blob: 76dd725358e8b666f2403f41816b4a1f13df14a8 [file] [log] [blame]
// Copyright (c) 2009 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.
#include "base/strings/utf_string_conversion_utils.h"
#include "base/third_party/icu/icu_utf.h"
#include "build/build_config.h"
namespace gurl_base {
// ReadUnicodeCharacter --------------------------------------------------------
bool ReadUnicodeCharacter(const char* src,
int32_t src_len,
int32_t* char_index,
base_icu::UChar32* code_point_out) {
base_icu::UChar32 code_point;
CBU8_NEXT(src, *char_index, src_len, code_point);
*code_point_out = code_point;
// The ICU macro above moves to the next char, we want to point to the last
// char consumed.
(*char_index)--;
// Validate the decoded value.
return IsValidCodepoint(code_point);
}
bool ReadUnicodeCharacter(const char16_t* src,
int32_t src_len,
int32_t* char_index,
base_icu::UChar32* code_point) {
if (CBU16_IS_SURROGATE(src[*char_index])) {
if (!CBU16_IS_SURROGATE_LEAD(src[*char_index]) ||
*char_index + 1 >= src_len ||
!CBU16_IS_TRAIL(src[*char_index + 1])) {
// Invalid surrogate pair.
return false;
}
// Valid surrogate pair.
*code_point = CBU16_GET_SUPPLEMENTARY(src[*char_index],
src[*char_index + 1]);
(*char_index)++;
} else {
// Not a surrogate, just one 16-bit word.
*code_point = src[*char_index];
}
return IsValidCodepoint(*code_point);
}
#if defined(WCHAR_T_IS_UTF32)
bool ReadUnicodeCharacter(const wchar_t* src,
int32_t src_len,
int32_t* char_index,
base_icu::UChar32* code_point) {
// Conversion is easy since the source is 32-bit.
*code_point = src[*char_index];
// Validate the value.
return IsValidCodepoint(*code_point);
}
#endif // defined(WCHAR_T_IS_UTF32)
// WriteUnicodeCharacter -------------------------------------------------------
size_t WriteUnicodeCharacter(uint32_t code_point, std::string* output) {
if (code_point <= 0x7f) {
// Fast path the common case of one byte.
output->push_back(static_cast<char>(code_point));
return 1;
}
// CBU8_APPEND_UNSAFE can append up to 4 bytes.
size_t char_offset = output->length();
size_t original_char_offset = char_offset;
output->resize(char_offset + CBU8_MAX_LENGTH);
CBU8_APPEND_UNSAFE(&(*output)[0], char_offset, code_point);
// CBU8_APPEND_UNSAFE will advance our pointer past the inserted character, so
// it will represent the new length of the string.
output->resize(char_offset);
return char_offset - original_char_offset;
}
size_t WriteUnicodeCharacter(uint32_t code_point, std::u16string* output) {
if (CBU16_LENGTH(code_point) == 1) {
// Thie code point is in the Basic Multilingual Plane (BMP).
output->push_back(static_cast<char16_t>(code_point));
return 1;
}
// Non-BMP characters use a double-character encoding.
size_t char_offset = output->length();
output->resize(char_offset + CBU16_MAX_LENGTH);
CBU16_APPEND_UNSAFE(&(*output)[0], char_offset, code_point);
return CBU16_MAX_LENGTH;
}
// Generalized Unicode converter -----------------------------------------------
template<typename CHAR>
void PrepareForUTF8Output(const CHAR* src,
size_t src_len,
std::string* output) {
output->clear();
if (src_len == 0)
return;
if (src[0] < 0x80) {
// Assume that the entire input will be ASCII.
output->reserve(src_len);
} else {
// Assume that the entire input is non-ASCII and will have 3 bytes per char.
output->reserve(src_len * 3);
}
}
// Instantiate versions we know callers will need.
#if !BUILDFLAG(IS_WIN)
// wchar_t and char16_t are the same thing on Windows.
template void PrepareForUTF8Output(const wchar_t*, size_t, std::string*);
#endif
template void PrepareForUTF8Output(const char16_t*, size_t, std::string*);
template<typename STRING>
void PrepareForUTF16Or32Output(const char* src,
size_t src_len,
STRING* output) {
output->clear();
if (src_len == 0)
return;
if (static_cast<unsigned char>(src[0]) < 0x80) {
// Assume the input is all ASCII, which means 1:1 correspondence.
output->reserve(src_len);
} else {
// Otherwise assume that the UTF-8 sequences will have 2 bytes for each
// character.
output->reserve(src_len / 2);
}
}
// Instantiate versions we know callers will need.
#if !BUILDFLAG(IS_WIN)
// std::wstring and std::u16string are the same thing on Windows.
template void PrepareForUTF16Or32Output(const char*, size_t, std::wstring*);
#endif
template void PrepareForUTF16Or32Output(const char*, size_t, std::u16string*);
} // namespace base