mirror of
https://github.com/ruby/ruby.git
synced 2022-11-09 12:17:21 -05:00
b5de8fcd81
* test/ruby/test_unicode_escape.rb (TestUnicodeEscape#test_basic): echo command doesn't work properly against non-ascii character on Windows with chcp 437. Instead we use ruby. [ruby-core:47709] [Bug #7076] git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@37050 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
270 lines
13 KiB
Ruby
270 lines
13 KiB
Ruby
# -*- coding: utf-8 -*-
|
|
|
|
require 'test/unit'
|
|
require_relative 'envutil'
|
|
|
|
class TestUnicodeEscape < Test::Unit::TestCase
|
|
def test_basic
|
|
assert_equal('Matz - 松本行弘',
|
|
"Matz - \u677E\u672C\u884C\u5F18")
|
|
assert_equal('Matz - まつもと ゆきひろ',
|
|
"Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D")
|
|
assert_equal('Matz - まつもと ゆきひろ',
|
|
"Matz - \u{307E}\u{3064}\u{3082}\u{3068} \u{3086}\u{304D}\u{3072}\u{308D}")
|
|
assert_equal('Matz - まつもと ゆきひろ',
|
|
"Matz - \u{307E 3064 3082 3068 20 3086 304D 3072 308D}")
|
|
assert_equal("Aoyama Gakuin University - \xE9\x9D\x92\xE5\xB1\xB1\xE5\xAD\xA6\xE9\x99\xA2\xE5\xA4\xA7\xE5\xAD\xA6",
|
|
"Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66")
|
|
assert_equal('Aoyama Gakuin University - 青山学院大学',
|
|
"Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66")
|
|
assert_equal('青山学院大学', "\u9752\u5C71\u5B66\u9662\u5927\u5B66")
|
|
assert_equal("Martin D\xC3\xBCrst", "Martin D\u00FCrst")
|
|
assert_equal('Martin Dürst', "Martin D\u00FCrst")
|
|
assert_equal('ü', "\u00FC")
|
|
assert_equal("Martin D\xC3\xBCrst", "Martin D\u{FC}rst")
|
|
assert_equal('Martin Dürst', "Martin D\u{FC}rst")
|
|
assert_equal('ü', "\u{FC}")
|
|
assert_equal('ü', %Q|\u{FC}|)
|
|
assert_equal('ü', %W{\u{FC}}[0])
|
|
|
|
# \u escapes in here documents
|
|
assert_equal('Matz - まつもと ゆきひろ', <<EOS.chop)
|
|
Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D
|
|
EOS
|
|
|
|
assert_equal('Matz - まつもと ゆきひろ', <<"EOS".chop)
|
|
Matz - \u{307E 3064 3082 3068} \u{3086 304D 3072 308D}
|
|
EOS
|
|
assert_not_equal('Matz - まつもと ゆきひろ', <<'EOS'.chop)
|
|
Matz - \u{307E 3064 3082 3068} \u{3086 304D 3072 308D}
|
|
EOS
|
|
|
|
# single-quoted things don't expand \u
|
|
assert_not_equal('ü', '\u{FC}')
|
|
assert_not_equal('ü', %q|\u{FC}|)
|
|
assert_not_equal('ü', %w{\u{FC}}[0])
|
|
assert_equal('\u00fc', "\\" + "u00fc")
|
|
|
|
# \u in %x strings
|
|
assert_match(/^("?)A\1$/, `echo "\u0041"`) #"
|
|
assert_match(/^("?)A\1$/, %x{echo "\u0041"}) #"
|
|
assert_match(/^("?)ü\1$/, `#{EnvUtil.rubybin} -e "puts \\"\u{FC}\\""`.force_encoding("utf-8")) #"
|
|
|
|
# \u in quoted symbols
|
|
assert_equal(:A, :"\u0041")
|
|
