darling-WTF/wtf/text/WTFString.cpp
2023-04-24 20:14:26 -07:00

1156 lines
33 KiB
C++

/*
* (C) 1999 Lars Knoll (knoll@kde.org)
* Copyright (C) 2004-2019 Apple Inc. All rights reserved.
* Copyright (C) 2007-2009 Torch Mobile, Inc.
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Library General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Library General Public License for more details.
*
* You should have received a copy of the GNU Library General Public License
* along with this library; see the file COPYING.LIB. If not, write to
* the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
* Boston, MA 02110-1301, USA.
*/
#include "config.h"
#include <wtf/text/WTFString.h>
#include <wtf/ASCIICType.h>
#include <wtf/DataLog.h>
#include <wtf/HexNumber.h>
#include <wtf/NeverDestroyed.h>
#include <wtf/Vector.h>
#include <wtf/dtoa.h>
#include <wtf/text/CString.h>
#include <wtf/text/IntegerToStringConversion.h>
#include <wtf/text/StringBuilder.h>
#include <wtf/text/StringToIntegerConversion.h>
#include <wtf/unicode/CharacterNames.h>
#include <wtf/unicode/UTF8Conversion.h>
namespace WTF {
using namespace Unicode;
// Construct a string with UTF-16 data.
String::String(const UChar* characters, unsigned length)
{
if (characters)
m_impl = StringImpl::create(characters, length);
}
// Construct a string with UTF-16 data, from a null-terminated source.
String::String(const UChar* nullTerminatedString)
{
if (nullTerminatedString)
m_impl = StringImpl::create(nullTerminatedString, lengthOfNullTerminatedString(nullTerminatedString));
}
// Construct a string with latin1 data.
String::String(const LChar* characters, unsigned length)
{
if (characters)
m_impl = StringImpl::create(characters, length);
}
String::String(const char* characters, unsigned length)
{
if (characters)
m_impl = StringImpl::create(reinterpret_cast<const LChar*>(characters), length);
}
// Construct a string with Latin-1 data, from a null-terminated source.
String::String(const LChar* nullTerminatedString)
{
if (nullTerminatedString)
m_impl = StringImpl::create(nullTerminatedString);
}
String::String(const char* nullTerminatedString)
{
if (nullTerminatedString)
m_impl = StringImpl::create(reinterpret_cast<const LChar*>(nullTerminatedString));
}
String::String(ASCIILiteral characters)
: m_impl(StringImpl::createFromLiteral(characters))
{
}
void String::append(const String& otherString)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
if (!m_impl) {
m_impl = otherString.m_impl;
return;
}
if (otherString.isEmpty())
return;
auto length = m_impl->length();
auto otherLength = otherString.m_impl->length();
if (otherLength > MaxLength - length)
CRASH();
if (m_impl->is8Bit() && otherString.m_impl->is8Bit()) {
LChar* data;
auto newImpl = StringImpl::createUninitialized(length + otherLength, data);
StringImpl::copyCharacters(data, m_impl->characters8(), length);
StringImpl::copyCharacters(data + length, otherString.m_impl->characters8(), otherLength);
m_impl = WTFMove(newImpl);
return;
}
UChar* data;
auto newImpl = StringImpl::createUninitialized(length + otherLength, data);
StringView(*m_impl).getCharactersWithUpconvert(data);
StringView(*otherString.m_impl).getCharactersWithUpconvert(data + length);
m_impl = WTFMove(newImpl);
}
void String::append(LChar character)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
if (!m_impl) {
m_impl = StringImpl::create(&character, 1);
return;
}
if (!is8Bit()) {
append(static_cast<UChar>(character));
return;
}
if (m_impl->length() >= MaxLength)
CRASH();
LChar* data;
auto newImpl = StringImpl::createUninitialized(m_impl->length() + 1, data);
StringImpl::copyCharacters(data, m_impl->characters8(), m_impl->length());
data[m_impl->length()] = character;
m_impl = WTFMove(newImpl);
}
void String::append(UChar character)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
if (!m_impl) {
m_impl = StringImpl::create(&character, 1);
return;
}
if (isLatin1(character) && is8Bit()) {
append(static_cast<LChar>(character));
return;
}
if (m_impl->length() >= MaxLength)
CRASH();
UChar* data;
auto newImpl = StringImpl::createUninitialized(m_impl->length() + 1, data);
StringView(*m_impl).getCharactersWithUpconvert(data);
data[m_impl->length()] = character;
m_impl = WTFMove(newImpl);
}
int codePointCompare(const String& a, const String& b)
{
return codePointCompare(a.impl(), b.impl());
}
