□□□□□□□ □□ □□□

□□□□□ □□□□□□□□
□□□□□□ □□□□□□□□ □□□□□□□□□□□ □□□□□□□□
□□□□□□□□□ □□□□□□ □□□
Unicode in C++
James McNellis (@JamesMcNellis)
Senior Software Development Engineer
Microsoft Visual C++
Before there was Unicode…
Single-Byte Encodings
ASCII

!"#$%&'()*+,-./0123456789:;<=>?
@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\]^_
`abcdefghijklmnopqrstuvwxyz{|}~
 ASCII

H e l l o ! \0
↓ ↓ ↓ ↓ ↓ ↓ ↓
48 65 6C 6C 6F 21 00
ASCII

 A 7-bit character encoding

 32 control characters
 95 printable characters
 Near-ubiquitous, but often with substitutions or extensions
 Great for English; not so great for most other languages
Extended ASCII

 IBM code pages, e.g. CP437 (“Latin US”), CP737 (“Greek”)

 Mac OS Roman
 DEC Multinational Character Set
 ISO/IEC 8859
 …and many more
Code page 437

ÇüéâäàåçêëèïîìÄÅÉæÆôöòûùÿÖÜ¢£¥₧ƒ
áíóúñѪº¿⌐¬½¼¡«»░▒▓│┤╡╢╖╕╣║╗╝╜╛┐
└┴┬├─┼╞╟╚╔╩╦╠═╬╧╨╤╥╙╘╒╓╫╪┘┌█▄▌▐▀
αßΓπΣσµτΦΘΩδ∞φε∩≡±≥≤⌠⌡÷≈°∙∙√ⁿ²■
ISO/IEC 8859

 Latin-1 Western European  Latin-6 Nordic

 Latin-2 Central European  Latin/Thai
 Latin-3 South European  Latin-7 Baltic Rim
 Latin-4 North European  Latin-8 Celtic
 Latin/Cyrillic  Latin-9 (revision of Latin-1)
 Latin/Arabic  Latin-10 South-Eastern European
 Latin/Greek
 Latin/Hebrew
 Latin-5 Turkish
ISO/IEC 8859-1 (“Latin 1”)

¡¢£¤¥¦§¨©ª«¬®¯°±²³´µ¶·¸¹º»¼½¾¿
ÀÁÂÃÄÅÆÇÈÉÊËÌÍÎÏÐÑÒÓÔÕÖ×ØÙÚÛÜÝÞß
àáâãäåæçèéêëìíîïðñòóôõö÷øùúûüýþÿ
ISO/IEC 8859-5 (“Latin/Cyrillic”)

ЁЂЃЄЅІЇЈЉЊЋЎЏАБВГДЕЖЗИЙКЛМНОП
РСТУФХЦЧШЩЪЫЬЭЮЯабвгдежзийклмноп
рстуфхцчшщъыьэюя№ёђѓєѕіїјљњћќ§ўџ
ISO/IEC 8859-5 (“Latin/Cyrillic”)

А л л о \0
↓ ↓ ↓ ↓ ↓
B0 DB DB DE 00
ISO/IEC 8859-5 (“Latin/Cyrillic”)

B0 DB DB DE 00
↓ ↓ ↓ ↓ ↓
А л л о \0
But You Have to be Careful…

B0 DB DB DE 00
↓ ↓ ↓ ↓ ↓
° Ϋ Ϋ ή \0
Single-Byte Encodings

 Many nice properties…

 Each character is the same size
 The encodings are compact
 String operations are generally straightforward
 …but there aren’t enough code points to represent all characters
 Different encodings can be used for different sets of characters…
 …but this doesn’t work for all languages,
 …and it makes text interchange difficult
Variable-Length Encodings
Shift-JIS

 Some characters are representable using a single byte; others require two bytes
 Two-byte characters consist of a lead byte and a trail byte
 The lead byte will always have the high bit set; the trail byte may have any value
 Starts from 7-bit ASCII, with a couple of substitutions (replaces \ with ¥ and ~ with ‾)
 The encoding form is very complex due to overlap
Shift-JIS
http://en.wikipedia.org/wiki/Shift_JIS
Shift-JIS

44 → D (Latin Capital D)
84 44 → Д (Cyrillic Capital De)
84 84 → т (Cyrillic Capital Te)
Shift-JIS

