Integrate internal changes

15 files changed, 3628 insertions, 1045 deletions

diff --git a/java/core/src/main/java/com/google/protobuf/ByteString.java b/java/core/src/main/java/com/google/protobuf/ByteString.java
index 3f3f9f3c..5b24976d 100644
--- a/java/core/src/main/java/com/google/protobuf/ByteString.java
+++ b/java/core/src/main/java/com/google/protobuf/ByteString.java
@@ -311,6 +311,18 @@ public abstract class ByteString implements Iterable<Byte>, Serializable {
   }
 
   /**
+   * Wraps the given bytes into a {@code ByteString}. Intended for internal only usage.
+   */
+  static ByteString wrap(ByteBuffer buffer) {
+    if (buffer.hasArray()) {
+      final int offset = buffer.arrayOffset();
+      return ByteString.wrap(buffer.array(), offset + buffer.position(), buffer.remaining());
+    } else {
+      return new NioByteString(buffer);
+    }
+  }
+
+  /**
    * Wraps the given bytes into a {@code ByteString}. Intended for internal only
    * usage to force a classload of ByteString before LiteralByteString.
    */
@@ -679,6 +691,7 @@ public abstract class ByteString implements Iterable<Byte>, Serializable {
    */
   abstract void writeTo(ByteOutput byteOutput) throws IOException;
 
+
   /**
    * Constructs a read-only {@code java.nio.ByteBuffer} whose content
    * is equal to the contents of this byte string.
@@ -820,6 +833,7 @@ public abstract class ByteString implements Iterable<Byte>, Serializable {
       return true;
     }
 
+
     /**
      * Check equality of the substring of given length of this object starting at
      * zero with another {@code ByteString} substring starting at offset.
diff --git a/java/core/src/main/java/com/google/protobuf/CodedInputStream.java b/java/core/src/main/java/com/google/protobuf/CodedInputStream.java
index e8860651..e461fa28 100644
--- a/java/core/src/main/java/com/google/protobuf/CodedInputStream.java
+++ b/java/core/src/main/java/com/google/protobuf/CodedInputStream.java
@@ -30,6 +30,14 @@
 
 package com.google.protobuf;
 
+import static com.google.protobuf.Internal.EMPTY_BYTE_ARRAY;
+import static com.google.protobuf.Internal.EMPTY_BYTE_BUFFER;
+import static com.google.protobuf.Internal.UTF_8;
+import static com.google.protobuf.Internal.checkNotNull;
+import static com.google.protobuf.WireFormat.FIXED_32_SIZE;
+import static com.google.protobuf.WireFormat.FIXED_64_SIZE;
+import static com.google.protobuf.WireFormat.MAX_VARINT_SIZE;
+
 import java.io.ByteArrayOutputStream;
 import java.io.IOException;
 import java.io.InputStream;
@@ -41,51 +49,55 @@ import java.util.List;
 /**
  * Reads and decodes protocol message fields.
  *
- * This class contains two kinds of methods:  methods that read specific
- * protocol message constructs and field types (e.g. {@link #readTag()} and
- * {@link #readInt32()}) and methods that read low-level values (e.g.
- * {@link #readRawVarint32()} and {@link #readRawBytes}).  If you are reading
- * encoded protocol messages, you should use the former methods, but if you are
- * reading some other format of your own design, use the latter.
+ * <p>This class contains two kinds of methods: methods that read specific protocol message
+ * constructs and field types (e.g. {@link #readTag()} and {@link #readInt32()}) and methods that
+ * read low-level values (e.g. {@link #readRawVarint32()} and {@link #readRawBytes}). If you are
+ * reading encoded protocol messages, you should use the former methods, but if you are reading some
+ * other format of your own design, use the latter.
  *
  * @author kenton@google.com Kenton Varda
  */
-public final class CodedInputStream {
-  /**
-   * Create a new CodedInputStream wrapping the given InputStream.
-   */
+public abstract class CodedInputStream {
+  private static final int DEFAULT_BUFFER_SIZE = 4096;
+  private static final int DEFAULT_RECURSION_LIMIT = 100;
+  private static final int DEFAULT_SIZE_LIMIT = 64 << 20; // 64MB
+
+  /** Visible for subclasses. See setRecursionLimit() */
+  int recursionDepth;
+
+  int recursionLimit = DEFAULT_RECURSION_LIMIT;
+
+  /** Visible for subclasses. See setSizeLimit() */
+  int sizeLimit = DEFAULT_SIZE_LIMIT;
+
+  /** Create a new CodedInputStream wrapping the given InputStream. */
   public static CodedInputStream newInstance(final InputStream input) {
-    return new CodedInputStream(input, BUFFER_SIZE);
+    return newInstance(input, DEFAULT_BUFFER_SIZE);
   }
-  
-  /**
-   * Create a new CodedInputStream wrapping the given InputStream.
-   */
+
+  /** Create a new CodedInputStream wrapping the given InputStream. */
   static CodedInputStream newInstance(final InputStream input, int bufferSize) {
-    return new CodedInputStream(input, bufferSize);
+    if (input == null) {
+      // TODO(nathanmittler): Ideally we should throw here. This is done for backward compatibility.
+      return newInstance(EMPTY_BYTE_ARRAY);
+    }
+    return new StreamDecoder(input, bufferSize);
   }
 
-  /**
-   * Create a new CodedInputStream wrapping the given byte array.
-   */
+  /** Create a new CodedInputStream wrapping the given byte array. */
   public static CodedInputStream newInstance(final byte[] buf) {
     return newInstance(buf, 0, buf.length);
   }
 
-  /**
-   * Create a new CodedInputStream wrapping the given byte array slice.
-   */
-  public static CodedInputStream newInstance(final byte[] buf, final int off,
-                                             final int len) {
+  /** Create a new CodedInputStream wrapping the given byte array slice. */
+  public static CodedInputStream newInstance(final byte[] buf, final int off, final int len) {
     return newInstance(buf, off, len, false /* bufferIsImmutable */);
   }
-  
-  /**
-   * Create a new CodedInputStream wrapping the given byte array slice.
-   */
+
+  /** Create a new CodedInputStream wrapping the given byte array slice. */
   static CodedInputStream newInstance(
       final byte[] buf, final int off, final int len, final boolean bufferIsImmutable) {
-    CodedInputStream result = new CodedInputStream(buf, off, len, bufferIsImmutable);
+    ArrayDecoder result = new ArrayDecoder(buf, off, len, bufferIsImmutable);
     try {
       // Some uses of CodedInputStream can be more efficient if they know
       // exactly how many bytes are available.  By pushing the end point of the
@@ -107,583 +119,358 @@ public final class CodedInputStream {
   }
 
   /**
-   * Create a new CodedInputStream wrapping the given ByteBuffer. The data
-   * starting from the ByteBuffer's current position to its limit will be read.
-   * The returned CodedInputStream may or may not share the underlying data
-   * in the ByteBuffer, therefore the ByteBuffer cannot be changed while the
-   * CodedInputStream is in use.
-   * Note that the ByteBuffer's position won't be changed by this function.
-   * Concurrent calls with the same ByteBuffer object are safe if no other
-   * thread is trying to alter the ByteBuffer's status.
+   * Create a new CodedInputStream wrapping the given ByteBuffer. The data starting from the
+   * ByteBuffer's current position to its limit will be read. The returned CodedInputStream may or
+   * may not share the underlying data in the ByteBuffer, therefore the ByteBuffer cannot be changed
+   * while the CodedInputStream is in use. Note that the ByteBuffer's position won't be changed by
+   * this function. Concurrent calls with the same ByteBuffer object are safe if no other thread is
+   * trying to alter the ByteBuffer's status.
    */
   public static CodedInputStream newInstance(ByteBuffer buf) {
+    return newInstance(buf, false /* bufferIsImmutable */);
+  }
+
+  /** Create a new CodedInputStream wrapping the given buffer. */
+  static CodedInputStream newInstance(ByteBuffer buf, boolean bufferIsImmutable) {
     if (buf.hasArray()) {
-      return newInstance(buf.array(), buf.arrayOffset() + buf.position(),
-          buf.remaining());
-    } else {
-      ByteBuffer temp = buf.duplicate();
-      byte[] buffer = new byte[temp.remaining()];
-      temp.get(buffer);
-      return newInstance(buffer);
+      return newInstance(
+          buf.array(), buf.arrayOffset() + buf.position(), buf.remaining(), bufferIsImmutable);
+    }
+
+    if (buf.isDirect() && UnsafeDirectNioDecoder.isSupported()) {
+      return new UnsafeDirectNioDecoder(buf, bufferIsImmutable);
     }
+
+    // The buffer is non-direct and does not expose the underlying array. Using the ByteBuffer API
+    // to access individual bytes is very slow, so just copy the buffer to an array.
+    // TODO(nathanmittler): Re-evaluate with Java 9
+    byte[] buffer = new byte[buf.remaining()];
+    buf.duplicate().get(buffer);
+    return newInstance(buffer, 0, buffer.length, true);
   }
 
+  /** Disable construction/inheritance outside of this class. */
+  private CodedInputStream() {}
+
   // -----------------------------------------------------------------
 
   /**
-   * Attempt to read a field tag, returning zero if we have reached EOF.
-   * Protocol message parsers use this to read tags, since a protocol message
-   * may legally end wherever a tag occurs, and zero is not a valid tag number.
+   * Attempt to read a field tag, returning zero if we have reached EOF. Protocol message parsers
+   * use this to read tags, since a protocol message may legally end wherever a tag occurs, and zero
+   * is not a valid tag number.
    */
-  public int readTag() throws IOException {
-    if (isAtEnd()) {
-      lastTag = 0;
-      return 0;
-    }
-
-    lastTag = readRawVarint32();
-    if (WireFormat.getTagFieldNumber(lastTag) == 0) {
-      // If we actually read zero (or any tag number corresponding to field
-      // number zero), that's not a valid tag.
-      throw InvalidProtocolBufferException.invalidTag();
-    }
-    return lastTag;
-  }
+  public abstract int readTag() throws IOException;
 
   /**
-   * Verifies that the last call to readTag() returned the given tag value.
-   * This is used to verify that a nested group ended with the correct
-   * end tag.
+   * Verifies that the last call to readTag() returned the given tag value. This is used to verify
+   * that a nested group ended with the correct end tag.
    *
-   * @throws InvalidProtocolBufferException {@code value} does not match the
-   *                                        last tag.
+   * @throws InvalidProtocolBufferException {@code value} does not match the last tag.
    */
-  public void checkLastTagWas(final int value)
-                              throws InvalidProtocolBufferException {
-    if (lastTag != value) {
-      throw InvalidProtocolBufferException.invalidEndTag();
-    }
-  }
+  public abstract void checkLastTagWas(final int value) throws InvalidProtocolBufferException;
 
-  public int getLastTag() {
-    return lastTag;
-  }
+  public abstract int getLastTag();
 
   /**
    * Reads and discards a single field, given its tag value.
    *
-   * @return {@code false} if the tag is an endgroup tag, in which case
-   *         nothing is skipped.  Otherwise, returns {@code true}.
+   * @return {@code false} if the tag is an endgroup tag, in which case nothing is skipped.
+   *     Otherwise, returns {@code true}.
    */
-  public boolean skipField(final int tag) throws IOException {
-    switch (WireFormat.getTagWireType(tag)) {
-      case WireFormat.WIRETYPE_VARINT:
-        skipRawVarint();
-        return true;
-      case WireFormat.WIRETYPE_FIXED64:
-        skipRawBytes(8);
-        return true;
-      case WireFormat.WIRETYPE_LENGTH_DELIMITED:
-        skipRawBytes(readRawVarint32());
-        return true;
-      case WireFormat.WIRETYPE_START_GROUP:
-        skipMessage();
-        checkLastTagWas(
-          WireFormat.makeTag(WireFormat.getTagFieldNumber(tag),
-                             WireFormat.WIRETYPE_END_GROUP));
-        return true;
-      case WireFormat.WIRETYPE_END_GROUP:
-        return false;
-      case WireFormat.WIRETYPE_FIXED32:
-        skipRawBytes(4);
-        return true;
-      default:
-        throw InvalidProtocolBufferException.invalidWireType();
-    }
-  }
+  public abstract boolean skipField(final int tag) throws IOException;
 
   /**
-   * Reads a single field and writes it to output in wire format,
-   * given its tag value.
+   * Reads a single field and writes it to output in wire format, given its tag value.
    *
-   * @return {@code false} if the tag is an endgroup tag, in which case
-   *         nothing is skipped.  Otherwise, returns {@code true}.
+   * @return {@code false} if the tag is an endgroup tag, in which case nothing is skipped.
+   *     Otherwise, returns {@code true}.
+   * @deprecated use {@code UnknownFieldSet} or {@code UnknownFieldSetLite} to skip to an output
+   *     stream.
    */
-  public boolean skipField(final int tag, final CodedOutputStream output)
-      throws IOException {
-    switch (WireFormat.getTagWireType(tag)) {
-      case WireFormat.WIRETYPE_VARINT: {
-        long value = readInt64();
-        output.writeRawVarint32(tag);
-        output.writeUInt64NoTag(value);
-        return true;
-      }
-      case WireFormat.WIRETYPE_FIXED64: {
-        long value = readRawLittleEndian64();
-        output.writeRawVarint32(tag);
-        output.writeFixed64NoTag(value);
-        return true;
-      }
-      case WireFormat.WIRETYPE_LENGTH_DELIMITED: {
-        ByteString value = readBytes();
-        output.writeRawVarint32(tag);
-        output.writeBytesNoTag(value);
-        return true;
-      }
-      case WireFormat.WIRETYPE_START_GROUP: {
-        output.writeRawVarint32(tag);
-        skipMessage(output);
-        int endtag = WireFormat.makeTag(WireFormat.getTagFieldNumber(tag),
-                                        WireFormat.WIRETYPE_END_GROUP);
-        checkLastTagWas(endtag);
-        output.writeRawVarint32(endtag);
-        return true;
-      }
-      case WireFormat.WIRETYPE_END_GROUP: {
-        return false;
-      }
-      case WireFormat.WIRETYPE_FIXED32: {
-        int value = readRawLittleEndian32();
-        output.writeRawVarint32(tag);
-        output.writeFixed32NoTag(value);
-        return true;
-      }
-      default:
-        throw InvalidProtocolBufferException.invalidWireType();
-    }
-  }
-
-  /**
-   * Reads and discards an entire message.  This will read either until EOF
-   * or until an endgroup tag, whichever comes first.
-   */
-  public void skipMessage() throws IOException {
-    while (true) {
-      final int tag = readTag();
-      if (tag == 0 || !skipField(tag)) {
-        return;
-      }
-    }
-  }
+  @Deprecated
+  public abstract boolean skipField(final int tag, final CodedOutputStream output)
+      throws IOException;
 
   /**
-   * Reads an entire message and writes it to output in wire format.
-   * This will read either until EOF or until an endgroup tag,
+   * Reads and discards an entire message. This will read either until EOF or until an endgroup tag,
    * whichever comes first.
    */
-  public void skipMessage(CodedOutputStream output) throws IOException {
-    while (true) {
-      final int tag = readTag();
-      if (tag == 0 || !skipField(tag, output)) {
-        return;
-      }
-    }
-  }
+  public abstract void skipMessage() throws IOException;
 
   /**
-   * Collects the bytes skipped and returns the data in a ByteBuffer.
+   * Reads an entire message and writes it to output in wire format. This will read either until EOF
+   * or until an endgroup tag, whichever comes first.
    */
-  private class SkippedDataSink implements RefillCallback {
-    private int lastPos = bufferPos;
-    private ByteArrayOutputStream byteArrayStream;
-
-    @Override
-    public void onRefill() {
-      if (byteArrayStream == null) {
-        byteArrayStream = new ByteArrayOutputStream();
-      }
-      byteArrayStream.write(buffer, lastPos, bufferPos - lastPos);
-      lastPos = 0;
-    }
-
-    /**
-     * Gets skipped data in a ByteBuffer. This method should only be
-     * called once.
-     */
-    ByteBuffer getSkippedData() {
-      if (byteArrayStream == null) {
-        return ByteBuffer.wrap(buffer, lastPos, bufferPos - lastPos);
-      } else {
-        byteArrayStream.write(buffer, lastPos, bufferPos);
-        return ByteBuffer.wrap(byteArrayStream.toByteArray());
-      }
-    }
-  }
+  public abstract void skipMessage(CodedOutputStream output) throws IOException;
 