void String::insert(const String& string, unsigned position)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
unsigned lengthToInsert = string.length();
if (!lengthToInsert) {
if (string.isNull())
return;
if (isNull())
m_impl = string.impl();
return;
}
if (position >= length()) {
append(string);
return;
}
if (lengthToInsert > MaxLength - length())
CRASH();
if (is8Bit() && string.is8Bit()) {
LChar* data;
auto newString = StringImpl::createUninitialized(length() + lengthToInsert, data);
StringView(*m_impl).substring(0, position).getCharactersWithUpconvert(data);
StringView(string).getCharactersWithUpconvert(data + position);
StringView(*m_impl).substring(position).getCharactersWithUpconvert(data + position + lengthToInsert);
m_impl = WTFMove(newString);
} else {
UChar* data;
auto newString = StringImpl::createUninitialized(length() + lengthToInsert, data);
StringView(*m_impl).substring(0, position).getCharactersWithUpconvert(data);
StringView(string).getCharactersWithUpconvert(data + position);
StringView(*m_impl).substring(position).getCharactersWithUpconvert(data + position + lengthToInsert);
m_impl = WTFMove(newString);
}
}
void String::append(const LChar* charactersToAppend, unsigned lengthToAppend)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
if (!m_impl) {
if (!charactersToAppend)
return;
m_impl = StringImpl::create(charactersToAppend, lengthToAppend);
return;
}
if (!lengthToAppend)
return;
ASSERT(charactersToAppend);
unsigned strLength = m_impl->length();
if (m_impl->is8Bit()) {
if (lengthToAppend > MaxLength - strLength)
CRASH();
LChar* data;
auto newImpl = StringImpl::createUninitialized(strLength + lengthToAppend, data);
StringImpl::copyCharacters(data, m_impl->characters8(), strLength);
StringImpl::copyCharacters(data + strLength, charactersToAppend, lengthToAppend);
m_impl = WTFMove(newImpl);
return;
}
if (lengthToAppend > MaxLength - strLength)
CRASH();
UChar* data;
auto newImpl = StringImpl::createUninitialized(length() + lengthToAppend, data);
StringImpl::copyCharacters(data, m_impl->characters16(), strLength);
StringImpl::copyCharacters(data + strLength, charactersToAppend, lengthToAppend);
m_impl = WTFMove(newImpl);
}
void String::append(const UChar* charactersToAppend, unsigned lengthToAppend)
{
// FIXME: This is extremely inefficient. So much so that we might want to take this out of String's API.
if (!m_impl) {
if (!charactersToAppend)
return;
m_impl = StringImpl::create(charactersToAppend, lengthToAppend);
return;
}
if (!lengthToAppend)
return;
unsigned strLength = m_impl->length();
ASSERT(charactersToAppend);
if (lengthToAppend > MaxLength - strLength)
CRASH();
UChar* data;
auto newImpl = StringImpl::createUninitialized(strLength + lengthToAppend, data);
if (m_impl->is8Bit())
StringImpl::copyCharacters(data, characters8(), strLength);
else
StringImpl::copyCharacters(data, characters16(), strLength);
StringImpl::copyCharacters(data + strLength, charactersToAppend, lengthToAppend);
m_impl = WTFMove(newImpl);
}
UChar32 String::characterStartingAt(unsigned i) const
{
if (!m_impl || i >= m_impl->length())
return 0;
return m_impl->characterStartingAt(i);
}
void String::truncate(unsigned position)
{
if (m_impl)
m_impl = m_impl->substring(0, position);
}
template<typename CharacterType> inline void String::removeInternal(const CharacterType* characters, unsigned position, unsigned lengthToRemove)
{
CharacterType* data;
auto newImpl = StringImpl::createUninitialized(length() - lengthToRemove, data);
StringImpl::copyCharacters(data, characters, position);
StringImpl::copyCharacters(data + position, characters + position + lengthToRemove, length() - lengthToRemove - position);
m_impl = WTFMove(newImpl);
}
void String::remove(unsigned position, unsigned lengthToRemove)
{
if (!lengthToRemove)
return;
auto length = this->length();
if (position >= length)
return;
lengthToRemove = std::min(lengthToRemove, length - position);
if (is8Bit())
removeInternal(characters8(), position, lengthToRemove);
else
removeInternal(characters16(), position, lengthToRemove);
}
String String::substring(unsigned position, unsigned length) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->substring(position, length) : String { };
}
String String::substringSharingImpl(unsigned offset, unsigned length) const
{
// FIXME: We used to check against a limit of Heap::minExtraCost / sizeof(UChar).
