mirror of
https://github.com/ruby/ruby.git
synced 2022-11-09 12:17:21 -05:00
d6ada9f14b
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@16821 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
271 lines
14 KiB
Ruby
271 lines
14 KiB
Ruby
# -*- encoding: ASCII-8BIT -*- # make sure this runs in binary mode
|
|
# some of the comments are in UTF-8
|
|
|
|
require 'test/unit'
|
|
class TestTranscode < Test::Unit::TestCase
|
|
def setup_really_needed? # trick to create all the necessary encodings
|
|
all_encodings = [ 'ISO-8859-1', 'ISO-8859-2',
|
|
'ISO-8859-3', 'ISO-8859-4',
|
|
'ISO-8859-5', 'ISO-8859-6',
|
|
'ISO-8859-7', 'ISO-8859-8',
|
|
'ISO-8859-9', 'ISO-8859-10',
|
|
'ISO-8859-11', 'ISO-8859-13',
|
|
'ISO-8859-14', 'ISO-8859-15',
|
|
'UTF-16BE'
|
|
]
|
|
all_encodings.each do |enc|
|
|
'abc'.encode(enc, 'UTF-8')
|
|
end
|
|
end
|
|
|
|
def test_errors
|
|
assert_raise(ArgumentError) { 'abc'.encode }
|
|
assert_raise(ArgumentError) { 'abc'.encode! }
|
|
assert_raise(ArgumentError) { 'abc'.encode('foo', 'bar') }
|
|
assert_raise(ArgumentError) { 'abc'.encode!('foo', 'bar') }
|
|
assert_raise(ArgumentError) { 'abc'.force_encoding('utf-8').encode('foo') }
|
|
assert_raise(ArgumentError) { 'abc'.force_encoding('utf-8').encode!('foo') }
|
|
assert_raise(RuntimeError) { "\x80".encode('utf-8','ASCII-8BIT') }
|
|
assert_raise(RuntimeError) { "\x80".encode('utf-8','US-ASCII') }
|
|
assert_raise(RuntimeError) { "\xA5".encode('utf-8','iso-8859-3') }
|
|
end
|
|
|
|
def test_arguments
|
|
assert_equal('abc', 'abc'.force_encoding('utf-8').encode('iso-8859-1'))
|
|
# check that encoding is kept when no conversion is done
|
|
assert_equal('abc'.force_encoding('Shift_JIS'), 'abc'.force_encoding('Shift_JIS').encode('Shift_JIS'))
|
|
assert_equal('abc'.force_encoding('Shift_JIS'), 'abc'.force_encoding('Shift_JIS').encode!('Shift_JIS'))
|
|
# assert that encoding is correctly set
|
|
assert_equal("D\u00FCrst".encoding, "D\xFCrst".force_encoding('iso-8859-1').encode('utf-8').encoding)
|
|
# check that Encoding can be used as parameter
|
|
assert_equal("D\u00FCrst", "D\xFCrst".encode('utf-8', Encoding.find('ISO-8859-1')))
|
|
assert_equal("D\u00FCrst", "D\xFCrst".encode(Encoding.find('utf-8'), 'ISO-8859-1'))
|
|
assert_equal("D\u00FCrst", "D\xFCrst".encode(Encoding.find('utf-8'), Encoding.find('ISO-8859-1')))
|
|
end
|
|
|
|
def test_length
|
|
assert_equal("\u20AC"*20, ("\xA4"*20).encode('utf-8', 'iso-8859-15'))
|
|
assert_equal("\u20AC"*20, ("\xA4"*20).encode!('utf-8', 'iso-8859-15'))
|
|
assert_equal("\u20AC"*2000, ("\xA4"*2000).encode('utf-8', 'iso-8859-15'))
|
|
assert_equal("\u20AC"*2000, ("\xA4"*2000).encode!('utf-8', 'iso-8859-15'))
|
|
assert_equal("\u20AC"*200000, ("\xA4"*200000).encode('utf-8', 'iso-8859-15'))
|
|
