path: root/Firestore/core/src/firebase/firestore/remote/serializer.cc

/*
 * Copyright 2018 Google
 *
 * 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.
 */

#include "Firestore/core/src/firebase/firestore/remote/serializer.h"

#include <pb_decode.h>
#include <pb_encode.h>

#include <functional>
#include <map>
#include <string>
#include <utility>

#include "Firestore/Protos/nanopb/google/firestore/v1beta1/document.pb.h"
#include "Firestore/core/include/firebase/firestore/timestamp.h"
#include "Firestore/core/src/firebase/firestore/model/resource_path.h"
#include "Firestore/core/src/firebase/firestore/timestamp_internal.h"
#include "Firestore/core/src/firebase/firestore/util/firebase_assert.h"

namespace firebase {
namespace firestore {
namespace remote {

using firebase::Timestamp;
using firebase::TimestampInternal;
using firebase::firestore::model::DatabaseId;
using firebase::firestore::model::DocumentKey;
using firebase::firestore::model::FieldValue;
using firebase::firestore::model::ObjectValue;
using firebase::firestore::model::ResourcePath;
using firebase::firestore::util::Status;
using firebase::firestore::util::StatusOr;

namespace {

class Writer;

class Reader;

void EncodeObject(Writer* writer, const ObjectValue& object_value);

ObjectValue::Map DecodeObject(Reader* reader);

/**
 * Represents a nanopb tag.
 *
 * field_number is one of the field tags that nanopb generates based off of
 * the proto messages. They're typically named in the format:
 * <parentNameSpace>_<childNameSpace>_<message>_<field>_tag, e.g.
 * google_firestore_v1beta1_Document_name_tag.
 */
struct Tag {
  pb_wire_type_t wire_type;
  uint32_t field_number;
};

/**
 * Docs TODO(rsgowman). But currently, this just wraps the underlying nanopb
 * pb_ostream_t. Also doc how to check status.
 */
class Writer {
 public:
  /**
   * Creates an output stream that writes to the specified vector. Note that
   * this vector pointer must remain valid for the lifetime of this Writer.
   *
   * (This is roughly equivalent to the nanopb function
   * pb_ostream_from_buffer())
   *
   * @param out_bytes where the output should be serialized to.
   */
  static Writer Wrap(std::vector<uint8_t>* out_bytes);

  /**
   * Creates a non-writing output stream used to calculate the size of
   * the serialized output.
   */
  static Writer Sizing() {
    return Writer(PB_OSTREAM_SIZING);
  }

  /**
   * Writes a message type to the output stream.
   *
   * This essentially wraps calls to nanopb's pb_encode_tag() method.
   */
  void WriteTag(Tag tag);

  /**
   * Writes a nanopb message to the output stream.
   *
   * This essentially wraps calls to nanopb's `pb_encode()` method. If we didn't
   * use `oneof`s in our protos, this would be the primary way of encoding
   * messages.
   */
  void WriteNanopbMessage(const pb_field_t fields[], const void* src_struct);

  void WriteSize(size_t size);
  void WriteNull();
  void WriteBool(bool bool_value);
  void WriteInteger(int64_t integer_value);

  void WriteString(const std::string& string_value);

  /**
   * Writes a message and its length.
   *
   * When writing a top level message, protobuf doesn't include the length
   * (since you can get that already from the length of the binary output.) But
   * when writing a sub/nested message, you must include the length in the
   * serialization.
   *
   * Call this method when writing a nested message. Provide a function to
   * write the message itself. This method will calculate the size of the
   * written message (using the provided function with a non-writing sizing
   * stream), write out the size (and perform sanity checks), and then serialize
   * the message by calling the provided function a second time.
   */
  void WriteNestedMessage(const std::function<void(Writer*)>& write_message_fn);

  size_t bytes_written() const {
    return stream_.bytes_written;
  }

  Status status() const {
    return status_;
  }

 private:
  Status status_ = Status::OK();

