Chromium Code Reviews| Index: ui/gfx/text_elider_unittest.cc |
| diff --git a/ui/gfx/text_elider_unittest.cc b/ui/gfx/text_elider_unittest.cc |
| index b0baa4226da611604a608d00ee9a3d7785e876f2..bdc7a531d4de0572d61fb97ea0746891dd3055d7 100644 |
| --- a/ui/gfx/text_elider_unittest.cc |
| +++ b/ui/gfx/text_elider_unittest.cc |
| @@ -1071,53 +1071,65 @@ TEST(TextEliderTest, MAYBE_ElideRectangleWide32) { |
| #define MAYBE_TruncateString TruncateString |
| #endif |
| TEST(TextEliderTest, MAYBE_TruncateString) { |
|
msw
2015/09/17 18:42:44
Adding a test with surrogate pairs, etc. would be
Peter Kasting
2015/09/19 02:43:25
Since the code can get those wrong today, let me d
msw
2015/09/21 18:28:10
Acknowledged.
|
| - base::string16 string = ASCIIToUTF16("foooooey bxxxar baz"); |
| + base::string16 str = ASCIIToUTF16("fooooey bxxxar baz "); |
| - // Tests that apply to both break behaviors: |
| + // Test breaking at character 0. |
| + EXPECT_EQ(base::string16(), TruncateString(str, 0, WORD_BREAK)); |
| + EXPECT_EQ(base::string16(), TruncateString(str, 0, CHARACTER_BREAK)); |
| - // Make sure it doesn't modify the string if length > string length. |
| - EXPECT_EQ(string, TruncateString(string, 100, WORD_BREAK)); |
| - EXPECT_EQ(string, TruncateString(string, 100, CHARACTER_BREAK)); |
| + // Test breaking at character 1. |
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str, 1, WORD_BREAK))); |
|
msw
2015/09/17 18:42:44
nit: maybe EXPECT_STREQ would make UTF16ToWide unn
Peter Kasting
2015/09/19 02:43:25
Nah, that just means you need to supply two pointe
|
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str, 1, CHARACTER_BREAK))); |
| - // Test no characters. |
| - EXPECT_EQ(L"", UTF16ToWide(TruncateString(string, 0, WORD_BREAK))); |
| - EXPECT_EQ(L"", UTF16ToWide(TruncateString(string, 0, CHARACTER_BREAK))); |
| + // Test breaking in the middle of the first word. |
| + EXPECT_EQ(L"f\x2026", UTF16ToWide(TruncateString(str, 2, WORD_BREAK))); |
| + EXPECT_EQ(L"f\x2026", UTF16ToWide(TruncateString(str, 2, CHARACTER_BREAK))); |
| - // Test 1 character. |
| - EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(string, 1, WORD_BREAK))); |
| - EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(string, 1, CHARACTER_BREAK))); |
| + // Test breaking in between words. |
| + EXPECT_EQ(L"fooooey\x2026", UTF16ToWide(TruncateString(str, 9, WORD_BREAK))); |
| + EXPECT_EQ(L"fooooey\x2026", |
| + UTF16ToWide(TruncateString(str, 9, CHARACTER_BREAK))); |
| - // Test completely truncates string if break is on initial whitespace. |
| - EXPECT_EQ(L"\x2026", |
| - UTF16ToWide(TruncateString(ASCIIToUTF16(" "), 2, WORD_BREAK))); |
| - EXPECT_EQ(L"\x2026", |
| - UTF16ToWide(TruncateString(ASCIIToUTF16(" "), 2, |
| - CHARACTER_BREAK))); |
| + // Test breaking at the start of a later word. |
|
msw
2015/09/17 18:42:44
It's probably worthwhile to test index 12 too (but
Peter Kasting
2015/09/19 02:43:25
Done. Expanded the tests below by a case as well.
|
| + EXPECT_EQ(L"fooooey\x2026", UTF16ToWide(TruncateString(str, 11, WORD_BREAK))); |
| + EXPECT_EQ(L"fooooey\x2026", |
| + UTF16ToWide(TruncateString(str, 11, CHARACTER_BREAK))); |
| - // Break-only-at-word-boundaries tests: |
| + // Test breaking in the middle of a word. |
| + EXPECT_EQ(L"fooooey\x2026", UTF16ToWide(TruncateString(str, 14, WORD_BREAK))); |
| + EXPECT_EQ(L"fooooey bx\x2026", |
| + UTF16ToWide(TruncateString(str, 14, CHARACTER_BREAK))); |
| - // Test adds ... at right spot when there is enough room to break at a |
| - // word boundary. |
| - EXPECT_EQ(L"foooooey\x2026", UTF16ToWide(TruncateString(string, 14, |
| - WORD_BREAK))); |
| + // Test breaking in whitespace at the end of the string. |
| + EXPECT_EQ(L"fooooey bxxxar baz\x2026", |
| + UTF16ToWide(TruncateString(str, 22, WORD_BREAK))); |
| + EXPECT_EQ(L"fooooey bxxxar baz\x2026", |
| + UTF16ToWide(TruncateString(str, 22, CHARACTER_BREAK))); |
| - // Test adds ... at right spot when there is not enough space in first word. |
| - EXPECT_EQ(L"f\x2026", UTF16ToWide(TruncateString(string, 2, WORD_BREAK))); |
| + // Test breaking at the end of the string. |
| + EXPECT_EQ(str, TruncateString(str, str.length(), WORD_BREAK)); |
| + EXPECT_EQ(str, TruncateString(str, str.length(), CHARACTER_BREAK)); |
| - // Test adds ... at right spot when there is not enough room to break at a |
| - // word boundary. |
| - EXPECT_EQ(L"foooooey\x2026", UTF16ToWide(TruncateString(string, 11, |
| - WORD_BREAK))); |
| + // Test breaking past the end of the string. |
| + EXPECT_EQ(str, TruncateString(str, str.length() + 10, WORD_BREAK)); |
| + EXPECT_EQ(str, TruncateString(str, str.length() + 10, CHARACTER_BREAK)); |
| - // Break-anywhere tests: |
| - // Test adds ... at right spot within a word. |
| - EXPECT_EQ(L"f\x2026", UTF16ToWide(TruncateString(string, 2, |
| - CHARACTER_BREAK))); |
| + // Tests of strings with leading whitespace: |
| + base::string16 str2 = ASCIIToUTF16(" foo"); |
| - // Test removes trailing whitespace if break falls between words. |
| - EXPECT_EQ(L"foooooey\x2026", UTF16ToWide(TruncateString(string, 12, |
| - CHARACTER_BREAK))); |
| + // Test breaking in leading whitespace. |
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str2, 2, WORD_BREAK))); |
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str2, 2, CHARACTER_BREAK))); |
| + |
| + // Test breaking at the beginning of the first word, with leading whitespace. |
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str2, 4, WORD_BREAK))); |
| + EXPECT_EQ(L"\x2026", UTF16ToWide(TruncateString(str2, 4, CHARACTER_BREAK))); |
| + |
| + // Test breaking in the middle of the first word, with leading whitespace. |
| + EXPECT_EQ(L" f\x2026", UTF16ToWide(TruncateString(str2, 5, WORD_BREAK))); |
| + EXPECT_EQ(L" f\x2026", |
| + UTF16ToWide(TruncateString(str2, 5, CHARACTER_BREAK))); |
| } |
| } // namespace gfx |