unsigned stringLength = this->length();
offset = std::min(offset, stringLength);
length = std::min(length, stringLength - offset);
if (!offset && length == stringLength)
return *this;
return StringImpl::createSubstringSharingImpl(*m_impl, offset, length);
}
String String::convertToASCIILowercase() const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToASCIILowercase() : String { };
}
String String::convertToASCIIUppercase() const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToASCIIUppercase() : String { };
}
String String::convertToLowercaseWithoutLocale() const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToLowercaseWithoutLocale() : String { };
}
String String::convertToLowercaseWithoutLocaleStartingAtFailingIndex8Bit(unsigned failingIndex) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToLowercaseWithoutLocaleStartingAtFailingIndex8Bit(failingIndex) : String { };
}
String String::convertToUppercaseWithoutLocale() const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToUppercaseWithoutLocale() : String { };
}
String String::convertToLowercaseWithLocale(const AtomString& localeIdentifier) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToLowercaseWithLocale(localeIdentifier) : String { };
}
String String::convertToUppercaseWithLocale(const AtomString& localeIdentifier) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->convertToUppercaseWithLocale(localeIdentifier) : String { };
}
String String::stripWhiteSpace() const
{
// FIXME: Should this function, and the many others like it, be inlined?
// FIXME: This function needs a new name. For one thing, "whitespace" is a single
// word so the "s" should be lowercase. For another, it's not clear from this name
// that the function uses the Unicode definition of whitespace. Most WebKit callers
// don't want that and eventually we should consider deleting this.
return m_impl ? m_impl->stripWhiteSpace() : String { };
}
String String::stripLeadingAndTrailingCharacters(CodeUnitMatchFunction predicate) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->stripLeadingAndTrailingCharacters(predicate) : String { };
}
String String::simplifyWhiteSpace() const
{
// FIXME: Should this function, and the many others like it, be inlined?
// FIXME: This function needs a new name. For one thing, "whitespace" is a single
// word so the "s" should be lowercase. For another, it's not clear from this name
// that the function uses the Unicode definition of whitespace. Most WebKit callers
// don't want that and eventually we should consider deleting this.
return m_impl ? m_impl->simplifyWhiteSpace() : String { };
}
String String::simplifyWhiteSpace(CodeUnitMatchFunction isWhiteSpace) const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->simplifyWhiteSpace(isWhiteSpace) : String { };
}
String String::foldCase() const
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->foldCase() : String { };
}
bool String::percentage(int& result) const
{
if (!m_impl || !m_impl->length())
return false;
if ((*m_impl)[m_impl->length() - 1] != '%')
return false;
if (m_impl->is8Bit())
result = charactersToIntStrict(m_impl->characters8(), m_impl->length() - 1);
else
result = charactersToIntStrict(m_impl->characters16(), m_impl->length() - 1);
return true;
}
Vector<UChar> String::charactersWithoutNullTermination() const
{
Vector<UChar> result;
if (m_impl) {
result.reserveInitialCapacity(length() + 1);
if (is8Bit()) {
const LChar* characters8 = m_impl->characters8();
for (size_t i = 0; i < length(); ++i)
result.uncheckedAppend(characters8[i]);
} else {
const UChar* characters16 = m_impl->characters16();
result.append(characters16, m_impl->length());
}
}
return result;
}
Vector<UChar> String::charactersWithNullTermination() const
{
auto result = charactersWithoutNullTermination();
result.append(0);
return result;
}
String String::number(int number)
{
return numberToStringSigned<String>(number);
}
String String::number(unsigned number)
{
return numberToStringUnsigned<String>(number);
}
String String::number(long number)
{
return numberToStringSigned<String>(number);
}
String String::number(unsigned long number)
{
return numberToStringUnsigned<String>(number);
}
String String::number(long long number)
{
return numberToStringSigned<String>(number);
}
String String::number(unsigned long long number)
{
return numberToStringUnsigned<String>(number);
}
String String::numberToStringFixedPrecision(float number, unsigned precision, TrailingZerosTruncatingPolicy trailingZerosTruncatingPolicy)
{
NumberToStringBuffer buffer;
return numberToFixedPrecisionString(number, precision, buffer, trailingZerosTruncatingPolicy == TruncateTrailingZeros);
}