 
   // -----------------------------------------------------------------
 
   /** Read a {@code double} field value from the stream. */
-  public double readDouble() throws IOException {
-    return Double.longBitsToDouble(readRawLittleEndian64());
-  }
+  public abstract double readDouble() throws IOException;
 
   /** Read a {@code float} field value from the stream. */
-  public float readFloat() throws IOException {
-    return Float.intBitsToFloat(readRawLittleEndian32());
-  }
+  public abstract float readFloat() throws IOException;
 
   /** Read a {@code uint64} field value from the stream. */
-  public long readUInt64() throws IOException {
-    return readRawVarint64();
-  }
+  public abstract long readUInt64() throws IOException;
 
   /** Read an {@code int64} field value from the stream. */
-  public long readInt64() throws IOException {
-    return readRawVarint64();
-  }
+  public abstract long readInt64() throws IOException;
 
   /** Read an {@code int32} field value from the stream. */
-  public int readInt32() throws IOException {
-    return readRawVarint32();
-  }
+  public abstract int readInt32() throws IOException;
 
   /** Read a {@code fixed64} field value from the stream. */
-  public long readFixed64() throws IOException {
-    return readRawLittleEndian64();
-  }
+  public abstract long readFixed64() throws IOException;
 
   /** Read a {@code fixed32} field value from the stream. */
-  public int readFixed32() throws IOException {
-    return readRawLittleEndian32();
-  }
+  public abstract int readFixed32() throws IOException;
 
   /** Read a {@code bool} field value from the stream. */
-  public boolean readBool() throws IOException {
-    return readRawVarint64() != 0;
-  }
+  public abstract boolean readBool() throws IOException;
 
   /**
-   * Read a {@code string} field value from the stream.
-   * If the stream contains malformed UTF-8,
+   * Read a {@code string} field value from the stream. If the stream contains malformed UTF-8,
    * replace the offending bytes with the standard UTF-8 replacement character.
    */
-  public String readString() throws IOException {
-    final int size = readRawVarint32();
-    if (size <= (bufferSize - bufferPos) && size > 0) {
-      // Fast path:  We already have the bytes in a contiguous buffer, so
-      //   just copy directly from it.
-      final String result = new String(buffer, bufferPos, size, Internal.UTF_8);
-      bufferPos += size;
-      return result;
-    } else if (size == 0) {
-      return "";
-    } else if (size <= bufferSize) {
-      refillBuffer(size);
-      String result = new String(buffer, bufferPos, size, Internal.UTF_8);
-      bufferPos += size;
-      return result;
-    } else {
-      // Slow path:  Build a byte array first then copy it.
-      return new String(readRawBytesSlowPath(size), Internal.UTF_8);
-    }
-  }
+  public abstract String readString() throws IOException;
 
   /**
-   * Read a {@code string} field value from the stream.
-   * If the stream contains malformed UTF-8,
+   * Read a {@code string} field value from the stream. If the stream contains malformed UTF-8,
    * throw exception {@link InvalidProtocolBufferException}.
    */
-  public String readStringRequireUtf8() throws IOException {
-    final int size = readRawVarint32();
-    final byte[] bytes;
-    final int oldPos = bufferPos;
-    final int pos;
-    if (size <= (bufferSize - oldPos) && size > 0) {
-      // Fast path:  We already have the bytes in a contiguous buffer, so
-      //   just copy directly from it.
-      bytes = buffer;
-      bufferPos = oldPos + size;
-      pos = oldPos;
-    } else if (size == 0) {
-      return "";
-    } else if (size <= bufferSize) {
-      refillBuffer(size);
-      bytes = buffer;
-      pos = 0;
-      bufferPos = pos + size;
-    } else {
-      // Slow path:  Build a byte array first then copy it.
-      bytes = readRawBytesSlowPath(size);
-      pos = 0;
-    }
-    // TODO(martinrb): We could save a pass by validating while decoding.
-    if (!Utf8.isValidUtf8(bytes, pos, pos + size)) {
-      throw InvalidProtocolBufferException.invalidUtf8();
-    }
-    return new String(bytes, pos, size, Internal.UTF_8);
-  }
+  public abstract String readStringRequireUtf8() throws IOException;
 
   /** Read a {@code group} field value from the stream. */
-  public void readGroup(final int fieldNumber,
-                        final MessageLite.Builder builder,
-                        final ExtensionRegistryLite extensionRegistry)
-      throws IOException {
-    if (recursionDepth >= recursionLimit) {
-      throw InvalidProtocolBufferException.recursionLimitExceeded();
-    }
-    ++recursionDepth;
-    builder.mergeFrom(this, extensionRegistry);
-    checkLastTagWas(
-      WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
-    --recursionDepth;
-  }
+  public abstract void readGroup(
+      final int fieldNumber,
+      final MessageLite.Builder builder,
+      final ExtensionRegistryLite extensionRegistry)
+      throws IOException;
 
 
   /** Read a {@code group} field value from the stream. */
-  public <T extends MessageLite> T readGroup(
-      final int fieldNumber,
-      final Parser<T> parser,
-      final ExtensionRegistryLite extensionRegistry)
-      throws IOException {
-    if (recursionDepth >= recursionLimit) {
-      throw InvalidProtocolBufferException.recursionLimitExceeded();
-    }
-    ++recursionDepth;
-    T result = parser.parsePartialFrom(this, extensionRegistry);
-    checkLastTagWas(
-      WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
-    --recursionDepth;
-    return result;
-  }
+  public abstract <T extends MessageLite> T readGroup(
+      final int fieldNumber, final Parser<T> parser, final ExtensionRegistryLite extensionRegistry)
+      throws IOException;
 
   /**
-   * Reads a {@code group} field value from the stream and merges it into the
-   * given {@link UnknownFieldSet}.
+   * Reads a {@code group} field value from the stream and merges it into the given {@link
+   * UnknownFieldSet}.
    *
-   * @deprecated UnknownFieldSet.Builder now implements MessageLite.Builder, so
-   *             you can just call {@link #readGroup}.
+   * @deprecated UnknownFieldSet.Builder now implements MessageLite.Builder, so you can just call
+   *     {@link #readGroup}.
    */
   @Deprecated
-  public void readUnknownGroup(final int fieldNumber,
-                               final MessageLite.Builder builder)
-      throws IOException {
-    // We know that UnknownFieldSet will ignore any ExtensionRegistry so it
-    // is safe to pass null here.  (We can't call
-    // ExtensionRegistry.getEmptyRegistry() because that would make this
-    // class depend on ExtensionRegistry, which is not part of the lite
-    // library.)
-    readGroup(fieldNumber, builder, null);
-  }
+  public abstract void readUnknownGroup(final int fieldNumber, final MessageLite.Builder builder)
+      throws IOException;
 
   /** Read an embedded message field value from the stream. */
-  public void readMessage(final MessageLite.Builder builder,
-                          final ExtensionRegistryLite extensionRegistry)
-      throws IOException {
-    final int length = readRawVarint32();
-    if (recursionDepth >= recursionLimit) {
-      throw InvalidProtocolBufferException.recursionLimitExceeded();
-    }
-    final int oldLimit = pushLimit(length);
-    ++recursionDepth;
-    builder.mergeFrom(this, extensionRegistry);
-    checkLastTagWas(0);
-    --recursionDepth;
-    popLimit(oldLimit);
-  }
+  public abstract void readMessage(
+      final MessageLite.Builder builder, final ExtensionRegistryLite extensionRegistry)
+      throws IOException;
 
 
   /** Read an embedded message field value from the stream. */
-  public <T extends MessageLite> T readMessage(
-      final Parser<T> parser,
-      final ExtensionRegistryLite extensionRegistry)
-      throws IOException {
-    int length = readRawVarint32();
-    if (recursionDepth >= recursionLimit) {
-      throw InvalidProtocolBufferException.recursionLimitExceeded();
-    }
-    final int oldLimit = pushLimit(length);
-    ++recursionDepth;
-    T result = parser.parsePartialFrom(this, extensionRegistry);
-    checkLastTagWas(0);
-    --recursionDepth;
-    popLimit(oldLimit);
-    return result;
-  }
+  public abstract <T extends MessageLite> T readMessage(
+      final Parser<T> parser, final ExtensionRegistryLite extensionRegistry) throws IOException;
 
   /** Read a {@code bytes} field value from the stream. */
-  public ByteString readBytes() throws IOException {
-    final int size = readRawVarint32();
-    if (size <= (bufferSize - bufferPos) && size > 0) {
-      // Fast path:  We already have the bytes in a contiguous buffer, so
-      //   just copy directly from it.
-      final ByteString result = bufferIsImmutable && enableAliasing
-          ? ByteString.wrap(buffer, bufferPos, size)
-          : ByteString.copyFrom(buffer, bufferPos, size);
-      bufferPos += size;
-      return result;
-    } else if (size == 0) {
-      return ByteString.EMPTY;
-    } else {
-      // Slow path:  Build a byte array first then copy it.
-      return ByteString.wrap(readRawBytesSlowPath(size));
-    }
-  }
+  public abstract ByteString readBytes() throws IOException;
 
   /** Read a {@code bytes} field value from the stream. */
-  public byte[] readByteArray() throws IOException {
-    final int size = readRawVarint32();
-    if (size <= (bufferSize - bufferPos) && size > 0) {
-      // Fast path: We already have the bytes in a contiguous buffer, so
-      // just copy directly from it.
-      final byte[] result =
-          Arrays.copyOfRange(buffer, bufferPos, bufferPos + size);
-      bufferPos += size;
-      return result;
-    } else {
-      // Slow path: Build a byte array first then copy it.
-      return readRawBytesSlowPath(size);
-    }
-  }
+  public abstract byte[] readByteArray() throws IOException;
 
   /** Read a {@code bytes} field value from the stream. */
-  public ByteBuffer readByteBuffer() throws IOException {
-    final int size = readRawVarint32();
-    if (size <= (bufferSize - bufferPos) && size > 0) {
-      // Fast path: We already have the bytes in a contiguous buffer.
-      // When aliasing is enabled, we can return a ByteBuffer pointing directly
-      // into the underlying byte array without copy if the CodedInputStream is
-      // constructed from a byte array. If aliasing is disabled or the input is
-      // from an InputStream or ByteString, we have to make a copy of the bytes.
-      ByteBuffer result = input == null && !bufferIsImmutable && enableAliasing
-          ? ByteBuffer.wrap(buffer, bufferPos, size).slice()
-          : ByteBuffer.wrap(Arrays.copyOfRange(
-              buffer, bufferPos, bufferPos + size));
-      bufferPos += size;
-      return result;
-    } else if (size == 0) {
-      return Internal.EMPTY_BYTE_BUFFER;
-    } else {
-      // Slow path: Build a byte array first then copy it.
-      return ByteBuffer.wrap(readRawBytesSlowPath(size));
-    }
-  }
+  public abstract ByteBuffer readByteBuffer() throws IOException;
 
   /** Read a {@code uint32} field value from the stream. */
-  public int readUInt32() throws IOException {
-    return readRawVarint32();
-  }
+  public abstract int readUInt32() throws IOException;
 
   /**
-   * Read an enum field value from the stream.  Caller is responsible
-   * for converting the numeric value to an actual enum.
+   * Read an enum field value from the stream. Caller is responsible for converting the numeric
+   * value to an actual enum.
    */
-  public int readEnum() throws IOException {
-    return readRawVarint32();
-  }
+  public abstract int readEnum() throws IOException;
 
   /** Read an {@code sfixed32} field value from the stream. */
-  public int readSFixed32() throws IOException {
-    return readRawLittleEndian32();
-  }
+  public abstract int readSFixed32() throws IOException;
 
   /** Read an {@code sfixed64} field value from the stream. */
-  public long readSFixed64() throws IOException {
-    return readRawLittleEndian64();
-  }
+  public abstract long readSFixed64() throws IOException;
 
   /** Read an {@code sint32} field value from the stream. */
-  public int readSInt32() throws IOException {
-    return decodeZigZag32(readRawVarint32());
-  }
+  public abstract int readSInt32() throws IOException;
 
   /** Read an {@code sint64} field value from the stream. */
-  public long readSInt64() throws IOException {
-    return decodeZigZag64(readRawVarint64());
-  }
+  public abstract long readSInt64() throws IOException;
 
   // =================================================================
 
+  /** Read a raw Varint from the stream. If larger than 32 bits, discard the upper bits. */
+  public abstract int readRawVarint32() throws IOException;
+
+  /** Read a raw Varint from the stream. */
+  public abstract long readRawVarint64() throws IOException;
+
+  /** Variant of readRawVarint64 for when uncomfortably close to the limit. */
+  /* Visible for testing */
+  abstract long readRawVarint64SlowPath() throws IOException;
+
+  /** Read a 32-bit little-endian integer from the stream. */
+  public abstract int readRawLittleEndian32() throws IOException;
+
+  /** Read a 64-bit little-endian integer from the stream. */
+  public abstract long readRawLittleEndian64() throws IOException;
+
+  // -----------------------------------------------------------------
+
   /**
-   * Read a raw Varint from the stream.  If larger than 32 bits, discard the
-   * upper bits.
+   * Enables {@link ByteString} aliasing of the underlying buffer, trading off on buffer pinning for
+   * data copies. Only valid for buffer-backed streams.
    */
-  public int readRawVarint32() throws IOException {
-    // See implementation notes for readRawVarint64
- fastpath: {
-      int pos = bufferPos;
-
-      if (bufferSize == pos) {
-        break fastpath;
-      }
+  public abstract void enableAliasing(boolean enabled);
 
-      final byte[] buffer = this.buffer;
-      int x;
-      if ((x = buffer[pos++]) >= 0) {
-        bufferPos = pos;
-        return x;
-      } else if (bufferSize - pos < 9) {
-        break fastpath;
-      } else if ((x ^= (buffer[pos++] << 7)) < 0) {
-        x ^= (~0 << 7);
-      } else if ((x ^= (buffer[pos++] << 14)) >= 0) {
-        x ^= (~0 << 7) ^ (~0 << 14);
-      } else if ((x ^= (buffer[pos++] << 21)) < 0) {
-        x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21);
-      } else {
-        int y = buffer[pos++];
-        x ^= y << 28;
-        x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21) ^ (~0 << 28);
-        if (y < 0 &&
-            buffer[pos++] < 0 &&
-            buffer[pos++] < 0 &&
-            buffer[pos++] < 0 &&
-            buffer[pos++] < 0 &&
-            buffer[pos++] < 0) {
-          break fastpath;  // Will throw malformedVarint()
-        }
-      }
-      bufferPos = pos;
-      return x;
+  /**
+   * Set the maximum message recursion depth. In order to prevent malicious messages from causing
+   * stack overflows, {@code CodedInputStream} limits how deeply messages may be nested. The default
+   * limit is 64.
+   *
+   * @return the old limit.
+   */
+  public final int setRecursionLimit(final int limit) {
+    if (limit < 0) {
+      throw new IllegalArgumentException("Recursion limit cannot be negative: " + limit);
     }
-    return (int) readRawVarint64SlowPath();
+    final int oldLimit = recursionLimit;
+    recursionLimit = limit;
+    return oldLimit;
   }
 