44 44 84 84 84 84 84 44
↓ ↓ ↓ ↓ ↓
D D т т Д
Shift-JIS

… 84 84 84 84 84 84 84 …
↑
p
Shift-JIS

 Note that this is just one technique; there are others

 E.g., using escape sequences to switch between “working sets” of characters
 See, for example, ISO/IEC 2022
Multi-Byte Encodings

 Substantially expanded code space

 With two bytes, up to 32,896 characters theoretically representable…
 …though real-world encodings tend to have lower limits
 Some byte-oriented string operations still work, e.g. strcpy
 Many disadvantages:
 Complex to parse (though it does depend on the particular encoding)
 Some common string operations require a linear scan over the string
 Many simple string operations become difficult or require special APIs
 Still have the problem that there are many different encoding standards
Unicode 1.0
Unicode Design Goals

 One Character Encoding to Rule Them All…

 Universal: Must be able to represent all characters likely to be used in text interchange
 Efficient: Plain text should be simple to parse
 Unambiguous: Any given Unicode code point always represents the same character

 A set of characters, not glyphs (so, not concerned with visual representation)
Unicode 1.0

 Each character is mapped to a 16-bit code point

 Up to 65,536 characters can be represented (in theory; fewer in practice)
 All of the characters in ISO/IEC 8859-1 (“Latin- 1”) map to the same code points
 The UCS-2 encoding is used: each code point is mapped to a single 16-bit code unit
 Unicode 1.0

H e l l o ! \0
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
 Unicode 1.0

H e l l o ! \0
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
 Unicode 1.0

X Δ Ж ヸ ᠼ ☃ ‫ڳ‬
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0058 U+0394 U+0436 U+30F8 U+183C U+2603 U+06B3
UCS-2

00 48 00 65 00 6C 00 6C 00 6F 00 21 00 00
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
H e l l o ! \0
UCS-2

00 48 00 65 00 6C 00 6C 00 6F 00 21 00 00
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+4800 U+6500 U+6C00 U+6C00 U+6F00 U+2100 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
䠀 攀 氀 氀 漀 ℀ \0
 UCS-2

FE FF 00 48 00 65 00 6C 00 6C 00 6F 00 21 00 00
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
U+FEFF U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
H e l l o ! \0
 UCS-2

FF FE 48 00 65 00 6C 00 6C 00 6F 00 21 00 00 00
↓ ↓ ↓ ↓ ↓ ↓ ↓ ↓
U+FEFF U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
H e l l o ! \0
UCS-2

 Advantages:
 A huge number of characters are representable
 Characters from different scripts are easily combinable in a single string
 Each code point is representable using a single code unit, so the encoding is simple

 Disadvantages:
 Multiple possible byte orderings, so byte order mark (BOM) is required for interchange
 Every character requires two bytes, so text for many languages require twice as much storage
 None of the byte-oriented string functions (like strcpy) work with UCS-2 strings
Code Space Usage
Code Space Usage

“With over 30,000 unallocated character

positions, the Unicode character encoding
provides sufficient space for foreseeable
future expansion.”
– Unicode 1.0 Introduction
 20,000
40,000
80,000
100,000
120,000

0
60,000
1.0 (1991)

2.0 (1996)

3.0 (1999)
Number of Characters

4.0 (2003)

5.0 (2006)

6.0 (2010)

7.0 (2014?)
Unicode Encodings Today
Expanding the Code Space

 It turns out we need to be able to encode more than 65,536 characters

 In Unicode 2.0, the code space was expanded

 Total code points: 1,112,064
 Maximum number of characters: 1,111,998

 We can’t store that many distinct values using 16 bits, so the encoding has to change too
 If 16 bits isn’t enough space, how about 32 bits?
UTF-32

00 00 00 48 00 00 00 69 00 00 00 21
↓ ↓ ↓
U+0048 U+0065 U+0021
↓ ↓ ↓
H i !
UTF-32

00 00 FE FF 00 00 00 48 00 00 00 69 00 00 00 21
↓ ↓ ↓ ↓
U+FEFF U+0048 U+0065 U+0021
↓ ↓ ↓
H i !
UTF-32

FF FE 00 00 48 00 00 00 69 00 00 00 21 00 00 00
↓ ↓ ↓ ↓
U+FEFF U+0048 U+0065 U+0021
↓ ↓ ↓
H i !
UTF-32

FF FE 00 00 48 00 00 00 69 00 00 00 30 F4 01 00
↓ ↓ ↓ ↓
U+FEFF U+0048 U+0065 U+1F430
↓ ↓ ↓
H i 🐰
UTF-32

 Advantages:
 A huge number of characters are representable
 Each code point is representable using a single code unit, so the encoding is simple
 …this isn’t all that useful in most practical string usage; we’ll see why later…

