1
0
Fork 0
mirror of https://github.com/ruby/ruby.git synced 2022-11-09 12:17:21 -05:00
ruby--ruby/test/ruby/test_m17n_comb.rb
akr acd98555b5 move a non-combinational test.
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@29479 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
2010-10-13 03:36:20 +00:00

1627 lines
43 KiB
Ruby

require 'test/unit'
require 'stringio'
require_relative 'allpairs'
class TestM17NComb < Test::Unit::TestCase
def assert_encoding(encname, actual, message=nil)
assert_equal(Encoding.find(encname), actual, message)
end
module AESU
def a(str) str.dup.force_encoding("ASCII-8BIT") end
def e(str) str.dup.force_encoding("EUC-JP") end
def s(str) str.dup.force_encoding("Shift_JIS") end
def u(str) str.dup.force_encoding("UTF-8") end
end
include AESU
extend AESU
def assert_strenc(bytes, enc, actual, message=nil)
assert_instance_of(String, actual, message)
enc = Encoding.find(enc) if String === enc
assert_equal(enc, actual.encoding, message)
assert_equal(a(bytes), a(actual), message)
end
def assert_warning(pat, mesg=nil)
begin
org_stderr = $stderr
$stderr = StringIO.new(warn = '')
yield
ensure
$stderr = org_stderr
end
assert_match(pat, warn, mesg)
end
def assert_regexp_generic_encoding(r)
assert(!r.fixed_encoding?)
%w[ASCII-8BIT EUC-JP Shift_JIS UTF-8].each {|ename|
# "\xc2\xa1" is a valid sequence for ASCII-8BIT, EUC-JP, Shift_JIS and UTF-8.
assert_nothing_raised { r =~ "\xc2\xa1".force_encoding(ename) }
}
end
def assert_regexp_fixed_encoding(r)
assert(r.fixed_encoding?)
%w[ASCII-8BIT EUC-JP Shift_JIS UTF-8].each {|ename|
enc = Encoding.find(ename)
if enc == r.encoding
assert_nothing_raised { r =~ "\xc2\xa1".force_encoding(enc) }
else
assert_raise(ArgumentError) { r =~ "\xc2\xa1".force_encoding(enc) }
end
}
end
def assert_regexp_generic_ascii(r)
assert_encoding("ASCII-8BIT", r.encoding)
assert_regexp_generic_encoding(r)
end
def assert_regexp_fixed_ascii8bit(r)
assert_encoding("ASCII-8BIT", r.encoding)
assert_regexp_fixed_encoding(r)
end
def assert_regexp_fixed_eucjp(r)
assert_encoding("EUC-JP", r.encoding)
assert_regexp_fixed_encoding(r)
end
def assert_regexp_fixed_sjis(r)
assert_encoding("Shift_JIS", r.encoding)
assert_regexp_fixed_encoding(r)
end
def assert_regexp_fixed_utf8(r)
assert_encoding("UTF-8", r.encoding)
assert_regexp_fixed_encoding(r)
end
STRINGS = [
a(""), e(""), s(""), u(""),
a("a"), e("a"), s("a"), u("a"),
a("."), e("."), s("."), u("."),
# single character
a("\x80"), a("\xff"),
e("\xa1\xa1"), e("\xfe\xfe"),
e("\x8e\xa1"), e("\x8e\xfe"),
e("\x8f\xa1\xa1"), e("\x8f\xfe\xfe"),
s("\x81\x40"), s("\xfc\xfc"),
s("\xa1"), s("\xdf"),
u("\xc2\x80"), u("\xf4\x8f\xbf\xbf"),
# same byte sequence
a("\xc2\xa1"), e("\xc2\xa1"), s("\xc2\xa1"), u("\xc2\xa1"),
s("\x81A"), # mutibyte character which contains "A"
s("\x81a"), # mutibyte character which contains "a"
# invalid
e("\xa1"), e("\x80"),
s("\x81"), s("\x80"),
u("\xc2"), u("\x80"),
# for transitivity test
u("\xe0\xa0\xa1"), e("\xe0\xa0\xa1"), s("\xe0\xa0\xa1"), # [ruby-dev:32693]
e("\xa1\xa1"), a("\xa1\xa1"), s("\xa1\xa1"), # [ruby-dev:36484]
#"aa".force_encoding("utf-16be"),
#"aaaa".force_encoding("utf-32be"),
#"aaa".force_encoding("utf-32be"),
]
def combination(*args, &b)
AllPairs.each(*args, &b)
#AllPairs.exhaustive_each(*args, &b)
end
def encdump(str)
d = str.dump
if /\.force_encoding\("[A-Za-z0-9.:_+-]*"\)\z/ =~ d
d
else
"#{d}.force_encoding(#{str.encoding.name.dump})"
end
end
def encdumpargs(args)
r = '('
args.each_with_index {|a, i|
r << ',' if 0 < i
if String === a
r << encdump(a)
else
r << a.inspect
end
}
r << ')'
r
end
def enccall(recv, meth, *args, &block)
desc = ''
if String === recv
desc << encdump(recv)
else
desc << recv.inspect
end
desc << '.' << meth.to_s
if !args.empty?
desc << '('
args.each_with_index {|a, i|
desc << ',' if 0 < i
if String === a
desc << encdump(a)
else
desc << a.inspect
end
}
desc << ')'
end
if block
desc << ' {}'
end
result = nil
assert_nothing_raised(desc) {
result = recv.send(meth, *args, &block)
}
result
end
def assert_str_enc_propagation(t, s1, s2)
if !s1.ascii_only?
