arkcompiler_ets_runtime/ecmascript/ecma_string.h
2024-05-10 18:27:47 +08:00

1549 lines
53 KiB
C++
Executable File

/*
* Copyright (c) 2021 Huawei Device Co., Ltd.
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#ifndef ECMASCRIPT_STRING_H
#define ECMASCRIPT_STRING_H
#include <cstddef>
#include <cstdint>
#include <cstring>
#include "ecmascript/base/utf_helper.h"
#include "ecmascript/common.h"
#include "ecmascript/ecma_macros.h"
#include "ecmascript/js_hclass.h"
#include "ecmascript/js_tagged_value.h"
#include "ecmascript/mem/barriers.h"
#include "ecmascript/mem/space.h"
#include "ecmascript/mem/tagged_object.h"
#include "libpandabase/macros.h"
#include "securec.h"
#include "unicode/locid.h"
namespace panda {
namespace ecmascript {
template<typename T>
class JSHandle;
class JSPandaFile;
class EcmaVM;
class LineEcmaString;
class ConstantString;
class TreeEcmaString;
class SlicedString;
class FlatStringInfo;
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define ECMA_STRING_CHECK_LENGTH_AND_TRHOW(vm, length) \
if ((length) >= MAX_STRING_LENGTH) { \
THROW_RANGE_ERROR_AND_RETURN((vm)->GetJSThread(), "Invalid string length", nullptr); \
}
class EcmaString : public TaggedObject {
/* Mix Hash Code: -- { 0 | [31 bits raw hash code] } computed through string
\ { 1 | [31 bits integer numbers] } fastpath for string to number
*/
public:
CAST_CHECK(EcmaString, IsString);
static constexpr uint32_t IS_INTEGER_MASK = 1U << 31;
static constexpr uint32_t STRING_COMPRESSED_BIT = 0x1;
static constexpr uint32_t STRING_INTERN_BIT = 0x2;
static constexpr size_t MAX_STRING_LENGTH = 0x40000000U; // 30 bits for string length, 2 bits for special meaning
static constexpr uint32_t STRING_LENGTH_SHIFT_COUNT = 2U;
static constexpr uint32_t MAX_INTEGER_HASH_NUMBER = 0x3B9AC9FF;
static constexpr uint32_t MAX_CACHED_INTEGER_SIZE = 9;
static constexpr size_t MIX_LENGTH_OFFSET = TaggedObjectSize();
// In last bit of mix_length we store if this string is compressed or not.
ACCESSORS_PRIMITIVE_FIELD(MixLength, uint32_t, MIX_LENGTH_OFFSET, MIX_HASHCODE_OFFSET)
// In last bit of mix_hash we store if this string is small-integer number or not.
ACCESSORS_PRIMITIVE_FIELD(MixHashcode, uint32_t, MIX_HASHCODE_OFFSET, SIZE)
enum CompressedStatus {
STRING_COMPRESSED,
STRING_UNCOMPRESSED,
};
enum IsIntegerStatus {
NOT_INTEGER = 0,
IS_INTEGER,
};
enum TrimMode : uint8_t {
TRIM,
TRIM_START,
TRIM_END,
};
enum ConcatOptStatus {
BEGIN_STRING_ADD = 1,
IN_STRING_ADD,
CONFIRMED_IN_STRING_ADD,
END_STRING_ADD,
INVALID_STRING_ADD,
HAS_BACKING_STORE,
};
private:
friend class EcmaStringAccessor;
friend class LineEcmaString;
friend class ConstantString;
friend class TreeEcmaString;
friend class SlicedString;
friend class FlatStringInfo;
friend class NameDictionary;
static constexpr int SMALL_STRING_SIZE = 128;
static EcmaString *CreateEmptyString(const EcmaVM *vm);
static EcmaString *CreateFromUtf8(const EcmaVM *vm, const uint8_t *utf8Data, uint32_t utf8Len,
bool canBeCompress, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE, bool isConstantString = false,
uint32_t idOffset = 0);
static EcmaString *CreateFromUtf8CompressedSubString(const EcmaVM *vm, const JSHandle<EcmaString> &string,
uint32_t offset, uint32_t utf8Len, MemSpaceType type = MemSpaceType::SEMI_SPACE);
static EcmaString *CreateUtf16StringFromUtf8(const EcmaVM *vm, const uint8_t *utf8Data, uint32_t utf8Len,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static EcmaString *CreateFromUtf16(const EcmaVM *vm, const uint16_t *utf16Data, uint32_t utf16Len,
bool canBeCompress, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static SlicedString *CreateSlicedString(const EcmaVM *vm, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static EcmaString *CreateLineString(const EcmaVM *vm, size_t length, bool compressed);
static EcmaString *CreateLineStringNoGC(const EcmaVM *vm, size_t length, bool compressed);
static EcmaString *CreateLineStringWithSpaceType(const EcmaVM *vm,
size_t length, bool compressed, MemSpaceType type);
static EcmaString *CreateTreeString(const EcmaVM *vm,
const JSHandle<EcmaString> &left, const JSHandle<EcmaString> &right, uint32_t length, bool compressed);
static EcmaString *CreateConstantString(const EcmaVM *vm, const uint8_t *utf8Data,
size_t length, bool compressed, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE, uint32_t idOffset = 0);
static EcmaString *Concat(const EcmaVM *vm, const JSHandle<EcmaString> &left,
const JSHandle<EcmaString> &right, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static EcmaString *CopyStringToOldSpace(const EcmaVM *vm, const JSHandle<EcmaString> &original,
uint32_t length, bool compressed);
static EcmaString *FastSubString(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length);
static EcmaString *GetSlicedString(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length);
static EcmaString *GetSubString(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length);
// require src is LineString
// not change src data structure
static inline EcmaString *FastSubUtf8String(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length);
// require src is LineString
// not change src data structure
static inline EcmaString *FastSubUtf16String(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length);
inline void TrimLineString(const JSThread *thread, uint32_t newLength);
inline bool IsUtf8() const
{
return (GetMixLength() & STRING_COMPRESSED_BIT) == STRING_COMPRESSED;
}
inline bool IsUtf16() const
{
return (GetMixLength() & STRING_COMPRESSED_BIT) == STRING_UNCOMPRESSED;
}
inline bool IsInteger()
{
return (GetHashcode() & IS_INTEGER_MASK) == IS_INTEGER_MASK;
}
// require is LineString
inline uint16_t *GetData() const;
inline const uint8_t *GetDataUtf8() const;
inline const uint16_t *GetDataUtf16() const;
// require is LineString
inline uint8_t *GetDataUtf8Writable();
inline uint16_t *GetDataUtf16Writable();
inline uint32_t GetLength() const
{
return GetMixLength() >> STRING_LENGTH_SHIFT_COUNT;
}
inline void SetLength(uint32_t length, bool compressed = false)
{
ASSERT(length < MAX_STRING_LENGTH);
// Use 0u for compressed/utf8 expression
SetMixLength((length << STRING_LENGTH_SHIFT_COUNT) | (compressed ? STRING_COMPRESSED : STRING_UNCOMPRESSED));
}
inline uint32_t GetRawHashcode() const
{
return GetMixHashcode() & (~IS_INTEGER_MASK);
}
static inline uint32_t MixHashcode(uint32_t hashcode, bool isInteger)
{
return isInteger ? (hashcode | IS_INTEGER_MASK) : (hashcode & (~IS_INTEGER_MASK));
}
inline void SetRawHashcode(uint32_t hashcode, bool isInteger = false)
{
// Use 0u for not integer string's expression
SetMixHashcode(MixHashcode(hashcode, isInteger));
}
inline size_t GetUtf8Length(bool modify = true) const;
inline void SetIsInternString()
{
SetMixLength(GetMixLength() | STRING_INTERN_BIT);
}
inline bool IsInternString() const
{
return (GetMixLength() & STRING_INTERN_BIT) != 0;
}
inline void ClearInternStringFlag()
{
SetMixLength(GetMixLength() & ~STRING_INTERN_BIT);
}
inline bool TryGetHashCode(uint32_t *hash)
{
uint32_t hashcode = GetMixHashcode();
if (hashcode == 0 && GetLength() != 0) {
return false;
}
*hash = hashcode;
return true;
}
inline uint32_t GetIntegerCode()
{
ASSERT(GetMixHashcode() & IS_INTEGER_MASK);
return GetRawHashcode();
}
// not change this data structure.