 Disadvantages:
 Two possible byte orderings, so byte order mark (BOM) is required
 Every character requires four bytes; wasting at least 11 bits per code point
 None of the byte-oriented string functions (like strcpy) work with UTF-32 strings
 Nothing that was written to work with UCS-2 works with UTF-32
 UTF-8

Bits of First Code Last Code Bytes in

Byte 1 Byte 2 Byte 3 Byte 4
Code Point Point Point Sequence

7 U+0000 U+007F 1 0xxxxxxx

11 U+0080 U+07FF 2 110xxxxx 10xxxxxx

16 U+08000 U+FFFF 3 1110xxxx 10xxxxxx 10xxxxxx

21 U+10000 U+10FFFF 4 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx

UTF-8

48 65 6C 6C 6F 21 00
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0048 U+0065 U+006C U+006C U+006F U+0021 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
H e l l o ! \0
 UTF-8

58 CE 94 D0 B6 E3 83 B8 E1 A0 BC F0 9F 90 B0 00
↓ ↓ ↓ ↓ ↓ ↓ ↓
U+0058 U+0394 U+0436 U+30F8 U+183C U+1F430 U+0000
↓ ↓ ↓ ↓ ↓ ↓ ↓
X Δ Ж ヸ ᠼ 🐰 \0
 UTF-8

Bits of First Code Last Code Bytes in

Byte 1 Byte 2 Byte 3 Byte 4
Code Point Point Point Sequence

7 U+0000 U+007F 1 0xxxxxxx

11 U+0080 U+07FF 2 110xxxxx 10xxxxxx

16 U+08000 U+FFFF 3 1110xxxx 10xxxxxx 10xxxxxx

21 U+10000 U+10FFFF 4 11110xxx 10xxxxxx 10xxxxxx 10xxxxxx

UTF-8

 Advantages:
 ASCII text has the same representation in UTF-8
 No byte order mark (BOM) is required, though there is an optional BOM (EF BB BF)
 Many byte-oriented string functions (strcpy, strcat, strlen, etc.) work with UTF-8 strings
 UTF-8 encoded text requires less storage than 16-bit and 32-bit encodings for most languages

 Disadvantages:
 It’s a variable-width encoding
 Text for a few languages requires more storage than 16-bit encodings
UTF-16

 UCS-2 is obsolete since it can’t encode all code points

 But many large systems and libraries were built using the 16-bit UCS-2 encoding
 UTF-16 supersedes UCS-2 and provides a 16-bit Unicode encoding

 For code points in [U+0000, U+FFFF], UTF-16 is equivalent to UCS-2

 Valid UCS-2 text is also valid UTF-16 text
 These are the most commonly used code points, called the Basic Multilingual Plane (BMP)

 2,048 code points in [U+D800, U+DFFF] are reserved for UTF-16

 These are used to encode code points outside of the Basic Multilingual Plane
 Code points in [U+10000, U+10FFFF] encoded using surrogate pairs
UTF-16

How do we encode code points in [U+10000, U+10FFFF]?

1. Subtract 0x010000 from the code point; this yields a 20 bit number in [0x0, 0xFFFFF]

2. Split the number in half, with the upper ten bits in one half; the lower ten bits in the second

3. The upper ten bits are added to 0xD800 to form the lead surrogate

4. The lower ten bits are added to 0xDC00 to form the trail surrogate
UTF-16

🍸
0x1F378
- 0x10000
0x0F378 → 0b00001111001101111000

0b00001111001101111000
↓ ↓
0x003C 0x0378
+ 0xD800 0xDC00
0xD83C 0xDF78
UTF-16

 Advantages:
 Some level of compatibility with UCS-2 and systems designed to use a 16-bit encoding
 Text for a few languages is smaller in UTF-16 than in UTF-8

 Disadvantages:
 Multiple possible byte orderings, so byte order mark (BOM) is required for interchange
 Every character requires at least two bytes
 None of the byte-oriented string functions (like strcpy) work with UTF-16 strings
 It’s a variable-width encoding, but it’s often misused as a fixed-width encoding
 Binary string comparison for UTF-16 produces different results than for UTF-8 and UTF-32
UTF-8, UTF-16, UTF-32