  /**
   * Creates a new Writer, based on the given nanopb pb_ostream_t. Note that
   * a shallow copy will be taken. (Non-null pointers within this struct must
   * remain valid for the lifetime of this Writer.)
   */
  explicit Writer(const pb_ostream_t& stream) : stream_(stream) {
  }

  /**
   * Writes a "varint" to the output stream.
   *
   * This essentially wraps calls to nanopb's pb_encode_varint() method.
   *
   * Note that (despite the value parameter type) this works for bool, enum,
   * int32, int64, uint32 and uint64 proto field types.
   *
   * Note: This is not expected to be called directly, but rather only
   * via the other Write* methods (i.e. WriteBool, WriteLong, etc)
   *
   * @param value The value to write, represented as a uint64_t.
   */
  void WriteVarint(uint64_t value);

  pb_ostream_t stream_;
};

/**
 * Docs TODO(rsgowman). But currently, this just wraps the underlying nanopb
 * pb_istream_t.
 */
class Reader {
 public:
  /**
   * Creates an input stream that reads from the specified bytes. Note that
   * this reference must remain valid for the lifetime of this Reader.
   *
   * (This is roughly equivalent to the nanopb function
   * pb_istream_from_buffer())
   *
   * @param bytes where the input should be deserialized from.
   */
  static Reader Wrap(const uint8_t* bytes, size_t length);

  /**
   * Reads a message type from the input stream.
   *
   * This essentially wraps calls to nanopb's pb_decode_tag() method.
   */
  Tag ReadTag();

  /**
   * Reads a nanopb message from the input stream.
   *
   * This essentially wraps calls to nanopb's pb_decode() method. If we didn't
   * use `oneof`s in our protos, this would be the primary way of decoding
   * messages.
   */
  void ReadNanopbMessage(const pb_field_t fields[], void* dest_struct);

  void ReadNull();
  bool ReadBool();
  int64_t ReadInteger();

  std::string ReadString();

  /**
   * Reads a message and its length.
   *
   * Analog to Writer::WriteNestedMessage(). See that methods docs for further
   * details.
   *
   * Call this method when reading a nested message. Provide a function to read
   * the message itself.
   */
  template <typename T>
  T ReadNestedMessage(const std::function<T(Reader*)>& read_message_fn);

  size_t bytes_left() const {
    return stream_.bytes_left;
  }

  Status status() const {
    return status_;
  }

  void set_status(Status status) {
    status_ = status;
  }

 private:
  /**
   * Creates a new Reader, based on the given nanopb pb_istream_t. Note that
   * a shallow copy will be taken. (Non-null pointers within this struct must
   * remain valid for the lifetime of this Reader.)
   */
  explicit Reader(pb_istream_t stream) : stream_(stream) {
  }

  /**
   * Reads a "varint" from the input stream.
   *
   * This essentially wraps calls to nanopb's pb_decode_varint() method.
   *
   * Note that (despite the return type) this works for bool, enum, int32,
   * int64, uint32 and uint64 proto field types.
   *
   * Note: This is not expected to be called direclty, but rather only via the
   * other Decode* methods (i.e. DecodeBool, DecodeLong, etc)
   *
   * @return The decoded varint as a uint64_t.
   */
  uint64_t ReadVarint();

  Status status_ = Status::OK();

  pb_istream_t stream_;
};

Writer Writer::Wrap(std::vector<uint8_t>* out_bytes) {
  // TODO(rsgowman): find a better home for this constant.
  // A document is defined to have a max size of 1MiB - 4 bytes.
  static const size_t kMaxDocumentSize = 1 * 1024 * 1024 - 4;

