src/google/protobuf/parse_context.cc

// Protocol Buffers - Google's data interchange format
// Copyright 2008 Google Inc.  All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd

#include "google/protobuf/parse_context.h"

#include <algorithm>
#include <climits>
#include <cstddef>
#include <cstdint>
#include <cstring>
#include <memory>
#include <string>
#include <utility>

#include "absl/base/optimization.h"
#include "absl/base/prefetch.h"
#include "absl/log/absl_check.h"
#include "absl/numeric/bits.h"
#include "absl/strings/cord.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/string_view.h"
#include "absl/types/span.h"
#include "google/protobuf/internal_visibility.h"
#include "google/protobuf/message_lite.h"
#include "google/protobuf/micro_string.h"
#include "google/protobuf/port.h"
#include "google/protobuf/repeated_field.h"
#include "google/protobuf/wire_format_lite.h"
#include "utf8_validity.h"


// Must be included last.
#include "google/protobuf/port_def.inc"

namespace google {
namespace protobuf {
namespace internal {

namespace {
bool ParsingEndsInBuffer(const char* ptr, const char* end, int depth) {
  while (ptr < end) {
    uint32_t tag;
    ptr = ReadTag(ptr, &tag);
    if (ptr == nullptr || ptr > end) return false;
    // ending on 0 tag is allowed and is the major reason for the necessity of
    // this function.
    if (tag == 0) return true;
    switch (tag & 7) {
      case 0: {  // Varint
        uint64_t val;
        ptr = VarintParse(ptr, &val);
        if (ptr == nullptr) return false;
        break;
      }
      case 1: {  // fixed64
        ptr += 8;
        break;
      }
      case 2: {  // len delim
        int32_t size = ReadSize(&ptr);
        if (ptr == nullptr || size > end - ptr) return false;
        ptr += size;
        break;
      }
      case 3: {  // start group
        depth++;
        break;
      }
      case 4: {                        // end group
        if (--depth < 0) return true;  // We exit early
        break;
      }
      case 5: {  // fixed32
        ptr += 4;
        break;
      }
      default:
        return false;  // Unknown wireformat
    }
  }
  return false;
}
}  // namespace

bool EpsCopyInputStream::IsRequestedLessThanOrEqualTo(int requested,
                                                      int available) {
  return static_cast<int64_t>(static_cast<uint32_t>(requested)) <=
         static_cast<int64_t>(available);
}

bool EpsCopyInputStream::CanReadFromPtr(int requested, const char* ptr) {
  return IsRequestedLessThanOrEqualTo(requested, BytesAvailable(ptr));
}

bool EpsCopyInputStream::HasEnoughTillLimit(int requested, const char* ptr) {
  return IsRequestedLessThanOrEqualTo(requested, BytesUntilLimit(ptr));
}

// Only call if at start of tag.
bool EpsCopyInputStream::ParseEndsInSlopRegion(const char* begin, int overrun,
                                               int depth) {
  constexpr int kSlopBytes = EpsCopyInputStream::kSlopBytes;
  ABSL_DCHECK_GE(overrun, 0);
  ABSL_DCHECK_LE(overrun, kSlopBytes);
  auto ptr = begin + overrun;
  auto end = begin + kSlopBytes;
  return ParsingEndsInBuffer(ptr, end, depth);
}