assert_equal(s1.encoding, t.encoding)
elsif !s2.ascii_only?
assert_equal(s2.encoding, t.encoding)
else
assert([s1.encoding, s2.encoding].include?(t.encoding))
end
end
def assert_same_result(expected_proc, actual_proc)
e = nil
begin
t = expected_proc.call
rescue
e = $!
end
if e
assert_raise(e.class) { actual_proc.call }
else
assert_equal(t, actual_proc.call)
end
end
def each_slice_call
combination(STRINGS, -2..2) {|s, nth|
yield s, nth
}
combination(STRINGS, -2..2, 0..2) {|s, nth, len|
yield s, nth, len
}
combination(STRINGS, STRINGS) {|s, substr|
yield s, substr
}
combination(STRINGS, -2..2, 0..2) {|s, first, last|
yield s, first..last
yield s, first...last
}
combination(STRINGS, STRINGS) {|s1, s2|
if !s2.valid_encoding?
next
end
yield s1, Regexp.new(Regexp.escape(s2))
}
combination(STRINGS, STRINGS, 0..2) {|s1, s2, nth|
if !s2.valid_encoding?
next
end
yield s1, Regexp.new(Regexp.escape(s2)), nth
}
end
ASCII_INCOMPATIBLE_ENCODINGS = %w[
UTF-16BE
UTF-16LE
UTF-32BE
UTF-32LE
]
def str_enc_compatible?(*strs)
encs = []
ascii_incompatible_encodings = {}
has_ascii_compatible = false
strs.each {|s|
encs << s.encoding if !s.ascii_only?
if /\A#{Regexp.union ASCII_INCOMPATIBLE_ENCODINGS}\z/o =~ s.encoding.name
ascii_incompatible_encodings[s.encoding] = true
else
has_ascii_compatible = true
end
}
if ascii_incompatible_encodings.empty?
encs.uniq!
encs.length <= 1
else
!has_ascii_compatible && ascii_incompatible_encodings.size == 1
end
end
# tests start
def test_str_new
STRINGS.each {|s|
t = String.new(s)
assert_strenc(a(s), s.encoding, t)
}
end
def test_str_plus
combination(STRINGS, STRINGS) {|s1, s2|
if s1.encoding != s2.encoding && !s1.ascii_only? && !s2.ascii_only?
assert_raise(Encoding::CompatibilityError) { s1 + s2 }
else
t = enccall(s1, :+, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert_equal(a(s1) + a(s2), a(t))
assert_str_enc_propagation(t, s1, s2)
end
}
end
def test_str_times
STRINGS.each {|s|
[0,1,2].each {|n|
t = s * n
assert(t.valid_encoding?) if s.valid_encoding?
assert_strenc(a(s) * n, s.encoding, t)
}
}
end
def test_sprintf_s
STRINGS.each {|s|
assert_strenc(a(s), s.encoding, "%s".force_encoding(s.encoding) % s)
if !s.empty? # xxx
t = enccall(a("%s"), :%, s)
assert_strenc(a(s), s.encoding, t)
end
}
end
def test_str_eq_reflexive
STRINGS.each {|s|
assert(s == s, "#{encdump s} == #{encdump s}")
}
end
def test_str_eq_symmetric
combination(STRINGS, STRINGS) {|s1, s2|
if s1 == s2
assert(s2 == s1, "#{encdump s2} == #{encdump s1}")
else
assert(!(s2 == s1), "!(#{encdump s2} == #{encdump s1})")
end
}
end
def test_str_eq_transitive
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
if s1 == s2 && s2 == s3
assert(s1 == s3, "transitive: #{encdump s1} == #{encdump s2} == #{encdump s3}")
end
}
end
def test_str_eq
combination(STRINGS, STRINGS) {|s1, s2|
desc_eq = "#{encdump s1} == #{encdump s2}"
if a(s1) == a(s2) and
(s1.ascii_only? && s2.ascii_only? or
s1.encoding == s2.encoding) then
assert(s1 == s2, desc_eq)
assert(!(s1 != s2))
assert_equal(0, s1 <=> s2)
assert(s1.eql?(s2), desc_eq)
else
assert(!(s1 == s2), "!(#{desc_eq})")
assert(s1 != s2)
assert_not_equal(0, s1 <=> s2)
assert(!s1.eql?(s2))
end
}
end
def test_str_concat
combination(STRINGS, STRINGS) {|s1, s2|
s = s1.dup
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
s << s2
assert(s.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert_equal(a(s), a(s1) + a(s2))
assert_str_enc_propagation(s, s1, s2)
else
assert_raise(Encoding::CompatibilityError) { s << s2 }
end
}
end
def test_str_aref
STRINGS.each {|s|
t = ''.force_encoding(s.encoding)
0.upto(s.length-1) {|i|
u = s[i]
assert(u.valid_encoding?) if s.valid_encoding?
t << u
}
assert_equal(t, s)
}
end
def test_str_aref_len
STRINGS.each {|s|
t = ''.force_encoding(s.encoding)
0.upto(s.length-1) {|i|
u = s[i,1]
assert(u.valid_encoding?) if s.valid_encoding?
t << u
}
assert_equal(t, s)
}
STRINGS.each {|s|
t = ''.force_encoding(s.encoding)
0.step(s.length-1, 2) {|i|
u = s[i,2]
assert(u.valid_encoding?) if s.valid_encoding?