  // Construct a nanopb output stream.
  //
  // Set the max_size to be the max document size (as an upper bound; one would
  // expect individual FieldValue's to be smaller than this).
  //
  // bytes_written is (always) initialized to 0. (NB: nanopb does not know or
  // care about the underlying output vector, so where we are in the vector
  // itself is irrelevant. i.e. don't use out_bytes->size())
  pb_ostream_t raw_stream = {
      /*callback=*/[](pb_ostream_t* stream, const pb_byte_t* buf,
                      size_t count) -> bool {
        auto* out_bytes = static_cast<std::vector<uint8_t>*>(stream->state);
        out_bytes->insert(out_bytes->end(), buf, buf + count);
        return true;
      },
      /*state=*/out_bytes,
      /*max_size=*/kMaxDocumentSize,
      /*bytes_written=*/0,
      /*errmsg=*/nullptr};
  return Writer(raw_stream);
}

Reader Reader::Wrap(const uint8_t* bytes, size_t length) {
  return Reader{pb_istream_from_buffer(bytes, length)};
}

// TODO(rsgowman): I've left the methods as near as possible to where they were
// before, which implies that the Writer methods are interspersed with the
// Reader methods. This should make it a bit easier to review. Refactor these to
// group the related methods together (probably within their own file rather
// than here).

void Writer::WriteTag(Tag tag) {
  if (!status_.ok()) return;

  if (!pb_encode_tag(&stream_, tag.wire_type, tag.field_number)) {
    FIREBASE_ASSERT_MESSAGE(false, PB_GET_ERROR(&stream_));
  }
}

Tag Reader::ReadTag() {
  Tag tag;
  if (!status_.ok()) return tag;

  bool eof;
  if (!pb_decode_tag(&stream_, &tag.wire_type, &tag.field_number, &eof)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
    return tag;
  }

  // nanopb code always returns a false status when setting eof.
  FIREBASE_ASSERT_MESSAGE(!eof, "nanopb set both ok status and eof to true");

  return tag;
}

void Writer::WriteNanopbMessage(const pb_field_t fields[],
                                const void* src_struct) {
  if (!status_.ok()) return;

  if (!pb_encode(&stream_, fields, src_struct)) {
    FIREBASE_ASSERT_MESSAGE(false, PB_GET_ERROR(&stream_));
  }
}

void Reader::ReadNanopbMessage(const pb_field_t fields[], void* dest_struct) {
  if (!status_.ok()) return;

  if (!pb_decode(&stream_, fields, dest_struct)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
  }
}

void Writer::WriteSize(size_t size) {
  return WriteVarint(size);
}

void Writer::WriteVarint(uint64_t value) {
  if (!status_.ok()) return;

  if (!pb_encode_varint(&stream_, value)) {
    FIREBASE_ASSERT_MESSAGE(false, PB_GET_ERROR(&stream_));
  }
}

/**
 * Note that (despite the return type) this works for bool, enum, int32, int64,
 * uint32 and uint64 proto field types.
 *
 * Note: This is not expected to be called directly, but rather only via the
 * other Decode* methods (i.e. DecodeBool, DecodeLong, etc)
 *
 * @return The decoded varint as a uint64_t.
 */
uint64_t Reader::ReadVarint() {
  if (!status_.ok()) return 0;

  uint64_t varint_value = 0;
  if (!pb_decode_varint(&stream_, &varint_value)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
  }
  return varint_value;
}

void Writer::WriteNull() {
  return WriteVarint(google_protobuf_NullValue_NULL_VALUE);
}

void Reader::ReadNull() {
  uint64_t varint = ReadVarint();
  if (!status_.ok()) return;

  if (varint != google_protobuf_NullValue_NULL_VALUE) {
    status_ = Status(FirestoreErrorCode::DataLoss,
                     "Input proto bytes cannot be parsed (invalid null value)");
  }
}

void Writer::WriteBool(bool bool_value) {
  return WriteVarint(bool_value);
}

bool Reader::ReadBool() {
  uint64_t varint = ReadVarint();
  if (!status_.ok()) return false;