 UTF-8
 More compact than UTF-32; usually more compact than UTF-16
 It’s a variable width encoding, so it’s more complex
 By far the most commonly used for storage and data transmission
 Dominant character encoding on the Internet

 UTF-32
 Simple, fixed width encoding
 Lots of wasted space because of the large 32-bit code units
 UTF-8 vs. UTF-16 Text Size

pnb 1443 1204 119.85 % // Panjabi, Western cmn 1563663 1055924 148.08 % // Mandarin Chinese
kor 67522 55332 122.03 % // Korean yue 115488 77874 148.30 % // Yue Chinese
mal 76590 60468 126.66 % // Malayalam wuu 150945 101588 148.59 % // Wu Chinese
tel 2642 2080 127.02 % // Telugu tha 18745 12610 148.65 % // Thai
hin 455546 349438 130.37 % // Hindi lzh 73285 48982 149.62 % // Literary Chinese
mar 1332509 1008604 132.11 % // Marathi bod 3925 2622 149.69 % // Tibetan
npi 460 348 132.18 % // Nepali jpn 9778643 6524720 149.87 % // Japanese
ben 27733 20802 133.32 % // Bengali
san 1465 1094 133.91 % // Sanskrit
kat 29970 22344 134.13 % // Georgian
ain 546 384 142.19 % // Ainu (Japan)
lao 1576 1084 145.39 % // Lao
khm 22676 15348 147.75 % // Central Khmer
Dynamic Composition
Dynamic Composition

ÀÁÂÃÄÅ
Dynamic Composition

A + ̈ → Ä
U+0041 U+0308
Dynamic Composition

e + ̃ + ̽ + ̪ → ̽e̪ ̃
U+0065 U+0303 U+033D U+032A
Dynamic Composition

e + ̃ + ̽ + ̪ → ẽ̪̽
U+0065 U+0303 U+033D U+032A

e + ̽ + ̃ + ̪ → ẽ̪̽
U+0065 U+033D U+0303 U+032A
 ◔_◔
2 x U+25D4 (CIRCLE WITH UPPER RIGHT
QUADRANT BLACK); ROLLING EYES

Unicode Types in C++

Unicode Code Unit Types

Code Unit Type

UTF-8 char
UTF-16 char16_t
UTF-32 char32_t
Unicode String Literals

char const a[]{u8"Hello, \u2603!"};

char const b[]{u8"Hello, ☃!"};
// sizeof(a) == sizeof(b) == 12

char16_t const c[]{u"Hello, \u2603!"};

char16_t const d[]{u"Hello, ☃!"};
// sizeof(c) == sizeof(d) == 20

char32_t const e[]{U"Hello, \u2603!"};

char32_t const f[]{U"Hello, ☃!"};
// sizeof(e) == sizeof(f) == 40
Unicode Character Literals (Sort Of)

char a{'x'}; char a{'\u0078'};

char b{'☃'}; // Wrong char b{'\u2603'}; // Wrong
char c{'🍸'}; // Wrong char c{'\U0001F378'}; // Wrong

char16_t d{u'x'}; char16_t d{u'\u0078'};

char16_t e{u'☃'}; char16_t e{u'\u2603'};
char16_t f{u'🍸'}; // Wrong char16_t f{u'\U0001F378'}; // Wrong

char32_t g{U'x'}; char32_t g{U'\U00000078'};

char32_t h{U'☃'}; char32_t h{U'\U00002603'};
char32_t i{U'🍸'}; char32_t i{U'\U0001F378'};
The Amazing C++ Unicode String Type

 [This slide intentionally left blank.]

std::basic_string specialization typedefs

Code Unit Type String Type

UTF-8 char std::string
UTF-16 char16_t std::u16string
UTF-32 char32_t std::u32string
std::basic_string

std::string a{u8"Hello, \u2603!"};

std::string b{u8"Hello, ☃!"};

std::u16string c[]{u"Hello, \u2603!"};

std::u16string d[]{u"Hello, ☃!"};

std::u32string e[]{U"Hello, \u2603!"};

std::u32string f[]{U"Hello, ☃!"};
std::basic_string

 You must remember that it’s just a dumb sequence of code units
 NOT code points, characters, text elements, grapheme clusters, etc.
 Iterators (using begin() and end()) iterate over code units
 size() does not return the number of code points; it returns the number of code units
 Functions like front, back, operator[], push_back, find(CharT) overloads operate on code units
std::basic_string

std::string a{"Hello, "};