assert_equal("\u20AC"*200000, ("\xA4"*200000).encode!('utf-8', 'iso-8859-15'))
|
|
end
|
|
|
|
def check_both_ways(utf8, raw, encoding)
|
|
assert_equal(utf8.force_encoding('utf-8'), raw.encode('utf-8', encoding))
|
|
assert_equal(raw.force_encoding(encoding), utf8.encode(encoding, 'utf-8'))
|
|
end
|
|
|
|
def test_encodings
|
|
check_both_ways("\u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D",
|
|
"\x82\xdc\x82\xc2\x82\xe0\x82\xc6 \x82\xe4\x82\xab\x82\xd0\x82\xeb", 'shift_jis') # まつもと ゆきひろ
|
|
check_both_ways("\u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D",
|
|
"\xa4\xde\xa4\xc4\xa4\xe2\xa4\xc8 \xa4\xe6\xa4\xad\xa4\xd2\xa4\xed", 'euc-jp')
|
|
check_both_ways("\u677E\u672C\u884C\u5F18", "\x8f\xbc\x96\x7b\x8d\x73\x8d\x4f", 'shift_jis') # 松本行弘
|
|
check_both_ways("\u677E\u672C\u884C\u5F18", "\xbe\xbe\xcb\xdc\xb9\xd4\xb9\xb0", 'euc-jp')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-1') # Dürst
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-2')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-3')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-4')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-9')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-10')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-13')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-14')
|
|
check_both_ways("D\u00FCrst", "D\xFCrst", 'iso-8859-15')
|
|
check_both_ways("r\u00E9sum\u00E9", "r\xE9sum\xE9", 'iso-8859-1') # résumé
|
|
check_both_ways("\u0065\u006C\u0151\u00ED\u0072\u00E1\u0073", "el\xF5\xEDr\xE1s", 'iso-8859-2') # előírás
|
|
check_both_ways("\u043F\u0435\u0440\u0435\u0432\u043E\u0434",
|
|
"\xDF\xD5\xE0\xD5\xD2\xDE\xD4", 'iso-8859-5') # перевод
|
|
check_both_ways("\u0643\u062A\u0628", "\xE3\xCA\xC8", 'iso-8859-6') # كتب
|
|
check_both_ways("\u65E5\u8A18", "\x93\xFA\x8BL", 'shift_jis') # 日記
|
|
check_both_ways("\u65E5\u8A18", "\xC6\xFC\xB5\xAD", 'euc-jp')
|
|
check_both_ways("\uC560\uC778\uAD6C\uD568\u0020\u6734\uC9C0\uC778",
|
|
"\xBE\xD6\xC0\xCE\xB1\xB8\xC7\xD4\x20\xDA\xD3\xC1\xF6\xC0\xCE", 'euc-kr') # 애인구함 朴지인
|
|
check_both_ways("\uC544\uD58F\uD58F\u0020\uB620\uBC29\uD6BD\uB2D8\u0020\uC0AC\uB791\uD716",
|
|
"\xBE\xC6\xC1\x64\xC1\x64\x20\x8C\x63\xB9\xE6\xC4\x4F\xB4\xD4\x20\xBB\xE7\xB6\xFB\xC5\x42", 'cp949') # 아햏햏 똠방횽님 사랑휖
|
|
end
|
|
|
|
def test_twostep
|
|
assert_equal("D\xFCrst".force_encoding('iso-8859-2'), "D\xFCrst".encode('iso-8859-2', 'iso-8859-1'))
|
|
end
|
|
|
|
def test_ascii_range
|
|
encodings = [
|
|
'US-ASCII', 'ASCII-8BIT',
|
|
'ISO-8859-1', 'ISO-8859-2', 'ISO-8859-3',
|
|
'ISO-8859-4', 'ISO-8859-5', 'ISO-8859-6',
|
|
'ISO-8859-7', 'ISO-8859-8', 'ISO-8859-9',
|
|
'ISO-8859-10', 'ISO-8859-11', 'ISO-8859-13',
|
|
'ISO-8859-14', 'ISO-8859-15',
|
|
'EUC-JP', 'SHIFT_JIS', 'EUC-KR'
|
|
]
|
|