t << u
}
assert_equal(t, s)
}
end
def test_str_aref_substr
combination(STRINGS, STRINGS) {|s1, s2|
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
t = enccall(s1, :[], s2)
if t != nil
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert_equal(s2, t)
assert_match(/#{Regexp.escape(a(s2))}/, a(s1))
if s1.valid_encoding?
assert_match(/#{Regexp.escape(s2)}/, s1)
end
end
else
assert_raise(Encoding::CompatibilityError) { s1[s2] }
end
}
end
def test_str_aref_range2
combination(STRINGS, -2..2, -2..2) {|s, first, last|
desc = "#{encdump s}[#{first}..#{last}]"
t = s[first..last]
if first < 0
first += s.length
if first < 0
assert_nil(t, desc)
next
end
end
if s.length < first
assert_nil(t, desc)
next
end
assert(t.valid_encoding?) if s.valid_encoding?
if last < 0
last += s.length
end
t2 = ''
first.upto(last) {|i|
c = s[i]
t2 << c if c
}
assert_equal(t2, t, desc)
}
end
def test_str_aref_range3
combination(STRINGS, -2..2, -2..2) {|s, first, last|
desc = "#{encdump s}[#{first}..#{last}]"
t = s[first...last]
if first < 0
first += s.length
if first < 0
assert_nil(t, desc)
next
end
end
if s.length < first
assert_nil(t, desc)
next
end
if last < 0
last += s.length
end
assert(t.valid_encoding?) if s.valid_encoding?
t2 = ''
first.upto(last-1) {|i|
c = s[i]
t2 << c if c
}
assert_equal(t2, t, desc)
}
end
def test_str_assign
combination(STRINGS, STRINGS) {|s1, s2|
(-2).upto(2) {|i|
t = s1.dup
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
if i < -s1.length || s1.length < i
assert_raise(IndexError) { t[i] = s2 }
else
t[i] = s2
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s2)))
if s1.valid_encoding? && s2.valid_encoding?
if i == s1.length && s2.empty?
assert_nil(t[i])
elsif i < 0
assert_equal(s2, t[i-s2.length+1,s2.length],
"t = #{encdump(s1)}; t[#{i}] = #{encdump(s2)}; t[#{i-s2.length+1},#{s2.length}]")
else
assert_equal(s2, t[i,s2.length],
"t = #{encdump(s1)}; t[#{i}] = #{encdump(s2)}; t[#{i},#{s2.length}]")
end
end
end
else
assert_raise(Encoding::CompatibilityError) { t[i] = s2 }
end
}
}
end
def test_str_assign_len
combination(STRINGS, -2..2, 0..2, STRINGS) {|s1, i, len, s2|
t = s1.dup
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
if i < -s1.length || s1.length < i
assert_raise(IndexError) { t[i,len] = s2 }
else
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
t[i,len] = s2
assert(a(t).index(a(s2)))
if s1.valid_encoding? && s2.valid_encoding?
if i == s1.length && s2.empty?
assert_nil(t[i])
elsif i < 0
if -i < len
len = -i
end
assert_equal(s2, t[i-s2.length+len,s2.length],
"t = #{encdump(s1)}; t[#{i},#{len}] = #{encdump(s2)}; t[#{i-s2.length+len},#{s2.length}]")
else
assert_equal(s2, t[i,s2.length],
"t = #{encdump(s1)}; t[#{i},#{len}] = #{encdump(s2)}; t[#{i},#{s2.length}]")
end
end
end
else
assert_raise(Encoding::CompatibilityError) { t[i,len] = s2 }
end
}
end
def test_str_assign_substr
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
t = s1.dup
encs = [
!s1.ascii_only? ? s1.encoding : nil,
!s2.ascii_only? ? s2.encoding : nil,
!s3.ascii_only? ? s3.encoding : nil].uniq.compact
if 1 < encs.length
assert_raise(Encoding::CompatibilityError, IndexError) { t[s2] = s3 }
else
if encs.empty?
encs = [
s1.encoding,
s2.encoding,
s3.encoding].uniq.reject {|e| e == Encoding.find("ASCII-8BIT") }
if encs.empty?
encs = [Encoding.find("ASCII-8BIT")]
end
end
if !t[s2]
else
enccall(t, :[]=, s2, s3)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding? && s3.valid_encoding?
end
end
}
end
def test_str_assign_range2
combination(STRINGS, -2..2, -2..2, STRINGS) {|s1, first, last, s2|
t = s1.dup
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
if first < -s1.length || s1.length < first
assert_raise(RangeError) { t[first..last] = s2 }
else
enccall(t, :[]=, first..last, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s2)))
if s1.valid_encoding? && s2.valid_encoding?
if first < 0
assert_equal(s2, t[s1.length+first, s2.length])
else
assert_equal(s2, t[first, s2.length])
end
end
end
else
assert_raise(Encoding::CompatibilityError, RangeError,
"t=#{encdump(s1)};t[#{first}..#{last}]=#{encdump(s2)}") {
t[first..last] = s2
}
end
}
end
def test_str_assign_range3
combination(STRINGS, -2..2, -2..2, STRINGS) {|s1, first, last, s2|
t = s1.dup
if s1.ascii_only? || s2.ascii_only? || s1.encoding == s2.encoding
if first < -s1.length || s1.length < first
assert_raise(RangeError) { t[first...last] = s2 }
else
enccall(t, :[]=, first...last, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s2)))
if s1.valid_encoding? && s2.valid_encoding?