const char* EpsCopyInputStream::NextBuffer(int overrun, int depth) {
  if (next_chunk_ == nullptr) return nullptr;  // We've reached end of stream.
  if (next_chunk_ != patch_buffer_) {
    ABSL_DCHECK(size_ > kSlopBytes);
    // The chunk is large enough to be used directly
    buffer_end_ = next_chunk_ + size_ - kSlopBytes;
    auto res = next_chunk_;
    next_chunk_ = patch_buffer_;
    if (aliasing_ == kOnPatch) aliasing_ = kNoDelta;
    return res;
  }
  // Move the slop bytes of previous buffer to start of the patch buffer.
  // Note we must use memmove because the previous buffer could be part of
  // patch_buffer_.
  std::memmove(patch_buffer_, buffer_end_, kSlopBytes);
  if (overall_limit_ > 0 &&
      (depth < 0 || !ParseEndsInSlopRegion(patch_buffer_, overrun, depth))) {
    const void* data;
    // ZeroCopyInputStream indicates Next may return 0 size buffers. Hence
    // we loop.
    while (StreamNext(&data)) {
      if (size_ > kSlopBytes) {
        // We got a large chunk
        std::memcpy(patch_buffer_ + kSlopBytes, data, kSlopBytes);
        next_chunk_ = static_cast<const char*>(data);
        buffer_end_ = patch_buffer_ + kSlopBytes;
        if (aliasing_ >= kNoDelta) aliasing_ = kOnPatch;
        return patch_buffer_;
      } else if (size_ > 0) {
        std::memcpy(patch_buffer_ + kSlopBytes, data, size_);
        next_chunk_ = patch_buffer_;
        buffer_end_ = patch_buffer_ + size_;
        if (aliasing_ >= kNoDelta) aliasing_ = kOnPatch;
        return patch_buffer_;
      }
      ABSL_DCHECK(size_ == 0) << size_;
    }
    overall_limit_ = 0;  // Next failed, no more needs for next
  }
  // End of stream or array
  if (aliasing_ == kNoDelta) {
    // If there is no more block and aliasing is true, the previous block
    // is still valid and we can alias. We have users relying on string_view's
    // obtained from protos to outlive the proto, when the parse was from an
    // array. This guarantees string_view's are always aliased if parsed from
    // an array.
    aliasing_ = reinterpret_cast<std::uintptr_t>(buffer_end_) -
                reinterpret_cast<std::uintptr_t>(patch_buffer_);
  }
  next_chunk_ = nullptr;
  buffer_end_ = patch_buffer_ + kSlopBytes;
  size_ = 0;
  return patch_buffer_;
}

const char* EpsCopyInputStream::Next() {
  ABSL_DCHECK(limit_ > kSlopBytes);
  auto p = NextBuffer(0 /* immaterial */, -1);
  if (p == nullptr) {
    limit_end_ = buffer_end_;
    // Distinguish ending on a pushed limit or ending on end-of-stream.
    SetEndOfStream();
    return nullptr;
  }
  limit_ -= buffer_end_ - p;  // Adjust limit_ relative to new anchor
  limit_end_ = buffer_end_ + std::min(0, limit_);
  return p;
}

std::pair<const char*, bool> EpsCopyInputStream::DoneFallback(int overrun,
                                                              int depth) {
  // Did we exceeded the limit (parse error).
  if (ABSL_PREDICT_FALSE(overrun > limit_)) return {nullptr, true};
  ABSL_DCHECK(overrun != limit_);  // Guaranteed by caller.
  ABSL_DCHECK(overrun < limit_);   // Follows from above
  // TODO Instead of this dcheck we could just assign, and remove
  // updating the limit_end from PopLimit, ie.
  // limit_end_ = buffer_end_ + (std::min)(0, limit_);
  // if (ptr < limit_end_) return {ptr, false};
  ABSL_DCHECK(limit_end_ == buffer_end_ + (std::min)(0, limit_));
  // At this point we know the following assertion holds.
  ABSL_DCHECK_GT(limit_, 0);
  ABSL_DCHECK(limit_end_ == buffer_end_);  // because limit_ > 0
  const char* p;
  do {
    // We are past the end of buffer_end_, in the slop region.
    ABSL_DCHECK_GE(overrun, 0);
    p = NextBuffer(overrun, depth);
    if (p == nullptr) {
      // We are at the end of the stream
      if (ABSL_PREDICT_FALSE(overrun != 0)) return {nullptr, true};
      ABSL_DCHECK_GT(limit_, 0);
      limit_end_ = buffer_end_;
      // Distinguish ending on a pushed limit or ending on end-of-stream.
      SetEndOfStream();
      return {buffer_end_, true};
    }
    limit_ -= buffer_end_ - p;  // Adjust limit_ relative to new anchor
    p += overrun;
    overrun = p - buffer_end_;
  } while (overrun >= 0);
  limit_end_ = buffer_end_ + std::min(0, limit_);
  return {p, false};
}

const char* EpsCopyInputStream::SkipFallback(const char* ptr, int size) {
  return AppendSize(ptr, size, [](const char* /*p*/, int /*s*/) {});
}

