summaryrefslogtreecommitdiffstats
path: root/base/utf_string_conversions_unittest.cc
diff options
context:
space:
mode:
Diffstat (limited to 'base/utf_string_conversions_unittest.cc')
-rw-r--r--base/utf_string_conversions_unittest.cc43
1 files changed, 0 insertions, 43 deletions
diff --git a/base/utf_string_conversions_unittest.cc b/base/utf_string_conversions_unittest.cc
index ca79ec5..19189971 100644
--- a/base/utf_string_conversions_unittest.cc
+++ b/base/utf_string_conversions_unittest.cc
@@ -226,47 +226,4 @@ TEST(UTFStringConversionsTest, ConvertMultiString) {
EXPECT_EQ(expected, converted);
}
-TEST(UTFStringConversionsTest, AdjustOffset) {
- struct UTF8ToWideCase {
- const char* utf8;
- size_t input_offset;
- size_t output_offset;
- } utf8_to_wide_cases[] = {
- {"", 0, std::wstring::npos},
- {"\xe4\xbd\xa0\xe5\xa5\xbd", 1, std::wstring::npos},
- {"\xe4\xbd\xa0\xe5\xa5\xbd", 3, 1},
- {"\xed\xb0\x80z", 3, 0},
- {"A\xF0\x90\x8C\x80z", 1, 1},
- {"A\xF0\x90\x8C\x80z", 2, std::wstring::npos},
-#if defined(WCHAR_T_IS_UTF16)
- {"A\xF0\x90\x8C\x80z", 5, 3},
-#elif defined(WCHAR_T_IS_UTF32)
- {"A\xF0\x90\x8C\x80z", 5, 2},
-#endif
- };
- for (size_t i = 0; i < ARRAYSIZE_UNSAFE(utf8_to_wide_cases); ++i) {
- size_t offset = utf8_to_wide_cases[i].input_offset;
- UTF8ToWideAndAdjustOffset(utf8_to_wide_cases[i].utf8, &offset);
- EXPECT_EQ(utf8_to_wide_cases[i].output_offset, offset);
- }
-
-#if defined(WCHAR_T_IS_UTF32)
- struct UTF16ToWideCase {
- const wchar_t* wide;
- size_t input_offset;
- size_t output_offset;
- } utf16_to_wide_cases[] = {
- {L"\xD840\xDC00\x4E00", 0, 0},
- {L"\xD840\xDC00\x4E00", 1, std::wstring::npos},
- {L"\xD840\xDC00\x4E00", 2, 1},
- };
- for (size_t i = 0; i < ARRAYSIZE_UNSAFE(utf16_to_wide_cases); ++i) {
- size_t offset = utf16_to_wide_cases[i].input_offset;
- UTF16ToWideAndAdjustOffset(BuildString16(utf16_to_wide_cases[i].wide),
- &offset);
- EXPECT_EQ(utf16_to_wide_cases[i].output_offset, offset);
- }
-#endif
-}
-
} // namaspace base