104 lines
4.4 KiB
C++
104 lines
4.4 KiB
C++
// Copyright (c) 2011 The Chromium Authors. All rights reserved.
|
|
// Use of this source code is governed by a BSD-style license that can be
|
|
// found in the LICENSE file.
|
|
|
|
#ifndef BASE_STRINGS_UTF_STRING_CONVERSION_UTILS_H_
|
|
#define BASE_STRINGS_UTF_STRING_CONVERSION_UTILS_H_
|
|
|
|
// Low-level UTF handling functions. Most code will want to use the functions
|
|
// in utf_string_conversions.h
|
|
|
|
#include <stddef.h>
|
|
#include <stdint.h>
|
|
|
|
#include "base/base_export.h"
|
|
#include "base/strings/string16.h"
|
|
|
|
namespace base {
|
|
|
|
inline bool IsValidCodepoint(uint32_t code_point) {
|
|
// Excludes code points that are not Unicode scalar values, i.e.
|
|
// surrogate code points ([0xD800, 0xDFFF]). Additionally, excludes
|
|
// code points larger than 0x10FFFF (the highest codepoint allowed).
|
|
// Non-characters and unassigned code points are allowed.
|
|
// https://unicode.org/glossary/#unicode_scalar_value
|
|
return code_point < 0xD800u ||
|
|
(code_point >= 0xE000u && code_point <= 0x10FFFFu);
|
|
}
|
|
|
|
inline bool IsValidCharacter(uint32_t code_point) {
|
|
// Excludes non-characters (U+FDD0..U+FDEF, and all code points
|
|
// ending in 0xFFFE or 0xFFFF) from the set of valid code points.
|
|
// https://unicode.org/faq/private_use.html#nonchar1
|
|
return code_point < 0xD800u || (code_point >= 0xE000u &&
|
|
code_point < 0xFDD0u) || (code_point > 0xFDEFu &&
|
|
code_point <= 0x10FFFFu && (code_point & 0xFFFEu) != 0xFFFEu);
|
|
}
|
|
|
|
// ReadUnicodeCharacter --------------------------------------------------------
|
|
|
|
// Reads a UTF-8 stream, placing the next code point into the given output
|
|
// |*code_point|. |src| represents the entire string to read, and |*char_index|
|
|
// is the character offset within the string to start reading at. |*char_index|
|
|
// will be updated to index the last character read, such that incrementing it
|
|
// (as in a for loop) will take the reader to the next character.
|
|
//
|
|
// Returns true on success. On false, |*code_point| will be invalid.
|
|
BASE_EXPORT bool ReadUnicodeCharacter(const char* src,
|
|
int32_t src_len,
|
|
int32_t* char_index,
|
|
uint32_t* code_point_out);
|
|
|
|
// Reads a UTF-16 character. The usage is the same as the 8-bit version above.
|
|
BASE_EXPORT bool ReadUnicodeCharacter(const char16* src,
|
|
int32_t src_len,
|
|
int32_t* char_index,
|
|
uint32_t* code_point);
|
|
|
|
#if defined(WCHAR_T_IS_UTF32)
|
|
// Reads UTF-32 character. The usage is the same as the 8-bit version above.
|
|
BASE_EXPORT bool ReadUnicodeCharacter(const wchar_t* src,
|
|
int32_t src_len,
|
|
int32_t* char_index,
|
|
uint32_t* code_point);
|
|
#endif // defined(WCHAR_T_IS_UTF32)
|
|
|
|
// WriteUnicodeCharacter -------------------------------------------------------
|
|
|
|
// Appends a UTF-8 character to the given 8-bit string. Returns the number of
|
|
// bytes written.
|
|
BASE_EXPORT size_t WriteUnicodeCharacter(uint32_t code_point,
|
|
std::string* output);
|
|
|
|
// Appends the given code point as a UTF-16 character to the given 16-bit
|
|
// string. Returns the number of 16-bit values written.
|
|
BASE_EXPORT size_t WriteUnicodeCharacter(uint32_t code_point, string16* output);
|
|
|
|
#if defined(WCHAR_T_IS_UTF32)
|
|
// Appends the given UTF-32 character to the given 32-bit string. Returns the
|
|
// number of 32-bit values written.
|
|
inline size_t WriteUnicodeCharacter(uint32_t code_point, std::wstring* output) {
|
|
// This is the easy case, just append the character.
|
|
output->push_back(code_point);
|
|
return 1;
|
|
}
|
|
#endif // defined(WCHAR_T_IS_UTF32)
|
|
|
|
// Generalized Unicode converter -----------------------------------------------
|
|
|
|
// Guesses the length of the output in UTF-8 in bytes, clears that output
|
|
// string, and reserves that amount of space. We assume that the input
|
|
// character types are unsigned, which will be true for UTF-16 and -32 on our
|
|
// systems.
|
|
template<typename CHAR>
|
|
void PrepareForUTF8Output(const CHAR* src, size_t src_len, std::string* output);
|
|
|
|
// Prepares an output buffer (containing either UTF-16 or -32 data) given some
|
|
// UTF-8 input that will be converted to it. See PrepareForUTF8Output().
|
|
template<typename STRING>
|
|
void PrepareForUTF16Or32Output(const char* src, size_t src_len, STRING* output);
|
|
|
|
} // namespace base
|
|
|
|
#endif // BASE_STRINGS_UTF_STRING_CONVERSION_UTILS_H_
|