const char* EpsCopyInputStream::ReadStringFallback(const char* ptr, int size,
                                                   std::string* str) {
  str->clear();
  if (ABSL_PREDICT_TRUE(HasEnoughTillLimit(size, ptr))) {
    // Reserve the string up to a static safe size. If strings are bigger than
    // this we proceed by growing the string as needed. This protects against
    // malicious payloads making protobuf hold on to a lot of memory.
    str->reserve(str->size() + std::min<int>(size, kSafeStringSize));
  }
  return AppendSize(ptr, size,
                    [str](const char* p, int s) { str->append(p, s); });
}

const char* EpsCopyInputStream::ReadArray(const char* ptr,
                                          absl::Span<char> out) {
  if (CanReadFromPtr(out.size(), ptr)) {
    memcpy(out.data(), ptr, out.size());
    return ptr + out.size();
  }
  return ReadArrayFallback(ptr, out);
}

const char* EpsCopyInputStream::ReadArrayFallback(const char* ptr,
                                                  absl::Span<char> out) {
  char* dst = out.data();
  return AppendSize(ptr, out.size(), [&dst](const char* p, int s) {
    memcpy(dst, p, s);
    dst += s;
  });
}

namespace {

// A valid UTF8 ranges in [1, 4]: https://en.wikipedia.org/wiki/UTF-8
constexpr size_t kUtfMax = 4;

void UnrolledMemcpy(char* dst, const char* src, size_t size) {
  PROTOBUF_ASSUME(size < 4);
  for (size_t i = 0; i < size; ++i) {
    dst[i] = src[i];
  }
}

struct LeftoverBuffer {
  absl::string_view view() const { return {buffer, size}; }
  bool empty() const { return size == 0; }

  void assign(absl::string_view from);
  void append(absl::string_view from);
  // Removes [0, pos) and moves data to the front if any.
  void remove_prefix(uint32_t pos);

  uint32_t size = 0;
  char buffer[kUtfMax];
};

void LeftoverBuffer::assign(absl::string_view from) {
  ABSL_DCHECK_LT(from.size(), kUtfMax);

  UnrolledMemcpy(buffer, from.data(), from.size());
  size = from.size();
}

void LeftoverBuffer::append(absl::string_view from) {
  ABSL_DCHECK_LE(size + from.size(), kUtfMax);

  UnrolledMemcpy(buffer + size, from.data(), from.size());
  size += from.size();
}

void LeftoverBuffer::remove_prefix(uint32_t pos) {
  ABSL_DCHECK_LE(pos, size);

  size -= pos;
  UnrolledMemcpy(buffer, buffer + pos, size);
}

// Returns true if "fragment" potentially prefixed with "leftover" is valid
// UTF8. Copied from `CordIsValid()`:
// http://google3/util/utf8/internal/unilib.cc;l=85;rcl=740640507
bool IsViewValidUTF8WithLeftover(absl::string_view fragment,
                                 LeftoverBuffer& leftover) {
  if (size_t leftover_size = leftover.size; leftover_size > 0) {
    ABSL_DCHECK_LT(leftover_size, kUtfMax);

    // Copy into the leftover buffer until it has kUtfMax bytes, and match code
    // points within that buffer, removing them from the prefix of the next
    // chunk.
    const size_t fill_size = kUtfMax - leftover_size;
    if (fragment.size() < fill_size) {
      // If the full fragment fits in the buffer, match and consume if possible.
      leftover.append(fragment);
      // Opportunistically validate but it's okay otherwise as we may be
      // building up to a valid UTF8.
      leftover.remove_prefix(
          utf8_range::SpanStructurallyValid(leftover.view()));
      return true;
    }