-  private void skipRawVarint() throws IOException {
-    if (bufferSize - bufferPos >= 10) {
-      final byte[] buffer = this.buffer;
-      int pos = bufferPos;
-      for (int i = 0; i < 10; i++) {
-        if (buffer[pos++] >= 0) {
-          bufferPos = pos;
-          return;
-        }
-      }
+  /**
+   * Only valid for {@link InputStream}-backed streams.
+   *
+   * <p>Set the maximum message size. In order to prevent malicious messages from exhausting memory
+   * or causing integer overflows, {@code CodedInputStream} limits how large a message may be. The
+   * default limit is 64MB. You should set this limit as small as you can without harming your app's
+   * functionality. Note that size limits only apply when reading from an {@code InputStream}, not
+   * when constructed around a raw byte array (nor with {@link ByteString#newCodedInput}).
+   *
+   * <p>If you want to read several messages from a single CodedInputStream, you could call {@link
+   * #resetSizeCounter()} after each one to avoid hitting the size limit.
+   *
+   * @return the old limit.
+   */
+  public final int setSizeLimit(final int limit) {
+    if (limit < 0) {
+      throw new IllegalArgumentException("Size limit cannot be negative: " + limit);
     }
-    skipRawVarintSlowPath();
+    final int oldLimit = sizeLimit;
+    sizeLimit = limit;
+    return oldLimit;
   }
 
-  private void skipRawVarintSlowPath() throws IOException {
-    for (int i = 0; i < 10; i++) {
-      if (readRawByte() >= 0) {
-        return;
-      }
-    }
-    throw InvalidProtocolBufferException.malformedVarint();
+  /**
+   * Resets the current size counter to zero (see {@link #setSizeLimit(int)}). Only valid for {@link
+   * InputStream}-backed streams.
+   */
+  public abstract void resetSizeCounter();
+
+  /**
+   * Sets {@code currentLimit} to (current position) + {@code byteLimit}. This is called when
+   * descending into a length-delimited embedded message.
+   *
+   * <p>Note that {@code pushLimit()} does NOT affect how many bytes the {@code CodedInputStream}
+   * reads from an underlying {@code InputStream} when refreshing its buffer. If you need to prevent
+   * reading past a certain point in the underlying {@code InputStream} (e.g. because you expect it
+   * to contain more data after the end of the message which you need to handle differently) then
+   * you must place a wrapper around your {@code InputStream} which limits the amount of data that
+   * can be read from it.
+   *
+   * @return the old limit.
+   */
+  public abstract int pushLimit(int byteLimit) throws InvalidProtocolBufferException;
+
+  /**
+   * Discards the current limit, returning to the previous limit.
+   *
+   * @param oldLimit The old limit, as returned by {@code pushLimit}.
+   */
+  public abstract void popLimit(final int oldLimit);
+
+  /**
+   * Returns the number of bytes to be read before the current limit. If no limit is set, returns
+   * -1.
+   */
+  public abstract int getBytesUntilLimit();
+
+  /**
+   * Returns true if the stream has reached the end of the input. This is the case if either the end
+   * of the underlying input source has been reached or if the stream has reached a limit created
+   * using {@link #pushLimit(int)}.
+   */
+  public abstract boolean isAtEnd() throws IOException;
+
+  /**
+   * The total bytes read up to the current position. Calling {@link #resetSizeCounter()} resets
+   * this value to zero.
+   */
+  public abstract int getTotalBytesRead();
+
+  /**
+   * Read one byte from the input.
+   *
+   * @throws InvalidProtocolBufferException The end of the stream or the current limit was reached.
+   */
+  public abstract byte readRawByte() throws IOException;
+
+  /**
+   * Read a fixed size of bytes from the input.
+   *
+   * @throws InvalidProtocolBufferException The end of the stream or the current limit was reached.
+   */
+  public abstract byte[] readRawBytes(final int size) throws IOException;
+
+  /**
+   * Reads and discards {@code size} bytes.
+   *
+   * @throws InvalidProtocolBufferException The end of the stream or the current limit was reached.
+   */
+  public abstract void skipRawBytes(final int size) throws IOException;
+
+  /**
+   * Decode a ZigZag-encoded 32-bit value. ZigZag encodes signed integers into values that can be
+   * efficiently encoded with varint. (Otherwise, negative values must be sign-extended to 64 bits
+   * to be varint encoded, thus always taking 10 bytes on the wire.)
+   *
+   * @param n An unsigned 32-bit integer, stored in a signed int because Java has no explicit
+   *     unsigned support.
+   * @return A signed 32-bit integer.
+   */
+  public static int decodeZigZag32(final int n) {
+    return (n >>> 1) ^ -(n & 1);
   }
 
   /**
-   * Reads a varint from the input one byte at a time, so that it does not
-   * read any bytes after the end of the varint.  If you simply wrapped the
-   * stream in a CodedInputStream and used {@link #readRawVarint32(InputStream)}
-   * then you would probably end up reading past the end of the varint since
-   * CodedInputStream buffers its input.
+   * Decode a ZigZag-encoded 64-bit value. ZigZag encodes signed integers into values that can be
+   * efficiently encoded with varint. (Otherwise, negative values must be sign-extended to 64 bits
+   * to be varint encoded, thus always taking 10 bytes on the wire.)
+   *
+   * @param n An unsigned 64-bit integer, stored in a signed int because Java has no explicit
+   *     unsigned support.
+   * @return A signed 64-bit integer.
    */
-  static int readRawVarint32(final InputStream input) throws IOException {
-    final int firstByte = input.read();
-    if (firstByte == -1) {
-      throw InvalidProtocolBufferException.truncatedMessage();
-    }
-    return readRawVarint32(firstByte, input);
+  public static long decodeZigZag64(final long n) {
+    return (n >>> 1) ^ -(n & 1);
   }
 
   /**
-   * Like {@link #readRawVarint32(InputStream)}, but expects that the caller
-   * has already read one byte.  This allows the caller to determine if EOF
-   * has been reached before attempting to read.
+   * Like {@link #readRawVarint32(InputStream)}, but expects that the caller has already read one
+   * byte. This allows the caller to determine if EOF has been reached before attempting to read.
    */
-  public static int readRawVarint32(
-      final int firstByte, final InputStream input) throws IOException {
+  public static int readRawVarint32(final int firstByte, final InputStream input)
+      throws IOException {
     if ((firstByte & 0x80) == 0) {
       return firstByte;
     }
@@ -713,590 +500,2396 @@ public final class CodedInputStream {
     throw InvalidProtocolBufferException.malformedVarint();
   }
 
-  /** Read a raw Varint from the stream. */
-  public long readRawVarint64() throws IOException {
-    // Implementation notes:
-    //
-    // Optimized for one-byte values, expected to be common.
-    // The particular code below was selected from various candidates
-    // empirically, by winning VarintBenchmark.
-    //
-    // Sign extension of (signed) Java bytes is usually a nuisance, but
-    // we exploit it here to more easily obtain the sign of bytes read.
-    // Instead of cleaning up the sign extension bits by masking eagerly,
-    // we delay until we find the final (positive) byte, when we clear all
-    // accumulated bits with one xor.  We depend on javac to constant fold.
- fastpath: {
-      int pos = bufferPos;
-
-      if (bufferSize == pos) {
-        break fastpath;
+  /**
+   * Reads a varint from the input one byte at a time, so that it does not read any bytes after the
+   * end of the varint. If you simply wrapped the stream in a CodedInputStream and used {@link
+   * #readRawVarint32(InputStream)} then you would probably end up reading past the end of the
+   * varint since CodedInputStream buffers its input.
+   */
+  static int readRawVarint32(final InputStream input) throws IOException {
+    final int firstByte = input.read();
+    if (firstByte == -1) {
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+    return readRawVarint32(firstByte, input);
+  }
+
+  /** A {@link CodedInputStream} implementation that uses a backing array as the input. */
+  private static final class ArrayDecoder extends CodedInputStream {
+    private final byte[] buffer;
+    private final boolean immutable;
+    private int limit;
+    private int bufferSizeAfterLimit;
+    private int pos;
+    private int startPos;
+    private int lastTag;
+    private boolean enableAliasing;
+
+    /** The absolute position of the end of the current message. */
+    private int currentLimit = Integer.MAX_VALUE;
+
+    private ArrayDecoder(final byte[] buffer, final int offset, final int len, boolean immutable) {
+      this.buffer = buffer;
+      limit = offset + len;
+      pos = offset;
+      startPos = pos;
+      this.immutable = immutable;
+    }
+
+    @Override
+    public int readTag() throws IOException {
+      if (isAtEnd()) {
+        lastTag = 0;
+        return 0;
       }
 
-      final byte[] buffer = this.buffer;
-      long x;
-      int y;
-      if ((y = buffer[pos++]) >= 0) {
-        bufferPos = pos;
-        return y;
-      } else if (bufferSize - pos < 9) {
-        break fastpath;
-      } else if ((y ^= (buffer[pos++] << 7)) < 0) {
-        x = y ^ (~0 << 7);
-      } else if ((y ^= (buffer[pos++] << 14)) >= 0) {
-        x = y ^ ((~0 << 7) ^ (~0 << 14));
-      } else if ((y ^= (buffer[pos++] << 21)) < 0) {
-        x = y ^ ((~0 << 7) ^ (~0 << 14) ^ (~0 << 21));
-      } else if ((x = ((long) y) ^ ((long) buffer[pos++] << 28)) >= 0L) {
-        x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28);
-      } else if ((x ^= ((long) buffer[pos++] << 35)) < 0L) {
-        x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35);
-      } else if ((x ^= ((long) buffer[pos++] << 42)) >= 0L) {
-        x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42);
-      } else if ((x ^= ((long) buffer[pos++] << 49)) < 0L) {
-        x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42)
-            ^ (~0L << 49);
-      } else {
-        x ^= ((long) buffer[pos++] << 56);
-        x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42)
-            ^ (~0L << 49) ^ (~0L << 56);
-        if (x < 0L) {
-          if (buffer[pos++] < 0L) {
-            break fastpath;  // Will throw malformedVarint()
+      lastTag = readRawVarint32();
+      if (WireFormat.getTagFieldNumber(lastTag) == 0) {
+        // If we actually read zero (or any tag number corresponding to field
+        // number zero), that's not a valid tag.
+        throw InvalidProtocolBufferException.invalidTag();
+      }
+      return lastTag;
+    }
+
+    @Override
+    public void checkLastTagWas(final int value) throws InvalidProtocolBufferException {
+      if (lastTag != value) {
+        throw InvalidProtocolBufferException.invalidEndTag();
+      }
+    }
+
+    @Override
+    public int getLastTag() {
+      return lastTag;
+    }
+
+    @Override
+    public boolean skipField(final int tag) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          skipRawVarint();
+          return true;
+        case WireFormat.WIRETYPE_FIXED64:
+          skipRawBytes(FIXED_64_SIZE);
+          return true;
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          skipRawBytes(readRawVarint32());
+          return true;
+        case WireFormat.WIRETYPE_START_GROUP:
+          skipMessage();
+          checkLastTagWas(
+              WireFormat.makeTag(WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP));
+          return true;
+        case WireFormat.WIRETYPE_END_GROUP:
+          return false;
+        case WireFormat.WIRETYPE_FIXED32:
+          skipRawBytes(FIXED_32_SIZE);
+          return true;
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
+      }
+    }
+
+    @Override
+    public boolean skipField(final int tag, final CodedOutputStream output) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          {
+            long value = readInt64();
+            output.writeRawVarint32(tag);
+            output.writeUInt64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_FIXED64:
+          {
+            long value = readRawLittleEndian64();
+            output.writeRawVarint32(tag);
+            output.writeFixed64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          {
+            ByteString value = readBytes();
+            output.writeRawVarint32(tag);
+            output.writeBytesNoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_START_GROUP:
+          {
+            output.writeRawVarint32(tag);
+            skipMessage(output);
+            int endtag =
+                WireFormat.makeTag(
+                    WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP);
+            checkLastTagWas(endtag);
+            output.writeRawVarint32(endtag);
+            return true;
           }
+        case WireFormat.WIRETYPE_END_GROUP:
+          {
+            return false;
+          }
+        case WireFormat.WIRETYPE_FIXED32:
+          {
+            int value = readRawLittleEndian32();
+            output.writeRawVarint32(tag);
+            output.writeFixed32NoTag(value);
+            return true;
+          }
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
+      }
+    }
+
+    @Override
+    public void skipMessage() throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag)) {
+          return;
         }
       }
-      bufferPos = pos;
-      return x;
     }
-    return readRawVarint64SlowPath();
-  }
 
-  /** Variant of readRawVarint64 for when uncomfortably close to the limit. */
-  /* Visible for testing */
-  long readRawVarint64SlowPath() throws IOException {
-    long result = 0;
-    for (int shift = 0; shift < 64; shift += 7) {
-      final byte b = readRawByte();
-      result |= (long) (b & 0x7F) << shift;
-      if ((b & 0x80) == 0) {
-        return result;
+    @Override
+    public void skipMessage(CodedOutputStream output) throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag, output)) {
+          return;
+        }
       }
     }
-    throw InvalidProtocolBufferException.malformedVarint();
-  }
 
-  /** Read a 32-bit little-endian integer from the stream. */
-  public int readRawLittleEndian32() throws IOException {
-    int pos = bufferPos;
 
-    // hand-inlined ensureAvailable(4);
-    if (bufferSize - pos < 4) {
-      refillBuffer(4);
-      pos = bufferPos;
+    // -----------------------------------------------------------------
+
+    @Override
+    public double readDouble() throws IOException {
+      return Double.longBitsToDouble(readRawLittleEndian64());
     }
 
-    final byte[] buffer = this.buffer;
-    bufferPos = pos + 4;
-    return (((buffer[pos]     & 0xff))       |
-            ((buffer[pos + 1] & 0xff) <<  8) |
-            ((buffer[pos + 2] & 0xff) << 16) |
-            ((buffer[pos + 3] & 0xff) << 24));
-  }
+    @Override
+    public float readFloat() throws IOException {
+      return Float.intBitsToFloat(readRawLittleEndian32());
+    }
 
-  /** Read a 64-bit little-endian integer from the stream. */
-  public long readRawLittleEndian64() throws IOException {
-    int pos = bufferPos;
-
-    // hand-inlined ensureAvailable(8);
-    if (bufferSize - pos < 8) {
-      refillBuffer(8);
-      pos = bufferPos;
-    }
-
-    final byte[] buffer = this.buffer;
-    bufferPos = pos + 8;
-    return ((((long) buffer[pos]     & 0xffL))       |
-            (((long) buffer[pos + 1] & 0xffL) <<  8) |
-            (((long) buffer[pos + 2] & 0xffL) << 16) |
-            (((long) buffer[pos + 3] & 0xffL) << 24) |
-            (((long) buffer[pos + 4] & 0xffL) << 32) |
-            (((long) buffer[pos + 5] & 0xffL) << 40) |
-            (((long) buffer[pos + 6] & 0xffL) << 48) |
-            (((long) buffer[pos + 7] & 0xffL) << 56));
-  }
+    @Override
+    public long readUInt64() throws IOException {
+      return readRawVarint64();
+    }
 
-  /**
-   * Decode a ZigZag-encoded 32-bit value.  ZigZag encodes signed integers
-   * into values that can be efficiently encoded with varint.  (Otherwise,
-   * negative values must be sign-extended to 64 bits to be varint encoded,
-   * thus always taking 10 bytes on the wire.)
-   *
-   * @param n An unsigned 32-bit integer, stored in a signed int because
-   *          Java has no explicit unsigned support.
-   * @return A signed 32-bit integer.
-   */
-  public static int decodeZigZag32(final int n) {
-    return (n >>> 1) ^ -(n & 1);
-  }
+    @Override
+    public long readInt64() throws IOException {
+      return readRawVarint64();
+    }
 