all_ascii = (0..127).to_a.pack 'C*'
|
|
encodings.each do |enc|
|
|
test_start = all_ascii
|
|
assert_equal(test_start, test_start.encode('UTF-8',enc).encode(enc).force_encoding('ASCII-8BIT'))
|
|
end
|
|
end
|
|
|
|
def test_all_bytes
|
|
encodings_8859 = [
|
|
'ISO-8859-1', 'ISO-8859-2',
|
|
#'ISO-8859-3', # not all bytes used
|
|
'ISO-8859-4', 'ISO-8859-5',
|
|
#'ISO-8859-6', # not all bytes used
|
|
#'ISO-8859-7', # not all bytes used
|
|
#'ISO-8859-8', # not all bytes used
|
|
'ISO-8859-9', 'ISO-8859-10',
|
|
#'ISO-8859-11', # not all bytes used
|
|
#'ISO-8859-12', # not available
|
|
'ISO-8859-13','ISO-8859-14','ISO-8859-15',
|
|
#'ISO-8859-16', # not available
|
|
]
|
|
all_bytes = (0..255).to_a.pack 'C*'
|
|
encodings_8859.each do |enc|
|
|
test_start = all_bytes
|
|
assert_equal(test_start, test_start.encode('UTF-8',enc).encode(enc).force_encoding('ASCII-8BIT'))
|
|
end
|
|
end
|
|
|
|
def check_utf_16_both_ways(utf8, raw)
|
|
copy = raw.dup
|
|
0.step(copy.length-1, 2) { |i| copy[i+1], copy[i] = copy[i], copy[i+1] }
|
|
check_both_ways(utf8, raw, 'utf-16be')
|
|
check_both_ways(utf8, copy, 'utf-16le')
|
|
end
|
|
|
|
def test_utf_16
|
|
check_utf_16_both_ways("abc", "\x00a\x00b\x00c")
|
|
check_utf_16_both_ways("\u00E9", "\x00\xE9");
|
|
check_utf_16_both_ways("\u00E9\u0070\u00E9\u0065", "\x00\xE9\x00\x70\x00\xE9\x00\x65") # épée
|
|
check_utf_16_both_ways("\u677E\u672C\u884C\u5F18", "\x67\x7E\x67\x2C\x88\x4C\x5F\x18") # 松本行弘
|
|
check_utf_16_both_ways("\u9752\u5C71\u5B66\u9662\u5927\u5B66", "\x97\x52\x5C\x71\x5B\x66\x96\x62\x59\x27\x5B\x66") # 青山学院大学
|
|
check_utf_16_both_ways("Martin D\u00FCrst", "\x00M\x00a\x00r\x00t\x00i\x00n\x00 \x00D\x00\xFC\x00r\x00s\x00t") # Martin Dürst
|
|
# BMP
|
|
check_utf_16_both_ways("\u0000", "\x00\x00")
|
|
check_utf_16_both_ways("\u007F", "\x00\x7F")
|
|
check_utf_16_both_ways("\u0080", "\x00\x80")
|
|
check_utf_16_both_ways("\u0555", "\x05\x55")
|
|
check_utf_16_both_ways("\u04AA", "\x04\xAA")
|
|
check_utf_16_both_ways("\u0333", "\x03\x33")
|
|
check_utf_16_both_ways("\u04CC", "\x04\xCC")
|
|
check_utf_16_both_ways("\u00F0", "\x00\xF0")
|
|
check_utf_16_both_ways("\u070F", "\x07\x0F")
|
|
check_utf_16_both_ways("\u07FF", "\x07\xFF")
|
|
check_utf_16_both_ways("\u0800", "\x08\x00")
|
|
check_utf_16_both_ways("\uD7FF", "\xD7\xFF")
|
|
check_utf_16_both_ways("\uE000", "\xE0\x00")
|
|
check_utf_16_both_ways("\uFFFF", "\xFF\xFF")
|
|
check_utf_16_both_ways("\u5555", "\x55\x55")
|
|
check_utf_16_both_ways("\uAAAA", "\xAA\xAA")
|
|
check_utf_16_both_ways("\u3333", "\x33\x33")
|
|
check_utf_16_both_ways("\uCCCC", "\xCC\xCC")
|
|
check_utf_16_both_ways("\uF0F0", "\xF0\xF0")
|
|
check_utf_16_both_ways("\u0F0F", "\x0F\x0F")
|
|
check_utf_16_both_ways("\uFF00", "\xFF\x00")
|
|
check_utf_16_both_ways("\u00FF", "\x00\xFF")