    // Otherwise, fill the buffer from the prefix of the fragment, match, and
    // remove the bytes in the *match* (not the unmatched part of the buffer)
    // that originally came from the current fragment.
    //
    // Note that fragment is big enough to fill leftover to the max UTF8 value.
    // It has to have a valid UTF8.
    leftover.append(fragment.substr(0, fill_size));
    const size_t leftover_valid =
        utf8_range::SpanStructurallyValid(leftover.view());
    if (leftover_valid == 0) {
      return false;
    }
    fragment.remove_prefix(leftover_valid - leftover_size);
  }

  const size_t valid = utf8_range::SpanStructurallyValid(fragment);
  fragment.remove_prefix(valid);
  // If the last Unicode char crosses to next fragment, length must be smaller
  // than kUtfMax.
  if (kUtfMax <= fragment.size()) {
    return false;
  }

  leftover.assign(fragment);

  // So far so good. Continue to the next fragment.
  return true;
}

}  // namespace

const char* EpsCopyInputStream::VerifyUTF8(const char* ptr, size_t size) {
  if (size <= static_cast<size_t>(BytesAvailable(ptr))) {
    return utf8_range::IsStructurallyValid({ptr, size}) ? ptr + size : nullptr;
  }
  return VerifyUTF8Fallback(ptr, size);
}

const char* EpsCopyInputStream::VerifyUTF8Fallback(const char* ptr,
                                                   size_t size) {
  // Copied the implementation of CordIsValid().
  LeftoverBuffer leftover;

  ptr = AppendSize(ptr, size, [&leftover](const char* p, int s) -> bool {
    return IsViewValidUTF8WithLeftover({p, static_cast<size_t>(s)}, leftover);
  });
  return leftover.empty() ? ptr : nullptr;
}

const char* EpsCopyInputStream::AppendStringFallback(const char* ptr, int size,
                                                     std::string* str) {
  if (ABSL_PREDICT_TRUE(
          IsRequestedLessThanOrEqualTo(size, BytesUntilLimit(ptr)))) {
    // Reserve the string up to a static safe size. If strings are bigger than
    // this we proceed by growing the string as needed. This protects against
    // malicious payloads making protobuf hold on to a lot of memory.
    str->reserve(str->size() + std::min<int>(size, kSafeStringSize));
  }
  return AppendSize(ptr, size,
                    [str](const char* p, int s) { str->append(p, s); });
}

const char* EpsCopyInputStream::ReadCordFallback(const char* ptr, int size,
                                                 absl::Cord* cord) {
  if (zcis_ == nullptr) {
    if (CanReadFromPtr(size, ptr)) {
      *cord = absl::string_view(ptr, size);
      return ptr + size;
    }
    return AppendSize(ptr, size, [cord](const char* p, int s) {
      cord->Append(absl::string_view(p, s));
    });
  }
  int new_limit = buffer_end_ - ptr + limit_;
  if (!IsRequestedLessThanOrEqualTo(size, new_limit)) return nullptr;
  new_limit -= size;
  int bytes_from_buffer = BytesAvailable(ptr);
  const bool in_patch_buf = reinterpret_cast<uintptr_t>(ptr) -
                                reinterpret_cast<uintptr_t>(patch_buffer_) <=
                            kPatchBufferSize;
  if (bytes_from_buffer > kPatchBufferSize || !in_patch_buf) {
    cord->Clear();
    StreamBackUp(bytes_from_buffer);
  } else if (bytes_from_buffer == kSlopBytes && next_chunk_ != nullptr &&
             // Only backup if next_chunk_ points to a valid buffer returned by
             // ZeroCopyInputStream. This happens when NextStream() returns a
             // chunk that's smaller than or equal to kSlopBytes.
             next_chunk_ != patch_buffer_) {
    cord->Clear();
    StreamBackUp(size_);
  } else {
    size -= bytes_from_buffer;
    ABSL_DCHECK_GT(size, 0);
    *cord = absl::string_view(ptr, bytes_from_buffer);
    if (next_chunk_ == patch_buffer_) {
      // We have read to end of the last buffer returned by
      // ZeroCopyInputStream. So the stream is in the right position.
    } else if (next_chunk_ == nullptr) {
      // There is no remaining chunks. We can't read size.
      SetEndOfStream();
      return nullptr;
    } else {
      // Next chunk is already loaded
      ABSL_DCHECK(size_ > kSlopBytes);
      StreamBackUp(size_ - kSlopBytes);
    }
  }
  if (size > overall_limit_) return nullptr;
  overall_limit_ -= size;
  if (!zcis_->ReadCord(cord, size)) return nullptr;
  ptr = InitFrom(zcis_);
  limit_ = new_limit - static_cast<int>(buffer_end_ - ptr);
  limit_end_ = buffer_end_ + (std::min)(0, limit_);
  return ptr;
}


