input encoding is UTF-8 or not.
#
# *Note* don't expect this return value is MatchData.
def isutf8(str)
RegexpUtf8.match( str )
end
module_function :isutf8
end
class String
# call-seq:
# String#kconv(out_code, in_code = Kconv::AUTO)
#
# Convert self
to out_code.
# out_code
and in_code
are given as constants of Kconv.
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want to decode them, use NKF.nkf.
def kconv(out_code, in_code=Kconv::AUTO)
Kconv::kconv(self, out_code, in_code)
end
#
# to Encoding
#
# call-seq:
# String#tojis -> string
#
# Convert self
to ISO-2022-JP
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-jxm0', str).
def tojis; Kconv.tojis(self) end
# call-seq:
# String#toeuc -> string
#
# Convert self
to EUC-JP
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-exm0', str).
def toeuc; Kconv.toeuc(self) end
# call-seq:
# String#tosjis -> string
#
# Convert self
to Shift_JIS
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-sxm0', str).
def tosjis; Kconv.tosjis(self) end
# call-seq:
# String#toutf8 -> string
#
# Convert self
to UTF-8
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-wxm0', str).
def toutf8; Kconv.toutf8(self) end
# call-seq:
# String#toutf16 -> string
#
# Convert self
to UTF-16
#
# *Note*
# This method decode MIME encoded string and
# convert halfwidth katakana to fullwidth katakana.
# If you don't want it, use NKF.nkf('-w16xm0', str).
def toutf16; Kconv.toutf16(self) end
#
# is Encoding
#
# call-seq:
# String#iseuc -> obj or nil
#
# Returns whether self
's encoding is EUC-JP or not.
#
# *Note* don't expect this return value is MatchData.
def iseuc; Kconv.iseuc(self) end
# call-seq:
# String#issjis -> obj or nil
#
# Returns whether self
's encoding is Shift_JIS or not.
#
# *Note* don't expect this return value is MatchData.
def issjis; Kconv.issjis(self) end
# call-seq:
# String#isutf8 -> obj or nil
#
# Returns whether self
's encoding is UTF-8 or not.
#
# *Note* don't expect this return value is MatchData.
def isutf8; Kconv.isutf8(self) end
end