-  /**
-   * Decode a ZigZag-encoded 64-bit value.  ZigZag encodes signed integers
-   * into values that can be efficiently encoded with varint.  (Otherwise,
-   * negative values must be sign-extended to 64 bits to be varint encoded,
-   * thus always taking 10 bytes on the wire.)
-   *
-   * @param n An unsigned 64-bit integer, stored in a signed int because
-   *          Java has no explicit unsigned support.
-   * @return A signed 64-bit integer.
-   */
-  public static long decodeZigZag64(final long n) {
-    return (n >>> 1) ^ -(n & 1);
-  }
+    @Override
+    public int readInt32() throws IOException {
+      return readRawVarint32();
+    }
 
-  // -----------------------------------------------------------------
+    @Override
+    public long readFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
 
-  private final byte[] buffer;
-  private final boolean bufferIsImmutable;
-  private int bufferSize;
-  private int bufferSizeAfterLimit;
-  private int bufferPos;
-  private final InputStream input;
-  private int lastTag;
-  private boolean enableAliasing = false;
+    @Override
+    public int readFixed32() throws IOException {
+      return readRawLittleEndian32();
+    }
 
-  /**
-   * The total number of bytes read before the current buffer.  The total
-   * bytes read up to the current position can be computed as
-   * {@code totalBytesRetired + bufferPos}.  This value may be negative if
-   * reading started in the middle of the current buffer (e.g. if the
-   * constructor that takes a byte array and an offset was used).
-   */
-  private int totalBytesRetired;
+    @Override
+    public boolean readBool() throws IOException {
+      return readRawVarint64() != 0;
+    }
 
-  /** The absolute position of the end of the current message. */
-  private int currentLimit = Integer.MAX_VALUE;
+    @Override
+    public String readString() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= (limit - pos)) {
+        // Fast path:  We already have the bytes in a contiguous buffer, so
+        //   just copy directly from it.
+        final String result = new String(buffer, pos, size, UTF_8);
+        pos += size;
+        return result;
+      }
 
-  /** See setRecursionLimit() */
-  private int recursionDepth;
-  private int recursionLimit = DEFAULT_RECURSION_LIMIT;
+      if (size == 0) {
+        return "";
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
 
-  /** See setSizeLimit() */
-  private int sizeLimit = DEFAULT_SIZE_LIMIT;
+    @Override
+    public String readStringRequireUtf8() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= (limit - pos)) {
+        // TODO(martinrb): We could save a pass by validating while decoding.
+        if (!Utf8.isValidUtf8(buffer, pos, pos + size)) {
+          throw InvalidProtocolBufferException.invalidUtf8();
+        }
+        final int tempPos = pos;
+        pos += size;
+        return new String(buffer, tempPos, size, UTF_8);
+      }
 
-  private static final int DEFAULT_RECURSION_LIMIT = 100;
-  private static final int DEFAULT_SIZE_LIMIT = 64 << 20;  // 64MB
-  private static final int BUFFER_SIZE = 4096;
-
-  private CodedInputStream(
-      final byte[] buffer, final int off, final int len, boolean bufferIsImmutable) {
-    this.buffer = buffer;
-    bufferSize = off + len;
-    bufferPos = off;
-    totalBytesRetired = -off;
-    input = null;
-    this.bufferIsImmutable = bufferIsImmutable;
-  }
+      if (size == 0) {
+        return "";
+      }
+      if (size <= 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
 
-  private CodedInputStream(final InputStream input, int bufferSize) {
-    buffer = new byte[bufferSize];
-    bufferSize = 0;
-    bufferPos = 0;
-    totalBytesRetired = 0;
-    this.input = input;
-    bufferIsImmutable = false;
-  }
+    @Override
+    public void readGroup(
+        final int fieldNumber,
+        final MessageLite.Builder builder,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
+    }
 
-  public void enableAliasing(boolean enabled) {
-    this.enableAliasing = enabled;
-  }
 
-  /**
-   * Set the maximum message recursion depth.  In order to prevent malicious
-   * messages from causing stack overflows, {@code CodedInputStream} limits
-   * how deeply messages may be nested.  The default limit is 64.
-   *
-   * @return the old limit.
-   */
-  public int setRecursionLimit(final int limit) {
-    if (limit < 0) {
-      throw new IllegalArgumentException(
-        "Recursion limit cannot be negative: " + limit);
+    @Override
+    public <T extends MessageLite> T readGroup(
+        final int fieldNumber,
+        final Parser<T> parser,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
+      return result;
     }
-    final int oldLimit = recursionLimit;
-    recursionLimit = limit;
-    return oldLimit;
-  }
 
-  /**
-   * Set the maximum message size.  In order to prevent malicious
-   * messages from exhausting memory or causing integer overflows,
-   * {@code CodedInputStream} limits how large a message may be.
-   * The default limit is 64MB.  You should set this limit as small
-   * as you can without harming your app's functionality.  Note that
-   * size limits only apply when reading from an {@code InputStream}, not
-   * when constructed around a raw byte array (nor with
-   * {@link ByteString#newCodedInput}).
-   * <p>
-   * If you want to read several messages from a single CodedInputStream, you
-   * could call {@link #resetSizeCounter()} after each one to avoid hitting the
-   * size limit.
-   *
-   * @return the old limit.
-   */
-  public int setSizeLimit(final int limit) {
-    if (limit < 0) {
-      throw new IllegalArgumentException(
-        "Size limit cannot be negative: " + limit);
+    @Deprecated
+    @Override
+    public void readUnknownGroup(final int fieldNumber, final MessageLite.Builder builder)
+        throws IOException {
+      readGroup(fieldNumber, builder, ExtensionRegistryLite.getEmptyRegistry());
     }
-    final int oldLimit = sizeLimit;
-    sizeLimit = limit;
-    return oldLimit;
-  }
 
-  /**
-   * Resets the current size counter to zero (see {@link #setSizeLimit(int)}).
-   */
-  public void resetSizeCounter() {
-    totalBytesRetired = -bufferPos;
-  }
+    @Override
+    public void readMessage(
+        final MessageLite.Builder builder, final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      final int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+    }
 
-  /**
-   * Sets {@code currentLimit} to (current position) + {@code byteLimit}.  This
-   * is called when descending into a length-delimited embedded message.
-   *
-   * <p>Note that {@code pushLimit()} does NOT affect how many bytes the
-   * {@code CodedInputStream} reads from an underlying {@code InputStream} when
-   * refreshing its buffer.  If you need to prevent reading past a certain
-   * point in the underlying {@code InputStream} (e.g. because you expect it to
-   * contain more data after the end of the message which you need to handle
-   * differently) then you must place a wrapper around your {@code InputStream}
-   * which limits the amount of data that can be read from it.
-   *
-   * @return the old limit.
-   */
-  public int pushLimit(int byteLimit) throws InvalidProtocolBufferException {
-    if (byteLimit < 0) {
-      throw InvalidProtocolBufferException.negativeSize();
+
+    @Override
+    public <T extends MessageLite> T readMessage(
+        final Parser<T> parser, final ExtensionRegistryLite extensionRegistry) throws IOException {
+      int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+      return result;
+    }
+
+    @Override
+    public ByteString readBytes() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= (limit - pos)) {
+        // Fast path:  We already have the bytes in a contiguous buffer, so
+        //   just copy directly from it.
+        final ByteString result =
+            immutable && enableAliasing
+                ? ByteString.wrap(buffer, pos, size)
+                : ByteString.copyFrom(buffer, pos, size);
+        pos += size;
+        return result;
+      }
+      if (size == 0) {
+        return ByteString.EMPTY;
+      }
+      // Slow path:  Build a byte array first then copy it.
+      return ByteString.wrap(readRawBytes(size));
+    }
+
+    @Override
+    public byte[] readByteArray() throws IOException {
+      final int size = readRawVarint32();
+      return readRawBytes(size);
     }
-    byteLimit += totalBytesRetired + bufferPos;
-    final int oldLimit = currentLimit;
-    if (byteLimit > oldLimit) {
+
+    @Override
+    public ByteBuffer readByteBuffer() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= (limit - pos)) {
+        // Fast path: We already have the bytes in a contiguous buffer.
+        // When aliasing is enabled, we can return a ByteBuffer pointing directly
+        // into the underlying byte array without copy if the CodedInputStream is
+        // constructed from a byte array. If aliasing is disabled or the input is
+        // from an InputStream or ByteString, we have to make a copy of the bytes.
+        ByteBuffer result =
+            !immutable && enableAliasing
+                ? ByteBuffer.wrap(buffer, pos, size).slice()
+                : ByteBuffer.wrap(Arrays.copyOfRange(buffer, pos, pos + size));
+        pos += size;
+        // TODO(nathanmittler): Investigate making the ByteBuffer be made read-only
+        return result;
+      }
+
+      if (size == 0) {
+        return EMPTY_BYTE_BUFFER;
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
       throw InvalidProtocolBufferException.truncatedMessage();
     }
-    currentLimit = byteLimit;
 
-    recomputeBufferSizeAfterLimit();
+    @Override
+    public int readUInt32() throws IOException {
+      return readRawVarint32();
+    }
 
-    return oldLimit;
-  }
+    @Override
+    public int readEnum() throws IOException {
+      return readRawVarint32();
+    }
 
-  private void recomputeBufferSizeAfterLimit() {
-    bufferSize += bufferSizeAfterLimit;
-    final int bufferEnd = totalBytesRetired + bufferSize;
-    if (bufferEnd > currentLimit) {
-      // Limit is in current buffer.
-      bufferSizeAfterLimit = bufferEnd - currentLimit;
-      bufferSize -= bufferSizeAfterLimit;
-    } else {
-      bufferSizeAfterLimit = 0;
+    @Override
+    public int readSFixed32() throws IOException {
+      return readRawLittleEndian32();
     }
-  }
 
-  /**
-   * Discards the current limit, returning to the previous limit.
-   *
-   * @param oldLimit The old limit, as returned by {@code pushLimit}.
-   */
-  public void popLimit(final int oldLimit) {
-    currentLimit = oldLimit;
-    recomputeBufferSizeAfterLimit();
-  }
+    @Override
+    public long readSFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
 
-  /**
-   * Returns the number of bytes to be read before the current limit.
-   * If no limit is set, returns -1.
-   */
-  public int getBytesUntilLimit() {
-    if (currentLimit == Integer.MAX_VALUE) {
-      return -1;
+    @Override
+    public int readSInt32() throws IOException {
+      return decodeZigZag32(readRawVarint32());
     }
 
-    final int currentAbsolutePosition = totalBytesRetired + bufferPos;
-    return currentLimit - currentAbsolutePosition;
-  }
+    @Override
+    public long readSInt64() throws IOException {
+      return decodeZigZag64(readRawVarint64());
+    }
 
-  /**
-   * Returns true if the stream has reached the end of the input.  This is the
-   * case if either the end of the underlying input source has been reached or
-   * if the stream has reached a limit created using {@link #pushLimit(int)}.
-   */
-  public boolean isAtEnd() throws IOException {
-    return bufferPos == bufferSize && !tryRefillBuffer(1);
-  }
+    // =================================================================
 
-  /**
-   * The total bytes read up to the current position. Calling
-   * {@link #resetSizeCounter()} resets this value to zero.
-   */
-  public int getTotalBytesRead() {
-      return totalBytesRetired + bufferPos;
-  }
+    @Override
+    public int readRawVarint32() throws IOException {
+      // See implementation notes for readRawVarint64
+      fastpath:
+      {
+        int tempPos = pos;
+
+        if (limit == tempPos) {
+          break fastpath;
+        }
 
-  private interface RefillCallback {
-    void onRefill();
-  }
+        final byte[] buffer = this.buffer;
+        int x;
+        if ((x = buffer[tempPos++]) >= 0) {
+          pos = tempPos;
+          return x;
+        } else if (limit - tempPos < 9) {
+          break fastpath;
+        } else if ((x ^= (buffer[tempPos++] << 7)) < 0) {
+          x ^= (~0 << 7);
+        } else if ((x ^= (buffer[tempPos++] << 14)) >= 0) {
+          x ^= (~0 << 7) ^ (~0 << 14);
+        } else if ((x ^= (buffer[tempPos++] << 21)) < 0) {
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21);
+        } else {
+          int y = buffer[tempPos++];
+          x ^= y << 28;
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21) ^ (~0 << 28);
+          if (y < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0) {
+            break fastpath; // Will throw malformedVarint()
+          }
+        }
+        pos = tempPos;
+        return x;
+      }
+      return (int) readRawVarint64SlowPath();
+    }
 
-  private RefillCallback refillCallback = null;
+    private void skipRawVarint() throws IOException {
+      if (limit - pos >= MAX_VARINT_SIZE) {
+        skipRawVarintFastPath();
+      } else {
+        skipRawVarintSlowPath();
+      }
+    }
 
-  /**
-   * Reads more bytes from the input, making at least {@code n} bytes available
-   * in the buffer.  Caller must ensure that the requested space is not yet
-   * available, and that the requested space is less than BUFFER_SIZE.
-   *
-   * @throws InvalidProtocolBufferException The end of the stream or the current
-   *                                        limit was reached.
-   */
-  private void refillBuffer(int n) throws IOException {
-    if (!tryRefillBuffer(n)) {
+    private void skipRawVarintFastPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (buffer[pos++] >= 0) {
+          return;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    private void skipRawVarintSlowPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (readRawByte() >= 0) {
+          return;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    @Override
+    public long readRawVarint64() throws IOException {
+      // Implementation notes:
+      //
+      // Optimized for one-byte values, expected to be common.
+      // The particular code below was selected from various candidates
+      // empirically, by winning VarintBenchmark.
+      //
+      // Sign extension of (signed) Java bytes is usually a nuisance, but
+      // we exploit it here to more easily obtain the sign of bytes read.
+      // Instead of cleaning up the sign extension bits by masking eagerly,
+      // we delay until we find the final (positive) byte, when we clear all
+      // accumulated bits with one xor.  We depend on javac to constant fold.
+      fastpath:
+      {
+        int tempPos = pos;
+
+        if (limit == tempPos) {
+          break fastpath;
+        }
+
+        final byte[] buffer = this.buffer;
+        long x;
+        int y;
+        if ((y = buffer[tempPos++]) >= 0) {
+          pos = tempPos;
+          return y;
+        } else if (limit - tempPos < 9) {
+          break fastpath;
+        } else if ((y ^= (buffer[tempPos++] << 7)) < 0) {
+          x = y ^ (~0 << 7);
+        } else if ((y ^= (buffer[tempPos++] << 14)) >= 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14));
+        } else if ((y ^= (buffer[tempPos++] << 21)) < 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14) ^ (~0 << 21));
+        } else if ((x = y ^ ((long) buffer[tempPos++] << 28)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28);
+        } else if ((x ^= ((long) buffer[tempPos++] << 35)) < 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35);
+        } else if ((x ^= ((long) buffer[tempPos++] << 42)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42);
+        } else if ((x ^= ((long) buffer[tempPos++] << 49)) < 0L) {
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49);
+        } else {
+          x ^= ((long) buffer[tempPos++] << 56);
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49)
+                  ^ (~0L << 56);
+          if (x < 0L) {
+            if (buffer[tempPos++] < 0L) {
+              break fastpath; // Will throw malformedVarint()
+            }
+          }
+        }
+        pos = tempPos;
+        return x;
+      }
+      return readRawVarint64SlowPath();
+    }
+
+    @Override
+    long readRawVarint64SlowPath() throws IOException {
+      long result = 0;
+      for (int shift = 0; shift < 64; shift += 7) {
+        final byte b = readRawByte();
+        result |= (long) (b & 0x7F) << shift;
+        if ((b & 0x80) == 0) {
+          return result;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    @Override
+    public int readRawLittleEndian32() throws IOException {
+      int tempPos = pos;
+
+      if (limit - tempPos < FIXED_32_SIZE) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      final byte[] buffer = this.buffer;
+      pos = tempPos + FIXED_32_SIZE;
+      return (((buffer[tempPos] & 0xff))
+          | ((buffer[tempPos + 1] & 0xff) << 8)
+          | ((buffer[tempPos + 2] & 0xff) << 16)
+          | ((buffer[tempPos + 3] & 0xff) << 24));
+    }
+
+    @Override
+    public long readRawLittleEndian64() throws IOException {
+      int tempPos = pos;
+
+      if (limit - tempPos < FIXED_64_SIZE) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      final byte[] buffer = this.buffer;
+      pos = tempPos + FIXED_64_SIZE;
+      return (((buffer[tempPos] & 0xffL))
+          | ((buffer[tempPos + 1] & 0xffL) << 8)
+          | ((buffer[tempPos + 2] & 0xffL) << 16)
+          | ((buffer[tempPos + 3] & 0xffL) << 24)
+          | ((buffer[tempPos + 4] & 0xffL) << 32)
+          | ((buffer[tempPos + 5] & 0xffL) << 40)
+          | ((buffer[tempPos + 6] & 0xffL) << 48)
+          | ((buffer[tempPos + 7] & 0xffL) << 56));
+    }
+
+    @Override
+    public void enableAliasing(boolean enabled) {
+      this.enableAliasing = enabled;
+    }
+
+    @Override
+    public void resetSizeCounter() {
+      startPos = pos;
+    }
+
+    @Override
+    public int pushLimit(int byteLimit) throws InvalidProtocolBufferException {
+      if (byteLimit < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      byteLimit += getTotalBytesRead();
+      final int oldLimit = currentLimit;
+      if (byteLimit > oldLimit) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+      currentLimit = byteLimit;
+
+      recomputeBufferSizeAfterLimit();
+
+      return oldLimit;
+    }
+
+    private void recomputeBufferSizeAfterLimit() {
+      limit += bufferSizeAfterLimit;
+      final int bufferEnd = limit - startPos;
+      if (bufferEnd > currentLimit) {
+        // Limit is in current buffer.
+        bufferSizeAfterLimit = bufferEnd - currentLimit;
+        limit -= bufferSizeAfterLimit;
+      } else {
+        bufferSizeAfterLimit = 0;
+      }
+    }
+
+    @Override
+    public void popLimit(final int oldLimit) {
+      currentLimit = oldLimit;
+      recomputeBufferSizeAfterLimit();
+    }
+
+    @Override
+    public int getBytesUntilLimit() {
+      if (currentLimit == Integer.MAX_VALUE) {
+        return -1;
+      }
+
+      return currentLimit - getTotalBytesRead();
+    }
+
+    @Override
+    public boolean isAtEnd() throws IOException {
+      return pos == limit;
+    }
+
+    @Override
+    public int getTotalBytesRead() {
+      return pos - startPos;
+    }
+
+    @Override
+    public byte readRawByte() throws IOException {
+      if (pos == limit) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+      return buffer[pos++];
+    }
+
+    @Override
+    public byte[] readRawBytes(final int length) throws IOException {
+      if (length > 0 && length <= (limit - pos)) {
+        final int tempPos = pos;
+        pos += length;
+        return Arrays.copyOfRange(buffer, tempPos, pos);
+      }
+
+      if (length <= 0) {
+        if (length == 0) {
+          return Internal.EMPTY_BYTE_ARRAY;
+        } else {
+          throw InvalidProtocolBufferException.negativeSize();
+        }
+      }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+
+    @Override
+    public void skipRawBytes(final int length) throws IOException {
+      if (length >= 0 && length <= (limit - pos)) {
+        // We have all the bytes we need already.
+        pos += length;
+        return;
+      }
+
+      if (length < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
       throw InvalidProtocolBufferException.truncatedMessage();
     }
   }
 
