You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
538 lines
22 KiB
538 lines
22 KiB
#include <plString.h> |
|
#include <HeadSpin.h> |
|
|
|
#include <gtest/gtest.h> |
|
#include <wchar.h> |
|
|
|
static const plUniChar test_data[] = { |
|
0x20, 0x7f, /* Normal ASCII chars */ |
|
0xff, 0x100, /* 8-bit boundary chars */ |
|
0x7fff, /* UTF-8 2-byte boundary */ |
|
0xffff, 0x10000, /* 16-bit boundary chars */ |
|
0x10020, 0x40000, /* Non-edge UTF-16 surrogate pairs */ |
|
0x10ffff, /* Highest Unicode character */ |
|
0 /* Null terminator */ |
|
}; |
|
|
|
/* UTF-8 version of above test data */ |
|
static const char utf8_test_data[] = |
|
"\x20" "\x7f" |
|
"\xc3\xbf" "\xc4\x80" |
|
"\xe7\xbf\xbf" |
|
"\xef\xbf\xbf" "\xf0\x90\x80\x80" |
|
"\xf0\x90\x80\xa0" "\xf1\x80\x80\x80" |
|
"\xf4\x8f\xbf\xbf"; |
|
static const size_t utf8_test_data_length = arrsize(utf8_test_data) - 1; |
|
|
|
/* UTF-16 version of test data */ |
|
static const uint16_t utf16_test_data[] = { |
|
0x20, 0x7f, |
|
0xff, 0x100, |
|
0x7fff, |
|
0xffff, |
|
/* surrogate pairs for chars >0xffff */ |
|
0xd800, 0xdc00, |
|
0xd800, 0xdc20, |
|
0xd8c0, 0xdc00, |
|
0xdbff, 0xdfff, |
|
0 |
|
}; |
|
|
|
/* Utility for comparing plUniChar buffers */ |
|
template <typename _Ch> |
|
static int T_strcmp(const _Ch *left, const _Ch *right) |
|
{ |
|
for ( ;; ) { |
|
if (*left != *right) |
|
return *left - *right; |
|
if (*left == 0) |
|
return (*right == 0) ? 0 : -1; |
|
if (*right == 0) |
|
return 1; |
|
|
|
++left; |
|
++right; |
|
} |
|
} |
|
|
|
TEST(plString, TestHelpers) |
|
{ |
|
/* Ensure the utilities for testing the module function properly */ |
|
EXPECT_EQ(0, T_strcmp("abc", "abc")); |
|
EXPECT_LT(0, T_strcmp("abc", "aba")); |
|
EXPECT_GT(0, T_strcmp("abc", "abe")); |
|
EXPECT_LT(0, T_strcmp("abc", "ab")); |
|
EXPECT_GT(0, T_strcmp("abc", "abcd")); |
|
EXPECT_EQ(0, T_strcmp("", "")); |
|
EXPECT_GT(0, T_strcmp("", "a")); |
|
EXPECT_LT(0, T_strcmp("a", "")); |
|
} |
|
|
|
TEST(plString, Utility) |
|
{ |
|
// Special plString::Null constant |
|
EXPECT_EQ(plString::Null, plString("")); |
|
EXPECT_EQ(plString::Null, plString{}); |
|
|
|
EXPECT_EQ(0, plString::Null.GetSize()); |
|
EXPECT_EQ(true, plString::Null.IsEmpty()); |
|
EXPECT_EQ(true, plString::Null.IsNull()); |
|
|
|
// Short and Long string length |
|
EXPECT_EQ(4, plString("1234").GetSize()); |
|
EXPECT_EQ(32, plString("12345678901234567890123456789012").GetSize()); |
|
|
|
// plStrings store data as UTF-8 internally |
|
EXPECT_EQ(utf8_test_data_length, plString(utf8_test_data).GetSize()); |
|
} |
|
|
|
TEST(plString, ConvertUtf8) |
|
{ |
|
// From UTF-8 to plString |
|
plString from_utf8 = plString::FromUtf8(utf8_test_data); |
|
EXPECT_STREQ(utf8_test_data, from_utf8.c_str()); |
|
EXPECT_EQ(utf8_test_data_length, from_utf8.GetSize()); |
|
plUnicodeBuffer unicode = from_utf8.GetUnicodeArray(); |
|
EXPECT_EQ(0, T_strcmp(test_data, unicode.GetData())); |
|
|
|
// From plString to UTF-8 |
|
plString to_utf8 = plString::FromUtf32(test_data); |
|
EXPECT_STREQ(utf8_test_data, to_utf8.c_str()); |
|
|
|
// Empty strings |
|
plString empty = plString::FromUtf8(""); |
|
EXPECT_EQ(0, empty.GetSize()); |
|
EXPECT_EQ(0, T_strcmp(empty.c_str(), "")); |
|
|
|
const plUniChar empty_data[] = { 0 }; |
|
empty = plString::FromUtf32(empty_data); |