assert_equal(:a, :"\u0061")
|
|
assert_equal(:ま, :ま)
|
|
assert_equal(:ü, :ü)
|
|
assert_equal(:"\u{41}", :"\u0041")
|
|
assert_equal(:ü, :"\u{fc}")
|
|
|
|
# the NUL character is allowed in symbols
|
|
bug = '[ruby-dev:41447]'
|
|
sym = "\0".to_sym
|
|
assert_nothing_raised(SyntaxError, bug) {assert_equal(sym, eval(%q(:"\u{0}")))}
|
|
assert_nothing_raised(SyntaxError, bug) {assert_equal(sym, eval(%q(:"\u0000")))}
|
|
assert_nothing_raised(SyntaxError, bug) {assert_equal("\u{fc}\0A".to_sym, eval(%q(:"\u{fc 0 0041}")))}
|
|
assert_nothing_raised(SyntaxError, bug) {assert_equal(sym, eval(%q(:"\x00")))}
|
|
assert_nothing_raised(SyntaxError, bug) {assert_equal(sym, eval(%q(:"\0")))}
|
|
end
|
|
|
|
def test_regexp
|
|
|
|
# Compare regexps to regexps
|
|
assert_not_equal(/Yukihiro Matsumoto - 松本行弘/,
|
|
/Yukihiro Matsumoto - \u677E\u672C\u884C\u5F18/)
|
|
assert_not_equal(/Yukihiro Matsumoto - 松本行弘/,
|
|
/Yukihiro Matsumoto - \u{677E 672C 884C 5F18}/)
|
|
assert_not_equal(/Matz - まつもと ゆきひろ/,
|
|
/Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D/)
|
|
assert_not_equal(/Aoyama Gakuin University - 青山学院大学/,
|
|
/Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66/)
|
|
assert_not_equal(/青山学院大学/, /\u9752\u5C71\u5B66\u9662\u5927\u5B66/)
|
|
assert_not_equal(/Martin Dürst/, /Martin D\u00FCrst/)
|
|
assert_not_equal(/ü/, /\u00FC/)
|
|
assert_not_equal(/Martin Dürst/, /Martin D\u{FC}rst/)
|
|
assert_not_equal(/ü/, /\u{FC}/)
|
|
assert_not_equal(/ü/, %r{\u{FC}})
|
|
assert_not_equal(/ü/i, %r{\u00FC}i)
|
|
|
|
assert_equal('Yukihiro Matsumoto - \u677E\u672C\u884C\u5F18',
|
|
/Yukihiro Matsumoto - \u677E\u672C\u884C\u5F18/.source)
|
|
assert_equal('Yukihiro Matsumoto - \u{677E 672C 884C 5F18}',
|
|
/Yukihiro Matsumoto - \u{677E 672C 884C 5F18}/.source)
|
|
assert_equal('Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D',
|
|
/Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D/.source)
|
|
assert_equal('Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66',
|
|
/Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66/.source)
|
|
assert_equal('\u9752\u5C71\u5B66\u9662\u5927\u5B66',
|
|
/\u9752\u5C71\u5B66\u9662\u5927\u5B66/.source)
|
|
assert_equal('Martin D\u00FCrst', /Martin D\u00FCrst/.source)
|
|
assert_equal('\u00FC', /\u00FC/.source)
|
|
assert_equal('Martin D\u{FC}rst', /Martin D\u{FC}rst/.source)
|
|
assert_equal('\u{FC}', /\u{FC}/.source)
|
|
assert_equal('\u{FC}', %r{\u{FC}}.source)
|
|
assert_equal('\u00FC', %r{\u00FC}i.source)
|
|
|
|
# match strings to regexps
|
|
assert_equal(0, "Yukihiro Matsumoto - 松本行弘" =~ /Yukihiro Matsumoto - \u677E\u672C\u884C\u5F18/)
|
|
