Make char16_t/char32_t string literals be UTF-16/32

Document Number: P1041R4
Date: 2019-02-18
Audience: Evolution Working Group
Reply-to: cpp@rmf.io

Changelog

Introduction

C++11 introduced character types suitable for code units of the UTF-16 and UTF-32 encoding forms, namely char16_t and char32_t. Along with this, it also introduced new string literals whose types are arrays of those two character types, prefixed with u and U, respectively. And last but not least, it also introduced UTF-8 string literals, prefixed with u8, with types arrays of const char. Of these three new string literal types, only one has a guarantee about the values that the elements of the array have; in other words, only one has a guaranteed encoding form, the UTF-8 string literals.

The standard text hints that the char16_t and char32_t string literals are intended to be encoded as, respectively, UTF-16 and UTF-32, but unlike it does for UTF-8 string literals, it never explicitly makes such a requirement.

Motivation

In defining char16_t string literals ([lex.string]/10), the standard makes a mention of “surrogate pairs”:

A string-literal that begins with u, such as u"asdf", is a char16_t string literal. A char16_t string literal has type “array of n const char16_t”, where n is the size of the string as defined below; it is initialized with the given characters. A single c-char may produce more than one char16_t character in the form of surrogate pairs.

Further down, when defining the size of char16_t string literals ([lex.string]/15), there is another mention of “surrogate pairs”:

The size of a char16_t string literal is the total number of escape sequences, universal-character-names, and other characters, plus one for each character requiring a surrogate pair, plus one for the terminating u'\0'. [Note: The size of a char16_­t string literal is the number of code units, not the number of characters. — end note]

For char32_t string literals, the definition of their size ([lex.string]/15) essentially limits the encoding form used to one that doesn’t have more than one code unit per character:

The size of a char32_t or wide string literal is the total number of escape sequences, universal-character-names, and other characters, plus one for the terminating U'\0' or L'\0'.

Additionally, the standard constrains the range of universal-character-names to the range that is supported by all of the UTF encoding forms discussed here:

Within char32_t and char16_t string literals, any universal-character-names shall be within the range 0x0 to 0x10FFFF.

All of these requirements, while never explicitly naming the UTF-16 or UTF-32 encoding forms, strongly imply that these are the encoding forms intended.

The C standard defines the __STDC_UTF_16__ and __STDC_UTF_32__ in the <uchar.h> header, which C++ defers to for the contents of <cuchar>. These macros are optional, which seems to imply some intent that the encodings for char16_t and char32_t literals be implementation-defined. However, it would be questionable for an implementation to pick any other encoding forms for these string literals: there is no well-known encoding form that uses a concept named “surrogate pair” other than UTF-16, and there is no well-known encoding form that encodes each character as a single 32-bit code unit other than UTF-32.

In practice, all implementations use UTF-16 and UTF-32 for these string literals. C++ should standardize this practice and make these requirements explicit instead of just hinting at them.

Proposal

This proposal renames “char16_t string literals” and “char32_t string literals” to “UTF-16 string literals” and “UTF-32 string literals”, to match the existing “UTF-8 string literals”, and explicitly requires the arrays created by those literals to have the values that correspond to the UTF-16 and UTF-32 (respectively) encodings of the given characters.

Technical Specifications