const char* EpsCopyInputStream::InitFrom(io::ZeroCopyInputStream* zcis) {
  zcis_ = zcis;
  const void* data;
  int size;
  limit_ = INT_MAX;
  if (zcis->Next(&data, &size)) {
    overall_limit_ -= size;
    if (size > kSlopBytes) {
      auto ptr = static_cast<const char*>(data);
      limit_ -= size - kSlopBytes;
      limit_end_ = buffer_end_ = ptr + size - kSlopBytes;
      next_chunk_ = patch_buffer_;
      if (aliasing_ == kOnPatch) aliasing_ = kNoDelta;
      return ptr;
    } else {
      limit_end_ = buffer_end_ = patch_buffer_ + kSlopBytes;
      next_chunk_ = patch_buffer_;
      auto ptr = patch_buffer_ + kPatchBufferSize - size;
      std::memcpy(ptr, data, size);
      return ptr;
    }
  }
  overall_limit_ = 0;
  next_chunk_ = nullptr;
  size_ = 0;
  limit_end_ = buffer_end_ = patch_buffer_;
  return patch_buffer_;
}

const char* ParseContext::ReadSizeAndPushLimitAndDepth(const char* ptr,
                                                       LimitToken* old_limit) {
  return ReadSizeAndPushLimitAndDepthInlined(ptr, old_limit);
}

const char* ParseContext::ParseMessage(MessageLite* msg, const char* ptr) {
  LimitToken old;
  ptr = ReadSizeAndPushLimitAndDepth(ptr, &old);
  if (ptr == nullptr) return ptr;
  auto old_depth = depth_;
  ptr = msg->_InternalParse(ptr, this);
  if (ptr != nullptr) ABSL_DCHECK_EQ(old_depth, depth_);
  depth_++;
  if (!PopLimit(std::move(old))) return nullptr;
  return ptr;
}

inline void WriteVarint(uint64_t val, std::string* s) {
  while (val >= 128) {
    uint8_t c = val | 0x80;
    s->push_back(c);
    val >>= 7;
  }
  s->push_back(val);
}

void WriteVarint(uint32_t num, uint64_t val, std::string* s) {
  WriteVarint(num << 3, s);
  WriteVarint(val, s);
}

void WriteLengthDelimited(uint32_t num, absl::string_view val, std::string* s) {
  WriteVarint((num << 3) + 2, s);
  WriteVarint(val.size(), s);
  s->append(val.data(), val.size());
}

std::pair<const char*, uint32_t> VarintParseSlow32(const char* p,
                                                   uint32_t res) {
  for (std::uint32_t i = 1; i < 5; i++) {
    uint32_t byte = static_cast<uint8_t>(p[i]);
    res += (byte - 1) << (7 * i);
    if (ABSL_PREDICT_TRUE(byte < 128)) {
      return {p + i + 1, res};
    }
  }
  // Accept >5 bytes
  for (std::uint32_t i = 5; i < 10; i++) {
    uint32_t byte = static_cast<uint8_t>(p[i]);
    if (ABSL_PREDICT_TRUE(byte < 128)) {
      return {p + i + 1, res};
    }
  }
  return {nullptr, 0};
}

std::pair<const char*, uint64_t> VarintParseSlow64(const char* p,
                                                   uint32_t res32) {
  uint64_t res = res32;
  for (std::uint32_t i = 1; i < 10; i++) {
    uint64_t byte = static_cast<uint8_t>(p[i]);
    res += (byte - 1) << (7 * i);
    if (ABSL_PREDICT_TRUE(byte < 128)) {
      return {p + i + 1, res};
    }
  }
  return {nullptr, 0};
}

std::pair<const char*, uint32_t> ReadTagFallback(const char* p, uint32_t res) {
  for (std::uint32_t i = 2; i < 5; i++) {
    uint32_t byte = static_cast<uint8_t>(p[i]);
    res += (byte - 1) << (7 * i);
    if (ABSL_PREDICT_TRUE(byte < 128)) {
      return {p + i + 1, res};
    }
  }
  return {nullptr, 0};
}