String String::numberToStringFixedPrecision(double number, unsigned precision, TrailingZerosTruncatingPolicy trailingZerosTruncatingPolicy)
{
NumberToStringBuffer buffer;
return numberToFixedPrecisionString(number, precision, buffer, trailingZerosTruncatingPolicy == TruncateTrailingZeros);
}
String String::number(float number)
{
NumberToStringBuffer buffer;
return numberToString(number, buffer);
}
String String::number(double number)
{
NumberToStringBuffer buffer;
return numberToString(number, buffer);
}
String String::numberToStringFixedWidth(double number, unsigned decimalPlaces)
{
NumberToStringBuffer buffer;
return numberToFixedWidthString(number, decimalPlaces, buffer);
}
int String::toIntStrict(bool* ok, int base) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toIntStrict(ok, base);
}
unsigned String::toUIntStrict(bool* ok, int base) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toUIntStrict(ok, base);
}
int64_t String::toInt64Strict(bool* ok, int base) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toInt64Strict(ok, base);
}
uint64_t String::toUInt64Strict(bool* ok, int base) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toUInt64Strict(ok, base);
}
intptr_t String::toIntPtrStrict(bool* ok, int base) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toIntPtrStrict(ok, base);
}
int String::toInt(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toInt(ok);
}
unsigned String::toUInt(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toUInt(ok);
}
int64_t String::toInt64(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toInt64(ok);
}
uint64_t String::toUInt64(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toUInt64(ok);
}
intptr_t String::toIntPtr(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0;
}
return m_impl->toIntPtr(ok);
}
double String::toDouble(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0.0;
}
return m_impl->toDouble(ok);
}
float String::toFloat(bool* ok) const
{
if (!m_impl) {
if (ok)
*ok = false;
return 0.0f;
}
return m_impl->toFloat(ok);
}
String String::isolatedCopy() const &
{
// FIXME: Should this function, and the many others like it, be inlined?
return m_impl ? m_impl->isolatedCopy() : String { };
}
String String::isolatedCopy() &&
{
if (isSafeToSendToAnotherThread()) {
// Since we know that our string is a temporary that will be destroyed
// we can just steal the m_impl from it, thus avoiding a copy.
return { WTFMove(*this) };
}
return m_impl ? m_impl->isolatedCopy() : String { };
}
bool String::isSafeToSendToAnotherThread() const
{
// AtomStrings are not safe to send between threads, as ~StringImpl()
// will try to remove them from the wrong AtomStringTable.
return isEmpty() || (m_impl->hasOneRef() && !m_impl->isAtom());
}
template<bool allowEmptyEntries>
inline Vector<String> String::splitInternal(const String& separator) const
{
Vector<String> result;
unsigned startPos = 0;
size_t endPos;
while ((endPos = find(separator, startPos)) != notFound) {
if (allowEmptyEntries || startPos != endPos)
result.append(substring(startPos, endPos - startPos));
startPos = endPos + separator.length();
}
if (allowEmptyEntries || startPos != length())
result.append(substring(startPos));
return result;
}
template<bool allowEmptyEntries>
inline void String::splitInternal(UChar separator, const SplitFunctor& functor) const
{
StringView view(*this);
unsigned startPos = 0;
size_t endPos;
while ((endPos = find(separator, startPos)) != notFound) {
if (allowEmptyEntries || startPos != endPos)
functor(view.substring(startPos, endPos - startPos));
startPos = endPos + 1;
}
if (allowEmptyEntries || startPos != length())
functor(view.substring(startPos));
}
template<bool allowEmptyEntries>
inline Vector<String> String::splitInternal(UChar separator) const
{
Vector<String> result;
splitInternal<allowEmptyEntries>(separator, [&result](StringView item) {
result.append(item.toString());
});
return result;
}
void String::split(UChar separator, const SplitFunctor& functor) const
{
splitInternal<false>(separator, functor);
}
Vector<String> String::split(UChar separator) const
{
return splitInternal<false>(separator);
}
Vector<String> String::split(const String& separator) const
{
return splitInternal<false>(separator);
}
void String::splitAllowingEmptyEntries(UChar separator, const SplitFunctor& functor) const
{
splitInternal<true>(separator, functor);
}
Vector<String> String::splitAllowingEmptyEntries(UChar separator) const
{
return splitInternal<true>(separator);
}
Vector<String> String::splitAllowingEmptyEntries(const String& separator) const
{
return splitInternal<true>(separator);
}
CString String::ascii() const
{
// Printable ASCII characters 32..127 and the null character are
// preserved, characters outside of this range are converted to '?'.