// if string is not flat, this func has low efficiency.
uint32_t PUBLIC_API GetHashcode()
{
uint32_t hashcode = GetMixHashcode();
// GetLength() == 0 means it's an empty array.No need to computeHashCode again when hashseed is 0.
if (hashcode == 0 && GetLength() != 0) {
hashcode = ComputeHashcode();
SetMixHashcode(hashcode);
}
return hashcode;
}
template<typename T>
inline static bool IsDecimalDigitChar(const T c)
{
return (c >= '0' && c <= '9');
}
static uint32_t ComputeIntegerHash(uint32_t *num, uint8_t c)
{
if (!IsDecimalDigitChar(c)) {
return false;
}
int charDate = c - '0';
*num = (*num) * 10 + charDate; // 10: decimal factor
return true;
}
bool HashIntegerString(uint32_t length, uint32_t *hash, uint32_t hashSeed) const;
template<typename T>
static bool HashIntegerString(const T *data, size_t size, uint32_t *hash, uint32_t hashSeed)
{
ASSERT(size >= 0);
if (hashSeed == 0) {
if (IsDecimalDigitChar(data[0]) && data[0] != '0') {
uint32_t num = data[0] - '0';
uint32_t i = 1;
do {
if (i == size) {
// compute mix hash
if (num <= MAX_INTEGER_HASH_NUMBER) {
*hash = MixHashcode(num, IS_INTEGER);
return true;
}
return false;
}
} while (ComputeIntegerHash(&num, data[i++]));
}
if (size == 1 && (data[0] == '0')) {
*hash = MixHashcode(0, IS_INTEGER);
return true;
}
} else {
if (IsDecimalDigitChar(data[0])) {
uint32_t num = hashSeed * 10 + (data[0] - '0'); // 10: decimal factor
uint32_t i = 1;
do {
if (i == size) {
// compute mix hash
if (num <= MAX_INTEGER_HASH_NUMBER) {
*hash = MixHashcode(num, IS_INTEGER);
return true;
}
return false;
}
} while (ComputeIntegerHash(&num, data[i++]));
}
}
return false;
}
// not change this data structure.
// if string is not flat, this func has low efficiency.
uint32_t PUBLIC_API ComputeHashcode() const;
std::pair<uint32_t, bool> PUBLIC_API ComputeRawHashcode() const;
uint32_t PUBLIC_API ComputeHashcode(uint32_t rawHashSeed, bool isInteger) const;
static uint32_t ComputeHashcodeUtf8(const uint8_t *utf8Data, size_t utf8Len, bool canBeCompress);
static uint32_t ComputeHashcodeUtf16(const uint16_t *utf16Data, uint32_t length);
template<bool verify = true>
uint16_t At(int32_t index) const;
// require is LineString
void WriteData(uint32_t index, uint16_t src);
// can change left and right data structure
static int32_t Compare(const EcmaVM *vm, const JSHandle<EcmaString> &left, const JSHandle<EcmaString> &right);
static bool IsSubStringAt(const EcmaVM *vm, const JSHandle<EcmaString>& left,
const JSHandle<EcmaString>& right, uint32_t offset);
// Check that two spans are equal. Should have the same length.
/* static */
template<typename T, typename T1>
static bool StringsAreEquals(Span<const T> &str1, Span<const T1> &str2)
{
ASSERT(str1.Size() <= str2.Size());
size_t size = str1.Size();
if (!std::is_same_v<T, T1>) {
for (size_t i = 0; i < size; i++) {
auto left = static_cast<uint16_t>(str1[i]);
auto right = static_cast<uint16_t>(str2[i]);
if (left != right) {
return false;
}
}
return true;
}
if (size < SMALL_STRING_SIZE) {
for (size_t i = 0; i < size; i++) {
if (str1[i] != str2[i]) {
return false;
}
}
return true;
}
return memcmp(str1.data(), str2.data(), size * sizeof(T)) == 0;
}
// Converts utf8Data to utf16 and compare it with given utf16_data.
static bool IsUtf8EqualsUtf16(const uint8_t *utf8Data, size_t utf8Len, const uint16_t *utf16Data,
uint32_t utf16Len);
// Compares string1 + string2 by bytes, It doesn't check canonical unicode equivalence.
bool EqualToSplicedString(const EcmaString *str1, const EcmaString *str2);
// Compares strings by bytes, It doesn't check canonical unicode equivalence.
static PUBLIC_API bool StringsAreEqual(const EcmaVM *vm, const JSHandle<EcmaString> &str1,
const JSHandle<EcmaString> &str2);
// Compares strings by bytes, It doesn't check canonical unicode equivalence.
static PUBLIC_API bool StringsAreEqual(EcmaString *str1, EcmaString *str2);
// Two strings have the same type of utf encoding format.
static bool StringsAreEqualDiffUtfEncoding(EcmaString *str1, EcmaString *str2);
static bool StringsAreEqualDiffUtfEncoding(const FlatStringInfo &str1, const FlatStringInfo &str2);
// Compares strings by bytes, It doesn't check canonical unicode equivalence.