|
EXPECT_EQ(0, empty.GetSize()); |
|
EXPECT_EQ(0, T_strcmp(empty.c_str(), "")); |
|
} |
|
|
|
TEST(plString, ConvertUtf16) |
|
{ |
|
// From UTF-16 to plString |
|
plString from_utf16 = plString::FromUtf16(utf16_test_data); |
|
EXPECT_EQ(utf8_test_data_length, from_utf16.GetSize()); |
|
plUnicodeBuffer unicode = from_utf16.GetUnicodeArray(); |
|
EXPECT_EQ(0, T_strcmp(test_data, unicode.GetData())); |
|
|
|
// From plString to UTF-16 |
|
plStringBuffer<uint16_t> to_utf16 = plString::FromUtf32(test_data).ToUtf16(); |
|
EXPECT_EQ(0, T_strcmp(utf16_test_data, to_utf16.GetData())); |
|
|
|
// Empty string |
|
const uint16_t empty_data[] = { 0 }; |
|
plString empty = plString::FromUtf16(empty_data); |
|
EXPECT_EQ(0, empty.GetSize()); |
|
EXPECT_EQ(0, T_strcmp(empty.c_str(), "")); |
|
} |
|
|
|
TEST(plString, ConvertIso8859_1) |
|
{ |
|
// From ISO-8859-1 to plString |
|
const char latin1[] = "\x20\x7e\xa0\xff"; |
|
const plUniChar unicode_cp0[] = { 0x20, 0x7e, 0xa0, 0xff, 0 }; |
|
static const size_t latin1_utf8_length = 6; |
|
plString from_latin1 = plString::FromIso8859_1(latin1); |
|
EXPECT_EQ(latin1_utf8_length, from_latin1.GetSize()); |
|
plUnicodeBuffer unicode = from_latin1.GetUnicodeArray(); |
|
EXPECT_EQ(0, T_strcmp(unicode_cp0, unicode.GetData())); |
|
|
|
// From plString to ISO-8859-1 |
|
plStringBuffer<char> to_latin1 = plString::FromUtf32(unicode_cp0).ToIso8859_1(); |
|
EXPECT_STREQ(latin1, to_latin1.GetData()); |
|
|
|
// Empty string |
|
plString empty = plString::FromIso8859_1(""); |
|
EXPECT_EQ(0, empty.GetSize()); |
|
EXPECT_EQ(0, T_strcmp(empty.c_str(), "")); |
|
} |
|
|
|
TEST(plString, ConvertWchar) |
|
{ |
|
// UTF-8 and UTF-16 are already tested, so just make sure we test |
|
// wchar_t and L"" conversions |
|
|
|
const wchar_t wtext[] = L"\x20\x7f\xff\u0100\uffff"; |
|
const plUniChar unicode_text[] = { 0x20, 0x7f, 0xff, 0x100, 0xffff, 0 }; |
|
static const size_t wtext_utf8_length = 9; |
|
plString from_wchar = plString::FromWchar(wtext); |
|
EXPECT_EQ(wtext_utf8_length, from_wchar.GetSize()); |
|
plUnicodeBuffer unicode = from_wchar.GetUnicodeArray(); |
|
EXPECT_EQ(0, T_strcmp(unicode_text, unicode.GetData())); |
|
|
|
// From plString to wchar_t |
|
plStringBuffer<wchar_t> to_wchar = plString::FromUtf32(unicode_text).ToWchar(); |
|
EXPECT_STREQ(wtext, to_wchar.GetData()); |
|
|
|
// Empty string |
|
plString empty = plString::FromWchar(L""); |
|
EXPECT_EQ(0, empty.GetSize()); |
|
EXPECT_EQ(0, T_strcmp(empty.c_str(), "")); |
|
} |
|
|
|
TEST(plString, ConvertInvalid) |
|
{ |
|
// The following should encode replacement characters for invalid chars |
|
const plUniChar unicode_replacement[] = { 0xfffd, 0 }; |
|
const char latin1_replacement[] = "?"; |
|
|
|
// Character outside of Unicode specification range |
|
const plUniChar too_big_c[] = { 0xffffff, 0 }; |
|
plUnicodeBuffer too_big = plString::FromUtf32(too_big_c).GetUnicodeArray(); |
|
EXPECT_EQ(0, T_strcmp(unicode_replacement, too_big.GetData())); |
|
|
|
// Invalid surrogate pairs |
|
const uint16_t incomplete_surr_c[] = { 0xd800, 0 }; |
|
plString incomplete_surr = plString::FromUtf16(incomplete_surr_c); |
|
EXPECT_EQ(0, T_strcmp(unicode_replacement, |
|
incomplete_surr.GetUnicodeArray().GetData())); |
|
|
|