unsigned length = this->length();
if (!length) {
char* characterBuffer;
return CString::newUninitialized(length, characterBuffer);
}
if (this->is8Bit()) {
const LChar* characters = this->characters8();
char* characterBuffer;
CString result = CString::newUninitialized(length, characterBuffer);
for (unsigned i = 0; i < length; ++i) {
LChar ch = characters[i];
characterBuffer[i] = ch && (ch < 0x20 || ch > 0x7f) ? '?' : ch;
}
return result;
}
const UChar* characters = this->characters16();
char* characterBuffer;
CString result = CString::newUninitialized(length, characterBuffer);
for (unsigned i = 0; i < length; ++i) {
UChar ch = characters[i];
characterBuffer[i] = ch && (ch < 0x20 || ch > 0x7f) ? '?' : ch;
}
return result;
}
CString String::latin1() const
{
// Basic Latin1 (ISO) encoding - Unicode characters 0..255 are
// preserved, characters outside of this range are converted to '?'.
unsigned length = this->length();
if (!length)
return CString("", 0);
if (is8Bit())
return CString(reinterpret_cast<const char*>(this->characters8()), length);
const UChar* characters = this->characters16();
char* characterBuffer;
CString result = CString::newUninitialized(length, characterBuffer);
for (unsigned i = 0; i < length; ++i) {
UChar ch = characters[i];
characterBuffer[i] = !isLatin1(ch) ? '?' : ch;
}
return result;
}
Expected<CString, UTF8ConversionError> String::tryGetUtf8(ConversionMode mode) const
{
return m_impl ? m_impl->tryGetUtf8(mode) : CString { "", 0 };
}
Expected<CString, UTF8ConversionError> String::tryGetUtf8() const
{
return tryGetUtf8(LenientConversion);
}
CString String::utf8(ConversionMode mode) const
{
Expected<CString, UTF8ConversionError> expectedString = tryGetUtf8(mode);
RELEASE_ASSERT(expectedString);
return expectedString.value();
}
CString String::utf8() const
{
return utf8(LenientConversion);
}
String String::make8BitFrom16BitSource(const UChar* source, size_t length)
{
if (!length)
return String();
LChar* destination;
String result = String::createUninitialized(length, destination);
copyLCharsFromUCharSource(destination, source, length);
return result;
}
String String::make16BitFrom8BitSource(const LChar* source, size_t length)
{
if (!length)
return String();
UChar* destination;
String result = String::createUninitialized(length, destination);
StringImpl::copyCharacters(destination, source, length);
return result;
}
template<bool replaceInvalidSequences>
String fromUTF8Impl(const LChar* stringStart, size_t length)
{
if (length > StringImplShape::MaxLength)
CRASH();
if (!stringStart)
return String();
if (!length)
return emptyString();
if (charactersAreAllASCII(stringStart, length))
return StringImpl::create(stringStart, length);
Vector<UChar, 1024> buffer(length);
UChar* bufferStart = buffer.data();
UChar* bufferCurrent = bufferStart;
const char* stringCurrent = reinterpret_cast<const char*>(stringStart);
constexpr auto function = replaceInvalidSequences ? convertUTF8ToUTF16ReplacingInvalidSequences : convertUTF8ToUTF16;
if (!function(stringCurrent, reinterpret_cast<const char*>(stringStart + length), &bufferCurrent, bufferCurrent + buffer.size(), nullptr))
return String();
unsigned utf16Length = bufferCurrent - bufferStart;
RELEASE_ASSERT_WITH_SECURITY_IMPLICATION(utf16Length <= length);
return StringImpl::create(bufferStart, utf16Length);
}
String String::fromUTF8(const LChar* stringStart, size_t length)
{
return fromUTF8Impl<false>(stringStart, length);
}
String String::fromUTF8ReplacingInvalidSequences(const LChar* characters, size_t length)
{
return fromUTF8Impl<true>(characters, length);
}
String String::fromUTF8(const LChar* string)
{
if (!string)