// not change str1 data structure.
// if str1 is not flat, this func has low efficiency.
static bool StringIsEqualUint8Data(const EcmaString *str1, const uint8_t *dataAddr, uint32_t dataLen,
bool canBeCompress);
// Compares strings by bytes, It doesn't check canonical unicode equivalence.
// not change str1 data structure.
// if str1 is not flat, this func has low efficiency.
static bool StringsAreEqualUtf16(const EcmaString *str1, const uint16_t *utf16Data, uint32_t utf16Len);
// can change receiver and search data structure
static int32_t IndexOf(const EcmaVM *vm,
const JSHandle<EcmaString> &receiver, const JSHandle<EcmaString> &search, int pos = 0);
// can change receiver and search data structure
static int32_t LastIndexOf(const EcmaVM *vm,
const JSHandle<EcmaString> &receiver, const JSHandle<EcmaString> &search, int pos = 0);
inline size_t CopyDataUtf8(uint8_t *buf, size_t maxLength, bool modify = true) const
{
if (maxLength == 0) {
return 1; // maxLength was -1 at napi
}
size_t length = GetLength();
if (length > maxLength) {
return 0;
}
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
buf[maxLength - 1] = '\0';
// Put comparison here so that internal usage and napi can use the same CopyDataRegionUtf8
return CopyDataRegionUtf8(buf, 0, length, maxLength, modify) + 1; // add place for zero in the end
}
// It allows user to copy into buffer even if maxLength < length
inline size_t WriteUtf8(uint8_t *buf, size_t maxLength, bool isWriteBuffer = false) const
{
if (maxLength == 0) {
return 1; // maxLength was -1 at napi
}
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
buf[maxLength - 1] = '\0';
return CopyDataRegionUtf8(buf, 0, GetLength(), maxLength, true, isWriteBuffer) + 1;
}
size_t CopyDataToUtf16(uint16_t *buf, uint32_t length, uint32_t bufLength) const
{
if (IsUtf16()) {
CVector<uint16_t> tmpBuf;
const uint16_t *data = EcmaString::GetUtf16DataFlat(this, tmpBuf);
if (length > bufLength) {
if (memcpy_s(buf, bufLength * sizeof(uint16_t), data, bufLength * sizeof(uint16_t)) != EOK) {
LOG_FULL(FATAL) << "memcpy_s failed when length > bufLength";
UNREACHABLE();
}
return bufLength;
}
if (memcpy_s(buf, bufLength * sizeof(uint16_t), data, length * sizeof(uint16_t)) != EOK) {
LOG_FULL(FATAL) << "memcpy_s failed";
UNREACHABLE();
}
return length;
}
CVector<uint8_t> tmpBuf;
const uint8_t *data = EcmaString::GetUtf8DataFlat(this, tmpBuf);
if (length > bufLength) {
return base::utf_helper::ConvertRegionUtf8ToUtf16(data, buf, bufLength, bufLength, 0);
}
return base::utf_helper::ConvertRegionUtf8ToUtf16(data, buf, length, bufLength, 0);
}
// It allows user to copy into buffer even if maxLength < length
inline size_t WriteUtf16(uint16_t *buf, uint32_t targetLength, uint32_t bufLength) const
{
if (bufLength == 0) {
return 0;
}
// Returns a number representing a valid backrest length.
return CopyDataToUtf16(buf, targetLength, bufLength);
}
size_t WriteOneByte(uint8_t *buf, size_t maxLength) const
{
if (maxLength == 0) {
return 0;
}
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
buf[maxLength - 1] = '\0';
uint32_t length = GetLength();
if (!IsUtf16()) {
CVector<uint8_t> tmpBuf;
const uint8_t *data = GetUtf8DataFlat(this, tmpBuf);
if (length > maxLength) {
length = maxLength;
}
if (memcpy_s(buf, maxLength, data, length) != EOK) {
LOG_FULL(FATAL) << "memcpy_s failed when write one byte";
UNREACHABLE();
}
return length;
}
CVector<uint16_t> tmpBuf;
const uint16_t *data = GetUtf16DataFlat(this, tmpBuf);
if (length > maxLength) {
return base::utf_helper::ConvertRegionUtf16ToLatin1(data, buf, maxLength, maxLength);
}
return base::utf_helper::ConvertRegionUtf16ToLatin1(data, buf, length, maxLength);
}
size_t CopyDataRegionUtf8(uint8_t *buf, size_t start, size_t length, size_t maxLength,
bool modify = true, bool isWriteBuffer = false) const
{
uint32_t len = GetLength();
if (start + length > len) {
return 0;
}
if (!IsUtf16()) {
if (length > std::numeric_limits<size_t>::max() / 2 - 1) { // 2: half
LOG_FULL(FATAL) << " length is higher than half of size_t::max";
UNREACHABLE();
}
CVector<uint8_t> tmpBuf;
const uint8_t *data = GetUtf8DataFlat(this, tmpBuf) + start;
// Only copy maxLength number of chars into buffer if length > maxLength
auto dataLen = std::min(length, maxLength);
std::copy(data, data + dataLen, buf);
return dataLen;
}
CVector<uint16_t> tmpBuf;
const uint16_t *data = GetUtf16DataFlat(this, tmpBuf);
if (length > maxLength) {
return base::utf_helper::ConvertRegionUtf16ToUtf8(data, buf, maxLength, maxLength, start,
modify, isWriteBuffer);
}
return base::utf_helper::ConvertRegionUtf16ToUtf8(data, buf, length, maxLength, start,
modify, isWriteBuffer);
}
inline uint32_t CopyDataUtf16(uint16_t *buf, uint32_t maxLength) const
{
return CopyDataRegionUtf16(buf, 0, GetLength(), maxLength);
}
uint32_t CopyDataRegionUtf16(uint16_t *buf, uint32_t start, uint32_t length, uint32_t maxLength) const
{
if (length > maxLength) {
return 0;
}
uint32_t len = GetLength();
if (start + length > len) {
return 0;
}
if (IsUtf16()) {
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
CVector<uint16_t> tmpBuf;
const uint16_t *data = EcmaString::GetUtf16DataFlat(this, tmpBuf);
if (memcpy_s(buf, maxLength * sizeof(uint16_t), data + start, length * sizeof(uint16_t)) != EOK) {
LOG_FULL(FATAL) << "memcpy_s failed";
