blob: c9dda73340608cdd827d5e7be57834826b1ca6cc [file] [log] [blame]
// Copyright 2014 The Chromium Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "net/base/net_string_util.h"
#include <string_view>
#include "base/i18n/case_conversion.h"
#include "base/i18n/i18n_constants.h"
#include "base/i18n/icu_string_conversions.h"
#include "base/strings/string_util.h"
#include "third_party/icu/source/common/unicode/ucnv.h"
namespace net {
const char* const kCharsetLatin1 = base::kCodepageLatin1;
bool ConvertToUtf8(std::string_view text,
const char* charset,
std::string* output) {
output->clear();
UErrorCode err = U_ZERO_ERROR;
UConverter* converter(ucnv_open(charset, &err));
if (U_FAILURE(err))
return false;
// A single byte in a legacy encoding can be expanded to 3 bytes in UTF-8.
// A 'two-byte character' in a legacy encoding can be expanded to 4 bytes
// in UTF-8. Therefore, the expansion ratio is 3 at most.
output->resize(text.length() * 3);
size_t output_length =
ucnv_toAlgorithmic(UCNV_UTF8, converter, output->data(), output->length(),
text.data(), text.length(), &err);
ucnv_close(converter);
if (U_FAILURE(err)) {
output->clear();
return false;
}
output->resize(output_length);
return true;
}
bool ConvertToUtf8AndNormalize(std::string_view text,
const char* charset,
std::string* output) {
return base::ConvertToUtf8AndNormalize(text, charset, output);
}
bool ConvertToUTF16(std::string_view text,
const char* charset,
std::u16string* output) {
return base::CodepageToUTF16(text, charset,
base::OnStringConversionError::FAIL, output);
}
bool ConvertToUTF16WithSubstitutions(std::string_view text,
const char* charset,
std::u16string* output) {
return base::CodepageToUTF16(
text, charset, base::OnStringConversionError::SUBSTITUTE, output);
}
bool ToUpper(std::u16string_view str, std::u16string* output) {
*output = base::i18n::ToUpper(str);
return true;
}
} // namespace net