// Wrong:
a.push_back('☃'); // U+2603

// Right:
char const snowman[]{u8"☃"}; // U+2603
a.insert(a.end(), begin(snowman), end(snowman));
std::basic_string

std::u16string a{u"A glass: "};

// Wrong:
a.push_back(u'🍸'); // U+1F378

// Right:
char16_t const glass[]{u"🍸"}; // U+1F378
a.insert(a.end(), begin(glass), end(glass));
std::basic_string

std::u32string a{U"A glass: "};

// Right:
a.push_back(U'🍸'); // U+1F378
“Length”
U+1F4CF (STRAIGHT RULER)
String Length

"Hello"
↑↑↑↑↑
12345
 String Length

"1 Ä 🍸"
1 A ¨ 🍸
U+0031 U+0041 U+0308 U+1F378

UTF-8 31 41 CC 88 F0 9F 8D B8
UTF-16 0031 0041 0308 D83C DF78
UTF-32 00000031 00000041 00000308 0001F378
 String Length: Number of Bytes

"1 Ä 🍸"
1 A ¨ 🍸
U+0031 U+0041 U+0308 U+1F378
Length
UTF-8 31 41 CC 88 F0 9F 8D B8 8
UTF-16 0031 0041 0308 D83C DF78 10
UTF-32 00000031 00000041 00000308 0001F378 16
 String Length: Number of Code Units

"1 Ä 🍸"
1 A ¨ 🍸
U+0031 U+0041 U+0308 U+1F378
Length
UTF-8 31 41 CC 88 F0 9F 8D B8 8
UTF-16 0031 0041 0308 D83C DF78 5
UTF-32 00000031 00000041 00000308 0001F378 4
 String Length: Number of Code Points

"1 Ä 🍸"
1 A ¨ 🍸
U+0031 U+0041 U+0308 U+1F378
Length
UTF-8 31 41 CC 88 F0 9F 8D B8 4
UTF-16 0031 0041 0308 D83C DF78 4
UTF-32 00000031 00000041 00000308 0001F378 4
 String Length: Number of Text Elements

"1 Ä 🍸"
1 A ¨ 🍸
U+0031 U+0041 U+0308 U+1F378
Length
UTF-8 31 41 CC 88 F0 9F 8D B8 3
UTF-16 0031 0041 0308 D83C DF78 3
UTF-32 00000031 00000041 00000308 0001F378 3
String Length

 At least four possible meanings of “string length”

 Number of bytes
 Number of code units
 Number of code points
 Number of text elements
 The # of code points and the # of text elements is the same regardless of encoding
 For UTF-8, the # of bytes is the same as the # of code units
 For UTF-32, the # of code units is the same as the # of code points
String Length

std::string s{u8"1Ä🍸"};

std::size_t number_of_code_units{s.size()};

for (auto&& c : s)
{
// Iterating over code units, not characters
}
“Equality”
U+2248 (ALMOST EQUAL TO)
Representational Equality

std::string a{"Hello, Aspen!"};

std::string b{"Hello, Aspen!"};

if (a == b)
{
std::cout << "The strings are equal.\n";
}
Representational Equality

std::string a{u8"1Ä🍸"};
std::string b{u8"1Ä🍸"};

if (a == b)
{
std::cout << "The strings are equal.\n";
}
Multiple Representations

Latin Capital Letter A; Combining Diaeresis

U+0041 U+0308

Ä
A + ¨
Ä
U+00C4
Latin Capital Letter A With Diaeresis
Multiple Representations

Latin Capital Letter A

U+0041

A
A
Ａ
U+FF21
Fullwidth Latin Capital Letter A
Multiple Representations

Latin Small Letter F; Latin Small Letter I

U+0066 U+0069

fi
f + i
fi
U+FB01
Latin Small Ligature Fi
Multiple Representations

Latin Capital Letter X; Latin Capital Letter I (x2)

U+0058 U+0049 U+0049

Ⅻ
X + I + I
Ⅻ
U+216B
Roman Numeral Twelve
Multiple Representations

Latin Capital Letter A

Combining Equals Sign Below

Â͇ Combining Circumflex Accent

U+0041 U+0347 U+0302

U+0041 U+0302 U+0347
Equivalent Sequences

Unicode 6.3 §2.12

Normalization

 Four Normalization Forms

 NFC: Canonical Composed
 NFD: Canonical Decomposed
 NFKC: Compatibility Composed
 NFKD: Compatibility Decomposed
 Normalization includes a well-defined ordering for combining marks
Normalization Using the Standard Library

 [This slide intentionally left blank.]