UNREACHABLE();
}
return length;
}
CVector<uint8_t> tmpBuf;
const uint8_t *data = EcmaString::GetUtf8DataFlat(this, tmpBuf);
return base::utf_helper::ConvertRegionUtf8ToUtf16(data, buf, len, maxLength, start);
}
std::u16string ToU16String(uint32_t len = 0);
std::unique_ptr<uint8_t[]> ToOneByteDataForced()
{
uint8_t *buf = nullptr;
auto length = GetLength();
if (IsUtf16()) {
auto size = length * sizeof(uint16_t);
buf = new uint8_t[size]();
CopyDataUtf16(reinterpret_cast<uint16_t *>(buf), length);
} else {
buf = new uint8_t[length + 1]();
CopyDataUtf8(buf, length + 1);
}
return std::unique_ptr<uint8_t[]>(buf);
}
Span<const uint8_t> ToUtf8Span(CVector<uint8_t> &buf, bool modify = true)
{
Span<const uint8_t> str;
uint32_t strLen = GetLength();
if (UNLIKELY(IsUtf16())) {
CVector<uint16_t> tmpBuf;
const uint16_t *data = EcmaString::GetUtf16DataFlat(this, tmpBuf);
ASSERT(base::utf_helper::Utf16ToUtf8Size(data, strLen, modify) > 0);
size_t len = base::utf_helper::Utf16ToUtf8Size(data, strLen, modify) - 1;
buf.reserve(len);
len = base::utf_helper::ConvertRegionUtf16ToUtf8(data, buf.data(), strLen, len, 0, modify);
str = Span<const uint8_t>(buf.data(), len);
} else {
const uint8_t *data = EcmaString::GetUtf8DataFlat(this, buf);
str = Span<const uint8_t>(data, strLen);
}
return str;
}
Span<const uint8_t> DebuggerToUtf8Span(CVector<uint8_t> &buf, bool modify = true)
{
Span<const uint8_t> str;
uint32_t strLen = GetLength();
if (UNLIKELY(IsUtf16())) {
CVector<uint16_t> tmpBuf;
const uint16_t *data = EcmaString::GetUtf16DataFlat(this, tmpBuf);
size_t len = base::utf_helper::Utf16ToUtf8Size(data, strLen, modify) - 1;
buf.reserve(len);
len = base::utf_helper::DebuggerConvertRegionUtf16ToUtf8(data, buf.data(), strLen, len, 0, modify);
str = Span<const uint8_t>(buf.data(), len);
} else {
const uint8_t *data = EcmaString::GetUtf8DataFlat(this, buf);
str = Span<const uint8_t>(data, strLen);
}
return str;
}
inline Span<const uint8_t> FastToUtf8Span() const;
bool TryToGetInteger(uint32_t *result)
{
if (!IsInteger()) {
return false;
}
ASSERT(GetLength() <= MAX_CACHED_INTEGER_SIZE);
*result = GetIntegerCode();
return true;
}
// using integer number set into hash
inline bool TryToSetIntegerHash(int32_t num)
{
uint32_t hashcode = GetMixHashcode();
if (hashcode == 0 && GetLength() != 0) {
SetRawHashcode(static_cast<uint32_t>(num), IS_INTEGER);
return true;
}
return false;
}
void WriteData(EcmaString *src, uint32_t start, uint32_t destSize, uint32_t length);
static bool CanBeCompressed(const uint8_t *utf8Data, uint32_t utf8Len);
static bool CanBeCompressed(const uint16_t *utf16Data, uint32_t utf16Len);
static bool CanBeCompressed(const EcmaString *string);
bool PUBLIC_API ToElementIndex(uint32_t *index);
bool ToInt(int32_t *index, bool *negative);
bool ToUInt64FromLoopStart(uint64_t *index, uint32_t loopStart, const uint8_t *data);
bool PUBLIC_API ToTypedArrayIndex(uint32_t *index);
template<bool isLower>
static EcmaString *ConvertCase(const EcmaVM *vm, const JSHandle<EcmaString> &src);
template<bool isLower>
static EcmaString *LocaleConvertCase(const EcmaVM *vm, const JSHandle<EcmaString> &src, const icu::Locale &locale);
template<typename T>
static EcmaString *TrimBody(const JSThread *thread, const JSHandle<EcmaString> &src, Span<T> &data, TrimMode mode);
static EcmaString *Trim(const JSThread *thread, const JSHandle<EcmaString> &src, TrimMode mode = TrimMode::TRIM);
// single char copy for loop
template<typename DstType, typename SrcType>
static void CopyChars(DstType *dst, SrcType *src, uint32_t count)
{
Span<SrcType> srcSp(src, count);
Span<DstType> dstSp(dst, count);
for (uint32_t i = 0; i < count; i++) {
dstSp[i] = srcSp[i];
}
}
// memory block copy
template<typename T>
static bool MemCopyChars(Span<T> &dst, size_t dstMax, Span<const T> &src, size_t count);
template<typename T>
static uint32_t ComputeHashForData(const T *data, size_t size, uint32_t hashSeed)
{
uint32_t hash = hashSeed;
Span<const T> sp(data, size);
for (auto c : sp) {
constexpr size_t SHIFT = 5;
hash = (hash << SHIFT) - hash + c;
}
return hash;
}
static bool IsASCIICharacter(uint16_t data)
{
if (data == 0) {
return false;
}
// \0 is not considered ASCII in Ecma-Modified-UTF8 [only modify '\u0000']
return data <= base::utf_helper::UTF8_1B_MAX;
}
template<typename T1, typename T2>
static int32_t IndexOf(Span<const T1> &lhsSp, Span<const T2> &rhsSp, int32_t pos, int32_t max);
template<typename T1, typename T2>
static int32_t LastIndexOf(Span<const T1> &lhsSp, Span<const T2> &rhsSp, int32_t pos);
bool IsFlat() const;
bool IsLineString() const
{
return GetClass()->IsLineString();
}
bool IsConstantString() const
{
return GetClass()->IsConstantString();
}
bool IsSlicedString() const
{
return GetClass()->IsSlicedString();
}
bool IsTreeString() const
{
return GetClass()->IsTreeString();
}
bool NotTreeString() const
{
return !IsTreeString();
}
bool IsLineOrConstantString() const
{
auto hclass = GetClass();
return hclass->IsLineString() || hclass->IsConstantString();
}
JSType GetStringType() const
{
JSType type = GetClass()->GetObjectType();
ASSERT(type >= JSType::STRING_FIRST && type <= JSType::STRING_LAST);
return type;
}
template <typename Char>
static void WriteToFlat(EcmaString *src, Char *buf, uint32_t maxLength);
template <typename Char>