assert_equal(0, "Yukihiro Matsumoto - \u677E\u672C\u884C\u5F18" =~ /Yukihiro Matsumoto - \u677E\u672C\u884C/)
|
|
assert_equal(0, "Yukihiro Matsumoto - 松本行弘" =~ /Yukihiro Matsumoto - \u{677E 672C 884C 5F18}/)
|
|
assert_equal(0, %Q{Yukihiro Matsumoto - \u{677E 672C 884C 5F18}} =~ /Yukihiro Matsumoto - \u{677E 672C 884C 5F18}/)
|
|
assert_equal(0, "Matz - まつもと ゆきひろ" =~ /Matz - \u307E\u3064\u3082\u3068 \u3086\u304D\u3072\u308D/)
|
|
assert_equal(0, "Aoyama Gakuin University - 青山学院大学" =~ /Aoyama Gakuin University - \u9752\u5C71\u5B66\u9662\u5927\u5B66/)
|
|
assert_equal(0, "青山学院大学" =~ /\u9752\u5C71\u5B66\u9662\u5927\u5B66/)
|
|
assert_equal(0, "Martin Dürst" =~ /Martin D\u00FCrst/)
|
|
assert_equal(0, "ü" =~ /\u00FC/)
|
|
assert_equal(0, "Martin Dürst" =~ /Martin D\u{FC}rst/)
|
|
assert_equal(0, "ü" =~ %r{\u{FC}})
|
|
assert_equal(0, "ü" =~ %r{\u00FC}i)
|
|
|
|
# Flip order of the two operands
|
|
assert_equal(0, /Martin D\u00FCrst/ =~ "Martin Dürst")
|
|
assert_equal(4, /\u00FC/ =~ "testü")
|
|
assert_equal(3, /Martin D\u{FC}rst/ =~ "fooMartin Dürstbar")
|
|
assert_equal(3, %r{\u{FC}} =~ "fooübar")
|
|
|
|
# Put \u in strings, literal character in regexp
|
|
assert_equal(0, "Martin D\u00FCrst" =~ /Martin Dürst/)
|
|
assert_equal(4, "test\u00FC" =~ /ü/)
|
|
assert_equal(3, "fooMartin D\u{FC}rstbar" =~ /Martin Dürst/)
|
|
assert_equal(3, %Q{foo\u{FC}bar} =~ %r<ü>)
|
|
|
|
assert_match(eval('/\u{2a}/'), "*")
|
|
assert_raise(SyntaxError) { eval('/\u{6666}/n') }
|
|
assert_raise(SyntaxError) { eval('/\u{6666}/e') }
|
|
assert_raise(SyntaxError) { eval('/\u{6666}/s') }
|
|
assert_nothing_raised { eval('/\u{6666}/u') }
|
|
end
|
|
|
|
def test_dynamic_regexp
|
|
assert_match(Regexp.new("Martin D\\u{FC}rst"), "Martin Dürst")
|
|
end
|
|
|
|
def test_syntax_variants
|
|
# all hex digits
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89AB\uCDEF")
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89AB\uCDEF")
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89ab\ucdef")
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89ab\ucdef")
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89aB\uCdEf")
|
|
assert_equal("\xC4\xA3\xE4\x95\xA7\xE8\xA6\xAB\xEC\xB7\xAF", "\u0123\u4567\u89aB\ucDEF")
|
|
end
|
|
|
|
def test_fulton
|
|
# examples from Hal Fulton's book (second edition), chapter 4
|
|
# precomposed e'pe'e
|
|
assert_equal('épée', "\u00E9\u0070\u00E9\u0065")
|
|
assert_equal('épée', "\u00E9p\u00E9e")
|
|
assert_equal("\xC3\xA9\x70\xC3\xA9\x65", "\u00E9\u0070\u00E9\u0065")
|
|
assert_equal("\xC3\xA9\x70\xC3\xA9\x65", "\u00E9p\u00E9e")
|
|
# decomposed e'pe'e
|
|
assert_equal('épée', "\u0065\u0301\u0070\u0065\u0301\u0065")