const uint16_t double_low_c[] = { 0xd800, 0xd801, 0 }; |
|
plString double_low = plString::FromUtf16(double_low_c); |
|
EXPECT_EQ(0, T_strcmp(unicode_replacement, double_low.GetUnicodeArray().GetData())); |
|
|
|
const uint16_t bad_combo_c[] = { 0xdc00, 0x20, 0 }; |
|
plString bad_combo = plString::FromUtf16(double_low_c); |
|
EXPECT_EQ(0, T_strcmp(unicode_replacement, bad_combo.GetUnicodeArray().GetData())); |
|
|
|
// ISO-8859-1 doesn't have \ufffd, so it uses '?' instead |
|
const plUniChar non_latin1_c[] = { 0x1ff, 0 }; |
|
plStringBuffer<char> non_latin1 = plString::FromUtf32(non_latin1_c).ToIso8859_1(); |
|
EXPECT_STREQ(latin1_replacement, non_latin1.GetData()); |
|
} |
|
|
|
TEST(plString, Concatenation) |
|
{ |
|
// If this changes, this test may need to be updated to match |
|
ASSERT_EQ(16, SSO_CHARS); |
|
|
|
plString expected_short = "xxxxyyy"; |
|
plString input1 = "xxxx"; |
|
plString input2 = "yyy"; |
|
|
|
plString expected_long = "xxxxxxxxxxyyyyyyyyy"; |
|
plString input3 = "xxxxxxxxxx"; |
|
plString input4 = "yyyyyyyyy"; |
|
|
|
// plString + plString |
|
EXPECT_EQ(expected_short, input1 + input2); |
|
EXPECT_EQ(expected_long, input3 + input4); |
|
EXPECT_EQ(input1, input1 + plString()); |
|
EXPECT_EQ(input1, plString() + input1); |
|
|
|
// plstring + const char* |
|
EXPECT_EQ(expected_short, input1 + input2.c_str()); |
|
EXPECT_EQ(expected_short, input1.c_str() + input2); |
|
EXPECT_EQ(expected_long, input3 + input4.c_str()); |
|
EXPECT_EQ(expected_long, input3.c_str() + input4); |
|
EXPECT_EQ(input1, input1 + ""); |
|
EXPECT_EQ(input1, "" + input1); |
|
} |
|
|
|
TEST(plString, Compare) |
|
{ |
|
// Same length, case sensitive |
|
EXPECT_EQ(0, plString("abc").Compare("abc", plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abc").Compare("abd", plString::kCaseSensitive)); |
|
EXPECT_LT(0, plString("abc").Compare("abb", plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abC").Compare("abc", plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("Abc").Compare("abc", plString::kCaseSensitive)); |
|
EXPECT_EQ(0, plString().Compare("", plString::kCaseSensitive)); |
|
|
|
// Same length, case insensitive |
|
EXPECT_EQ(0, plString("abc").Compare("abc", plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abc").Compare("ABC", plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString("abc").Compare("abD", plString::kCaseInsensitive)); |
|
EXPECT_LT(0, plString("abc").Compare("abB", plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString().Compare("", plString::kCaseInsensitive)); |
|
|
|
// Mismatched length, case sensitive |
|
EXPECT_LT(0, plString("abc").Compare("ab", plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abc").Compare("abcd", plString::kCaseSensitive)); |
|
EXPECT_LT(0, plString("abc").Compare("", plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("").Compare("abc", plString::kCaseSensitive)); |
|
|
|
// Mismatched length, case insensitive |
|
EXPECT_LT(0, plString("abc").Compare("Ab", plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString("abc").Compare("Abcd", plString::kCaseInsensitive)); |
|
EXPECT_LT(0, plString("abc").Compare("", plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString().Compare("abc", plString::kCaseInsensitive)); |
|
} |
|
|
|
TEST(plString, CompareN) |
|