  switch (varint) {
    case 0:
      return false;
    case 1:
      return true;
    default:
      status_ =
          Status(FirestoreErrorCode::DataLoss,
                 "Input proto bytes cannot be parsed (invalid bool value)");
      return false;
  }
}

void Writer::WriteInteger(int64_t integer_value) {
  return WriteVarint(integer_value);
}

int64_t Reader::ReadInteger() {
  return ReadVarint();
}

void Writer::WriteString(const std::string& string_value) {
  if (!status_.ok()) return;

  if (!pb_encode_string(
          &stream_, reinterpret_cast<const pb_byte_t*>(string_value.c_str()),
          string_value.length())) {
    FIREBASE_ASSERT_MESSAGE(false, PB_GET_ERROR(&stream_));
  }
}

std::string Reader::ReadString() {
  if (!status_.ok()) return "";

  pb_istream_t substream;
  if (!pb_make_string_substream(&stream_, &substream)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
    pb_close_string_substream(&stream_, &substream);
    return "";
  }

  std::string result(substream.bytes_left, '\0');
  if (!pb_read(&substream, reinterpret_cast<pb_byte_t*>(&result[0]),
               substream.bytes_left)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
    pb_close_string_substream(&stream_, &substream);
    return "";
  }

  // NB: future versions of nanopb read the remaining characters out of the
  // substream (and return false if that fails) as an additional safety
  // check within pb_close_string_substream. Unfortunately, that's not present
  // in the current version (0.38).  We'll make a stronger assertion and check
  // to make sure there *are* no remaining characters in the substream.
  FIREBASE_ASSERT_MESSAGE(
      substream.bytes_left == 0,
      "Bytes remaining in substream after supposedly reading all of them.");

  pb_close_string_substream(&stream_, &substream);

  return result;
}

void EncodeTimestamp(Writer* writer, const Timestamp& timestamp_value) {
  google_protobuf_Timestamp timestamp_proto =
      google_protobuf_Timestamp_init_zero;
  timestamp_proto.seconds = timestamp_value.seconds();
  timestamp_proto.nanos = timestamp_value.nanoseconds();
  writer->WriteNanopbMessage(google_protobuf_Timestamp_fields,
                             &timestamp_proto);
}

Timestamp DecodeTimestamp(Reader* reader) {
  google_protobuf_Timestamp timestamp_proto =
      google_protobuf_Timestamp_init_zero;
  reader->ReadNanopbMessage(google_protobuf_Timestamp_fields, &timestamp_proto);

  // The Timestamp ctor will assert if we provide values outside the valid
  // range. However, since we're decoding, a single corrupt byte could cause
  // this to occur, so we'll verify the ranges before passing them in since we'd
  // rather not abort in these situations.
  if (timestamp_proto.seconds < TimestampInternal::Min().seconds()) {
    reader->set_status(Status(
        FirestoreErrorCode::DataLoss,
        "Invalid message: timestamp beyond the earliest supported date"));
    return {};
  } else if (TimestampInternal::Max().seconds() < timestamp_proto.seconds) {
    reader->set_status(
        Status(FirestoreErrorCode::DataLoss,
               "Invalid message: timestamp behond the latest supported date"));
    return {};
  } else if (timestamp_proto.nanos < 0 || timestamp_proto.nanos > 999999999) {
    reader->set_status(Status(
        FirestoreErrorCode::DataLoss,
        "Invalid message: timestamp nanos must be between 0 and 999999999"));
    return {};
  }
  return Timestamp{timestamp_proto.seconds, timestamp_proto.nanos};
}