Ordering
U+003C (LESS-THAN SIGN)
Ordering

std::string a{"a"};
std::string b{"b"};

if (a < b)
{
std::cout << "a is ordered before b.\n";
}
Ordering

std::string bear {u8"🐻"};

std::string whale{u8"🐳"};

if (bear < whale)

{
std::cout << "the bear is ordered before the whale.\n";
}
Ordering

std::vector<std::string> v{"c", "X", "b", "Y", "a", "Z"};

// Alphabetize the strings...

std::sort(v.begin(), v.end());

for (auto&& s : v) { std::cout << s << ' '; }

Ordering

std::vector<std::string> v{"c", "X", "b", "Y", "a", "Z"};

// Alphabetize the strings...

std::sort(v.begin(), v.end());

for (auto&& s : v) { std::cout << s << ' '; }

// Output: X Y Z a b c
String Collation

std::vector<std::string> v{"c", "X", "b", "Y", "a", "Z"};

std::locale en_us{"en-US"};
std::sort(v.begin(), v.end(), en_us);

for (auto&& s : v) { std::cout << s << ' '; }

// Output: a b c X Y Z
String Collation

std::vector<std::string> v{"z", "ä", "b", "a"};

std::locale en_us{"de-DE"};
std::sort(v.begin(), v.end(), en_us);

for (auto&& s : v) { std::cout << s << ' '; }

// Output: a ä b z
String Collation

std::vector<std::string> v{"z", "ä", "b", "a"};

std::locale en_us{"sv-SE"};
std::sort(v.begin(), v.end(), en_us);

for (auto&& s : v) { std::cout << s << ' '; }

// Output: a b z ä
Unicode Collation using the Standard Library

 [This slide intentionally left blank.]

Other Text Operations
Text Manipulation

std::locale loc{"en-US"};

char lowercase_a{'a'};
char uppercase_a{std::toupper(lowercase_a, loc)}; // A
Text Manipulation

 Most Standard Library manipulation functions (like toupper) are code unit based
 But Unicode text manipulation does not work well with this model
 Consider ß in German: the uppercase form is SS, which is two characters
Text Manipulation

Even the classification functions are unusable…

template <typename CharT> bool isspace(CharT c, locale const& loc);

template <typename CharT> bool isalpha(CharT c, locale const& loc);
template <typename CharT> bool isalnum(CharT c, locale const& loc);

…except possibly for UTF-32 (char32_t).

Encoding Conversions using <codecvt>

std::string utf8{u8"1Ä🍸"};

std::wstring_convert<
std::codecvt_utf8<char32_t>, char32_t
> utf32_converter;

std::u32string utf32{utf32_converter.from_bytes(utf8)};
I/O Conversions using <codecvt>

// Write UTF-8 data with BOM:

std::ofstream("text.txt")
<<"\xef\xbb\xbf\x31\x41\xcc\x88\xf0\x9f\x8d\xb8";

// Read the UTF8 file, skipping BOM:

std::basic_ifstream<char32_t> f("text.txt");
f.imbue(std::locale(f.getloc(),
new std::codecvt_utf8<char32_t, 0x10ffff, std::consume_header>));

for (char32_t c{}; f.get(c); ) { std::cout << std::hex << c << ' '; }
// Output: 31 41 308 1f378
International Components for
Unicode (ICU)
International Components for Unicode

 Huge set of features—probably everything you’d ever need

 Very widely-used, so it’s well-tested in a lot of real-world software
 Runs on many platforms; has permissive license

 It has its own string type, UnicodeString, which uses UTF-16

 It doesn’t work well with other string types (like std::string) in most contexts
 It’s not “modern C++” by any definition of that term
 Originally written for Java; ported to C++
UChar32 and UChar

// UTF-32 code units

UChar32 rook_utf32{static_cast<UChar32>(U'♖')}; // U+2656
UChar32 cake_utf32{static_cast<UChar32>(U'🍰')}; // U+1F370