return String();
return fromUTF8(string, strlen(reinterpret_cast<const char*>(string)));
}
String String::fromUTF8(const CString& s)
{
return fromUTF8(s.data());
}
String String::fromUTF8WithLatin1Fallback(const LChar* string, size_t size)
{
String utf8 = fromUTF8(string, size);
if (!utf8)
return String(string, size);
return utf8;
}
String String::fromCodePoint(UChar32 codePoint)
{
UChar buffer[2];
uint8_t length = 0;
UBool error = false;
U16_APPEND(buffer, length, 2, codePoint, error);
return error ? String() : String(buffer, length);
}
// String Operations
template<typename CharacterType>
static unsigned lengthOfCharactersAsInteger(const CharacterType* data, size_t length)
{
size_t i = 0;
// Allow leading spaces.
for (; i != length; ++i) {
if (!isSpaceOrNewline(data[i]))
break;
}
// Allow sign.
if (i != length && (data[i] == '+' || data[i] == '-'))
++i;
// Allow digits.
for (; i != length; ++i) {
if (!isASCIIDigit(data[i]))
break;
}
return i;
}
int charactersToIntStrict(const LChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<int>(data, length, ok, base);
}
int charactersToIntStrict(const UChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<int>(data, length, ok, base);
}
unsigned charactersToUIntStrict(const LChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<unsigned>(data, length, ok, base);
}
unsigned charactersToUIntStrict(const UChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<unsigned>(data, length, ok, base);
}
int64_t charactersToInt64Strict(const LChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<int64_t>(data, length, ok, base);
}
int64_t charactersToInt64Strict(const UChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<int64_t>(data, length, ok, base);
}
uint64_t charactersToUInt64Strict(const LChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<uint64_t>(data, length, ok, base);
}
uint64_t charactersToUInt64Strict(const UChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<uint64_t>(data, length, ok, base);
}
intptr_t charactersToIntPtrStrict(const LChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<intptr_t>(data, length, ok, base);
}
intptr_t charactersToIntPtrStrict(const UChar* data, size_t length, bool* ok, int base)
{
return toIntegralType<intptr_t>(data, length, ok, base);
}
int charactersToInt(const LChar* data, size_t length, bool* ok)
{
return toIntegralType<int>(data, lengthOfCharactersAsInteger<LChar>(data, length), ok, 10);
}
int charactersToInt(const UChar* data, size_t length, bool* ok)
{
return toIntegralType<int>(data, lengthOfCharactersAsInteger(data, length), ok, 10);
}
unsigned charactersToUInt(const LChar* data, size_t length, bool* ok)
{
return toIntegralType<unsigned>(data, lengthOfCharactersAsInteger<LChar>(data, length), ok, 10);
}
unsigned charactersToUInt(const UChar* data, size_t length, bool* ok)
{
return toIntegralType<unsigned>(data, lengthOfCharactersAsInteger<UChar>(data, length), ok, 10);
}
int64_t charactersToInt64(const LChar* data, size_t length, bool* ok)
{
return toIntegralType<int64_t>(data, lengthOfCharactersAsInteger<LChar>(data, length), ok, 10);
}
int64_t charactersToInt64(const UChar* data, size_t length, bool* ok)
{
return toIntegralType<int64_t>(data, lengthOfCharactersAsInteger<UChar>(data, length), ok, 10);
}
uint64_t charactersToUInt64(const LChar* data, size_t length, bool* ok)
{
return toIntegralType<uint64_t>(data, lengthOfCharactersAsInteger<LChar>(data, length), ok, 10);
}
uint64_t charactersToUInt64(const UChar* data, size_t length, bool* ok)
{
return toIntegralType<uint64_t>(data, lengthOfCharactersAsInteger<UChar>(data, length), ok, 10);
}