if first < 0
assert_equal(s2, t[s1.length+first, s2.length])
else
assert_equal(s2, t[first, s2.length])
end
end
end
else
assert_raise(Encoding::CompatibilityError, RangeError,
"t=#{encdump(s1)};t[#{first}...#{last}]=#{encdump(s2)}") {
t[first...last] = s2
}
end
}
end
def test_str_cmp
combination(STRINGS, STRINGS) {|s1, s2|
desc = "#{encdump s1} <=> #{encdump s2}"
r = s1 <=> s2
if s1 == s2
assert_equal(0, r, desc)
else
assert_not_equal(0, r, desc)
end
}
end
def test_str_capitalize
STRINGS.each {|s|
begin
t1 = s.capitalize
rescue ArgumentError
assert(!s.valid_encoding?)
next
end
assert(t1.valid_encoding?) if s.valid_encoding?
assert(t1.casecmp(s))
t2 = s.dup
t2.capitalize!
assert_equal(t1, t2)
r = s.downcase
r = enccall(r, :sub, /\A[a-z]/) {|ch| a(ch).upcase }
assert_equal(r, t1)
}
end
def test_str_casecmp
combination(STRINGS, STRINGS) {|s1, s2|
#puts "#{encdump(s1)}.casecmp(#{encdump(s2)})"
begin
r = s1.casecmp(s2)
rescue ArgumentError
assert(!s1.valid_encoding? || !s2.valid_encoding?)
next
end
#assert_equal(s1.upcase <=> s2.upcase, r)
}
end
def test_str_center
combination(STRINGS, [0,1,2,3,10]) {|s1, width|
t = s1.center(width)
assert(a(t).index(a(s1)))
}
combination(STRINGS, [0,1,2,3,10], STRINGS) {|s1, width, s2|
if s2.empty?
assert_raise(ArgumentError) { s1.center(width, s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.center(width, s2) }
next
end
t = enccall(s1, :center, width, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s1)))
assert_str_enc_propagation(t, s1, s2) if (t != s1)
}
end
def test_str_ljust
combination(STRINGS, [0,1,2,3,10]) {|s1, width|
t = s1.ljust(width)
assert(a(t).index(a(s1)))
}
combination(STRINGS, [0,1,2,3,10], STRINGS) {|s1, width, s2|
if s2.empty?
assert_raise(ArgumentError) { s1.ljust(width, s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.ljust(width, s2) }
next
end
t = enccall(s1, :ljust, width, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s1)))
assert_str_enc_propagation(t, s1, s2) if (t != s1)
}
end
def test_str_rjust
combination(STRINGS, [0,1,2,3,10]) {|s1, width|
t = s1.rjust(width)
assert(a(t).index(a(s1)))
}
combination(STRINGS, [0,1,2,3,10], STRINGS) {|s1, width, s2|
if s2.empty?
assert_raise(ArgumentError) { s1.rjust(width, s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.rjust(width, s2) }
next
end
t = enccall(s1, :rjust, width, s2)
assert(t.valid_encoding?) if s1.valid_encoding? && s2.valid_encoding?
assert(a(t).index(a(s1)))
assert_str_enc_propagation(t, s1, s2) if (t != s1)
}
end
def test_str_chomp
combination(STRINGS, STRINGS) {|s1, s2|
if !s1.ascii_only? && !s2.ascii_only? && !Encoding.compatible?(s1,s2)
if s1.bytesize > s2.bytesize
assert_raise(Encoding::CompatibilityError) { s1.chomp(s2) }
end
next
end
t = enccall(s1, :chomp, s2)
assert(t.valid_encoding?, "#{encdump(s1)}.chomp(#{encdump(s2)})") if s1.valid_encoding? && s2.valid_encoding?
assert_equal(s1.encoding, t.encoding)
t2 = s1.dup
t2.chomp!(s2)
assert_equal(t, t2)
}
end
def test_str_chop
STRINGS.each {|s|
s = s.dup
desc = "#{encdump s}.chop"
t = nil
assert_nothing_raised(desc) { t = s.chop }
assert(t.valid_encoding?) if s.valid_encoding?
assert(a(s).index(a(t)))
t2 = s.dup
t2.chop!
assert_equal(t, t2)
}
end
def test_str_clear
STRINGS.each {|s|
t = s.dup
t.clear
assert(t.valid_encoding?)
assert(t.empty?)
}
end
def test_str_clone
STRINGS.each {|s|
t = s.clone
assert_equal(s, t)
assert_equal(s.encoding, t.encoding)
assert_equal(a(s), a(t))
}
end
def test_str_dup
STRINGS.each {|s|
t = s.dup
assert_equal(s, t)
assert_equal(s.encoding, t.encoding)
assert_equal(a(s), a(t))
}
end
def test_str_count
combination(STRINGS, STRINGS) {|s1, s2|
if !s1.valid_encoding? || !s2.valid_encoding?
assert_raise(ArgumentError, Encoding::CompatibilityError) { s1.count(s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.count(s2) }
next
end
n = enccall(s1, :count, s2)
n0 = a(s1).count(a(s2))
assert_operator(n, :<=, n0)
}
end
def test_str_crypt
combination(STRINGS, STRINGS) {|str, salt|
if a(salt).length < 2
assert_raise(ArgumentError) { str.crypt(salt) }
next
end
t = str.crypt(salt)
assert_equal(a(str).crypt(a(salt)), t, "#{encdump(str)}.crypt(#{encdump(salt)})")
assert_encoding('ASCII-8BIT', t.encoding)
}
end
def test_str_delete
combination(STRINGS, STRINGS) {|s1, s2|
if s1.empty?