{ |
|
// Same length, case sensitive |
|
EXPECT_EQ(0, plString("abcXX").CompareN("abcYY", 3, plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abcXX").CompareN("abdYY", 3, plString::kCaseSensitive)); |
|
EXPECT_LT(0, plString("abcXX").CompareN("abbYY", 3, plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abCXX").CompareN("abcYY", 3, plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("AbcXX").CompareN("abcYY", 3, plString::kCaseSensitive)); |
|
|
|
// Same length, case insensitive |
|
EXPECT_EQ(0, plString("abcXX").CompareN("abcYY", 3, plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abcXX").CompareN("ABCYY", 3, plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString("abcXX").CompareN("abDYY", 3, plString::kCaseInsensitive)); |
|
EXPECT_LT(0, plString("abcXX").CompareN("abBYY", 3, plString::kCaseInsensitive)); |
|
|
|
// Mismatched length, case sensitive |
|
EXPECT_LT(0, plString("abc").CompareN("ab", 3, plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("abc").CompareN("abcd", 4, plString::kCaseSensitive)); |
|
EXPECT_LT(0, plString("abc").CompareN("", 3, plString::kCaseSensitive)); |
|
EXPECT_GT(0, plString("").CompareN("abc", 3, plString::kCaseSensitive)); |
|
|
|
// Mismatched length, case insensitive |
|
EXPECT_LT(0, plString("abc").CompareN("Ab", 3, plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString("abc").CompareN("Abcd", 4, plString::kCaseInsensitive)); |
|
EXPECT_LT(0, plString("abc").CompareN("", 3, plString::kCaseInsensitive)); |
|
EXPECT_GT(0, plString().CompareN("abc", 3, plString::kCaseInsensitive)); |
|
} |
|
|
|
TEST(plString, FindChar) |
|
{ |
|
// Available char, case sensitive |
|
EXPECT_EQ(0, plString("Aaaaaaaa").Find('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(0, plString("AaaaAaaa").Find('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("aaaaAaaa").Find('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(7, plString("aaaaaaaA").Find('A', plString::kCaseSensitive)); |
|
|
|
// Available char, case insensitive |
|
EXPECT_EQ(0, plString("Abbbbbbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("AbbbAbbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbAbbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbbA").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abbbbbbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abbbabbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbabbb").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbba").Find('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("Abbbbbbb").Find('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("AbbbAbbb").Find('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbAbbb").Find('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbbA").Find('a', plString::kCaseInsensitive)); |
|
|
|
// Unavailable char |
|
EXPECT_EQ(-1, plString("AaaaAaaa").Find('C', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString("caaacaaa").Find('C', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString("AaaaAaaa").Find('C', plString::kCaseInsensitive)); |
|
|
|
// Empty string |
|
EXPECT_EQ(-1, plString().Find('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString().Find('A', plString::kCaseInsensitive)); |
|
} |
|
|
|
TEST(plString, FindLast) |
|
{ |
|