|
|
# outer planes
|
|
check_utf_16_both_ways("\u{10000}", "\xD8\x00\xDC\x00")
|
|
check_utf_16_both_ways("\u{FFFFF}", "\xDB\xBF\xDF\xFF")
|
|
check_utf_16_both_ways("\u{100000}", "\xDB\xC0\xDC\x00")
|
|
check_utf_16_both_ways("\u{10FFFF}", "\xDB\xFF\xDF\xFF")
|
|
check_utf_16_both_ways("\u{105555}", "\xDB\xD5\xDD\x55")
|
|
check_utf_16_both_ways("\u{55555}", "\xD9\x15\xDD\x55")
|
|
check_utf_16_both_ways("\u{AAAAA}", "\xDA\x6A\xDE\xAA")
|
|
check_utf_16_both_ways("\u{33333}", "\xD8\x8C\xDF\x33")
|
|
check_utf_16_both_ways("\u{CCCCC}", "\xDA\xF3\xDC\xCC")
|
|
check_utf_16_both_ways("\u{8F0F0}", "\xD9\xFC\xDC\xF0")
|
|
check_utf_16_both_ways("\u{F0F0F}", "\xDB\x83\xDF\x0F")
|
|
check_utf_16_both_ways("\u{8FF00}", "\xD9\xFF\xDF\x00")
|
|
check_utf_16_both_ways("\u{F00FF}", "\xDB\x80\xDC\xFF")
|
|
end
|
|
|
|
def check_utf_32_both_ways(utf8, raw)
|
|
copy = raw.dup
|
|
0.step(copy.length-1, 4) do |i|
|
|
copy[i+3], copy[i+2], copy[i+1], copy[i] = copy[i], copy[i+1], copy[i+2], copy[i+3]
|
|
end
|
|
check_both_ways(utf8, raw, 'utf-32be')
|
|
#check_both_ways(utf8, copy, 'utf-32le')
|
|
end
|
|
|
|
def test_utf_32
|
|
check_utf_32_both_ways("abc", "\x00\x00\x00a\x00\x00\x00b\x00\x00\x00c")
|
|
check_utf_32_both_ways("\u00E9", "\x00\x00\x00\xE9");
|
|
check_utf_32_both_ways("\u00E9\u0070\u00E9\u0065",
|
|
"\x00\x00\x00\xE9\x00\x00\x00\x70\x00\x00\x00\xE9\x00\x00\x00\x65") # épée
|
|
check_utf_32_both_ways("\u677E\u672C\u884C\u5F18",
|
|
"\x00\x00\x67\x7E\x00\x00\x67\x2C\x00\x00\x88\x4C\x00\x00\x5F\x18") # 松本行弘
|
|
check_utf_32_both_ways("\u9752\u5C71\u5B66\u9662\u5927\u5B66",
|
|
"\x00\x00\x97\x52\x00\x00\x5C\x71\x00\x00\x5B\x66\x00\x00\x96\x62\x00\x00\x59\x27\x00\x00\x5B\x66") # 青山学院大学
|
|
check_utf_32_both_ways("Martin D\u00FCrst",
|
|
"\x00\x00\x00M\x00\x00\x00a\x00\x00\x00r\x00\x00\x00t\x00\x00\x00i\x00\x00\x00n\x00\x00\x00 \x00\x00\x00D\x00\x00\x00\xFC\x00\x00\x00r\x00\x00\x00s\x00\x00\x00t") # Martin Dürst
|
|
# BMP
|
|
check_utf_32_both_ways("\u0000", "\x00\x00\x00\x00")
|
|
check_utf_32_both_ways("\u007F", "\x00\x00\x00\x7F")
|
|
check_utf_32_both_ways("\u0080", "\x00\x00\x00\x80")
|
|
check_utf_32_both_ways("\u0555", "\x00\x00\x05\x55")
|
|
check_utf_32_both_ways("\u04AA", "\x00\x00\x04\xAA")
|
|
check_utf_32_both_ways("\u0333", "\x00\x00\x03\x33")
|
|
check_utf_32_both_ways("\u04CC", "\x00\x00\x04\xCC")
|
|
check_utf_32_both_ways("\u00F0", "\x00\x00\x00\xF0")
|
|
check_utf_32_both_ways("\u070F", "\x00\x00\x07\x0F")
|
|
check_utf_32_both_ways("\u07FF", "\x00\x00\x07\xFF")
|
|
check_utf_32_both_ways("\u0800", "\x00\x00\x08\x00")
|
|
check_utf_32_both_ways("\uD7FF", "\x00\x00\xD7\xFF")
|
|
check_utf_32_both_ways("\uE000", "\x00\x00\xE0\x00")
|
|
check_utf_32_both_ways("\uFFFF", "\x00\x00\xFF\xFF")
|
|
check_utf_32_both_ways("\u5555", "\x00\x00\x55\x55")