|
|
assert_equal('épée', "e\u0301pe\u0301e")
|
|
assert_equal("\x65\xCC\x81\x70\x65\xCC\x81\x65", "\u0065\u0301\u0070\u0065\u0301\u0065")
|
|
assert_equal("\x65\xCC\x81\x70\x65\xCC\x81\x65", "e\u0301pe\u0301e")
|
|
# combinations of NFC/D, NFKC/D
|
|
assert_equal('öffnen', "\u00F6\u0066\u0066\u006E\u0065\u006E")
|
|
assert_equal("\xC3\xB6ffnen", "\u00F6\u0066\u0066\u006E\u0065\u006E")
|
|
assert_equal('öffnen', "\u00F6ffnen")
|
|
assert_equal("\xC3\xB6ffnen", "\u00F6ffnen")
|
|
assert_equal('öffnen', "\u006F\u0308\u0066\u0066\u006E\u0065\u006E")
|
|
assert_equal("\x6F\xCC\x88ffnen", "\u006F\u0308\u0066\u0066\u006E\u0065\u006E")
|
|
assert_equal('öffnen', "o\u0308ffnen")
|
|
assert_equal("\x6F\xCC\x88ffnen", "o\u0308ffnen")
|
|
assert_equal('öffnen', "\u00F6\uFB00\u006E\u0065\u006E")
|
|
assert_equal("\xC3\xB6\xEF\xAC\x80nen", "\u00F6\uFB00\u006E\u0065\u006E")
|
|
assert_equal('öffnen', "\u00F6\uFB00nen")
|
|
assert_equal("\xC3\xB6\xEF\xAC\x80nen", "\u00F6\uFB00nen")
|
|
assert_equal('öffnen', "\u006F\u0308\uFB00\u006E\u0065\u006E")
|
|
assert_equal("\x6F\xCC\x88\xEF\xAC\x80nen", "\u006F\u0308\uFB00\u006E\u0065\u006E")
|
|
assert_equal('öffnen', "o\u0308\uFB00nen")
|
|
assert_equal("\x6F\xCC\x88\xEF\xAC\x80nen", "o\u0308\uFB00nen")
|
|
# German sharp s (sz)
|
|
assert_equal('Straße', "\u0053\u0074\u0072\u0061\u00DF\u0065")
|
|
assert_equal("\x53\x74\x72\x61\xC3\x9F\x65", "\u0053\u0074\u0072\u0061\u00DF\u0065")
|
|
assert_equal('Straße', "Stra\u00DFe")
|
|
assert_equal("\x53\x74\x72\x61\xC3\x9F\x65", "Stra\u00DFe")
|
|
assert_equal('Straße', "\u{53}\u{74}\u{72}\u{61}\u{DF}\u{65}")
|
|
assert_equal("\x53\x74\x72\x61\xC3\x9F\x65", "\u{53}\u{74}\u{72}\u{61}\u{DF}\u{65}")
|
|
assert_equal("\x53\x74\x72\x61\xC3\x9F\x65", "\u{53 74 72 61 DF 65}")
|
|
assert_equal('Straße', "Stra\u{DF}e")
|
|
assert_equal("\x53\x74\x72\x61\xC3\x9F\x65", "Stra\u{DF}e")
|
|
end
|
|
|
|
def test_edge_cases
|
|
# start and end of each outer plane
|
|
assert_equal("\xF4\x8F\xBF\xBF", "\u{10FFFF}")
|
|
assert_equal("\xF4\x80\x80\x80", "\u{100000}")
|
|
assert_equal("\xF3\xBF\xBF\xBF", "\u{FFFFF}")
|
|
assert_equal("\xF3\xB0\x80\x80", "\u{F0000}")
|
|
assert_equal("\xF3\xAF\xBF\xBF", "\u{EFFFF}")
|
|
assert_equal("\xF3\xA0\x80\x80", "\u{E0000}")
|
|
assert_equal("\xF3\x9F\xBF\xBF", "\u{DFFFF}")
|
|
assert_equal("\xF3\x90\x80\x80", "\u{D0000}")
|
|
assert_equal("\xF3\x8F\xBF\xBF", "\u{CFFFF}")
|
|
assert_equal("\xF3\x80\x80\x80", "\u{C0000}")
|
|
assert_equal("\xF2\xBF\xBF\xBF", "\u{BFFFF}")
|
|
assert_equal("\xF2\xB0\x80\x80", "\u{B0000}")
|
|
assert_equal("\xF2\xAF\xBF\xBF", "\u{AFFFF}")
|
|
assert_equal("\xF2\xA0\x80\x80", "\u{A0000}")