// Available char, case sensitive |
|
EXPECT_EQ(0, plString("Aaaaaaaa").FindLast('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("AaaaAaaa").FindLast('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("aaaaAaaa").FindLast('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(7, plString("aaaaaaaA").FindLast('A', plString::kCaseSensitive)); |
|
|
|
// Available char, case insensitive |
|
EXPECT_EQ(0, plString("Abbbbbbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("AbbbAbbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbAbbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbbA").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abbbbbbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("abbbabbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbabbb").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbba").FindLast('A', plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("Abbbbbbb").FindLast('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("AbbbAbbb").FindLast('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("bbbbAbbb").FindLast('a', plString::kCaseInsensitive)); |
|
EXPECT_EQ(7, plString("bbbbbbbA").FindLast('a', plString::kCaseInsensitive)); |
|
|
|
// Unavailable char |
|
EXPECT_EQ(-1, plString("AaaaAaaa").FindLast('C', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString("caaacaaa").FindLast('C', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString("AaaaAaaa").FindLast('C', plString::kCaseInsensitive)); |
|
|
|
// Empty string |
|
EXPECT_EQ(-1, plString().FindLast('A', plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString().FindLast('A', plString::kCaseInsensitive)); |
|
} |
|
|
|
TEST(plString, FindString) |
|
{ |
|
// Available string, case sensitive |
|
EXPECT_EQ(0, plString("ABCDabcd").Find("ABCD", plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("abcdABCDABCDabcd").Find("ABCD", plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("abcdABCDabcd").Find("ABCD", plString::kCaseSensitive)); |
|
EXPECT_EQ(4, plString("abcdABCD").Find("ABCD", plString::kCaseSensitive)); |
|
|
|
// Available string, case insensitive |
|
EXPECT_EQ(0, plString("ABCDxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCDABCDxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCDxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCD").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("abcdxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxabcdABCDxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxabcdxxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxabcd").Find("ABCD", plString::kCaseInsensitive)); |
|
EXPECT_EQ(0, plString("ABCDxxxx").Find("abcd", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCDabcdxxxx").Find("abcd", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCDxxxx").Find("abcd", plString::kCaseInsensitive)); |
|
EXPECT_EQ(4, plString("xxxxABCD").Find("abcd", plString::kCaseInsensitive)); |
|
|
|
// Unavailable string |
|
EXPECT_EQ(-1, plString("xxxx").Find("ABCD", plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString("xxxx").Find("ABCD", plString::kCaseInsensitive)); |
|
|
|
// Empty string |
|
EXPECT_EQ(-1, plString().Find("AAAA", plString::kCaseSensitive)); |
|
EXPECT_EQ(-1, plString().Find("AAAA", plString::kCaseInsensitive)); |
|