static void WriteToFlatWithPos(EcmaString *src, Char *buf, uint32_t length, uint32_t pos);
static const uint8_t *PUBLIC_API GetUtf8DataFlat(const EcmaString *src, CVector<uint8_t> &buf);
static const uint16_t *PUBLIC_API GetUtf16DataFlat(const EcmaString *src, CVector<uint16_t> &buf);
// string must be not flat
static EcmaString *SlowFlatten(const EcmaVM *vm, const JSHandle<EcmaString> &string, MemSpaceType type);
PUBLIC_API static EcmaString *Flatten(const EcmaVM *vm, const JSHandle<EcmaString> &string,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static FlatStringInfo FlattenAllString(const EcmaVM *vm, const JSHandle<EcmaString> &string,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE);
static EcmaString *FlattenNoGC(const EcmaVM *vm, EcmaString *string);
static EcmaString *ToLower(const EcmaVM *vm, const JSHandle<EcmaString> &src);
static EcmaString *ToUpper(const EcmaVM *vm, const JSHandle<EcmaString> &src);
static EcmaString *ToLocaleLower(const EcmaVM *vm, const JSHandle<EcmaString> &src, const icu::Locale &locale);
static EcmaString *ToLocaleUpper(const EcmaVM *vm, const JSHandle<EcmaString> &src, const icu::Locale &locale);
static EcmaString *TryToLower(const EcmaVM *vm, const JSHandle<EcmaString> &src);
static EcmaString *ConvertUtf8ToLowerOrUpper(const EcmaVM *vm, const JSHandle<EcmaString> &srcFlat,
bool toLower, uint32_t startIndex = 0);
};
// The LineEcmaString abstract class captures sequential string values, only LineEcmaString can store chars data
class LineEcmaString : public EcmaString {
public:
static constexpr uint32_t MAX_LENGTH = (1 << 28) - 16;
static constexpr uint32_t INIT_LENGTH_TIMES = 4;
// DATA_OFFSET: the string data stored after the string header.
// Data can be stored in utf8 or utf16 form according to compressed bit.
static constexpr size_t DATA_OFFSET = EcmaString::SIZE; // DATA_OFFSET equal to Empty String size
CAST_CHECK(LineEcmaString, IsLineString);
DECL_VISIT_ARRAY(DATA_OFFSET, 0, GetPointerLength());
static LineEcmaString *Cast(EcmaString *str)
{
return static_cast<LineEcmaString *>(str);
}
static LineEcmaString *Cast(const EcmaString *str)
{
return LineEcmaString::Cast(const_cast<EcmaString *>(str));
}
static size_t ComputeSizeUtf8(uint32_t utf8Len)
{
return DATA_OFFSET + utf8Len;
}
static size_t ComputeSizeUtf16(uint32_t utf16Len)
{
return DATA_OFFSET + utf16Len * sizeof(uint16_t);
}
static size_t ObjectSize(EcmaString *str)
{
uint32_t length = str->GetLength();
return str->IsUtf16() ? ComputeSizeUtf16(length) : ComputeSizeUtf8(length);
}
static size_t DataSize(EcmaString *str)
{
uint32_t length = str->GetLength();
return str->IsUtf16() ? length * sizeof(uint16_t) : length;
}
size_t GetPointerLength()
{
size_t byteSize = DataSize(this);
return AlignUp(byteSize, static_cast<size_t>(MemAlignment::MEM_ALIGN_OBJECT)) / sizeof(JSTaggedType);
}
uint16_t *GetData() const
{
return reinterpret_cast<uint16_t *>(ToUintPtr(this) + DATA_OFFSET);
}
template<bool verify = true>
uint16_t Get(int32_t index) const
{
int32_t length = static_cast<int32_t>(GetLength());
if (verify) {
if ((index < 0) || (index >= length)) {
return 0;
}
}
if (!IsUtf16()) {
Span<const uint8_t> sp(GetDataUtf8(), length);
return sp[index];
}
Span<const uint16_t> sp(GetDataUtf16(), length);
return sp[index];
}
void Set(uint32_t index, uint16_t src)
{
ASSERT(index < GetLength());
if (IsUtf8()) {
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
*(reinterpret_cast<uint8_t *>(GetData()) + index) = static_cast<uint8_t>(src);
} else {
// NOLINTNEXTLINE(cppcoreguidelines-pro-bounds-pointer-arithmetic)
*(GetData() + index) = src;
}
}
};
static_assert((LineEcmaString::DATA_OFFSET % static_cast<uint8_t>(MemAlignment::MEM_ALIGN_OBJECT)) == 0);
class ConstantString : public EcmaString {
public:
static constexpr size_t RELOCTAED_DATA_OFFSET = EcmaString::SIZE;
// ConstantData is the pointer of const string in the pandafile.
// String in pandafile is encoded by the utf8 format.
// EntityId is normally the uint32_t index in the pandafile.
// When the pandafile is to be removed, EntityId will become -1.
// The real string data will be reloacted into bytearray and stored in RelocatedData.
// ConstantData will also point at data of bytearray data.
ACCESSORS(RelocatedData, RELOCTAED_DATA_OFFSET, ENTITY_ID_OFFSET);
ACCESSORS_PRIMITIVE_FIELD(EntityId, int64_t, ENTITY_ID_OFFSET, CONSTANT_DATA_OFFSET);
ACCESSORS_NATIVE_FIELD(ConstantData, uint8_t, CONSTANT_DATA_OFFSET, LAST_OFFSET);
DEFINE_ALIGN_SIZE(LAST_OFFSET);
CAST_CHECK(ConstantString, IsConstantString);
DECL_VISIT_OBJECT(RELOCTAED_DATA_OFFSET, ENTITY_ID_OFFSET);
static ConstantString *Cast(EcmaString *str)
{
return static_cast<ConstantString *>(str);
}
static ConstantString *Cast(const EcmaString *str)
{
return ConstantString::Cast(const_cast<EcmaString *>(str));
}
static size_t ObjectSize()
{
return ConstantString::SIZE;
}
uint32_t GetEntityIdU32() const
{
ASSERT(GetEntityId() >= 0);
return static_cast<uint32_t>(GetEntityId());
}
template<bool verify = true>
uint16_t Get(int32_t index) const
{
int32_t length = static_cast<int32_t>(GetLength());
if (verify) {
if ((index < 0) || (index >= length)) {
return 0;
}
}
ASSERT(IsUtf8());
Span<const uint8_t> sp(GetConstantData(), length);
return sp[index];
}
};
// The substrings of another string use SlicedString to describe.