std::pair<const char*, int32_t> ReadSizeFallback(const char* p, uint32_t res) {
  // Both byte 0 and byte 1 are guaranteed to have bit 7 set upon
  // fallback entry.
  ABSL_DCHECK(static_cast<uint8_t>(p[0]) >= 128);
  ABSL_DCHECK(static_cast<uint8_t>(p[1]) >= 128);
  res += (static_cast<uint8_t>(p[1]) - 1) << 7;
  for (std::uint32_t i = 2; i < 4; i++) {
    uint32_t byte = static_cast<uint8_t>(p[i]);
    res += (byte - 1) << (7 * i);
    if (ABSL_PREDICT_TRUE(byte < 128)) {
      return {p + i + 1, res};
    }
  }
  std::uint32_t byte = static_cast<uint8_t>(p[4]);
  if (ABSL_PREDICT_FALSE(byte >= 8)) return {nullptr, 0};  // size >= 2gb
  res += (byte - 1) << 28;
  // Protect against sign integer overflow in PushLimit. Limits are relative
  // to buffer ends and ptr could potential be kSlopBytes beyond a buffer end.
  // To protect against overflow we reject limits absurdly close to INT_MAX.
  if (ABSL_PREDICT_FALSE(res > INT_MAX - ParseContext::kSlopBytes)) {
    return {nullptr, 0};
  }
  return {p + 5, res};
}

const char* StringParser(const char* begin, const char* end, void* object,
                         ParseContext*) {
  auto str = static_cast<std::string*>(object);
  str->append(begin, end - begin);
  return end;
}

// Defined in wire_format_lite.cc
void PrintUTF8ErrorLog(absl::string_view message_name,
                       absl::string_view field_name, const char* operation_str,
                       bool emit_stacktrace);

bool VerifyUTF8(absl::string_view str, const char* field_name) {
  if (!utf8_range::IsStructurallyValid(str)) {
    PrintUTF8ErrorLog("", field_name, "parsing", false);
    return false;
  }
  return true;
}

const char* InlineGreedyStringParser(std::string* s, const char* ptr,
                                     ParseContext* ctx) {
  int size = ReadSize(&ptr);
  if (!ptr) return nullptr;
  return ctx->ReadString(ptr, size, s);
}

void WireFormatStringSink::Flush(const char* ptr) {
  ABSL_CHECK_GE(ptr, prev);
  absl::StrAppend(&data,
                  absl::string_view(prev, static_cast<size_t>(ptr - prev)));
}
void WireFormatStringSink::Append(absl::string_view view) {
  absl::StrAppend(&data, view);
  prev = view.data() + view.size();
}

template <typename SinkT>
[[nodiscard]] const char* EpsCopyInputStream::ReadArrayMaybeFlush(
    const char* ptr, absl::Span<char> out, SinkT& sink) {
  char* dst = out.data();
  return AdvancePtrMaybeFlush<char>(
      ptr, out.size(), sink, [&](absl::string_view view) {
        memcpy(dst, view.data(), view.size());
        dst += view.size();
        ABSL_DCHECK_LE(dst, out.data() + out.size());
        return true;
      });
}

template <typename SinkT>
const char* ParseContext::VerifyUTF8MaybeFlushFallback(const char* ptr,
                                                       int64_t size,
                                                       SinkT& sink) {
  // Copied the implementation of CordIsValid().
  LeftoverBuffer leftover;

  ptr = AdvancePtrMaybeFlush<char>(
      ptr, size, sink, [&leftover](absl::string_view view) -> bool {
        return IsViewValidUTF8WithLeftover(view, leftover);
      });
  return leftover.empty() ? ptr : nullptr;
}

template const char* EpsCopyInputStream::ReadArrayMaybeFlush(
    const char* ptr, absl::Span<char> out, WireFormatNoOpSink& sink);
template const char* EpsCopyInputStream::ReadArrayMaybeFlush(
    const char* ptr, absl::Span<char> out, WireFormatStringSink& sink);