intptr_t charactersToIntPtr(const LChar* data, size_t length, bool* ok)
{
return toIntegralType<intptr_t>(data, lengthOfCharactersAsInteger<LChar>(data, length), ok, 10);
}
intptr_t charactersToIntPtr(const UChar* data, size_t length, bool* ok)
{
return toIntegralType<intptr_t>(data, lengthOfCharactersAsInteger<UChar>(data, length), ok, 10);
}
enum TrailingJunkPolicy { DisallowTrailingJunk, AllowTrailingJunk };
template<typename CharacterType, TrailingJunkPolicy policy>
static inline double toDoubleType(const CharacterType* data, size_t length, bool* ok, size_t& parsedLength)
{
size_t leadingSpacesLength = 0;
while (leadingSpacesLength < length && isASCIISpace(data[leadingSpacesLength]))
++leadingSpacesLength;
double number = parseDouble(data + leadingSpacesLength, length - leadingSpacesLength, parsedLength);
if (!parsedLength) {
if (ok)
*ok = false;
return 0.0;
}
parsedLength += leadingSpacesLength;
if (ok)
*ok = policy == AllowTrailingJunk || parsedLength == length;
return number;
}
double charactersToDouble(const LChar* data, size_t length, bool* ok)
{
size_t parsedLength;
return toDoubleType<LChar, DisallowTrailingJunk>(data, length, ok, parsedLength);
}
double charactersToDouble(const UChar* data, size_t length, bool* ok)
{
size_t parsedLength;
return toDoubleType<UChar, DisallowTrailingJunk>(data, length, ok, parsedLength);
}
float charactersToFloat(const LChar* data, size_t length, bool* ok)
{
// FIXME: This will return ok even when the string fits into a double but not a float.
size_t parsedLength;
return static_cast<float>(toDoubleType<LChar, DisallowTrailingJunk>(data, length, ok, parsedLength));
}
float charactersToFloat(const UChar* data, size_t length, bool* ok)
{
// FIXME: This will return ok even when the string fits into a double but not a float.
size_t parsedLength;
return static_cast<float>(toDoubleType<UChar, DisallowTrailingJunk>(data, length, ok, parsedLength));
}
float charactersToFloat(const LChar* data, size_t length, size_t& parsedLength)
{
// FIXME: This will return ok even when the string fits into a double but not a float.
return static_cast<float>(toDoubleType<LChar, AllowTrailingJunk>(data, length, nullptr, parsedLength));
}
float charactersToFloat(const UChar* data, size_t length, size_t& parsedLength)
{
// FIXME: This will return ok even when the string fits into a double but not a float.
return static_cast<float>(toDoubleType<UChar, AllowTrailingJunk>(data, length, nullptr, parsedLength));
}
const String& emptyString()
{
static NeverDestroyed<String> emptyString(StringImpl::empty());
return emptyString;
}
const String& nullString()
{
static NeverDestroyed<String> nullString;
return nullString;
}
} // namespace WTF
#ifndef NDEBUG
// For use in the debugger.
String* string(const char*);
Vector<char> asciiDebug(StringImpl* impl);
Vector<char> asciiDebug(String& string);
void String::show() const
{
dataLogF("%s\n", asciiDebug(impl()).data());
}
String* string(const char* s)
{
// Intentionally leaks memory!
return new String(s);
}
Vector<char> asciiDebug(StringImpl* impl)
{
if (!impl)
return asciiDebug(String("[null]"_s).impl());
StringBuilder buffer;
for (unsigned i = 0; i < impl->length(); ++i) {
UChar ch = (*impl)[i];
if (isASCIIPrintable(ch)) {
if (ch == '\\')
buffer.append(ch);
buffer.append(ch);
} else {
buffer.append('\\');
buffer.append('u');
buffer.append(hex(ch, 4));
}
}
CString narrowString = buffer.toString().ascii();
Vector<char> result;
result.append(reinterpret_cast<const char*>(narrowString.data()), narrowString.length() + 1);
return result;
}
Vector<char> asciiDebug(String& string)
{
return asciiDebug(string.impl());
}
#endif