|
|
check_utf_32_both_ways("\uAAAA", "\x00\x00\xAA\xAA")
|
|
check_utf_32_both_ways("\u3333", "\x00\x00\x33\x33")
|
|
check_utf_32_both_ways("\uCCCC", "\x00\x00\xCC\xCC")
|
|
check_utf_32_both_ways("\uF0F0", "\x00\x00\xF0\xF0")
|
|
check_utf_32_both_ways("\u0F0F", "\x00\x00\x0F\x0F")
|
|
check_utf_32_both_ways("\uFF00", "\x00\x00\xFF\x00")
|
|
check_utf_32_both_ways("\u00FF", "\x00\x00\x00\xFF")
|
|
# outer planes
|
|
check_utf_32_both_ways("\u{10000}", "\x00\x01\x00\x00")
|
|
check_utf_32_both_ways("\u{FFFFF}", "\x00\x0F\xFF\xFF")
|
|
check_utf_32_both_ways("\u{100000}","\x00\x10\x00\x00")
|
|
check_utf_32_both_ways("\u{10FFFF}","\x00\x10\xFF\xFF")
|
|
check_utf_32_both_ways("\u{105555}","\x00\x10\x55\x55")
|
|
check_utf_32_both_ways("\u{55555}", "\x00\x05\x55\x55")
|
|
check_utf_32_both_ways("\u{AAAAA}", "\x00\x0A\xAA\xAA")
|
|
check_utf_32_both_ways("\u{33333}", "\x00\x03\x33\x33")
|
|
check_utf_32_both_ways("\u{CCCCC}", "\x00\x0C\xCC\xCC")
|
|
check_utf_32_both_ways("\u{8F0F0}", "\x00\x08\xF0\xF0")
|
|
check_utf_32_both_ways("\u{F0F0F}", "\x00\x0F\x0F\x0F")
|
|
check_utf_32_both_ways("\u{8FF00}", "\x00\x08\xFF\x00")
|
|
check_utf_32_both_ways("\u{F00FF}", "\x00\x0F\x00\xFF")
|
|
end
|
|
|
|
def test_invalid_ignore
|
|
# arguments only
|
|
assert_nothing_raised { 'abc'.encode('utf-8', invalid: :ignore) }
|
|
# check handling of UTF-8 ill-formed subsequences
|
|
assert_equal("\x00\x41\x00\x3E\x00\x42".force_encoding('UTF-16BE'),
|
|
"\x41\xC2\x3E\x42".encode('UTF-16BE', 'UTF-8', invalid: :ignore))
|
|
assert_equal("\x00\x41\x00\xF1\x00\x42".force_encoding('UTF-16BE'),
|
|
"\x41\xC2\xC3\xB1\x42".encode('UTF-16BE', 'UTF-8', invalid: :ignore))
|
|
assert_equal("\x00\x42".force_encoding('UTF-16BE'),
|
|
"\xF0\x80\x80\x42".encode('UTF-16BE', 'UTF-8', invalid: :ignore))
|
|
assert_equal(''.force_encoding('UTF-16BE'),
|
|
"\x82\xAB".encode('UTF-16BE', 'UTF-8', invalid: :ignore))
|
|
end
|
|
|
|
def test_iso_2022_jp
|
|
assert_raise(RuntimeError) { "\x1b(A".encode("utf-8", "iso-2022-jp") }
|
|
assert_raise(RuntimeError) { "\x1b$(A".encode("utf-8", "iso-2022-jp") }
|
|
assert_equal("\uff71\uff72\uff73\uff74\uff75",
|
|
"\x1b(I12345\x1b(B".force_encoding("iso-2022-jp").encode("utf-8"))
|
|
assert_equal("\u9299", "\x1b$(Dd!\x1b(B".encode("utf-8", "iso-2022-jp"))
|
|
assert_raise(RuntimeError) { "\x1b$C".encode("utf-8", "iso-2022-jp") }
|
|
assert_raise(RuntimeError) { "\x1e".encode("utf-8", "iso-2022-jp") }
|
|
assert_raise(RuntimeError) { "\x80".encode("utf-8", "iso-2022-jp") }
|
|
|
|
assert_equal("\x1b(I12345\x1b(B".force_encoding("iso-2022-jp"),
|
|
"\uff71\uff72\uff73\uff74\uff75".encode("iso-2022-jp"))
|
|
assert_equal("\x1b$(Dd!\x1b(B".force_encoding("iso-2022-jp"), "\u9299".encode("iso-2022-jp"))
|
|
end
|
|
end
|