template const char* ParseContext::VerifyUTF8MaybeFlushFallback(
    const char* ptr, int64_t size, WireFormatNoOpSink& sink);
template const char* ParseContext::VerifyUTF8MaybeFlushFallback(
    const char* ptr, int64_t size, WireFormatStringSink& sink);


template <typename T, bool sign>
inline const char* VarintParser(const MessageLite& parent, void* object,
                                const char* ptr, ParseContext* ctx) {
  return ctx->ReadPackedVarint(ptr, [&parent, object](uint64_t varint) {
    T val;
    if (sign) {
      if (sizeof(T) == 8) {
        val = WireFormatLite::ZigZagDecode64(varint);
      } else {
        val = WireFormatLite::ZigZagDecode32(varint);
      }
    } else {
      val = varint;
    }
    static_cast<RepeatedField<T>*>(object)->InternalAddWithArena(
        InternalVisibility(), &parent, val);
  });
}

const char* PackedInt32Parser(const MessageLite& parent, void* object,
                              const char* ptr, ParseContext* ctx) {
  return VarintParser<int32_t, false>(parent, object, ptr, ctx);
}
const char* PackedUInt32Parser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  return VarintParser<uint32_t, false>(parent, object, ptr, ctx);
}
const char* PackedInt64Parser(const MessageLite& parent, void* object,
                              const char* ptr, ParseContext* ctx) {
  return VarintParser<int64_t, false>(parent, object, ptr, ctx);
}
const char* PackedUInt64Parser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  return VarintParser<uint64_t, false>(parent, object, ptr, ctx);
}
const char* PackedSInt32Parser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  return VarintParser<int32_t, true>(parent, object, ptr, ctx);
}
const char* PackedSInt64Parser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  return VarintParser<int64_t, true>(parent, object, ptr, ctx);
}

const char* PackedEnumParser(const MessageLite& parent, void* object,
                             const char* ptr, ParseContext* ctx) {
  return VarintParser<int, false>(parent, object, ptr, ctx);
}

const char* PackedBoolParser(const MessageLite& parent, void* object,
                             const char* ptr, ParseContext* ctx) {
  return VarintParser<bool, false>(parent, object, ptr, ctx);
}

template <typename T>
inline const char* FixedParser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  int size = ReadSize(&ptr);
  return ctx->ReadPackedFixed(parent, ptr, size,
                              static_cast<RepeatedField<T>*>(object));
}

const char* PackedFixed32Parser(const MessageLite& parent, void* object,
                                const char* ptr, ParseContext* ctx) {
  return FixedParser<uint32_t>(parent, object, ptr, ctx);
}
const char* PackedSFixed32Parser(const MessageLite& parent, void* object,
                                 const char* ptr, ParseContext* ctx) {
  return FixedParser<int32_t>(parent, object, ptr, ctx);
}
const char* PackedFixed64Parser(const MessageLite& parent, void* object,
                                const char* ptr, ParseContext* ctx) {
  return FixedParser<uint64_t>(parent, object, ptr, ctx);
}
const char* PackedSFixed64Parser(const MessageLite& parent, void* object,
                                 const char* ptr, ParseContext* ctx) {
  return FixedParser<int64_t>(parent, object, ptr, ctx);
}
const char* PackedFloatParser(const MessageLite& parent, void* object,
                              const char* ptr, ParseContext* ctx) {
  return FixedParser<float>(parent, object, ptr, ctx);
}
const char* PackedDoubleParser(const MessageLite& parent, void* object,
                               const char* ptr, ParseContext* ctx) {
  return FixedParser<double>(parent, object, ptr, ctx);
}

class UnknownFieldLiteParserHelper {
 public:
  explicit UnknownFieldLiteParserHelper(std::string* unknown)
      : unknown_(unknown) {}