class SlicedString : public EcmaString {
public:
static constexpr uint32_t MIN_SLICED_ECMASTRING_LENGTH = 13;
static constexpr size_t PARENT_OFFSET = EcmaString::SIZE;
ACCESSORS(Parent, PARENT_OFFSET, STARTINDEX_OFFSET);
ACCESSORS_PRIMITIVE_FIELD(StartIndex, uint32_t, STARTINDEX_OFFSET, BACKING_STORE_FLAG);
ACCESSORS_PRIMITIVE_FIELD(HasBackingStore, uint32_t, BACKING_STORE_FLAG, SIZE);
DECL_VISIT_OBJECT(PARENT_OFFSET, STARTINDEX_OFFSET);
CAST_CHECK(SlicedString, IsSlicedString);
private:
friend class EcmaString;
static SlicedString *Cast(EcmaString *str)
{
return static_cast<SlicedString *>(str);
}
static SlicedString *Cast(const EcmaString *str)
{
return SlicedString::Cast(const_cast<EcmaString *>(str));
}
static size_t ObjectSize()
{
return SlicedString::SIZE;
}
// Minimum length for a sliced string
template<bool verify = true>
uint16_t Get(int32_t index) const
{
int32_t length = static_cast<int32_t>(GetLength());
if (verify) {
if ((index < 0) || (index >= length)) {
return 0;
}
}
EcmaString *parent = EcmaString::Cast(GetParent());
if (parent->IsLineString()) {
if (parent->IsUtf8()) {
Span<const uint8_t> sp(parent->GetDataUtf8() + GetStartIndex(), length);
return sp[index];
}
Span<const uint16_t> sp(parent->GetDataUtf16() + GetStartIndex(), length);
return sp[index];
}
Span<const uint8_t> sp(ConstantString::Cast(parent)->GetConstantData() + GetStartIndex(), length);
return sp[index];
}
};
class TreeEcmaString : public EcmaString {
public:
// Minimum length for a tree string
static constexpr uint32_t MIN_TREE_ECMASTRING_LENGTH = 13;
static constexpr size_t FIRST_OFFSET = EcmaString::SIZE;
ACCESSORS(First, FIRST_OFFSET, SECOND_OFFSET);
ACCESSORS(Second, SECOND_OFFSET, SIZE);
DECL_VISIT_OBJECT(FIRST_OFFSET, SIZE);
CAST_CHECK(TreeEcmaString, IsTreeString);
static TreeEcmaString *Cast(EcmaString *str)
{
return static_cast<TreeEcmaString *>(str);
}
static TreeEcmaString *Cast(const EcmaString *str)
{
return TreeEcmaString::Cast(const_cast<EcmaString *>(str));
}
bool IsFlat() const
{
auto strSecond = EcmaString::Cast(GetSecond());
return strSecond->GetLength() == 0;
}
template<bool verify = true>
uint16_t Get(int32_t index) const
{
int32_t length = static_cast<int32_t>(GetLength());
if (verify) {
if ((index < 0) || (index >= length)) {
return 0;
}
}
if (IsFlat()) {
EcmaString *first = EcmaString::Cast(GetFirst());
return first->At<verify>(index);
}
EcmaString *string = const_cast<TreeEcmaString *>(this);
while (true) {
if (string->IsTreeString()) {
EcmaString *first = EcmaString::Cast(TreeEcmaString::Cast(string)->GetFirst());
if (static_cast<int32_t>(first->GetLength()) > index) {
string = first;
} else {
index -= static_cast<int32_t>(first->GetLength());
string = EcmaString::Cast(TreeEcmaString::Cast(string)->GetSecond());
}
} else {
return string->At<verify>(index);
}
}
UNREACHABLE();
}
};
class FlatStringInfo {
public:
FlatStringInfo(EcmaString *string, uint32_t startIndex, uint32_t length) : string_(string),
startIndex_(startIndex),
length_(length) {}
bool IsUtf8() const
{
return string_->IsUtf8();
}
bool IsUtf16() const
{
return string_->IsUtf16();
}
EcmaString *GetString() const
{
return string_;
}
void SetString(EcmaString *string)
{
string_ = string;
}
uint32_t GetStartIndex() const
{
return startIndex_;
}
void SetStartIndex(uint32_t index)
{
startIndex_ = index;
}
uint32_t GetLength() const
{
return length_;
}
const uint8_t *GetDataUtf8() const;
const uint16_t *GetDataUtf16() const;
uint8_t *GetDataUtf8Writable() const;
uint16_t *GetDataUtf16Writable() const;
std::u16string ToU16String(uint32_t len = 0);
private:
EcmaString *string_ {nullptr};
uint32_t startIndex_ {0};
uint32_t length_ {0};
};
// if you want to use functions of EcmaString, please not use directly,
// and use functions of EcmaStringAccessor alternatively.