assert_equal(s1, s1.delete(s2))
next
end
if !s1.valid_encoding? || !s2.valid_encoding?
assert_raise(ArgumentError, Encoding::CompatibilityError) { s1.delete(s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.delete(s2) }
next
end
t = enccall(s1, :delete, s2)
assert(t.valid_encoding?)
assert_equal(t.encoding, s1.encoding)
assert_operator(t.length, :<=, s1.length)
t2 = s1.dup
t2.delete!(s2)
assert_equal(t, t2)
}
end
def test_str_downcase
STRINGS.each {|s|
if !s.valid_encoding?
assert_raise(ArgumentError) { s.downcase }
next
end
t = s.downcase
assert(t.valid_encoding?)
assert_equal(t.encoding, s.encoding)
assert(t.casecmp(s))
t2 = s.dup
t2.downcase!
assert_equal(t, t2)
}
end
def test_str_dump
STRINGS.each {|s|
t = s.dump
assert(t.valid_encoding?)
assert(t.ascii_only?)
u = eval(t)
assert_equal(a(s), a(u))
}
end
def test_str_each_line
combination(STRINGS, STRINGS) {|s1, s2|
if !s1.valid_encoding? || !s2.valid_encoding?
assert_raise(ArgumentError, Encoding::CompatibilityError) { s1.each_line(s2) {} }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.each_line(s2) {} }
next
end
lines = []
enccall(s1, :each_line, s2) {|line|
assert(line.valid_encoding?)
assert_equal(s1.encoding, line.encoding)
lines << line
}
next if lines.size == 0
s2 = lines.join('')
assert_equal(s1.encoding, s2.encoding)
assert_equal(s1, s2)
}
end
def test_str_each_byte
STRINGS.each {|s|
bytes = []
s.each_byte {|b|
bytes << b
}
a(s).split(//).each_with_index {|ch, i|
assert_equal(ch.ord, bytes[i])
}
}
end
def test_str_empty?
STRINGS.each {|s|
if s.length == 0
assert(s.empty?)
else
assert(!s.empty?)
end
}
end
def test_str_hex
STRINGS.each {|s|
t = s.hex
t2 = a(s)[/\A[0-9a-fA-Fx]*/].hex
assert_equal(t2, t)
}
end
def test_str_include?
combination(STRINGS, STRINGS) {|s1, s2|
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.include?(s2) }
assert_raise(Encoding::CompatibilityError) { s1.index(s2) }
assert_raise(Encoding::CompatibilityError) { s1.rindex(s2) }
next
end
t = enccall(s1, :include?, s2)
if t
assert(a(s1).include?(a(s2)))
assert(s1.index(s2))
assert(s1.rindex(s2))
else
assert(!s1.index(s2))
assert(!s1.rindex(s2), "!#{encdump(s1)}.rindex(#{encdump(s2)})")
end
if s2.empty?
assert_equal(true, t)
next
end
if !s1.valid_encoding? || !s2.valid_encoding?
assert_equal(false, t, "#{encdump s1}.include?(#{encdump s2})")
next
end
if t && s1.valid_encoding? && s2.valid_encoding?
assert_match(/#{Regexp.escape(s2)}/, s1)
else
assert_no_match(/#{Regexp.escape(s2)}/, s1)
end
}
end
def test_str_index
combination(STRINGS, STRINGS, -2..2) {|s1, s2, pos|
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.index(s2) }
next
end
t = enccall(s1, :index, s2, pos)
if s2.empty?
if pos < 0 && pos+s1.length < 0
assert_equal(nil, t, "#{encdump s1}.index(#{encdump s2}, #{pos})");
elsif pos < 0
assert_equal(s1.length+pos, t, "#{encdump s1}.index(#{encdump s2}, #{pos})");
elsif s1.length < pos
assert_equal(nil, t, "#{encdump s1}.index(#{encdump s2}, #{pos})");
else
assert_equal(pos, t, "#{encdump s1}.index(#{encdump s2}, #{pos})");
end
next
end
if !s1.valid_encoding? || !s2.valid_encoding?
assert_equal(nil, t, "#{encdump s1}.index(#{encdump s2}, #{pos})");
next
end
if t
re = /#{Regexp.escape(s2)}/
assert(re.match(s1, pos))
assert_equal($`.length, t, "#{encdump s1}.index(#{encdump s2}, #{pos})")
else
assert_no_match(/#{Regexp.escape(s2)}/, s1[pos..-1])
end
}
end
def test_str_rindex
combination(STRINGS, STRINGS, -2..2) {|s1, s2, pos|
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.rindex(s2) }
next
end
t = enccall(s1, :rindex, s2, pos)
if s2.empty?
if pos < 0 && pos+s1.length < 0
assert_equal(nil, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
elsif pos < 0
assert_equal(s1.length+pos, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
elsif s1.length < pos
assert_equal(s1.length, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
else
assert_equal(pos, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
end
next
end
if !s1.valid_encoding? || !s2.valid_encoding?