|
|
assert_equal("\xF2\x9F\xBF\xBF", "\u{9FFFF}")
|
|
assert_equal("\xF2\x90\x80\x80", "\u{90000}")
|
|
assert_equal("\xF2\x8F\xBF\xBF", "\u{8FFFF}")
|
|
assert_equal("\xF2\x80\x80\x80", "\u{80000}")
|
|
assert_equal("\xF1\xBF\xBF\xBF", "\u{7FFFF}")
|
|
assert_equal("\xF1\xB0\x80\x80", "\u{70000}")
|
|
assert_equal("\xF1\xAF\xBF\xBF", "\u{6FFFF}")
|
|
assert_equal("\xF1\xA0\x80\x80", "\u{60000}")
|
|
assert_equal("\xF1\x9F\xBF\xBF", "\u{5FFFF}")
|
|
assert_equal("\xF1\x90\x80\x80", "\u{50000}")
|
|
assert_equal("\xF1\x8F\xBF\xBF", "\u{4FFFF}")
|
|
assert_equal("\xF1\x80\x80\x80", "\u{40000}")
|
|
assert_equal("\xF0\xBF\xBF\xBF", "\u{3FFFF}")
|
|
assert_equal("\xF0\xB0\x80\x80", "\u{30000}")
|
|
assert_equal("\xF0\xAF\xBF\xBF", "\u{2FFFF}")
|
|
assert_equal("\xF0\xA0\x80\x80", "\u{20000}")
|
|
assert_equal("\xF0\x9F\xBF\xBF", "\u{1FFFF}")
|
|
assert_equal("\xF0\x90\x80\x80", "\u{10000}")
|
|
# BMP
|
|
assert_equal("\xEF\xBF\xBF", "\uFFFF")
|
|
assert_equal("\xEE\x80\x80", "\uE000")
|
|
assert_equal("\xED\x9F\xBF", "\uD7FF")
|
|
assert_equal("\xE0\xA0\x80", "\u0800")
|
|
assert_equal("\xDF\xBF", "\u07FF")
|
|
assert_equal("\xC2\x80", "\u0080")
|
|
assert_equal("\x7F", "\u007F")
|
|
assert_equal("\x00", "\u0000")
|
|
end
|
|
|
|
def test_chars
|
|
assert_equal(?\u0041, ?A)
|
|
assert_equal(?\u{79}, ?\x79)
|
|
assert_equal(?\u{0}, ?\000)
|
|
assert_equal(?\u0000, ?\000)
|
|
end
|
|
|
|
# Tests to make sure that disallowed cases fail
|
|
def test_fail
|
|
assert_raise(SyntaxError) { eval %q("\uabc") } # too short
|
|
assert_raise(SyntaxError) { eval %q("\uab") } # too short
|
|
assert_raise(SyntaxError) { eval %q("\ua") } # too short
|
|
assert_raise(SyntaxError) { eval %q("\u") } # too short
|
|
assert_raise(SyntaxError) { eval %q("\u{110000}") } # too high
|
|
assert_raise(SyntaxError) { eval %q("\u{abcdeff}") } # too long
|
|
assert_raise(SyntaxError) { eval %q("\ughij") } # bad hex digits
|
|
assert_raise(SyntaxError) { eval %q("\u{ghij}") } # bad hex digits
|
|
|
|
assert_raise(SyntaxError) { eval %q("\u{123 456 }")} # extra space
|
|
assert_raise(SyntaxError) { eval %q("\u{ 123 456}")} # extra space
|
|
assert_raise(SyntaxError) { eval %q("\u{123 456}")} # extra space
|
|
|
|
# The utf-8 encoding object currently does not object to codepoints
|
|
# in the surrogate blocks, so these do not raise an error.
|
|
# assert_raise(SyntaxError) { "\uD800" } # surrogate block
|
|
# assert_raise(SyntaxError) { "\uDCBA" } # surrogate block
|
|
# assert_raise(SyntaxError) { "\uDFFF" } # surrogate block
|
|
# assert_raise(SyntaxError) { "\uD847\uDD9A" } # surrogate pair
|
|
|
|
end
|
|
end
|