// eg: EcmaString *str = ***; str->GetLength() -----> EcmaStringAccessor(str).GetLength()
class PUBLIC_API EcmaStringAccessor {
public:
explicit EcmaStringAccessor(EcmaString *string);
explicit EcmaStringAccessor(TaggedObject *obj);
explicit EcmaStringAccessor(JSTaggedValue value);
explicit EcmaStringAccessor(const JSHandle<EcmaString> &strHandle);
static EcmaString *CreateLineString(const EcmaVM *vm, size_t length, bool compressed)
{
return EcmaString::CreateLineString(vm, length, compressed);
}
static EcmaString *CreateEmptyString(const EcmaVM *vm)
{
return EcmaString::CreateEmptyString(vm);
}
static EcmaString *CreateFromUtf8(const EcmaVM *vm, const uint8_t *utf8Data, uint32_t utf8Len, bool canBeCompress,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE, bool isConstantString = false,
uint32_t idOffset = 0)
{
return EcmaString::CreateFromUtf8(vm, utf8Data, utf8Len, canBeCompress, type, isConstantString, idOffset);
}
static EcmaString *CreateFromUtf8CompressedSubString(const EcmaVM *vm, const JSHandle<EcmaString> &string,
uint32_t offset, uint32_t utf8Len,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::CreateFromUtf8CompressedSubString(vm, string, offset, utf8Len, type);
}
static EcmaString *CreateConstantString(const EcmaVM *vm, const uint8_t *utf8Data, size_t length,
bool compressed, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE, uint32_t idOffset = 0)
{
return EcmaString::CreateConstantString(vm, utf8Data, length, compressed, type, idOffset);
}
static EcmaString *CreateUtf16StringFromUtf8(const EcmaVM *vm, const uint8_t *utf8Data, uint32_t utf8Len,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::CreateUtf16StringFromUtf8(vm, utf8Data, utf8Len, type);
}
static EcmaString *CreateFromUtf16(const EcmaVM *vm, const uint16_t *utf16Data, uint32_t utf16Len,
bool canBeCompress, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::CreateFromUtf16(vm, utf16Data, utf16Len, canBeCompress, type);
}
static EcmaString *Concat(const EcmaVM *vm, const JSHandle<EcmaString> &str1Handle,
const JSHandle<EcmaString> &str2Handle, MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::Concat(vm, str1Handle, str2Handle, type);
}
static EcmaString *CopyStringToOldSpace(const EcmaVM *vm, const JSHandle<EcmaString> &original,
uint32_t length, bool compressed)
{
return EcmaString::CopyStringToOldSpace(vm, original, length, compressed);
}
// can change src data structure
static EcmaString *FastSubString(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length)
{
return EcmaString::FastSubString(vm, src, start, length);
}
// get
static EcmaString *GetSubString(const EcmaVM *vm,
const JSHandle<EcmaString> &src, uint32_t start, uint32_t length)
{
return EcmaString::GetSubString(vm, src, start, length);
}
bool IsUtf8() const
{
return string_->IsUtf8();
}
bool IsUtf16() const
{
return string_->IsUtf16();
}
uint32_t GetLength() const
{
return string_->GetLength();
}
// require is LineString
inline size_t GetUtf8Length() const;
size_t ObjectSize() const
{
if (string_->IsLineString()) {
return LineEcmaString::ObjectSize(string_);
} if (string_->IsConstantString()) {
return ConstantString::ObjectSize();
} else {
return TreeEcmaString::SIZE;
}
}
// For TreeString, the calculation result is size of LineString correspondingly.
size_t GetFlatStringSize() const
{
if (string_->IsConstantString()) {
return ConstantString::ObjectSize();
}
return LineEcmaString::ObjectSize(string_);
}
bool IsInternString() const
{
return string_->IsInternString();
}
void SetInternString()
{
string_->SetIsInternString();
}
void ClearInternString()
{
string_->ClearInternStringFlag();
}
// require is LineString
// It's Utf8 format, but without 0 in the end.
inline const uint8_t *GetDataUtf8();
// require is LineString
inline const uint16_t *GetDataUtf16();
// not change string data structure.
// if string is not flat, this func has low efficiency.
std::u16string ToU16String(uint32_t len = 0)
{
return string_->ToU16String(len);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
std::unique_ptr<uint8_t[]> ToOneByteDataForced()
{
return string_->ToOneByteDataForced();
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
Span<const uint8_t> ToUtf8Span(CVector<uint8_t> &buf)
{
return string_->ToUtf8Span(buf);
}
// only for string is flat and using UTF8 encoding
inline Span<const uint8_t> FastToUtf8Span();
// Using string's hash to figure out whether the string can be converted to integer
inline bool TryToGetInteger(uint32_t *result)
{
return string_->TryToGetInteger(result);
}
inline bool TryToSetIntegerHash(int32_t num)
{
return string_->TryToSetIntegerHash(num);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
std::string ToStdString(StringConvertedUsage usage = StringConvertedUsage::PRINT);
std::string DebuggerToStdString(StringConvertedUsage usage = StringConvertedUsage::PRINT);
// not change string data structure.
// if string is not flat, this func has low efficiency.
CString ToCString(StringConvertedUsage usage = StringConvertedUsage::LOGICOPERATION);
// not change string data structure.
// if string is not flat, this func has low efficiency.
uint32_t WriteToFlatUtf8(uint8_t *buf, uint32_t maxLength, bool isWriteBuffer = false)
{
return string_->WriteUtf8(buf, maxLength, isWriteBuffer);
}
uint32_t WriteToUtf16(uint16_t *buf, uint32_t bufLength)
{
return string_->WriteUtf16(buf, GetLength(), bufLength);
}
uint32_t WriteToOneByte(uint8_t *buf, uint32_t maxLength)
{
return string_->WriteOneByte(buf, maxLength);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
uint32_t WriteToFlatUtf16(uint16_t *buf, uint32_t maxLength) const
{
return string_->CopyDataUtf16(buf, maxLength);
}
template <typename Char>
static void WriteToFlatWithPos(EcmaString *src, Char *buf, uint32_t length, uint32_t pos)
{
src->WriteToFlatWithPos(src, buf, length, pos);
}
template <typename Char>
static void WriteToFlat(EcmaString *src, Char *buf, uint32_t maxLength)
{
src->WriteToFlat(src, buf, maxLength);
}
// require dst is LineString
// not change src data structure.
// if src is not flat, this func has low efficiency.
inline static void ReadData(EcmaString * dst, EcmaString *src, uint32_t start, uint32_t destSize, uint32_t length);
// not change src data structure.
// if src is not flat, this func has low efficiency.
template<bool verify = true>
uint16_t Get(uint32_t index) const
{
return string_->At<verify>(index);
}
// require string is LineString.
void Set(uint32_t index, uint16_t src)
{
return string_->WriteData(index, src);
}
// not change src data structure.
// if src is not flat, this func has low efficiency.
uint32_t GetHashcode()
{
return string_->GetHashcode();
}
uint32_t GetRawHashcode()
{
return string_->GetRawHashcode();
}
// not change src data structure.