assert_equal(nil, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
next
end
if t
#puts "#{encdump s1}.rindex(#{encdump s2}, #{pos}) => #{t}"
assert(a(s1).index(a(s2)))
pos2 = pos
pos2 += s1.length if pos < 0
re = /\A(.{0,#{pos2}})#{Regexp.escape(s2)}/m
m = enccall(re, :match, s1)
assert(m, "#{re.inspect}.match(#{encdump(s1)})")
assert_equal(m[1].length, t, "#{encdump s1}.rindex(#{encdump s2}, #{pos})")
else
re = /#{Regexp.escape(s2)}/
n = re =~ s1
if n
if pos < 0
assert_operator(n, :>, s1.length+pos)
else
assert_operator(n, :>, pos)
end
end
end
}
end
def test_str_insert
combination(STRINGS, 0..2, STRINGS) {|s1, nth, s2|
t1 = s1.dup
t2 = s1.dup
begin
t1[nth, 0] = s2
rescue Encoding::CompatibilityError, IndexError => e1
end
begin
t2.insert(nth, s2)
rescue Encoding::CompatibilityError, IndexError => e2
end
assert_equal(t1, t2, "t=#{encdump s1}; t.insert(#{nth},#{encdump s2}); t")
assert_equal(e1.class, e2.class, "begin #{encdump s1}.insert(#{nth},#{encdump s2}); rescue ArgumentError, IndexError => e; e end")
}
combination(STRINGS, -2..-1, STRINGS) {|s1, nth, s2|
next if s1.length + nth < 0
next unless s1.valid_encoding?
next unless s2.valid_encoding?
t1 = s1.dup
begin
t1.insert(nth, s2)
slen = s2.length
assert_equal(t1[nth-slen+1,slen], s2, "t=#{encdump s1}; t.insert(#{nth},#{encdump s2}); t")
rescue Encoding::CompatibilityError, IndexError => e
end
}
end
def test_str_intern
STRINGS.each {|s|
if /\0/ =~ a(s)
assert_raise(ArgumentError) { s.intern }
elsif s.valid_encoding?
sym = s.intern
assert_equal(s, sym.to_s, "#{encdump s}.intern.to_s")
assert_equal(sym, s.to_sym)
else
assert_raise(EncodingError) { s.intern }
end
}
end
def test_str_length
STRINGS.each {|s|
assert_operator(s.length, :<=, s.bytesize)
}
end
def test_str_oct
STRINGS.each {|s|
t = s.oct
t2 = a(s)[/\A[0-9a-fA-FxX]*/].oct
assert_equal(t2, t)
}
end
def test_str_replace
combination(STRINGS, STRINGS) {|s1, s2|
t = s1.dup
t.replace s2
assert_equal(s2, t)
assert_equal(s2.encoding, t.encoding)
}
end
def test_str_reverse
STRINGS.each {|s|
t = s.reverse
assert_equal(s.bytesize, t.bytesize)
if !s.valid_encoding?
assert_operator(t.length, :<=, s.length)
next
end
assert_equal(s, t.reverse)
}
end
def test_str_scan
combination(STRINGS, STRINGS) {|s1, s2|
if !s2.valid_encoding?
assert_raise(RegexpError) { s1.scan(s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
if s1.valid_encoding?
assert_raise(Encoding::CompatibilityError) { s1.scan(s2) }
else
assert_raise(ArgumentError, /invalid byte sequence/) { s1.scan(s2) }
end
next
end
if !s1.valid_encoding?
assert_raise(ArgumentError) { s1.scan(s2) }
next
end
r = enccall(s1, :scan, s2)
r.each {|t|
assert_equal(s2, t)
}
}
end
def test_str_slice
each_slice_call {|obj, *args|
assert_same_result(lambda { obj[*args] }, lambda { obj.slice(*args) })
}
end
def test_str_slice!
each_slice_call {|s, *args|
desc_slice = "#{encdump s}.slice#{encdumpargs args}"
desc_slice_bang = "#{encdump s}.slice!#{encdumpargs args}"
t = s.dup
begin
r = t.slice!(*args)
rescue
e = $!
end
if e
assert_raise(e.class, desc_slice) { s.slice(*args) }
next
end
if !r
assert_nil(s.slice(*args), desc_slice)
next
end
assert_equal(s.slice(*args), r, desc_slice_bang)
assert_equal(s.bytesize, r.bytesize + t.bytesize)
if args.length == 1 && String === args[0]
assert_equal(args[0].encoding, r.encoding,
"#{encdump s}.slice!#{encdumpargs args}.encoding")
else
assert_equal(s.encoding, r.encoding,
"#{encdump s}.slice!#{encdumpargs args}.encoding")
end
if [s, *args].all? {|o| !(String === o) || o.valid_encoding? }
assert(r.valid_encoding?)
assert(t.valid_encoding?)
assert_equal(s.length, r.length + t.length)
end
}
end
def test_str_split
combination(STRINGS, STRINGS) {|s1, s2|
if !s2.valid_encoding?
assert_raise(ArgumentError, RegexpError) { s1.split(s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(ArgumentError, Encoding::CompatibilityError) { s1.split(s2) }
next
end
if !s1.valid_encoding?
assert_raise(ArgumentError) { s1.split(s2) }
next
end
t = enccall(s1, :split, s2)
t.each {|r|
assert(a(s1).include?(a(r)))
assert_equal(s1.encoding, r.encoding)
}
assert(a(s1).include?(t.map {|u| a(u) }.join(a(s2))))
if s1.valid_encoding? && s2.valid_encoding?
t.each {|r|
assert(r.valid_encoding?)