// Named '..Impl' so as to not conflict with Serializer::EncodeFieldValue.
// TODO(rsgowman): Refactor to use a helper class that wraps the stream struct.
// This will help with error handling, and should eliminate the issue of two
// 'EncodeFieldValue' methods.
void EncodeFieldValueImpl(Writer* writer, const FieldValue& field_value) {
  // TODO(rsgowman): some refactoring is in order... but will wait until after a
  // non-varint, non-fixed-size (i.e. string) type is present before doing so.
  switch (field_value.type()) {
    case FieldValue::Type::Null:
      writer->WriteTag(
          {PB_WT_VARINT, google_firestore_v1beta1_Value_null_value_tag});
      writer->WriteNull();
      break;

    case FieldValue::Type::Boolean:
      writer->WriteTag(
          {PB_WT_VARINT, google_firestore_v1beta1_Value_boolean_value_tag});
      writer->WriteBool(field_value.boolean_value());
      break;

    case FieldValue::Type::Integer:
      writer->WriteTag(
          {PB_WT_VARINT, google_firestore_v1beta1_Value_integer_value_tag});
      writer->WriteInteger(field_value.integer_value());
      break;

    case FieldValue::Type::String:
      writer->WriteTag(
          {PB_WT_STRING, google_firestore_v1beta1_Value_string_value_tag});
      writer->WriteString(field_value.string_value());
      break;

    case FieldValue::Type::Timestamp:
      writer->WriteTag(
          {PB_WT_STRING, google_firestore_v1beta1_Value_timestamp_value_tag});
      writer->WriteNestedMessage([&field_value](Writer* writer) {
        EncodeTimestamp(writer, field_value.timestamp_value());
      });
      break;

    case FieldValue::Type::Object:
      writer->WriteTag(
          {PB_WT_STRING, google_firestore_v1beta1_Value_map_value_tag});
      EncodeObject(writer, field_value.object_value());
      break;

    default:
      // TODO(rsgowman): implement the other types
      abort();
  }
}

FieldValue DecodeFieldValueImpl(Reader* reader) {
  Tag tag = reader->ReadTag();
  if (!reader->status().ok()) return FieldValue::NullValue();

  // Ensure the tag matches the wire type
  switch (tag.field_number) {
    case google_firestore_v1beta1_Value_null_value_tag:
    case google_firestore_v1beta1_Value_boolean_value_tag:
    case google_firestore_v1beta1_Value_integer_value_tag:
      if (tag.wire_type != PB_WT_VARINT) {
        reader->set_status(
            Status(FirestoreErrorCode::DataLoss,
                   "Input proto bytes cannot be parsed (mismatch between "
                   "the wiretype and the field number (tag))"));
      }
      break;

    case google_firestore_v1beta1_Value_string_value_tag:
    case google_firestore_v1beta1_Value_timestamp_value_tag:
    case google_firestore_v1beta1_Value_map_value_tag:
      if (tag.wire_type != PB_WT_STRING) {
        reader->set_status(
            Status(FirestoreErrorCode::DataLoss,
                   "Input proto bytes cannot be parsed (mismatch between "
                   "the wiretype and the field number (tag))"));
      }
      break;

    default:
      // We could get here for one of two reasons; either because the input
      // bytes are corrupt, or because we're attempting to parse a tag that we
      // haven't implemented yet. Long term, the latter reason should become
      // less likely (especially in production), so we'll assume former.

      // TODO(rsgowman): While still in development, we'll contradict the above
      // and assume the latter. Remove the following assertion when we're
      // confident that we're handling all the tags in the protos.
      FIREBASE_ASSERT_MESSAGE(
          false,
          "Unhandled message field number (tag): %i. (Or possibly "
          "corrupt input bytes)",
          tag.field_number);
      reader->set_status(Status(
          FirestoreErrorCode::DataLoss,
          "Input proto bytes cannot be parsed (invalid field number (tag))"));
  }

  if (!reader->status().ok()) return FieldValue::NullValue();