   /**
-   * Tries to read more bytes from the input, making at least {@code n} bytes
-   * available in the buffer.  Caller must ensure that the requested space is
-   * not yet available, and that the requested space is less than BUFFER_SIZE.
-   *
-   * @return {@code true} if the bytes could be made available; {@code false}
-   *         if the end of the stream or the current limit was reached.
+   * A {@link CodedInputStream} implementation that uses a backing direct ByteBuffer as the input.
+   * Requires the use of {@code sun.misc.Unsafe} to perform fast reads on the buffer.
    */
-  private boolean tryRefillBuffer(int n) throws IOException {
-    if (bufferPos + n <= bufferSize) {
-      throw new IllegalStateException(
-          "refillBuffer() called when " + n +
-          " bytes were already available in buffer");
+  private static final class UnsafeDirectNioDecoder extends CodedInputStream {
+    /** The direct buffer that is backing this stream. */
+    private final ByteBuffer buffer;
+
+    /**
+     * If {@code true}, indicates that the buffer is backing a {@link ByteString} and is therefore
+     * considered to be an immutable input source.
+     */
+    private final boolean immutable;
+
+    /** The unsafe address of the content of {@link #buffer}. */
+    private final long address;
+
+    /** The unsafe address of the current read limit of the buffer. */
+    private long limit;
+
+    /** The unsafe address of the current read position of the buffer. */
+    private long pos;
+
+    /** The unsafe address of the starting read position. */
+    private long startPos;
+
+    /** The amount of available data in the buffer beyond {@link #limit}. */
+    private int bufferSizeAfterLimit;
+
+    /** The last tag that was read from this stream. */
+    private int lastTag;
+
+    /**
+     * If {@code true}, indicates that calls to read {@link ByteString} or {@code byte[]}
+     * <strong>may</strong> return slices of the underlying buffer, rather than copies.
+     */
+    private boolean enableAliasing;
+
+    /** The absolute position of the end of the current message. */
+    private int currentLimit = Integer.MAX_VALUE;
+
+    static boolean isSupported() {
+      return UnsafeUtil.hasUnsafeByteBufferOperations();
     }
 
-    if (totalBytesRetired + bufferPos + n > currentLimit) {
-      // Oops, we hit a limit.
-      return false;
+    private UnsafeDirectNioDecoder(ByteBuffer buffer, boolean immutable) {
+      this.buffer = buffer;
+      address = UnsafeUtil.addressOffset(buffer);
+      limit = address + buffer.limit();
+      pos = address + buffer.position();
+      startPos = pos;
+      this.immutable = immutable;
     }
 
-    if (refillCallback != null) {
-      refillCallback.onRefill();
+    @Override
+    public int readTag() throws IOException {
+      if (isAtEnd()) {
+        lastTag = 0;
+        return 0;
+      }
+
+      lastTag = readRawVarint32();
+      if (WireFormat.getTagFieldNumber(lastTag) == 0) {
+        // If we actually read zero (or any tag number corresponding to field
+        // number zero), that's not a valid tag.
+        throw InvalidProtocolBufferException.invalidTag();
+      }
+      return lastTag;
     }
 
-    if (input != null) {
-      int pos = bufferPos;
-      if (pos > 0) {
-        if (bufferSize > pos) {
-          System.arraycopy(buffer, pos, buffer, 0, bufferSize - pos);
-        }
-        totalBytesRetired += pos;
-        bufferSize -= pos;
-        bufferPos = 0;
+    @Override
+    public void checkLastTagWas(final int value) throws InvalidProtocolBufferException {
+      if (lastTag != value) {
+        throw InvalidProtocolBufferException.invalidEndTag();
       }
+    }
 
-      int bytesRead = input.read(buffer, bufferSize, buffer.length - bufferSize);
-      if (bytesRead == 0 || bytesRead < -1 || bytesRead > buffer.length) {
-        throw new IllegalStateException(
-            "InputStream#read(byte[]) returned invalid result: " + bytesRead +
-            "\nThe InputStream implementation is buggy.");
+    @Override
+    public int getLastTag() {
+      return lastTag;
+    }
+
+    @Override
+    public boolean skipField(final int tag) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          skipRawVarint();
+          return true;
+        case WireFormat.WIRETYPE_FIXED64:
+          skipRawBytes(FIXED_64_SIZE);
+          return true;
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          skipRawBytes(readRawVarint32());
+          return true;
+        case WireFormat.WIRETYPE_START_GROUP:
+          skipMessage();
+          checkLastTagWas(
+              WireFormat.makeTag(WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP));
+          return true;
+        case WireFormat.WIRETYPE_END_GROUP:
+          return false;
+        case WireFormat.WIRETYPE_FIXED32:
+          skipRawBytes(FIXED_32_SIZE);
+          return true;
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
       }
-      if (bytesRead > 0) {
-        bufferSize += bytesRead;
-        // Integer-overflow-conscious check against sizeLimit
-        if (totalBytesRetired + n - sizeLimit > 0) {
-          throw InvalidProtocolBufferException.sizeLimitExceeded();
+    }
+
+    @Override
+    public boolean skipField(final int tag, final CodedOutputStream output) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          {
+            long value = readInt64();
+            output.writeRawVarint32(tag);
+            output.writeUInt64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_FIXED64:
+          {
+            long value = readRawLittleEndian64();
+            output.writeRawVarint32(tag);
+            output.writeFixed64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          {
+            ByteString value = readBytes();
+            output.writeRawVarint32(tag);
+            output.writeBytesNoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_START_GROUP:
+          {
+            output.writeRawVarint32(tag);
+            skipMessage(output);
+            int endtag =
+                WireFormat.makeTag(
+                    WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP);
+            checkLastTagWas(endtag);
+            output.writeRawVarint32(endtag);
+            return true;
+          }
+        case WireFormat.WIRETYPE_END_GROUP:
+          {
+            return false;
+          }
+        case WireFormat.WIRETYPE_FIXED32:
+          {
+            int value = readRawLittleEndian32();
+            output.writeRawVarint32(tag);
+            output.writeFixed32NoTag(value);
+            return true;
+          }
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
+      }
+    }
+
+    @Override
+    public void skipMessage() throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag)) {
+          return;
         }
-        recomputeBufferSizeAfterLimit();
-        return (bufferSize >= n) ? true : tryRefillBuffer(n);
       }
     }
 
-    return false;
-  }
+    @Override
+    public void skipMessage(CodedOutputStream output) throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag, output)) {
+          return;
+        }
+      }
+    }
 
-  /**
-   * Read one byte from the input.
-   *
-   * @throws InvalidProtocolBufferException The end of the stream or the current
-   *                                        limit was reached.
-   */
-  public byte readRawByte() throws IOException {
-    if (bufferPos == bufferSize) {
-      refillBuffer(1);
+
+    // -----------------------------------------------------------------
+
+    @Override
+    public double readDouble() throws IOException {
+      return Double.longBitsToDouble(readRawLittleEndian64());
     }
-    return buffer[bufferPos++];
-  }
 
-  /**
-   * Read a fixed size of bytes from the input.
-   *
-   * @throws InvalidProtocolBufferException The end of the stream or the current
-   *                                        limit was reached.
-   */
-  public byte[] readRawBytes(final int size) throws IOException {
-    final int pos = bufferPos;
-    if (size <= (bufferSize - pos) && size > 0) {
-      bufferPos = pos + size;
-      return Arrays.copyOfRange(buffer, pos, pos + size);
-    } else {
-      return readRawBytesSlowPath(size);
+    @Override
+    public float readFloat() throws IOException {
+      return Float.intBitsToFloat(readRawLittleEndian32());
     }
-  }
 
-  /**
-   * Exactly like readRawBytes, but caller must have already checked the fast
-   * path: (size <= (bufferSize - pos) && size > 0)
-   */
-  private byte[] readRawBytesSlowPath(final int size) throws IOException {
-    if (size <= 0) {
+    @Override
+    public long readUInt64() throws IOException {
+      return readRawVarint64();
+    }
+
+    @Override
+    public long readInt64() throws IOException {
+      return readRawVarint64();
+    }
+
+    @Override
+    public int readInt32() throws IOException {
+      return readRawVarint32();
+    }
+
+    @Override
+    public long readFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
+
+    @Override
+    public int readFixed32() throws IOException {
+      return readRawLittleEndian32();
+    }
+
+    @Override
+    public boolean readBool() throws IOException {
+      return readRawVarint64() != 0;
+    }
+
+    @Override
+    public String readString() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= remaining()) {
+        // TODO(nathanmittler): Is there a way to avoid this copy?
+        byte[] bytes = copyToArray(pos, pos + size);
+        String result = new String(bytes, UTF_8);
+        pos += size;
+        return result;
+      }
+
       if (size == 0) {
-        return Internal.EMPTY_BYTE_ARRAY;
-      } else {
+        return "";
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+
+    @Override
+    public String readStringRequireUtf8() throws IOException {
+      final int size = readRawVarint32();
+      if (size >= 0 && size <= remaining()) {
+        // TODO(nathanmittler): Is there a way to avoid this copy?
+        byte[] bytes = copyToArray(pos, pos + size);
+        // TODO(martinrb): We could save a pass by validating while decoding.
+        if (!Utf8.isValidUtf8(bytes)) {
+          throw InvalidProtocolBufferException.invalidUtf8();
+        }
+
+        String result = new String(bytes, UTF_8);
+        pos += size;
+        return result;
+      }
+
+      if (size == 0) {
+        return "";
+      }
+      if (size <= 0) {
         throw InvalidProtocolBufferException.negativeSize();
       }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+
+    @Override
+    public void readGroup(
+        final int fieldNumber,
+        final MessageLite.Builder builder,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
     }
 
-    // Verify that the message size so far has not exceeded sizeLimit.
-    int currentMessageSize = totalBytesRetired + bufferPos + size;
-    if (currentMessageSize > sizeLimit) {
-      throw InvalidProtocolBufferException.sizeLimitExceeded();
+
+    @Override
+    public <T extends MessageLite> T readGroup(
+        final int fieldNumber,
+        final Parser<T> parser,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
+      return result;
     }
 
-    // Verify that the message size so far has not exceeded currentLimit.
-    if (currentMessageSize > currentLimit) {
-      // Read to the end of the stream anyway.
-      skipRawBytes(currentLimit - totalBytesRetired - bufferPos);
+    @Deprecated
+    @Override
+    public void readUnknownGroup(final int fieldNumber, final MessageLite.Builder builder)
+        throws IOException {
+      readGroup(fieldNumber, builder, ExtensionRegistryLite.getEmptyRegistry());
+    }
+
+    @Override
+    public void readMessage(
+        final MessageLite.Builder builder, final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      final int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+    }
+
+
+    @Override
+    public <T extends MessageLite> T readMessage(
+        final Parser<T> parser, final ExtensionRegistryLite extensionRegistry) throws IOException {
+      int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+      return result;
+    }
+
+    @Override
+    public ByteString readBytes() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= remaining()) {
+        ByteBuffer result;
+        if (immutable && enableAliasing) {
+          result = slice(pos, pos + size);
+        } else {
+          result = copy(pos, pos + size);
+        }
+        pos += size;
+        return ByteString.wrap(result);
+      }
+
+      if (size == 0) {
+        return ByteString.EMPTY;
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
       throw InvalidProtocolBufferException.truncatedMessage();
     }
 
-    // We need the input stream to proceed.
-    if (input == null) {
+    @Override
+    public byte[] readByteArray() throws IOException {
+      return readRawBytes(readRawVarint32());
+    }
+
+    @Override
+    public ByteBuffer readByteBuffer() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= remaining()) {
+        ByteBuffer result;
+        // "Immutable" implies that buffer is backing a ByteString.
+        // Disallow slicing in this case to prevent the caller from modifying the contents
+        // of the ByteString.
+        if (!immutable && enableAliasing) {
+          result = slice(pos, pos + size);
+        } else {
+          result = copy(pos, pos + size);
+        }
+        pos += size;
+        // TODO(nathanmittler): Investigate making the ByteBuffer be made read-only
+        return result;
+      }
+
+      if (size == 0) {
+        return EMPTY_BYTE_BUFFER;
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
       throw InvalidProtocolBufferException.truncatedMessage();
     }
 
-    final int originalBufferPos = bufferPos;
-    final int bufferedBytes = bufferSize - bufferPos;
+    @Override
+    public int readUInt32() throws IOException {
+      return readRawVarint32();
+    }
+
+    @Override
+    public int readEnum() throws IOException {
+      return readRawVarint32();
+    }
 