}
end
}
end
def test_str_squeeze
combination(STRINGS, STRINGS) {|s1, s2|
if !s1.valid_encoding? || !s2.valid_encoding?
assert_raise(ArgumentError, Encoding::CompatibilityError, "#{encdump s1}.squeeze(#{encdump s2})") { s1.squeeze(s2) }
next
end
if !s1.ascii_only? && !s2.ascii_only? && s1.encoding != s2.encoding
assert_raise(Encoding::CompatibilityError) { s1.squeeze(s2) }
next
end
t = enccall(s1, :squeeze, s2)
assert_operator(t.length, :<=, s1.length)
t2 = s1.dup
t2.squeeze!(s2)
assert_equal(t, t2)
}
end
def test_str_strip
STRINGS.each {|s|
if !s.valid_encoding?
assert_raise(ArgumentError, "#{encdump s}.strip") { s.strip }
next
end
t = s.strip
l = s.lstrip
r = s.rstrip
assert_operator(l.length, :<=, s.length)
assert_operator(r.length, :<=, s.length)
assert_operator(t.length, :<=, l.length)
assert_operator(t.length, :<=, r.length)
t2 = s.dup
t2.strip!
assert_equal(t, t2)
l2 = s.dup
l2.lstrip!
assert_equal(l, l2)
r2 = s.dup
r2.rstrip!
assert_equal(r, r2)
}
end
def test_str_sum
STRINGS.each {|s|
assert_equal(a(s).sum, s.sum)
}
end
def test_str_swapcase
STRINGS.each {|s|
if !s.valid_encoding?
assert_raise(ArgumentError, "#{encdump s}.swapcase") { s.swapcase }
next
end
t1 = s.swapcase
assert(t1.valid_encoding?) if s.valid_encoding?
assert(t1.casecmp(s))
t2 = s.dup
t2.swapcase!
assert_equal(t1, t2)
t3 = t1.swapcase
assert_equal(s, t3);
}
end
def test_str_to_f
STRINGS.each {|s|
assert_nothing_raised { s.to_f }
}
end
def test_str_to_i
STRINGS.each {|s|
assert_nothing_raised { s.to_i }
2.upto(36) {|radix|
assert_nothing_raised { s.to_i(radix) }
}
}
end
def test_str_to_s
STRINGS.each {|s|
assert_same(s, s.to_s)
assert_same(s, s.to_str)
}
end
def test_tr
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
desc = "#{encdump s1}.tr(#{encdump s2}, #{encdump s3})"
if s1.empty?
assert_equal(s1, s1.tr(s2, s3), desc)
next
end
if !str_enc_compatible?(s1, s2, s3)
assert_raise(Encoding::CompatibilityError, desc) { s1.tr(s2, s3) }
next
end
if !s1.valid_encoding?
assert_raise(ArgumentError, desc) { s1.tr(s2, s3) }
next
end
if s2.empty?
t = enccall(s1, :tr, s2, s3)
assert_equal(s1, t, desc)
next
end
if !s2.valid_encoding? || !s3.valid_encoding?
assert_raise(ArgumentError, desc) { s1.tr(s2, s3) }
next
end
t = enccall(s1, :tr, s2, s3)
assert_operator(s1.length, :>=, t.length, desc)
}
end
def test_tr_s
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
desc = "#{encdump s1}.tr_s(#{encdump s2}, #{encdump s3})"
if s1.empty?
assert_equal(s1, s1.tr_s(s2, s3), desc)
next
end
if !s1.valid_encoding?
assert_raise(ArgumentError, Encoding::CompatibilityError, desc) { s1.tr_s(s2, s3) }
next
end
if !str_enc_compatible?(s1, s2, s3)
assert_raise(Encoding::CompatibilityError, desc) { s1.tr(s2, s3) }
next
end
if s2.empty?
t = enccall(s1, :tr_s, s2, s3)
assert_equal(s1, t, desc)
next
end
if !s2.valid_encoding? || !s3.valid_encoding?
assert_raise(ArgumentError, desc) { s1.tr_s(s2, s3) }
next
end
t = enccall(s1, :tr_s, s2, s3)
assert_operator(s1.length, :>=, t.length, desc)
}
end
def test_str_upcase
STRINGS.each {|s|
desc = "#{encdump s}.upcase"
if !s.valid_encoding?
assert_raise(ArgumentError, desc) { s.upcase }
next
end
t1 = s.upcase
assert(t1.valid_encoding?)
assert(t1.casecmp(s))
t2 = s.dup
t2.upcase!
assert_equal(t1, t2)
}
end
def test_str_succ
STRINGS.each {|s0|
next if s0.empty?
s = s0.dup
n = 300
h = {}
n.times {|i|
if h[s]
assert(false, "#{encdump s} cycle with succ #{i-h[s]} times")
end
h[s] = i
assert_operator(s.length, :<=, s0.length + Math.log2(i+1) + 1, "#{encdump s0} succ #{i} times => #{encdump s}")
#puts encdump(s)
t = s.succ
if s.valid_encoding?
assert(t.valid_encoding?, "#{encdump s}.succ.valid_encoding?")
end
s = t
}
}
end
def test_str_hash
combination(STRINGS, STRINGS) {|s1, s2|
if s1.eql?(s2)
assert_equal(s1.hash, s2.hash, "#{encdump s1}.hash == #{encdump s2}.dump")
end
}
end
def test_marshal
STRINGS.each {|s|
m = Marshal.dump(s)
t = Marshal.load(m)
assert_equal(s, t)
}
end
def test_str_sub
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
if !s2.valid_encoding?
assert_raise(RegexpError) { Regexp.new(Regexp.escape(s2)) }
next
end
r2 = Regexp.new(Regexp.escape(s2))
[
[
"#{encdump s1}.sub(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { s1.sub(r2, s3) },
false
],
[
"#{encdump s1}.sub(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { s1.sub(r2) { s3 } },
false
],
[
"#{encdump s1}.gsub(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { s1.gsub(r2, s3) },
true
],
[
"#{encdump s1}.gsub(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { s1.gsub(r2) { s3 } },
true
]
].each {|desc, doit, g|
if !s1.valid_encoding?