  switch (tag.field_number) {
    case google_firestore_v1beta1_Value_null_value_tag:
      reader->ReadNull();
      return FieldValue::NullValue();
    case google_firestore_v1beta1_Value_boolean_value_tag:
      return FieldValue::BooleanValue(reader->ReadBool());
    case google_firestore_v1beta1_Value_integer_value_tag:
      return FieldValue::IntegerValue(reader->ReadInteger());
    case google_firestore_v1beta1_Value_string_value_tag:
      return FieldValue::StringValue(reader->ReadString());
    case google_firestore_v1beta1_Value_timestamp_value_tag:
      return FieldValue::TimestampValue(
          reader->ReadNestedMessage<Timestamp>(DecodeTimestamp));
    case google_firestore_v1beta1_Value_map_value_tag:
      return FieldValue::ObjectValueFromMap(DecodeObject(reader));

    default:
      // This indicates an internal error as we've already ensured that this is
      // a valid field_number.
      FIREBASE_ASSERT_MESSAGE(
          false,
          "Somehow got an unexpected field number (tag) after verifying that "
          "the field number was expected.");
  }
}

void Writer::WriteNestedMessage(
    const std::function<void(Writer*)>& write_message_fn) {
  if (!status_.ok()) return;

  // First calculate the message size using a non-writing substream.
  Writer sizer = Writer::Sizing();
  write_message_fn(&sizer);
  status_ = sizer.status();
  if (!status_.ok()) return;
  size_t size = sizer.bytes_written();

  // Write out the size to the output writer.
  WriteSize(size);
  if (!status_.ok()) return;

  // If this stream is itself a sizing stream, then we don't need to actually
  // parse field_value a second time; just update the bytes_written via a call
  // to pb_write. (If we try to write the contents into a sizing stream, it'll
  // fail since sizing streams don't actually have any buffer space.)
  if (stream_.callback == nullptr) {
    if (!pb_write(&stream_, nullptr, size)) {
      FIREBASE_ASSERT_MESSAGE(false, PB_GET_ERROR(&stream_));
    }
    return;
  }

  // Ensure the output stream has enough space
  if (stream_.bytes_written + size > stream_.max_size) {
    FIREBASE_ASSERT_MESSAGE(
        false,
        "Insufficient space in the output stream to write the given message");
  }

  // Use a substream to verify that a callback doesn't write more than what it
  // did the first time. (Use an initializer rather than setting fields
  // individually like nanopb does. This gives us a *chance* of noticing if
  // nanopb adds new fields.)
  Writer writer({stream_.callback, stream_.state,
                 /*max_size=*/size, /*bytes_written=*/0,
                 /*errmsg=*/nullptr});
  write_message_fn(&writer);
  status_ = writer.status();
  if (!status_.ok()) return;

  stream_.bytes_written += writer.stream_.bytes_written;
  stream_.state = writer.stream_.state;
  stream_.errmsg = writer.stream_.errmsg;

  if (writer.bytes_written() != size) {
    // submsg size changed
    FIREBASE_ASSERT_MESSAGE(
        false, "Parsing the nested message twice yielded different sizes");
  }
}

template <typename T>
T Reader::ReadNestedMessage(const std::function<T(Reader*)>& read_message_fn) {
  // Implementation note: This is roughly modeled on pb_decode_delimited,
  // adjusted to account for the oneof in FieldValue.

  if (!status_.ok()) return T();

  pb_istream_t raw_substream;
  if (!pb_make_string_substream(&stream_, &raw_substream)) {
    status_ = Status(FirestoreErrorCode::DataLoss, PB_GET_ERROR(&stream_));
    pb_close_string_substream(&stream_, &raw_substream);
    return T();
  }
  Reader substream(raw_substream);

  // If this fails, we *won't* return right away so that we can cleanup the
  // substream (although technically, that turns out not to matter; no resource
  // leaks occur if we don't do this.)
  // TODO(rsgowman): Consider RAII here. (Watch out for Reader class which also
  // wraps streams.)
  T message = read_message_fn(&substream);
  status_ = substream.status();