  void AddVarint(uint32_t num, uint64_t value) {
    if (unknown_ == nullptr) return;
    WriteVarint(num * 8, unknown_);
    WriteVarint(value, unknown_);
  }
  void AddFixed64(uint32_t num, uint64_t value) {
    if (unknown_ == nullptr) return;
    WriteVarint(num * 8 + 1, unknown_);
    char buffer[8];
    // TODO: Remove this suppression.
    (void)io::CodedOutputStream::WriteLittleEndian64ToArray(
        value, reinterpret_cast<uint8_t*>(buffer));
    unknown_->append(buffer, 8);
  }
  const char* ParseLengthDelimited(uint32_t num, const char* ptr,
                                   ParseContext* ctx) {
    int size = ReadSize(&ptr);
    GOOGLE_PROTOBUF_PARSER_ASSERT(ptr);
    if (unknown_ == nullptr) return ctx->Skip(ptr, size);
    WriteVarint(num * 8 + 2, unknown_);
    WriteVarint(size, unknown_);
    return ctx->AppendString(ptr, size, unknown_);
  }
  const char* ParseGroup(uint32_t num, const char* ptr, ParseContext* ctx) {
    if (unknown_) WriteVarint(num * 8 + 3, unknown_);
    ptr = ctx->ParseGroupInlined(ptr, num * 8 + 3, [&](const char* ptr) {
      return WireFormatParser(*this, ptr, ctx);
    });
    GOOGLE_PROTOBUF_PARSER_ASSERT(ptr);
    if (unknown_) WriteVarint(num * 8 + 4, unknown_);
    return ptr;
  }
  void AddFixed32(uint32_t num, uint32_t value) {
    if (unknown_ == nullptr) return;
    WriteVarint(num * 8 + 5, unknown_);
    char buffer[4];
    // TODO: Remove this suppression.
    (void)io::CodedOutputStream::WriteLittleEndian32ToArray(
        value, reinterpret_cast<uint8_t*>(buffer));
    unknown_->append(buffer, 4);
  }

 private:
  std::string* unknown_;
};

const char* UnknownGroupLiteParse(std::string* unknown, const char* ptr,
                                  ParseContext* ctx) {
  UnknownFieldLiteParserHelper field_parser(unknown);
  return WireFormatParser(field_parser, ptr, ctx);
}

const char* UnknownFieldParse(uint32_t tag, std::string* unknown,
                              const char* ptr, ParseContext* ctx) {
  UnknownFieldLiteParserHelper field_parser(unknown);
  return FieldParser(tag, field_parser, ptr, ctx);
}

const char* EpsCopyInputStream::ReadMicroStringFallback(
    const char* ptr, int size, MicroString& str, size_t inline_capacity,
    const MessageLite& parent) {
  str.SetInChunks(
      size, parent,
      [&](auto append) {
        ptr = AppendSize(ptr, size, [&](const char* p, int s) {
          append(absl::string_view(p, s));
        });
      },
      inline_capacity);
  return ptr;
}

int CountVarintsAssumingLargeArray(const char* ptr, const char* end) {
  // The number of varints is the number of bytes with the highest bit clear.
  // This is easier to compute as the total number of bytes, minus the number
  // of bytes with the highest bit set.
  int num_varints = end - ptr;
  ABSL_DCHECK_GE(num_varints, int{sizeof(uint64_t)});

  // Count in whole blocks, except for the last one.
  const char* const limit = end - sizeof(uint64_t);
  while (ptr < limit) {
    num_varints -=
        absl::popcount(EndianHelper<8>::Load(ptr) & 0x8080808080808080);
    ptr += sizeof(uint64_t);
  }

  // Count in the last, possibly incomplete block.
  return num_varints -
         absl::popcount(EndianHelper<8>::Load(limit) &
                        (0x8080808080808080 << ((ptr - limit) * 8)));
}

bool VerifyBoolsAssumingLargeArray(const char* ptr, const char* end) {
  ABSL_DCHECK_GE(end - ptr, int{sizeof(uint64_t)});

  // Verify whole blocks, except for the last one.
  uint64_t bit_or = 0;
  const char* const limit = end - sizeof(uint64_t);
  while (ptr < limit) {
    uint64_t block;
    std::memcpy(&block, ptr, 8);
    bit_or |= block;
    ptr += 8;
  }
  // Verify the last, possibly incomplete block.
  uint64_t block;
  std::memcpy(&block, limit, 8);
  bit_or |= block;

  return (bit_or & ~0x0101010101010101) == 0;
}

}  // namespace internal
}  // namespace protobuf
}  // namespace google

#include "google/protobuf/port_undef.inc"