assert_raise(ArgumentError, desc) { doit.call }
next
end
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { doit.call }
next
end
if !enccall(s1, :include?, s2)
assert_equal(s1, doit.call)
next
end
if !str_enc_compatible?(g ? s1.gsub(r2, '') : s1.sub(r2, ''), s3)
assert_raise(Encoding::CompatibilityError, desc) { doit.call }
next
end
t = nil
assert_nothing_raised(desc) {
t = doit.call
}
if s2 == s3
assert_equal(s1, t, desc)
else
assert_not_equal(s1, t, desc)
end
}
}
end
def test_str_sub!
combination(STRINGS, STRINGS, STRINGS) {|s1, s2, s3|
if !s2.valid_encoding?
assert_raise(RegexpError) { Regexp.new(Regexp.escape(s2)) }
next
end
r2 = Regexp.new(Regexp.escape(s2))
[
[
"t=#{encdump s1}.dup;t.sub!(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { t=s1.dup; [t, t.sub!(r2, s3)] },
false
],
[
"t=#{encdump s1}.dup;t.sub!(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { t=s1.dup; [t, t.sub!(r2) { s3 }] },
false
],
[
"t=#{encdump s1}.dup;t.gsub!(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { t=s1.dup; [t, t.gsub!(r2, s3)] },
true
],
[
"t=#{encdump s1}.dup;t.gsub!(Regexp.new(#{encdump s2}), #{encdump s3})",
lambda { t=s1.dup; [t, t.gsub!(r2) { s3 }] },
true
]
].each {|desc, doit, g|
if !s1.valid_encoding?
assert_raise(ArgumentError, desc) { doit.call }
next
end
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { doit.call }
next
end
if !enccall(s1, :include?, s2)
assert_equal([s1, nil], doit.call)
next
end
if !str_enc_compatible?(g ? s1.gsub(r2, '') : s1.sub(r2, ''), s3)
assert_raise(Encoding::CompatibilityError, desc) { doit.call }
next
end
t = ret = nil
assert_nothing_raised(desc) {
t, ret = doit.call
}
assert(ret)
if s2 == s3
assert_equal(s1, t, desc)
else
assert_not_equal(s1, t, desc)
end
}
}
end
def test_str_bytes
STRINGS.each {|s1|
ary = []
s1.bytes.each {|b|
ary << b
}
assert_equal(s1.unpack("C*"), ary)
}
end
def test_str_bytesize
STRINGS.each {|s1|
assert_equal(s1.unpack("C*").length, s1.bytesize)
}
end
def test_str_chars
STRINGS.each {|s1|
ary = []
s1.chars.each {|c|
ary << c
}
expected = []
s1.length.times {|i|
expected << s1[i]
}
assert_equal(expected, ary)
}
end
def test_str_chr
STRINGS.each {|s1|
if s1.empty?
assert_equal("", s1.chr)
next
end
assert_equal(s1[0], s1.chr)
}
end
def test_str_end_with?
combination(STRINGS, STRINGS) {|s1, s2|
desc = "#{encdump s1}.end_with?(#{encdump s2})"
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { s1.end_with?(s2) }
next
end
if s1.length < s2.length
assert_equal(false, enccall(s1, :end_with?, s2), desc)
next
end
if s1[s1.length-s2.length, s2.length] == s2
assert_equal(true, enccall(s1, :end_with?, s2), desc)
next
end
assert_equal(false, enccall(s1, :end_with?, s2), desc)
}
end
def test_str_start_with?
combination(STRINGS, STRINGS) {|s1, s2|
desc = "#{encdump s1}.start_with?(#{encdump s2})"
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { s1.start_with?(s2) }
next
end
s1 = s1.dup.force_encoding("ASCII-8BIT")
s2 = s2.dup.force_encoding("ASCII-8BIT")
if s1.length < s2.length
assert_equal(false, enccall(s1, :start_with?, s2), desc)
next
end
if s1[0, s2.length] == s2
assert_equal(true, enccall(s1, :start_with?, s2), desc)
next
end
assert_equal(false, enccall(s1, :start_with?, s2), desc)
}
end
def test_str_ord
STRINGS.each {|s1|
if s1.empty?
assert_raise(ArgumentError) { s1.ord }
next
end
if !s1.valid_encoding?
assert_raise(ArgumentError) { s1.ord }
next
end
assert_equal(s1[0].ord, s1.ord)
}
end
def test_str_partition
combination(STRINGS, STRINGS) {|s1, s2|
desc = "#{encdump s1}.partition(#{encdump s2})"
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { s1.partition(s2) }
next
end
i = enccall(s1, :index, s2)
if !i
assert_equal([s1, "", ""], s1.partition(s2), desc)
next
end
assert_equal([s1[0,i], s2, s1[(i+s2.length)..-1]], s1.partition(s2), desc)
}
end
def test_str_rpartition
combination(STRINGS, STRINGS) {|s1, s2|
desc = "#{encdump s1}.rpartition(#{encdump s2})"
if !str_enc_compatible?(s1, s2)
assert_raise(Encoding::CompatibilityError, desc) { s1.rpartition(s2) }
next
end
i = enccall(s1, :rindex, s2)
if !i
assert_equal(["", "", s1], s1.rpartition(s2), desc)
next
end
assert_equal([s1[0,i], s2, s1[(i+s2.length)..-1]], s1.rpartition(s2), desc)
}
end
end