  // NB: future versions of nanopb read the remaining characters out of the
  // substream (and return false if that fails) as an additional safety
  // check within pb_close_string_substream. Unfortunately, that's not present
  // in the current version (0.38).  We'll make a stronger assertion and check
  // to make sure there *are* no remaining characters in the substream.
  FIREBASE_ASSERT_MESSAGE(
      substream.bytes_left() == 0,
      "Bytes remaining in substream after supposedly reading all of them.");

  pb_close_string_substream(&stream_, &substream.stream_);

  return message;
}

/**
 * Encodes a 'FieldsEntry' object, within a FieldValue's map_value type.
 *
 * In protobuf, maps are implemented as a repeated set of key/values. For
 * instance, this:
 *   message Foo {
 *     map<string, Value> fields = 1;
 *   }
 * would be written (in proto text format) as:
 *   {
 *     fields: {key:"key string 1", value:{<Value message here>}}
 *     fields: {key:"key string 2", value:{<Value message here>}}
 *     ...
 *   }
 *
 * This method writes an individual entry from that list. It is expected that
 * this method will be called once for each entry in the map.
 *
 * @param kv The individual key/value pair to write.
 */
void EncodeFieldsEntry(Writer* writer, const ObjectValue::Map::value_type& kv) {
  // Write the key (string)
  writer->WriteTag(
      {PB_WT_STRING, google_firestore_v1beta1_MapValue_FieldsEntry_key_tag});
  writer->WriteString(kv.first);

  // Write the value (FieldValue)
  writer->WriteTag(
      {PB_WT_STRING, google_firestore_v1beta1_MapValue_FieldsEntry_value_tag});
  writer->WriteNestedMessage(
      [&kv](Writer* writer) { EncodeFieldValueImpl(writer, kv.second); });
}

ObjectValue::Map::value_type DecodeFieldsEntry(Reader* reader) {
  Tag tag = reader->ReadTag();
  if (!reader->status().ok()) return {};

  // TODO(rsgowman): figure out error handling: We can do better than a failed
  // assertion.
  FIREBASE_ASSERT(tag.field_number ==
                  google_firestore_v1beta1_MapValue_FieldsEntry_key_tag);
  FIREBASE_ASSERT(tag.wire_type == PB_WT_STRING);
  std::string key = reader->ReadString();

  tag = reader->ReadTag();
  if (!reader->status().ok()) return {};
  FIREBASE_ASSERT(tag.field_number ==
                  google_firestore_v1beta1_MapValue_FieldsEntry_value_tag);
  FIREBASE_ASSERT(tag.wire_type == PB_WT_STRING);

  FieldValue value =
      reader->ReadNestedMessage<FieldValue>(DecodeFieldValueImpl);

  return ObjectValue::Map::value_type{key, value};
}

void EncodeObject(Writer* writer, const ObjectValue& object_value) {
  return writer->WriteNestedMessage([&object_value](Writer* writer) {
    // Write each FieldsEntry (i.e. key-value pair.)
    for (const auto& kv : object_value.internal_value) {
      writer->WriteTag({PB_WT_STRING,
                        google_firestore_v1beta1_MapValue_FieldsEntry_key_tag});
      writer->WriteNestedMessage(
          [&kv](Writer* writer) { return EncodeFieldsEntry(writer, kv); });
    }
  });
}

ObjectValue::Map DecodeObject(Reader* reader) {
  if (!reader->status().ok()) return ObjectValue::Map();

  return reader->ReadNestedMessage<ObjectValue::Map>(
      [](Reader* reader) -> ObjectValue::Map {
        ObjectValue::Map result;
        if (!reader->status().ok()) return result;

        while (reader->bytes_left()) {
          Tag tag = reader->ReadTag();
          if (!reader->status().ok()) return result;
          FIREBASE_ASSERT(tag.field_number ==
                          google_firestore_v1beta1_MapValue_fields_tag);
          FIREBASE_ASSERT(tag.wire_type == PB_WT_STRING);