|
|
// Unicode substring |
|
plString haystack; |
|
haystack = plString("xxxx") + plString::FromUtf32(test_data); |
|
EXPECT_EQ(4, haystack.Find(utf8_test_data, plString::kCaseSensitive)); |
|
EXPECT_EQ(4, haystack.Find(utf8_test_data, plString::kCaseInsensitive)); |
|
|
|
haystack = plString::FromUtf32(test_data) + plString("xxxx"); |
|
EXPECT_EQ(0, haystack.Find(utf8_test_data, plString::kCaseSensitive)); |
|
EXPECT_EQ(0, haystack.Find(utf8_test_data, plString::kCaseInsensitive)); |
|
|
|
haystack = plString("xxxx") + plString::FromUtf32(test_data) + plString("xxxx"); |
|
EXPECT_EQ(4, haystack.Find(utf8_test_data, plString::kCaseSensitive)); |
|
EXPECT_EQ(4, haystack.Find(utf8_test_data, plString::kCaseInsensitive)); |
|
} |
|
|
|
//TODO: test regex functions |
|
|
|
TEST(plString, Trim) |
|
{ |
|
EXPECT_EQ(plString("xxx "), plString(" xxx ").TrimLeft(" \t\r\n")); |
|
EXPECT_EQ(plString("xxx\t"), plString("\txxx\t").TrimLeft(" \t\r\n")); |
|
EXPECT_EQ(plString("xxx\r\n"), plString("\r\nxxx\r\n").TrimLeft(" \t\r\n")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").TrimLeft("abc")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").TrimLeft("x")); |
|
|
|
EXPECT_EQ(plString(" xxx"), plString(" xxx ").TrimRight(" \t\r\n")); |
|
EXPECT_EQ(plString("\txxx"), plString("\txxx\t").TrimRight(" \t\r\n")); |
|
EXPECT_EQ(plString("\r\nxxx"), plString("\r\nxxx\r\n").TrimRight(" \t\r\n")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").TrimRight("abc")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").TrimRight("x")); |
|
|
|
EXPECT_EQ(plString("xxx"), plString(" xxx ").Trim(" \t\r\n")); |
|
EXPECT_EQ(plString("xxx"), plString("\txxx\t").Trim(" \t\r\n")); |
|
EXPECT_EQ(plString("xxx"), plString("\r\nxxx\r\n").Trim(" \t\r\n")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").Trim("abc")); |
|
EXPECT_EQ(plString(" xxx "), plString(" xxx ").Trim("x")); |
|
} |
|
|
|
TEST(plString, Substrings) |
|
{ |
|
EXPECT_EQ(plString("AAA"), plString("AAA").Left(3)); |
|
EXPECT_EQ(plString("AAA"), plString("AAAxxxx").Left(3)); |
|
EXPECT_EQ(plString("A"), plString("A").Left(3)); |
|
EXPECT_EQ(plString(""), plString("").Left(3)); |
|
|
|
EXPECT_EQ(plString("AAA"), plString("AAA").Right(3)); |
|
EXPECT_EQ(plString("AAA"), plString("xxxxAAA").Right(3)); |
|
EXPECT_EQ(plString("A"), plString("A").Right(3)); |
|
EXPECT_EQ(plString(""), plString("").Right(3)); |
|
|
|
EXPECT_EQ(plString("AAA"), plString("AAAxxxx").Substr(0, 3)); |
|
EXPECT_EQ(plString("AAA"), plString("xxxxAAA").Substr(4, 3)); |
|
EXPECT_EQ(plString("AAA"), plString("xxAAAxx").Substr(2, 3)); |
|
|
|
EXPECT_EQ(plString(""), plString("AAAA").Substr(2, 0)); |
|
EXPECT_EQ(plString("AA"), plString("AAAA").Substr(2, 4)); |
|
EXPECT_EQ(plString(""), plString("AAAA").Substr(6, 4)); |
|
EXPECT_EQ(plString("AAAA"), plString("AAAA").Substr(0, 4)); |
|
EXPECT_EQ(plString(""), plString("").Substr(0, 4)); |
|
|
|
// Negative indexes start from the right |
|
EXPECT_EQ(plString("AAA"), plString("xxxxAAA").Substr(-3, 3)); |
|
EXPECT_EQ(plString("AAA"), plString("xxAAAxx").Substr(-5, 3)); |
|
EXPECT_EQ(plString("AAA"), plString("xxxxAAA").Substr(-3, 6)); |
|
EXPECT_EQ(plString("AAA"), plString("AAAxxxx").Substr(-10, 3)); |
|
EXPECT_EQ(plString("AAA"), plString("AAA").Substr(-10, 10)); |