-    // Mark the current buffer consumed.
-    totalBytesRetired += bufferSize;
-    bufferPos = 0;
-    bufferSize = 0;
+    @Override
+    public int readSFixed32() throws IOException {
+      return readRawLittleEndian32();
+    }
 
-    // Determine the number of bytes we need to read from the input stream.
-    int sizeLeft = size - bufferedBytes;
-    // TODO(nathanmittler): Consider using a value larger than BUFFER_SIZE.
-    if (sizeLeft < BUFFER_SIZE || sizeLeft <= input.available()) {
-      // Either the bytes we need are known to be available, or the required buffer is
-      // within an allowed threshold - go ahead and allocate the buffer now.
-      final byte[] bytes = new byte[size];
+    @Override
+    public long readSFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
 
-      // Copy all of the buffered bytes to the result buffer.
-      System.arraycopy(buffer, originalBufferPos, bytes, 0, bufferedBytes);
+    @Override
+    public int readSInt32() throws IOException {
+      return decodeZigZag32(readRawVarint32());
+    }
+
+    @Override
+    public long readSInt64() throws IOException {
+      return decodeZigZag64(readRawVarint64());
+    }
+
+    // =================================================================
 
-      // Fill the remaining bytes from the input stream.
-      int pos = bufferedBytes;
-      while (pos < bytes.length) {
-        int n = input.read(bytes, pos, size - pos);
-        if (n == -1) {
-          throw InvalidProtocolBufferException.truncatedMessage();
+    @Override
+    public int readRawVarint32() throws IOException {
+      // See implementation notes for readRawVarint64
+      fastpath:
+      {
+        long tempPos = pos;
+
+        if (limit == tempPos) {
+          break fastpath;
+        }
+
+        int x;
+        if ((x = UnsafeUtil.getByte(tempPos++)) >= 0) {
+          pos = tempPos;
+          return x;
+        } else if (limit - tempPos < 9) {
+          break fastpath;
+        } else if ((x ^= (UnsafeUtil.getByte(tempPos++) << 7)) < 0) {
+          x ^= (~0 << 7);
+        } else if ((x ^= (UnsafeUtil.getByte(tempPos++) << 14)) >= 0) {
+          x ^= (~0 << 7) ^ (~0 << 14);
+        } else if ((x ^= (UnsafeUtil.getByte(tempPos++) << 21)) < 0) {
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21);
+        } else {
+          int y = UnsafeUtil.getByte(tempPos++);
+          x ^= y << 28;
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21) ^ (~0 << 28);
+          if (y < 0
+              && UnsafeUtil.getByte(tempPos++) < 0
+              && UnsafeUtil.getByte(tempPos++) < 0
+              && UnsafeUtil.getByte(tempPos++) < 0
+              && UnsafeUtil.getByte(tempPos++) < 0
+              && UnsafeUtil.getByte(tempPos++) < 0) {
+            break fastpath; // Will throw malformedVarint()
+          }
         }
-        totalBytesRetired += n;
-        pos += n;
+        pos = tempPos;
+        return x;
       }
+      return (int) readRawVarint64SlowPath();
+    }
 
-      return bytes;
+    private void skipRawVarint() throws IOException {
+      if (remaining() >= MAX_VARINT_SIZE) {
+        skipRawVarintFastPath();
+      } else {
+        skipRawVarintSlowPath();
+      }
     }
 
-    // The size is very large.  For security reasons, we can't allocate the
-    // entire byte array yet.  The size comes directly from the input, so a
-    // maliciously-crafted message could provide a bogus very large size in
-    // order to trick the app into allocating a lot of memory.  We avoid this
-    // by allocating and reading only a small chunk at a time, so that the
-    // malicious message must actually *be* extremely large to cause
-    // problems.  Meanwhile, we limit the allowed size of a message elsewhere.
-    final List<byte[]> chunks = new ArrayList<byte[]>();
-
-    while (sizeLeft > 0) {
-      // TODO(nathanmittler): Consider using a value larger than BUFFER_SIZE.
-      final byte[] chunk = new byte[Math.min(sizeLeft, BUFFER_SIZE)];
-      int pos = 0;
-      while (pos < chunk.length) {
-        final int n = input.read(chunk, pos, chunk.length - pos);
-        if (n == -1) {
-          throw InvalidProtocolBufferException.truncatedMessage();
+    private void skipRawVarintFastPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (UnsafeUtil.getByte(pos++) >= 0) {
+          return;
         }
-        totalBytesRetired += n;
-        pos += n;
       }
-      sizeLeft -= chunk.length;
-      chunks.add(chunk);
+      throw InvalidProtocolBufferException.malformedVarint();
     }
 
-    // OK, got everything.  Now concatenate it all into one buffer.
-    final byte[] bytes = new byte[size];
+    private void skipRawVarintSlowPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (readRawByte() >= 0) {
+          return;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
 
-    // Start by copying the leftover bytes from this.buffer.
-    System.arraycopy(buffer, originalBufferPos, bytes, 0, bufferedBytes);
+    @Override
+    public long readRawVarint64() throws IOException {
+      // Implementation notes:
+      //
+      // Optimized for one-byte values, expected to be common.
+      // The particular code below was selected from various candidates
+      // empirically, by winning VarintBenchmark.
+      //
+      // Sign extension of (signed) Java bytes is usually a nuisance, but
+      // we exploit it here to more easily obtain the sign of bytes read.
+      // Instead of cleaning up the sign extension bits by masking eagerly,
+      // we delay until we find the final (positive) byte, when we clear all
+      // accumulated bits with one xor.  We depend on javac to constant fold.
+      fastpath:
+      {
+        long tempPos = pos;
+
+        if (limit == tempPos) {
+          break fastpath;
+        }
 
-    // And now all the chunks.
-    int pos = bufferedBytes;
-    for (final byte[] chunk : chunks) {
-      System.arraycopy(chunk, 0, bytes, pos, chunk.length);
-      pos += chunk.length;
+        long x;
+        int y;
+        if ((y = UnsafeUtil.getByte(tempPos++)) >= 0) {
+          pos = tempPos;
+          return y;
+        } else if (limit - tempPos < 9) {
+          break fastpath;
+        } else if ((y ^= (UnsafeUtil.getByte(tempPos++) << 7)) < 0) {
+          x = y ^ (~0 << 7);
+        } else if ((y ^= (UnsafeUtil.getByte(tempPos++) << 14)) >= 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14));
+        } else if ((y ^= (UnsafeUtil.getByte(tempPos++) << 21)) < 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14) ^ (~0 << 21));
+        } else if ((x = y ^ ((long) UnsafeUtil.getByte(tempPos++) << 28)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28);
+        } else if ((x ^= ((long) UnsafeUtil.getByte(tempPos++) << 35)) < 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35);
+        } else if ((x ^= ((long) UnsafeUtil.getByte(tempPos++) << 42)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42);
+        } else if ((x ^= ((long) UnsafeUtil.getByte(tempPos++) << 49)) < 0L) {
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49);
+        } else {
+          x ^= ((long) UnsafeUtil.getByte(tempPos++) << 56);
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49)
+                  ^ (~0L << 56);
+          if (x < 0L) {
+            if (UnsafeUtil.getByte(tempPos++) < 0L) {
+              break fastpath; // Will throw malformedVarint()
+            }
+          }
+        }
+        pos = tempPos;
+        return x;
+      }
+      return readRawVarint64SlowPath();
     }
 
-    // Done.
-    return bytes;
-  }
+    @Override
+    long readRawVarint64SlowPath() throws IOException {
+      long result = 0;
+      for (int shift = 0; shift < 64; shift += 7) {
+        final byte b = readRawByte();
+        result |= (long) (b & 0x7F) << shift;
+        if ((b & 0x80) == 0) {
+          return result;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
 
-  /**
-   * Reads and discards {@code size} bytes.
-   *
-   * @throws InvalidProtocolBufferException The end of the stream or the current
-   *                                        limit was reached.
-   */
-  public void skipRawBytes(final int size) throws IOException {
-    if (size <= (bufferSize - bufferPos) && size >= 0) {
-      // We have all the bytes we need already.
-      bufferPos += size;
-    } else {
-      skipRawBytesSlowPath(size);
+    @Override
+    public int readRawLittleEndian32() throws IOException {
+      long tempPos = pos;
+
+      if (limit - tempPos < FIXED_32_SIZE) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      pos = tempPos + FIXED_32_SIZE;
+      return (((UnsafeUtil.getByte(tempPos) & 0xff))
+          | ((UnsafeUtil.getByte(tempPos + 1) & 0xff) << 8)
+          | ((UnsafeUtil.getByte(tempPos + 2) & 0xff) << 16)
+          | ((UnsafeUtil.getByte(tempPos + 3) & 0xff) << 24));
+    }
+
+    @Override
+    public long readRawLittleEndian64() throws IOException {
+      long tempPos = pos;
+
+      if (limit - tempPos < FIXED_64_SIZE) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      pos = tempPos + FIXED_64_SIZE;
+      return (((UnsafeUtil.getByte(tempPos) & 0xffL))
+          | ((UnsafeUtil.getByte(tempPos + 1) & 0xffL) << 8)
+          | ((UnsafeUtil.getByte(tempPos + 2) & 0xffL) << 16)
+          | ((UnsafeUtil.getByte(tempPos + 3) & 0xffL) << 24)
+          | ((UnsafeUtil.getByte(tempPos + 4) & 0xffL) << 32)
+          | ((UnsafeUtil.getByte(tempPos + 5) & 0xffL) << 40)
+          | ((UnsafeUtil.getByte(tempPos + 6) & 0xffL) << 48)
+          | ((UnsafeUtil.getByte(tempPos + 7) & 0xffL) << 56));
+    }
+
+    @Override
+    public void enableAliasing(boolean enabled) {
+      this.enableAliasing = enabled;
+    }
+
+    @Override
+    public void resetSizeCounter() {
+      startPos = pos;
+    }
+
+    @Override
+    public int pushLimit(int byteLimit) throws InvalidProtocolBufferException {
+      if (byteLimit < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      byteLimit += getTotalBytesRead();
+      final int oldLimit = currentLimit;
+      if (byteLimit > oldLimit) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+      currentLimit = byteLimit;
+
+      recomputeBufferSizeAfterLimit();
+
+      return oldLimit;
+    }
+
+    @Override
+    public void popLimit(final int oldLimit) {
+      currentLimit = oldLimit;
+      recomputeBufferSizeAfterLimit();
+    }
+
+    @Override
+    public int getBytesUntilLimit() {
+      if (currentLimit == Integer.MAX_VALUE) {
+        return -1;
+      }
+
+      return currentLimit - getTotalBytesRead();
+    }
+
+    @Override
+    public boolean isAtEnd() throws IOException {
+      return pos == limit;
+    }
+
+    @Override
+    public int getTotalBytesRead() {
+      return (int) (pos - startPos);
+    }
+
+    @Override
+    public byte readRawByte() throws IOException {
+      if (pos == limit) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+      return UnsafeUtil.getByte(pos++);
+    }
+
+    @Override
+    public byte[] readRawBytes(final int length) throws IOException {
+      if (length >= 0 && length <= remaining()) {
+        byte[] bytes = new byte[length];
+        slice(pos, pos + length).get(bytes);
+        pos += length;
+        return bytes;
+      }
+
+      if (length <= 0) {
+        if (length == 0) {
+          return EMPTY_BYTE_ARRAY;
+        } else {
+          throw InvalidProtocolBufferException.negativeSize();
+        }
+      }
+
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+
+    @Override
+    public void skipRawBytes(final int length) throws IOException {
+      if (length >= 0 && length <= remaining()) {
+        // We have all the bytes we need already.
+        pos += length;
+        return;
+      }
+
+      if (length < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      throw InvalidProtocolBufferException.truncatedMessage();
+    }
+
+    private void recomputeBufferSizeAfterLimit() {
+      limit += bufferSizeAfterLimit;
+      final int bufferEnd = (int) (limit - startPos);
+      if (bufferEnd > currentLimit) {
+        // Limit is in current buffer.
+        bufferSizeAfterLimit = bufferEnd - currentLimit;
+        limit -= bufferSizeAfterLimit;
+      } else {
+        bufferSizeAfterLimit = 0;
+      }
+    }
+
+    private int remaining() {
+      return (int) (limit - pos);
+    }
+
+    private int bufferPos(long pos) {
+      return (int) (pos - address);
+    }
+
+    private ByteBuffer slice(long begin, long end) throws IOException {
+      int prevPos = buffer.position();
+      int prevLimit = buffer.limit();
+      try {
+        buffer.position(bufferPos(begin));
+        buffer.limit(bufferPos(end));
+        return buffer.slice();
+      } catch (IllegalArgumentException e) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      } finally {
+        buffer.position(prevPos);
+        buffer.limit(prevLimit);
+      }
+    }
+
+    private ByteBuffer copy(long begin, long end) throws IOException {
+      return ByteBuffer.wrap(copyToArray(begin, end));
+    }
+
+    private byte[] copyToArray(long begin, long end) throws IOException {
+      int prevPos = buffer.position();
+      int prevLimit = buffer.limit();
+      try {
+        buffer.position(bufferPos(begin));
+        buffer.limit(bufferPos(end));
+        byte[] bytes = new byte[(int) (end - begin)];
+        buffer.get(bytes);
+        return bytes;
+      } catch (IllegalArgumentException e) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      } finally {
+        buffer.position(prevPos);
+        buffer.limit(prevLimit);
+      }
     }
   }
 
   /**
-   * Exactly like skipRawBytes, but caller must have already checked the fast
-   * path: (size <= (bufferSize - pos) && size >= 0)
+   * Implementation of {@link CodedInputStream} that uses an {@link InputStream} as the data source.
    */
-  private void skipRawBytesSlowPath(final int size) throws IOException {
-    if (size < 0) {
-      throw InvalidProtocolBufferException.negativeSize();
+  private static final class StreamDecoder extends CodedInputStream {
+    private final InputStream input;
+    private final byte[] buffer;
+    /** bufferSize represents how many bytes are currently filled in the buffer */
+    private int bufferSize;
+
+    private int bufferSizeAfterLimit;
+    private int pos;
+    private int lastTag;
+
+    /**
+     * The total number of bytes read before the current buffer. The total bytes read up to the
+     * current position can be computed as {@code totalBytesRetired + pos}. This value may be
+     * negative if reading started in the middle of the current buffer (e.g. if the constructor that
+     * takes a byte array and an offset was used).
+     */
+    private int totalBytesRetired;
+
+    /** The absolute position of the end of the current message. */
+    private int currentLimit = Integer.MAX_VALUE;
+
+    private StreamDecoder(final InputStream input, int bufferSize) {
+      checkNotNull(input, "input");
+      this.input = input;
+      this.buffer = new byte[bufferSize];
+      this.bufferSize = 0;
+      pos = 0;
+      totalBytesRetired = 0;
     }
 
-    if (totalBytesRetired + bufferPos + size > currentLimit) {
-      // Read to the end of the stream anyway.
-      skipRawBytes(currentLimit - totalBytesRetired - bufferPos);
-      // Then fail.
-      throw InvalidProtocolBufferException.truncatedMessage();
+    @Override
+    public int readTag() throws IOException {
+      if (isAtEnd()) {
+        lastTag = 0;
+        return 0;
+      }
+
+      lastTag = readRawVarint32();
+      if (WireFormat.getTagFieldNumber(lastTag) == 0) {
+        // If we actually read zero (or any tag number corresponding to field
+        // number zero), that's not a valid tag.
+        throw InvalidProtocolBufferException.invalidTag();
+      }
+      return lastTag;
     }
 
-    // Skipping more bytes than are in the buffer.  First skip what we have.
-    int pos = bufferSize - bufferPos;
-    bufferPos = bufferSize;
+    @Override
+    public void checkLastTagWas(final int value) throws InvalidProtocolBufferException {
+      if (lastTag != value) {
+        throw InvalidProtocolBufferException.invalidEndTag();
+      }
+    }
 