// UTF-16 code units

UChar rook_utf16{u'♖'}; // U+2656
UChar cake_utf16[2]{0xD83C, 0xDF70}; // U+1F370

// UTF-8 code units

char rook_utf8[4]{u8"♖"}; // U+2656
char cake_utf8[5]{u8"🍰"}; // U+1F370
UnicodeString

UChar32 cake_utf32c{static_cast<UChar32>(U'🍰')};
UChar32 cake_utf32s[2]{static_cast<UChar32>(U'🍰'), 0};
UChar cake_utf16 [3]{0xD83C, 0xDF70, 0};
char cake_utf8 [5]{u8"🍰"};

UnicodeString cake_1{cake_utf32c};
UnicodeString cake_2{UnicodeString::fromUTF32(cake_utf32s, 1)};
UnicodeString cake_2{cake_utf16};
UnicodeString cake_3{UnicodeString::fromUTF8(cake_utf8)};
UnicodeString

 compare, compareCodePointOrder, caseCompare

 length, countChar32
 startsWith, endsWith, indexOf, lastIndexOf
 append, insert, replace, findAndReplace, reverse
 padLeading, pad, padTrailing
 toUpper, toLower, toTitle
 toUTF8String, toUTF32
Normalization

UErrorCode status{U_ZERO_ERROR};
Normalizer2 const& normalizer{*icu::Normalizer2::getInstance(
nullptr, "nfc", UNORM2_COMPOSE, status)};

// U+0041 U+0308 (Latin Capital Letter A, Combining Diaeresis)

UnicodeString source{UnicodeString::fromUTF8(u8"\u0041\u0308")};
bool is_normalized{normalizer.isNormalized(source, status)};

UnicodeString result{};
normalizer.normalize(source, result, status);
// U+00C4 (Latin Capital Letter A with Diaeresis)
Collation

UErrorCode status{U_ZERO_ERROR};
std::unique_ptr<Collator> collator{
Collator::createInstance(Locale("en", "US"), status)};

UnicodeString a{UnicodeString::fromUTF8(u8"a")};
UnicodeString z{UnicodeString::fromUTF8(u8"Z")};

UCollationResult result{collator->compare(a, z, status)};

Regular Expressions

UErrorCode status{U_ZERO_ERROR};
std::unique_ptr<RegexMatcher> matcher{new RegexMatcher{
UnicodeString::fromUTF8(u8R"(\p{Number})"), 0, status}};

matcher->reset(UnicodeString::fromUTF8(u8"Ⅻ⅝"));
while (matcher->find())
{
// Handle match
}
Boost.Locale and Boost.Regex
Boost.Locale

 Introduced in Boost 1.48

 A modern C++ API that can use various backends (with varying levels of feature support)
 ICU
 std (C++ Standard Library, plus workarounds for common bugs)
 posix
 winapi
 Not nearly as feature-filled as ICU, but supports most commonly used features
Boost.Locale

boost::locale::generator gen{};

std::locale loc{gen.generate("en-US.UTF-8")};
Normalization

// Latin Capital Letter A With Diaeresis

std::string a{u8"\u00C4"};
std::string b{u8"A\u0308"};

std::string a_decomp{boost::locale::normalize(a, norm_nfd, loc)};

std::string b_comp {boost::locale::normalize(b, norm_nfc, loc)};

assert(a_decomp == b);
assert(b_comp == a);
Collation

std::string a{u8"A"};
std::string accented_a{u8"A\u0308"};

int result{std::use_facet<boost::locale::collator<char>>(loc)
.compare(boost::locale::collator_base::primary, a, accented_a)};

typedef std::set<std::string, boost::locale::comparator<char>> set_t;

set_t s{boost::locale::comparator<char>{
loc, boost::locale::collator_base::primary}};
Collation Comparison Levels

Supports all five Unicode comparison levels:

 L1 (Primary): Ignores case and accents; compares base characters only
 L2 (Secondary): Ignores case but considers accents
 L3 (Tertiary): Considers both case and accents
 L4 (Quaternary): Considers case, accents, and punctuation
 Ln (Identical): Considers case, accents, and punctuation; requires identical code points
Conversions

std::string a{u8"1Ä🍸"};
std::u16string b{boost::locale::conv::utf_to_utf<char16_t>(a)};
std::u32string c{boost::locale::conv::utf_to_utf<char32_t>(a)};
std::u32string d{boost::locale::conv::utf_to_utf<char32_t>(b)};
std::string e{boost::locale::conv::utf_to_utf<char >(b)};
Conversions