|
} |
|
|
|
TEST(plString, Replace) |
|
{ |
|
EXPECT_EQ(plString("xxYYxx"), plString("xxAAxx").Replace("A", "Y")); |
|
EXPECT_EQ(plString("xxAAxx"), plString("xxAAxx").Replace("XX", "Y")); |
|
EXPECT_EQ(plString("xxxx"), plString("xxAAxx").Replace("A", "")); |
|
EXPECT_EQ(plString("xxREPLACExx"), plString("xxFINDxx").Replace("FIND", "REPLACE")); |
|
EXPECT_EQ(plString("xxREPLACExxREPLACExx"), plString("xxFINDxxFINDxx").Replace("FIND", "REPLACE")); |
|
|
|
EXPECT_EQ(plString("YYxx"), plString("AAxx").Replace("A", "Y")); |
|
EXPECT_EQ(plString("AAxx"), plString("AAxx").Replace("XX", "Y")); |
|
EXPECT_EQ(plString("xx"), plString("AAxx").Replace("A", "")); |
|
EXPECT_EQ(plString("REPLACExx"), plString("FINDxx").Replace("FIND", "REPLACE")); |
|
EXPECT_EQ(plString("REPLACExxREPLACExx"), plString("FINDxxFINDxx").Replace("FIND", "REPLACE")); |
|
|
|
EXPECT_EQ(plString("xxYY"), plString("xxAA").Replace("A", "Y")); |
|
EXPECT_EQ(plString("xxAA"), plString("xxAA").Replace("XX", "Y")); |
|
EXPECT_EQ(plString("xx"), plString("xxAA").Replace("A", "")); |
|
EXPECT_EQ(plString("xxREPLACE"), plString("xxFIND").Replace("FIND", "REPLACE")); |
|
EXPECT_EQ(plString("xxREPLACExxREPLACE"), plString("xxFINDxxFIND").Replace("FIND", "REPLACE")); |
|
|
|
EXPECT_EQ(plString("YY"), plString("AA").Replace("A", "Y")); |
|
EXPECT_EQ(plString("AA"), plString("AA").Replace("XX", "Y")); |
|
EXPECT_EQ(plString(""), plString("AA").Replace("A", "")); |
|
EXPECT_EQ(plString("REPLACE"), plString("FIND").Replace("FIND", "REPLACE")); |
|
EXPECT_EQ(plString("REPLACExxREPLACE"), plString("FINDxxFIND").Replace("FIND", "REPLACE")); |
|
} |
|
|
|
TEST(plString, CaseConvert) |
|
{ |
|
/* Edge cases: |
|
* '@' = 'A' - 1 |
|
* '[' = 'Z' + 1 |
|
* '`' = 'a' - 1 |
|
* '{' = 'z' + 1 |
|
*/ |
|
|
|
EXPECT_EQ(plString("AAZZ"), plString("aazz").ToUpper()); |
|
EXPECT_EQ(plString("AAZZ"), plString("AAZZ").ToUpper()); |
|
EXPECT_EQ(plString("@AZ[`AZ{"), plString("@AZ[`az{").ToUpper()); |
|
EXPECT_EQ(plString(""), plString("").ToUpper()); |
|
|
|
EXPECT_EQ(plString("aazz"), plString("aazz").ToLower()); |
|
EXPECT_EQ(plString("aazz"), plString("AAZZ").ToLower()); |
|
EXPECT_EQ(plString("@az[`az{"), plString("@AZ[`az{").ToLower()); |
|
EXPECT_EQ(plString(""), plString("").ToLower()); |
|
} |
|
|
|
TEST(PlStringTest,Tokenize) |
|
{ |
|
std::vector<plString> expected; |
|
expected.push_back(plString("a")); |
|
expected.push_back(plString("b")); |
|
expected.push_back(plString("c")); |
|
expected.push_back(plString("d")); |
|
expected.push_back(plString("è")); |
|
|
|
plString input = plString("a\t\tb\n;c-d;è"); |
|
std::vector<plString> output = input.Tokenize("\t\n-;"); |
|
EXPECT_EQ(expected,output); |
|
|
|
} |
|
|
|
TEST(PlStringTest,Split) |
|
{ |
|
std::vector<plString> expected; |
|
expected.push_back(plString("a")); |
|
expected.push_back(plString("b")); |
|
expected.push_back(plString("c")); |
|
expected.push_back(plString("d")); |
|
expected.push_back(plString("è")); |
|
|
|
plString input = plString("a-b-c-d-è"); |
|
std::vector<plString> output = input.Split("-",4); |
|
EXPECT_EQ(expected,output); |
|
|
|
} |
|
|
|
TEST(PlStringTest,Fill) |
|
{ |
|
plString expected = plString("aaaaa"); |
|
plString output = plString::Fill(5,'a'); |
|
EXPECT_EQ(expected,output); |
|
} |
|
|
|
|