          ObjectValue::Map::value_type fv =
              reader->ReadNestedMessage<ObjectValue::Map::value_type>(
                  DecodeFieldsEntry);

          // Sanity check: ensure that this key doesn't already exist in the
          // map.
          // TODO(rsgowman): figure out error handling: We can do better than a
          // failed assertion.
          if (!reader->status().ok()) return result;
          FIREBASE_ASSERT(result.find(fv.first) == result.end());

          // Add this key,fieldvalue to the results map.
          result.emplace(std::move(fv));
        }
        return result;
      });
}

/**
 * Creates the prefix for a fully qualified resource path, without a local path
 * on the end.
 */
ResourcePath EncodeDatabaseId(const DatabaseId& database_id) {
  return ResourcePath{"projects", database_id.project_id(), "databases",
                      database_id.database_id()};
}

/**
 * Encodes a databaseId and resource path into the following form:
 * /projects/$projectId/database/$databaseId/documents/$path
 */
std::string EncodeResourceName(const DatabaseId& database_id,
                               const ResourcePath& path) {
  return EncodeDatabaseId(database_id)
      .Append("documents")
      .Append(path)
      .CanonicalString();
}

/**
 * Validates that a path has a prefix that looks like a valid encoded
 * databaseId.
 */
bool IsValidResourceName(const ResourcePath& path) {
  // Resource names have at least 4 components (project ID, database ID)
  // and commonly the (root) resource type, e.g. documents
  return path.size() >= 4 && path[0] == "projects" && path[2] == "databases";
}

/**
 * Decodes a fully qualified resource name into a resource path and validates
 * that there is a project and database encoded in the path. There are no
 * guarantees that a local path is also encoded in this resource name.
 */
ResourcePath DecodeResourceName(absl::string_view encoded) {
  ResourcePath resource = ResourcePath::FromString(encoded);
  FIREBASE_ASSERT_MESSAGE(IsValidResourceName(resource),
                          "Tried to deserialize invalid key %s",
                          resource.CanonicalString().c_str());
  return resource;
}

/**
 * Decodes a fully qualified resource name into a resource path and validates
 * that there is a project and database encoded in the path along with a local
 * path.
 */
ResourcePath ExtractLocalPathFromResourceName(
    const ResourcePath& resource_name) {
  FIREBASE_ASSERT_MESSAGE(
      resource_name.size() > 4 && resource_name[4] == "documents",
      "Tried to deserialize invalid key %s",
      resource_name.CanonicalString().c_str());
  return resource_name.PopFirst(5);
}

}  // namespace

Status Serializer::EncodeFieldValue(const FieldValue& field_value,
                                    std::vector<uint8_t>* out_bytes) {
  Writer writer = Writer::Wrap(out_bytes);
  EncodeFieldValueImpl(&writer, field_value);
  return writer.status();
}

StatusOr<FieldValue> Serializer::DecodeFieldValue(const uint8_t* bytes,
                                                  size_t length) {
  Reader reader = Reader::Wrap(bytes, length);
  FieldValue fv = DecodeFieldValueImpl(&reader);
  if (reader.status().ok()) {
    return fv;
  } else {
    return reader.status();
  }
}

std::string Serializer::EncodeKey(const DocumentKey& key) const {
  return EncodeResourceName(database_id_, key.path());
}

DocumentKey Serializer::DecodeKey(absl::string_view name) const {
  ResourcePath resource = DecodeResourceName(name);
  FIREBASE_ASSERT_MESSAGE(resource[1] == database_id_.project_id(),
                          "Tried to deserialize key from different project.");
  FIREBASE_ASSERT_MESSAGE(resource[3] == database_id_.database_id(),
                          "Tried to deserialize key from different database.");
  return DocumentKey{ExtractLocalPathFromResourceName(resource)};
}

}  // namespace remote
}  // namespace firestore
}  // namespace firebase