-    // Keep refilling the buffer until we get to the point we wanted to skip to.
-    // This has the side effect of ensuring the limits are updated correctly.
-    refillBuffer(1);
-    while (size - pos > bufferSize) {
-      pos += bufferSize;
-      bufferPos = bufferSize;
-      refillBuffer(1);
+    @Override
+    public int getLastTag() {
+      return lastTag;
+    }
+
+    @Override
+    public boolean skipField(final int tag) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          skipRawVarint();
+          return true;
+        case WireFormat.WIRETYPE_FIXED64:
+          skipRawBytes(FIXED_64_SIZE);
+          return true;
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          skipRawBytes(readRawVarint32());
+          return true;
+        case WireFormat.WIRETYPE_START_GROUP:
+          skipMessage();
+          checkLastTagWas(
+              WireFormat.makeTag(WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP));
+          return true;
+        case WireFormat.WIRETYPE_END_GROUP:
+          return false;
+        case WireFormat.WIRETYPE_FIXED32:
+          skipRawBytes(FIXED_32_SIZE);
+          return true;
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
+      }
+    }
+
+    @Override
+    public boolean skipField(final int tag, final CodedOutputStream output) throws IOException {
+      switch (WireFormat.getTagWireType(tag)) {
+        case WireFormat.WIRETYPE_VARINT:
+          {
+            long value = readInt64();
+            output.writeRawVarint32(tag);
+            output.writeUInt64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_FIXED64:
+          {
+            long value = readRawLittleEndian64();
+            output.writeRawVarint32(tag);
+            output.writeFixed64NoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_LENGTH_DELIMITED:
+          {
+            ByteString value = readBytes();
+            output.writeRawVarint32(tag);
+            output.writeBytesNoTag(value);
+            return true;
+          }
+        case WireFormat.WIRETYPE_START_GROUP:
+          {
+            output.writeRawVarint32(tag);
+            skipMessage(output);
+            int endtag =
+                WireFormat.makeTag(
+                    WireFormat.getTagFieldNumber(tag), WireFormat.WIRETYPE_END_GROUP);
+            checkLastTagWas(endtag);
+            output.writeRawVarint32(endtag);
+            return true;
+          }
+        case WireFormat.WIRETYPE_END_GROUP:
+          {
+            return false;
+          }
+        case WireFormat.WIRETYPE_FIXED32:
+          {
+            int value = readRawLittleEndian32();
+            output.writeRawVarint32(tag);
+            output.writeFixed32NoTag(value);
+            return true;
+          }
+        default:
+          throw InvalidProtocolBufferException.invalidWireType();
+      }
+    }
+
+    @Override
+    public void skipMessage() throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag)) {
+          return;
+        }
+      }
+    }
+
+    @Override
+    public void skipMessage(CodedOutputStream output) throws IOException {
+      while (true) {
+        final int tag = readTag();
+        if (tag == 0 || !skipField(tag, output)) {
+          return;
+        }
+      }
+    }
+
+    /** Collects the bytes skipped and returns the data in a ByteBuffer. */
+    private class SkippedDataSink implements RefillCallback {
+      private int lastPos = pos;
+      private ByteArrayOutputStream byteArrayStream;
+
+      @Override
+      public void onRefill() {
+        if (byteArrayStream == null) {
+          byteArrayStream = new ByteArrayOutputStream();
+        }
+        byteArrayStream.write(buffer, lastPos, pos - lastPos);
+        lastPos = 0;
+      }
+
+      /** Gets skipped data in a ByteBuffer. This method should only be called once. */
+      ByteBuffer getSkippedData() {
+        if (byteArrayStream == null) {
+          return ByteBuffer.wrap(buffer, lastPos, pos - lastPos);
+        } else {
+          byteArrayStream.write(buffer, lastPos, pos);
+          return ByteBuffer.wrap(byteArrayStream.toByteArray());
+        }
+      }
+    }
+
+
+    // -----------------------------------------------------------------
+
+    @Override
+    public double readDouble() throws IOException {
+      return Double.longBitsToDouble(readRawLittleEndian64());
+    }
+
+    @Override
+    public float readFloat() throws IOException {
+      return Float.intBitsToFloat(readRawLittleEndian32());
+    }
+
+    @Override
+    public long readUInt64() throws IOException {
+      return readRawVarint64();
+    }
+
+    @Override
+    public long readInt64() throws IOException {
+      return readRawVarint64();
+    }
+
+    @Override
+    public int readInt32() throws IOException {
+      return readRawVarint32();
+    }
+
+    @Override
+    public long readFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
+
+    @Override
+    public int readFixed32() throws IOException {
+      return readRawLittleEndian32();
     }
 
-    bufferPos = size - pos;
+    @Override
+    public boolean readBool() throws IOException {
+      return readRawVarint64() != 0;
+    }
+
+    @Override
+    public String readString() throws IOException {
+      final int size = readRawVarint32();
+      if (size > 0 && size <= (bufferSize - pos)) {
+        // Fast path:  We already have the bytes in a contiguous buffer, so
+        //   just copy directly from it.
+        final String result = new String(buffer, pos, size, UTF_8);
+        pos += size;
+        return result;
+      }
+      if (size == 0) {
+        return "";
+      }
+      if (size <= bufferSize) {
+        refillBuffer(size);
+        String result = new String(buffer, pos, size, UTF_8);
+        pos += size;
+        return result;
+      }
+      // Slow path:  Build a byte array first then copy it.
+      return new String(readRawBytesSlowPath(size), UTF_8);
+    }
+
+    @Override
+    public String readStringRequireUtf8() throws IOException {
+      final int size = readRawVarint32();
+      final byte[] bytes;
+      final int oldPos = pos;
+      final int tempPos;
+      if (size <= (bufferSize - oldPos) && size > 0) {
+        // Fast path:  We already have the bytes in a contiguous buffer, so
+        //   just copy directly from it.
+        bytes = buffer;
+        pos = oldPos + size;
+        tempPos = oldPos;
+      } else if (size == 0) {
+        return "";
+      } else if (size <= bufferSize) {
+        refillBuffer(size);
+        bytes = buffer;
+        tempPos = 0;
+        pos = tempPos + size;
+      } else {
+        // Slow path:  Build a byte array first then copy it.
+        bytes = readRawBytesSlowPath(size);
+        tempPos = 0;
+      }
+      // TODO(martinrb): We could save a pass by validating while decoding.
+      if (!Utf8.isValidUtf8(bytes, tempPos, tempPos + size)) {
+        throw InvalidProtocolBufferException.invalidUtf8();
+      }
+      return new String(bytes, tempPos, size, UTF_8);
+    }
+
+    @Override
+    public void readGroup(
+        final int fieldNumber,
+        final MessageLite.Builder builder,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
+    }
+
+
+    @Override
+    public <T extends MessageLite> T readGroup(
+        final int fieldNumber,
+        final Parser<T> parser,
+        final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(WireFormat.makeTag(fieldNumber, WireFormat.WIRETYPE_END_GROUP));
+      --recursionDepth;
+      return result;
+    }
+
+    @Deprecated
+    @Override
+    public void readUnknownGroup(final int fieldNumber, final MessageLite.Builder builder)
+        throws IOException {
+      readGroup(fieldNumber, builder, ExtensionRegistryLite.getEmptyRegistry());
+    }
+
+    @Override
+    public void readMessage(
+        final MessageLite.Builder builder, final ExtensionRegistryLite extensionRegistry)
+        throws IOException {
+      final int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      builder.mergeFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+    }
+
+
+    @Override
+    public <T extends MessageLite> T readMessage(
+        final Parser<T> parser, final ExtensionRegistryLite extensionRegistry) throws IOException {
+      int length = readRawVarint32();
+      if (recursionDepth >= recursionLimit) {
+        throw InvalidProtocolBufferException.recursionLimitExceeded();
+      }
+      final int oldLimit = pushLimit(length);
+      ++recursionDepth;
+      T result = parser.parsePartialFrom(this, extensionRegistry);
+      checkLastTagWas(0);
+      --recursionDepth;
+      popLimit(oldLimit);
+      return result;
+    }
+
+    @Override
+    public ByteString readBytes() throws IOException {
+      final int size = readRawVarint32();
+      if (size <= (bufferSize - pos) && size > 0) {
+        // Fast path:  We already have the bytes in a contiguous buffer, so
+        //   just copy directly from it.
+        final ByteString result = ByteString.copyFrom(buffer, pos, size);
+        pos += size;
+        return result;
+      }
+      if (size == 0) {
+        return ByteString.EMPTY;
+      }
+      // Slow path:  Build a byte array first then copy it.
+      return ByteString.wrap(readRawBytesSlowPath(size));
+    }
+
+    @Override
+    public byte[] readByteArray() throws IOException {
+      final int size = readRawVarint32();
+      if (size <= (bufferSize - pos) && size > 0) {
+        // Fast path: We already have the bytes in a contiguous buffer, so
+        // just copy directly from it.
+        final byte[] result = Arrays.copyOfRange(buffer, pos, pos + size);
+        pos += size;
+        return result;
+      } else {
+        // Slow path: Build a byte array first then copy it.
+        return readRawBytesSlowPath(size);
+      }
+    }
+
+    @Override
+    public ByteBuffer readByteBuffer() throws IOException {
+      final int size = readRawVarint32();
+      if (size <= (bufferSize - pos) && size > 0) {
+        // Fast path: We already have the bytes in a contiguous buffer.
+        ByteBuffer result = ByteBuffer.wrap(Arrays.copyOfRange(buffer, pos, pos + size));
+        pos += size;
+        return result;
+      }
+      if (size == 0) {
+        return Internal.EMPTY_BYTE_BUFFER;
+      }
+      // Slow path: Build a byte array first then copy it.
+      return ByteBuffer.wrap(readRawBytesSlowPath(size));
+    }
+
+    @Override
+    public int readUInt32() throws IOException {
+      return readRawVarint32();
+    }
+
+    @Override
+    public int readEnum() throws IOException {
+      return readRawVarint32();
+    }
+
+    @Override
+    public int readSFixed32() throws IOException {
+      return readRawLittleEndian32();
+    }
+
+    @Override
+    public long readSFixed64() throws IOException {
+      return readRawLittleEndian64();
+    }
+
+    @Override
+    public int readSInt32() throws IOException {
+      return decodeZigZag32(readRawVarint32());
+    }
+
+    @Override
+    public long readSInt64() throws IOException {
+      return decodeZigZag64(readRawVarint64());
+    }
+
+    // =================================================================
+
+    @Override
+    public int readRawVarint32() throws IOException {
+      // See implementation notes for readRawVarint64
+      fastpath:
+      {
+        int tempPos = pos;
+
+        if (bufferSize == tempPos) {
+          break fastpath;
+        }
+
+        final byte[] buffer = this.buffer;
+        int x;
+        if ((x = buffer[tempPos++]) >= 0) {
+          pos = tempPos;
+          return x;
+        } else if (bufferSize - tempPos < 9) {
+          break fastpath;
+        } else if ((x ^= (buffer[tempPos++] << 7)) < 0) {
+          x ^= (~0 << 7);
+        } else if ((x ^= (buffer[tempPos++] << 14)) >= 0) {
+          x ^= (~0 << 7) ^ (~0 << 14);
+        } else if ((x ^= (buffer[tempPos++] << 21)) < 0) {
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21);
+        } else {
+          int y = buffer[tempPos++];
+          x ^= y << 28;
+          x ^= (~0 << 7) ^ (~0 << 14) ^ (~0 << 21) ^ (~0 << 28);
+          if (y < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0
+              && buffer[tempPos++] < 0) {
+            break fastpath; // Will throw malformedVarint()
+          }
+        }
+        pos = tempPos;
+        return x;
+      }
+      return (int) readRawVarint64SlowPath();
+    }
+
+    private void skipRawVarint() throws IOException {
+      if (bufferSize - pos >= MAX_VARINT_SIZE) {
+        skipRawVarintFastPath();
+      } else {
+        skipRawVarintSlowPath();
+      }
+    }
+
+    private void skipRawVarintFastPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (buffer[pos++] >= 0) {
+          return;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    private void skipRawVarintSlowPath() throws IOException {
+      for (int i = 0; i < MAX_VARINT_SIZE; i++) {
+        if (readRawByte() >= 0) {
+          return;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    @Override
+    public long readRawVarint64() throws IOException {
+      // Implementation notes:
+      //
+      // Optimized for one-byte values, expected to be common.
+      // The particular code below was selected from various candidates
+      // empirically, by winning VarintBenchmark.
+      //
+      // Sign extension of (signed) Java bytes is usually a nuisance, but
+      // we exploit it here to more easily obtain the sign of bytes read.
+      // Instead of cleaning up the sign extension bits by masking eagerly,
+      // we delay until we find the final (positive) byte, when we clear all
+      // accumulated bits with one xor.  We depend on javac to constant fold.
+      fastpath:
+      {
+        int tempPos = pos;
+
+        if (bufferSize == tempPos) {
+          break fastpath;
+        }
+
+        final byte[] buffer = this.buffer;
+        long x;
+        int y;
+        if ((y = buffer[tempPos++]) >= 0) {
+          pos = tempPos;
+          return y;
+        } else if (bufferSize - tempPos < 9) {
+          break fastpath;
+        } else if ((y ^= (buffer[tempPos++] << 7)) < 0) {
+          x = y ^ (~0 << 7);
+        } else if ((y ^= (buffer[tempPos++] << 14)) >= 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14));
+        } else if ((y ^= (buffer[tempPos++] << 21)) < 0) {
+          x = y ^ ((~0 << 7) ^ (~0 << 14) ^ (~0 << 21));
+        } else if ((x = y ^ ((long) buffer[tempPos++] << 28)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28);
+        } else if ((x ^= ((long) buffer[tempPos++] << 35)) < 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35);
+        } else if ((x ^= ((long) buffer[tempPos++] << 42)) >= 0L) {
+          x ^= (~0L << 7) ^ (~0L << 14) ^ (~0L << 21) ^ (~0L << 28) ^ (~0L << 35) ^ (~0L << 42);
+        } else if ((x ^= ((long) buffer[tempPos++] << 49)) < 0L) {
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49);
+        } else {
+          x ^= ((long) buffer[tempPos++] << 56);
+          x ^=
+              (~0L << 7)
+                  ^ (~0L << 14)
+                  ^ (~0L << 21)
+                  ^ (~0L << 28)
+                  ^ (~0L << 35)
+                  ^ (~0L << 42)
+                  ^ (~0L << 49)
+                  ^ (~0L << 56);
+          if (x < 0L) {
+            if (buffer[tempPos++] < 0L) {
+              break fastpath; // Will throw malformedVarint()
+            }
+          }
+        }
+        pos = tempPos;
+        return x;
+      }
+      return readRawVarint64SlowPath();
+    }
+
+    @Override
+    long readRawVarint64SlowPath() throws IOException {
+      long result = 0;
+      for (int shift = 0; shift < 64; shift += 7) {
+        final byte b = readRawByte();
+        result |= (long) (b & 0x7F) << shift;
+        if ((b & 0x80) == 0) {
+          return result;
+        }
+      }
+      throw InvalidProtocolBufferException.malformedVarint();
+    }
+
+    @Override
+    public int readRawLittleEndian32() throws IOException {
+      int tempPos = pos;
+
+      if (bufferSize - tempPos < FIXED_32_SIZE) {
+        refillBuffer(FIXED_32_SIZE);
+        tempPos = pos;
+      }
+
+      final byte[] buffer = this.buffer;
+      pos = tempPos + FIXED_32_SIZE;
+      return (((buffer[tempPos] & 0xff))
+          | ((buffer[tempPos + 1] & 0xff) << 8)
+          | ((buffer[tempPos + 2] & 0xff) << 16)
+          | ((buffer[tempPos + 3] & 0xff) << 24));
+    }
+
+    @Override
+    public long readRawLittleEndian64() throws IOException {
+      int tempPos = pos;
+
+      if (bufferSize - tempPos < FIXED_64_SIZE) {
+        refillBuffer(FIXED_64_SIZE);
+        tempPos = pos;
+      }
+
+      final byte[] buffer = this.buffer;
+      pos = tempPos + FIXED_64_SIZE;
+      return (((buffer[tempPos] & 0xffL))
+          | ((buffer[tempPos + 1] & 0xffL) << 8)
+          | ((buffer[tempPos + 2] & 0xffL) << 16)
+          | ((buffer[tempPos + 3] & 0xffL) << 24)
+          | ((buffer[tempPos + 4] & 0xffL) << 32)
+          | ((buffer[tempPos + 5] & 0xffL) << 40)
+          | ((buffer[tempPos + 6] & 0xffL) << 48)
+          | ((buffer[tempPos + 7] & 0xffL) << 56));
+    }
+
+    // -----------------------------------------------------------------
+
+    @Override
+    public void enableAliasing(boolean enabled) {
+      // TODO(nathanmittler): Ideally we should throw here. Do nothing for backward compatibility.
+    }
+
+    @Override
+    public void resetSizeCounter() {
+      totalBytesRetired = -pos;
+    }
+
+    @Override
+    public int pushLimit(int byteLimit) throws InvalidProtocolBufferException {
+      if (byteLimit < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+      byteLimit += totalBytesRetired + pos;
+      final int oldLimit = currentLimit;
+      if (byteLimit > oldLimit) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+      currentLimit = byteLimit;
+
+      recomputeBufferSizeAfterLimit();
+
+      return oldLimit;
+    }
+
+    private void recomputeBufferSizeAfterLimit() {
+      bufferSize += bufferSizeAfterLimit;
+      final int bufferEnd = totalBytesRetired + bufferSize;
+      if (bufferEnd > currentLimit) {
+        // Limit is in current buffer.
+        bufferSizeAfterLimit = bufferEnd - currentLimit;
+        bufferSize -= bufferSizeAfterLimit;
+      } else {
+        bufferSizeAfterLimit = 0;
+      }
+    }
+
+    @Override
+    public void popLimit(final int oldLimit) {
+      currentLimit = oldLimit;
+      recomputeBufferSizeAfterLimit();
+    }
+
+    @Override
+    public int getBytesUntilLimit() {
+      if (currentLimit == Integer.MAX_VALUE) {
+        return -1;
+      }
+
+      final int currentAbsolutePosition = totalBytesRetired + pos;
+      return currentLimit - currentAbsolutePosition;
+    }
+
+    @Override
+    public boolean isAtEnd() throws IOException {
+      return pos == bufferSize && !tryRefillBuffer(1);
+    }
+
+    @Override
+    public int getTotalBytesRead() {
+      return totalBytesRetired + pos;
+    }
+
+    private interface RefillCallback {
+      void onRefill();
+    }
+
+    private RefillCallback refillCallback = null;
+
+    /**
+     * Reads more bytes from the input, making at least {@code n} bytes available in the buffer.
+     * Caller must ensure that the requested space is not yet available, and that the requested
+     * space is less than BUFFER_SIZE.
+     *
+     * @throws InvalidProtocolBufferException The end of the stream or the current limit was
+     *     reached.
+     */
+    private void refillBuffer(int n) throws IOException {
+      if (!tryRefillBuffer(n)) {
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+    }
+
+    /**
+     * Tries to read more bytes from the input, making at least {@code n} bytes available in the
+     * buffer. Caller must ensure that the requested space is not yet available, and that the
+     * requested space is less than BUFFER_SIZE.
+     *
+     * @return {@code true} if the bytes could be made available; {@code false} if the end of the
+     *     stream or the current limit was reached.
+     */
+    private boolean tryRefillBuffer(int n) throws IOException {
+      if (pos + n <= bufferSize) {
+        throw new IllegalStateException(
+            "refillBuffer() called when " + n + " bytes were already available in buffer");
+      }
+
+      if (totalBytesRetired + pos + n > currentLimit) {
+        // Oops, we hit a limit.
+        return false;
+      }
+
+      if (refillCallback != null) {
+        refillCallback.onRefill();
+      }
+
+      int tempPos = pos;
+      if (tempPos > 0) {
+        if (bufferSize > tempPos) {
+          System.arraycopy(buffer, tempPos, buffer, 0, bufferSize - tempPos);
+        }
+        totalBytesRetired += tempPos;
+        bufferSize -= tempPos;
+        pos = 0;
+      }
+
+      int bytesRead = input.read(buffer, bufferSize, buffer.length - bufferSize);
+      if (bytesRead == 0 || bytesRead < -1 || bytesRead > buffer.length) {
+        throw new IllegalStateException(
+            "InputStream#read(byte[]) returned invalid result: "
+                + bytesRead
+                + "\nThe InputStream implementation is buggy.");
+      }
+      if (bytesRead > 0) {
+        bufferSize += bytesRead;
+        // Integer-overflow-conscious check against sizeLimit
+        if (totalBytesRetired + n - sizeLimit > 0) {
+          throw InvalidProtocolBufferException.sizeLimitExceeded();
+        }
+        recomputeBufferSizeAfterLimit();
+        return (bufferSize >= n) ? true : tryRefillBuffer(n);
+      }
+
+      return false;
+    }
+
+    @Override
+    public byte readRawByte() throws IOException {
+      if (pos == bufferSize) {
+        refillBuffer(1);
+      }
+      return buffer[pos++];
+    }
+
+    @Override
+    public byte[] readRawBytes(final int size) throws IOException {
+      final int tempPos = pos;
+      if (size <= (bufferSize - tempPos) && size > 0) {
+        pos = tempPos + size;
+        return Arrays.copyOfRange(buffer, tempPos, tempPos + size);
+      } else {
+        return readRawBytesSlowPath(size);
+      }
+    }
+
+    /**
+     * Exactly like readRawBytes, but caller must have already checked the fast path: (size <=
+     * (bufferSize - pos) && size > 0)
+     */
+    private byte[] readRawBytesSlowPath(final int size) throws IOException {
+      if (size == 0) {
+        return Internal.EMPTY_BYTE_ARRAY;
+      }
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+
+      // Verify that the message size so far has not exceeded sizeLimit.
+      int currentMessageSize = totalBytesRetired + pos + size;
+      if (currentMessageSize > sizeLimit) {
+        throw InvalidProtocolBufferException.sizeLimitExceeded();
+      }
+
+      // Verify that the message size so far has not exceeded currentLimit.
+      if (currentMessageSize > currentLimit) {
+        // Read to the end of the stream anyway.
+        skipRawBytes(currentLimit - totalBytesRetired - pos);
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      final int originalBufferPos = pos;
+      final int bufferedBytes = bufferSize - pos;
+
+      // Mark the current buffer consumed.
+      totalBytesRetired += bufferSize;
+      pos = 0;
+      bufferSize = 0;
+
+      // Determine the number of bytes we need to read from the input stream.
+      int sizeLeft = size - bufferedBytes;
+      // TODO(nathanmittler): Consider using a value larger than DEFAULT_BUFFER_SIZE.
+      if (sizeLeft < DEFAULT_BUFFER_SIZE || sizeLeft <= input.available()) {
+        // Either the bytes we need are known to be available, or the required buffer is
+        // within an allowed threshold - go ahead and allocate the buffer now.
+        final byte[] bytes = new byte[size];
+
+        // Copy all of the buffered bytes to the result buffer.
+        System.arraycopy(buffer, originalBufferPos, bytes, 0, bufferedBytes);
+
+        // Fill the remaining bytes from the input stream.
+        int tempPos = bufferedBytes;
+        while (tempPos < bytes.length) {
+          int n = input.read(bytes, tempPos, size - tempPos);
+          if (n == -1) {
+            throw InvalidProtocolBufferException.truncatedMessage();
+          }
+          totalBytesRetired += n;
+          tempPos += n;
+        }
+
+        return bytes;
+      }
+
+      // The size is very large.  For security reasons, we can't allocate the
+      // entire byte array yet.  The size comes directly from the input, so a
+      // maliciously-crafted message could provide a bogus very large size in
+      // order to trick the app into allocating a lot of memory.  We avoid this
+      // by allocating and reading only a small chunk at a time, so that the
+      // malicious message must actually *be* extremely large to cause
+      // problems.  Meanwhile, we limit the allowed size of a message elsewhere.
+      final List<byte[]> chunks = new ArrayList<byte[]>();
+
+      while (sizeLeft > 0) {
+        // TODO(nathanmittler): Consider using a value larger than DEFAULT_BUFFER_SIZE.
+        final byte[] chunk = new byte[Math.min(sizeLeft, DEFAULT_BUFFER_SIZE)];
+        int tempPos = 0;
+        while (tempPos < chunk.length) {
+          final int n = input.read(chunk, tempPos, chunk.length - tempPos);
+          if (n == -1) {
+            throw InvalidProtocolBufferException.truncatedMessage();
+          }
+          totalBytesRetired += n;
+          tempPos += n;
+        }
+        sizeLeft -= chunk.length;
+        chunks.add(chunk);
+      }
+
+      // OK, got everything.  Now concatenate it all into one buffer.
+      final byte[] bytes = new byte[size];
+
+      // Start by copying the leftover bytes from this.buffer.
+      System.arraycopy(buffer, originalBufferPos, bytes, 0, bufferedBytes);
+
+      // And now all the chunks.
+      int tempPos = bufferedBytes;
+      for (final byte[] chunk : chunks) {
+        System.arraycopy(chunk, 0, bytes, tempPos, chunk.length);
+        tempPos += chunk.length;
+      }
+
+      // Done.
+      return bytes;
+    }
+
+    @Override
+    public void skipRawBytes(final int size) throws IOException {
+      if (size <= (bufferSize - pos) && size >= 0) {
+        // We have all the bytes we need already.
+        pos += size;
+      } else {
+        skipRawBytesSlowPath(size);
+      }
+    }
+
+    /**
+     * Exactly like skipRawBytes, but caller must have already checked the fast path: (size <=
+     * (bufferSize - pos) && size >= 0)
+     */
+    private void skipRawBytesSlowPath(final int size) throws IOException {
+      if (size < 0) {
+        throw InvalidProtocolBufferException.negativeSize();
+      }
+
+      if (totalBytesRetired + pos + size > currentLimit) {
+        // Read to the end of the stream anyway.
+        skipRawBytes(currentLimit - totalBytesRetired - pos);
+        // Then fail.
+        throw InvalidProtocolBufferException.truncatedMessage();
+      }
+
+      // Skipping more bytes than are in the buffer.  First skip what we have.
+      int tempPos = bufferSize - pos;
+      pos = bufferSize;
+
+      // Keep refilling the buffer until we get to the point we wanted to skip to.
+      // This has the side effect of ensuring the limits are updated correctly.
+      refillBuffer(1);
+      while (size - tempPos > bufferSize) {
+        tempPos += bufferSize;
+        pos = bufferSize;
+        refillBuffer(1);
+      }
+
+      pos = size - tempPos;
+    }
   }
 }
diff --git a/java/core/src/main/java/com/google/protobuf/CodedOutputStream.java b/java/core/src/main/java/com/google/protobuf/CodedOutputStream.java
index e5515285..3e32c2c5 100644
--- a/java/core/src/main/java/com/google/protobuf/CodedOutputStream.java
+++ b/java/core/src/main/java/com/google/protobuf/CodedOutputStream.java
@@ -30,10 +30,12 @@
 