// А л л о
std::string s{"\xB0\xDB\xDB\xDE"};
std::string utf8_s{boost::locale::conv::to_utf<char>(s, "ISO-8859-5")};
Text Manipulation

 Provides string-based case manipulation (to_upper, to_lower, to_title, fold_case)

Boundary Analysis

namespace ba = boost::locale::boundary;

std::string subject{u8"1Ä🍸"};

ba::segment_index<std::string::const_iterator> map(
ba::character, subject.begin(), subject.end(), loc);

size_t const byte_length{subject.size()}; // 8

size_t const text_length{std::distance(map.begin(), map.end())}; // 3
Boost.Regex

boost::u32regex r{boost::make_u32regex(u8R"(\p{Number})")};

std::string subject{u8"Ⅻ⅝"};

typedef boost::u32regex_token_iterator<...> iterator_type;

iterator_type first{boost::make_u32regex_token_iterator(subject, r)};
iterator_type last{};
for (auto it(first); it != last; ++it)
{
// Process matches
}
Proposals for Standardization
N3336: Adapting Standard Library Strings
and I/O to a Unicode World

 www.open-std.org/jtc1/sc22/wg21/docs/papers/2012/n3336.html

 Retrofit some new features onto existing Standard Library stuff

 New overloads of basic_string constructors, operator=, etc., to do encoding conversions
 New overloads of formatted I/O operators for I/O streams to handle different string types
 New encoding conversion iterators
N3572: Unicode Support in the Standard
Library

 www.open-std.org/jtc1/sc22/wg21/docs/papers/2013/n3572.html

 Proposes new Unicode-aware functionality for the Standard Library

 A new encoded_string type that supports various Unicode and other encodings
 Provides a slimmed-down std::string-like interface
 All operations are on code points, not code units
 Including iteration, size(), push_back(), etc.
 Iterators are bidirectional only
 Operators to support mixing/matching different encoded_string specializations
 Hashing and comparison produce correct results for canonically equivalent strings
Resources 📖
U+1F4D6 (OPEN BOOK)

 Unicode Explained, by Jukka K. Korpela: http://shop.oreilly.com/product/9780596101213.do

 The Unicode Standard: http://www.unicode.org/standard/standard.html
 International Components for Unicode (ICU): http://site.icu-project.org/
 Boost.Locale: http://www.boost.org/doc/libs/1_55_0/libs/locale/doc/html/index.html
 Boost.Nowide: http://cppcms.com/files/nowide/html/ (not part of Boost…yet?)
 http://www.utf8everywhere.org/
 “Should UTF-16 be considered harmful?” http://programmers.stackexchange.com/q/102205/206
Questions?

U+1F64B (HAPPY PERSON RAISING ONE HAND)

Questions?

U+0000 (HAPPY PERSON RAISING ONE HAND?)

Unicode in C
Unicode Support in C90

 [This slide intentionally left blank.]

Unicode Support in C90

 wchar_t
 mbtowc, wctomb
 mbstowcs, wcstombs
Unicode Support in C99

 wchar_t
 mbtowc, wctomb
 mbstowcs, wcstombs
 wchar_t-equivalents for many I/O and string handling function
Unicode Support in C11

 New code unit types: char16_t and char32_t

 Actual encoding is implementation-defined
 __STDC_UTF_16__ and __STDC_UTF_32__
 New string literal prefixes: u”Hello!” and U”Hello!”
 <uchar.h>: mbrtoc16, c16rtomb, mbrtoc32, c32rtomb
mbrtoc32

setlocale(LC_CTYPE, "en_US.UTF-8");

char const utf8_c[5] = "\U0001f378";

char32_t utf32_c = 0;
mbstate_t state = { 0 };
mbrtoc32(&utf32_c, utf8_c, 4, &state);

printf("0x%8x\n", utf32_c);
mbrtoc32

setlocale(LC_CTYPE, "en_US.UTF-8");

char const utf8_c[5] = "\U0001f378";

char16_t utf16_c[2] = { 0 };
mbstate_t state = { 0 };
mbrtoc16(&utf16_c[0], utf8_c, 4, &state);
mbrtoc16(&utf16_c[1], utf8_c, 4, &state);

printf("0x%4x 0x%4x\n", utf16_c[0], utf16_c[1]);