// if src is not flat, this func has low efficiency.
std::pair<uint32_t, bool> ComputeRawHashcode()
{
return string_->ComputeRawHashcode();
}
uint32_t ComputeHashcode()
{
return string_->ComputeHashcode();
}
uint32_t ComputeHashcode(uint32_t rawHashSeed, bool isInteger)
{
return string_->ComputeHashcode(rawHashSeed, isInteger);
}
static uint32_t ComputeHashcodeUtf8(const uint8_t *utf8Data, size_t utf8Len, bool canBeCompress)
{
return EcmaString::ComputeHashcodeUtf8(utf8Data, utf8Len, canBeCompress);
}
static uint32_t ComputeHashcodeUtf16(const uint16_t *utf16Data, uint32_t length)
{
return EcmaString::ComputeHashcodeUtf16(utf16Data, length);
}
// can change receiver and search data structure
static int32_t IndexOf(const EcmaVM *vm,
const JSHandle<EcmaString> &receiver, const JSHandle<EcmaString> &search, int pos = 0)
{
return EcmaString::IndexOf(vm, receiver, search, pos);
}
// can change receiver and search data structure
static int32_t LastIndexOf(const EcmaVM *vm,
const JSHandle<EcmaString> &receiver, const JSHandle<EcmaString> &search, int pos = 0)
{
return EcmaString::LastIndexOf(vm, receiver, search, pos);
}
// can change receiver and search data structure
static int32_t Compare(const EcmaVM *vm, const JSHandle<EcmaString>& left, const JSHandle<EcmaString>& right)
{
return EcmaString::Compare(vm, left, right);
}
// can change receiver and search data structure
static bool IsSubStringAt(const EcmaVM *vm, const JSHandle<EcmaString>& left,
const JSHandle<EcmaString>& right, uint32_t offset = 0)
{
return EcmaString::IsSubStringAt(vm, left, right, offset);
}
// can change str1 and str2 data structure
static bool StringsAreEqual(const EcmaVM *vm, const JSHandle<EcmaString> &str1, const JSHandle<EcmaString> &str2)
{
return EcmaString::StringsAreEqual(vm, str1, str2);
}
// not change str1 and str2 data structure.
// if str1 or str2 is not flat, this func has low efficiency.
static bool StringsAreEqual(EcmaString *str1, EcmaString *str2)
{
return EcmaString::StringsAreEqual(str1, str2);
}
// not change str1 and str2 data structure.
// if str1 or str2 is not flat, this func has low efficiency.
static bool StringsAreEqualDiffUtfEncoding(EcmaString *str1, EcmaString *str2)
{
return EcmaString::StringsAreEqualDiffUtfEncoding(str1, str2);
}
// not change str1 data structure.
// if str1 is not flat, this func has low efficiency.
static bool StringIsEqualUint8Data(const EcmaString *str1, const uint8_t *dataAddr, uint32_t dataLen,
bool canBeCompress)
{
return EcmaString::StringIsEqualUint8Data(str1, dataAddr, dataLen, canBeCompress);
}
// not change str1 data structure.
// if str1 is not flat, this func has low efficiency.
static bool StringsAreEqualUtf16(const EcmaString *str1, const uint16_t *utf16Data, uint32_t utf16Len)
{
return EcmaString::StringsAreEqualUtf16(str1, utf16Data, utf16Len);
}
// require str1 and str2 are LineString.
// not change string data structure.
// if string is not flat, this func has low efficiency.
bool EqualToSplicedString(const EcmaString *str1, const EcmaString *str2)
{
return string_->EqualToSplicedString(str1, str2);
}
static bool CanBeCompressed(const uint8_t *utf8Data, uint32_t utf8Len)
{
return EcmaString::CanBeCompressed(utf8Data, utf8Len);
}
static bool CanBeCompressed(const uint16_t *utf16Data, uint32_t utf16Len)
{
return EcmaString::CanBeCompressed(utf16Data, utf16Len);
}
// require string is LineString
static bool CanBeCompressed(const EcmaString *string)
{
return EcmaString::CanBeCompressed(string);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
bool ToElementIndex(uint32_t *index)
{
return string_->ToElementIndex(index);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
bool ToInt(int32_t *index, bool *negative)
{
return string_->ToInt(index, negative);
}
// not change string data structure.
// if string is not flat, this func has low efficiency.
bool PUBLIC_API ToTypedArrayIndex(uint32_t *index)
{
return string_->ToTypedArrayIndex(index);
}
static EcmaString *ToLower(const EcmaVM *vm, const JSHandle<EcmaString> &src)
{
return EcmaString::ToLower(vm, src);
}
static EcmaString *TryToLower(const EcmaVM *vm, const JSHandle<EcmaString> &src)
{
return EcmaString::TryToLower(vm, src);
}
static EcmaString *ToUpper(const EcmaVM *vm, const JSHandle<EcmaString> &src)
{
return EcmaString::ToUpper(vm, src);
}
static EcmaString *ToLocaleLower(const EcmaVM *vm, const JSHandle<EcmaString> &src, const icu::Locale &locale)
{
return EcmaString::ToLocaleLower(vm, src, locale);
}
static EcmaString *ToLocaleUpper(const EcmaVM *vm, const JSHandle<EcmaString> &src, const icu::Locale &locale)
{
return EcmaString::ToLocaleUpper(vm, src, locale);
}
static EcmaString *Trim(const JSThread *thread,
const JSHandle<EcmaString> &src, EcmaString::TrimMode mode = EcmaString::TrimMode::TRIM)
{
return EcmaString::Trim(thread, src, mode);
}
bool IsFlat() const
{
return string_->IsFlat();
}
bool IsLineString() const
{
return string_->IsLineString();
}
bool IsConstantString() const
{
return string_->IsConstantString();
}
bool IsSlicedString() const
{
return string_->IsSlicedString();
}
bool IsLineOrConstantString() const
{
return string_->IsLineOrConstantString();
}
bool IsTreeString() const
{
return string_->IsTreeString();
}
bool NotTreeString() const
{
return string_->NotTreeString();
}
PUBLIC_API static EcmaString *Flatten(const EcmaVM *vm, const JSHandle<EcmaString> &string,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::Flatten(vm, string, type);
}
static FlatStringInfo FlattenAllString(const EcmaVM *vm, const JSHandle<EcmaString> &string,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::FlattenAllString(vm, string, type);
}
static EcmaString *SlowFlatten(const EcmaVM *vm, const JSHandle<EcmaString> &string,
MemSpaceType type = MemSpaceType::SHARED_OLD_SPACE)
{
return EcmaString::SlowFlatten(vm, string, type);
}
static EcmaString *FlattenNoGC(const EcmaVM *vm, EcmaString *string)
{
return EcmaString::FlattenNoGC(vm, string);
}
static const uint8_t *GetUtf8DataFlat(const EcmaString *src, CVector<uint8_t> &buf)
{
return EcmaString::GetUtf8DataFlat(src, buf);
}
static const uint16_t *GetUtf16DataFlat(const EcmaString *src, CVector<uint16_t> &buf)
{
return EcmaString::GetUtf16DataFlat(src, buf);
}
static JSTaggedValue StringToList(JSThread *thread, JSHandle<JSTaggedValue> &str);
private:
EcmaString *string_ {nullptr};
};
} // namespace ecmascript
} // namespace panda
#endif // ECMASCRIPT_STRING_H