 package com.google.protobuf;
 
+import static com.google.protobuf.WireFormat.FIXED_32_SIZE;
+import static com.google.protobuf.WireFormat.FIXED_64_SIZE;
+import static com.google.protobuf.WireFormat.MAX_VARINT_SIZE;
 import static java.lang.Math.max;
 
 import com.google.protobuf.Utf8.UnpairedSurrogateException;
-
 import java.io.IOException;
 import java.io.OutputStream;
 import java.nio.BufferOverflowException;
@@ -59,10 +61,6 @@ public abstract class CodedOutputStream extends ByteOutput {
   private static final boolean HAS_UNSAFE_ARRAY_OPERATIONS = UnsafeUtil.hasUnsafeArrayOperations();
   private static final long ARRAY_BASE_OFFSET = UnsafeUtil.getArrayBaseOffset();
 
-  private static final int FIXED_32_SIZE = 4;
-  private static final int FIXED_64_SIZE = 8;
-  private static final int MAX_VARINT_SIZE = 10;
-
   /**
    * @deprecated Use {@link #computeFixed32SizeNoTag(int)} instead.
    */
@@ -134,14 +132,27 @@ public abstract class CodedOutputStream extends ByteOutput {
     return new ArrayEncoder(flatArray, offset, length);
   }
 
-  /**
-   * Create a new {@code CodedOutputStream} that writes to the given {@link ByteBuffer}.
-   */
-  public static CodedOutputStream newInstance(ByteBuffer byteBuffer) {
-    if (byteBuffer.hasArray()) {
-      return new NioHeapEncoder(byteBuffer);
+  /** Create a new {@code CodedOutputStream} that writes to the given {@link ByteBuffer}. */
+  public static CodedOutputStream newInstance(ByteBuffer buffer) {
+    if (buffer.hasArray()) {
+      return new HeapNioEncoder(buffer);
     }
-    return new NioEncoder(byteBuffer);
+    if (buffer.isDirect() && !buffer.isReadOnly()) {
+      return UnsafeDirectNioEncoder.isSupported()
+          ? newUnsafeInstance(buffer)
+          : newSafeInstance(buffer);
+    }
+    throw new IllegalArgumentException("ByteBuffer is read-only");
+  }
+
+  /** For testing purposes only. */
+  static CodedOutputStream newUnsafeInstance(ByteBuffer buffer) {
+    return new UnsafeDirectNioEncoder(buffer);
+  }
+
+  /** For testing purposes only. */
+  static CodedOutputStream newSafeInstance(ByteBuffer buffer) {
+    return new SafeDirectNioEncoder(buffer);
   }
 
   /**
@@ -979,6 +990,10 @@ public abstract class CodedOutputStream extends ByteOutput {
       super(MESSAGE);
     }
 
+    OutOfSpaceException(String explanationMessage) {
+      super(MESSAGE + ": " + explanationMessage);
+    }
+
     OutOfSpaceException(Throwable cause) {
       super(MESSAGE, cause);
     }
@@ -1486,11 +1501,11 @@ public abstract class CodedOutputStream extends ByteOutput {
    * A {@link CodedOutputStream} that writes directly to a heap {@link ByteBuffer}. Writes are
    * done directly to the underlying array. The buffer position is only updated after a flush.
    */
-  private static final class NioHeapEncoder extends ArrayEncoder {
+  private static final class HeapNioEncoder extends ArrayEncoder {
     private final ByteBuffer byteBuffer;
     private int initialPosition;
 
-    NioHeapEncoder(ByteBuffer byteBuffer) {
+    HeapNioEncoder(ByteBuffer byteBuffer) {
       super(byteBuffer.array(), byteBuffer.arrayOffset() + byteBuffer.position(),
           byteBuffer.remaining());
       this.byteBuffer = byteBuffer;
@@ -1505,14 +1520,15 @@ public abstract class CodedOutputStream extends ByteOutput {
   }
 
   /**
-   * A {@link CodedOutputStream} that writes directly to a {@link ByteBuffer}.
+   * A {@link CodedOutputStream} that writes directly to a direct {@link ByteBuffer}, using only
+   * safe operations..
    */
-  private static final class NioEncoder extends CodedOutputStream {
+  private static final class SafeDirectNioEncoder extends CodedOutputStream {
     private final ByteBuffer originalBuffer;
     private final ByteBuffer buffer;
     private final int initialPosition;
 
-    NioEncoder(ByteBuffer buffer) {
+    SafeDirectNioEncoder(ByteBuffer buffer) {
       this.originalBuffer = buffer;
       this.buffer = buffer.duplicate().order(ByteOrder.LITTLE_ENDIAN);
       initialPosition = buffer.position();
@@ -1815,6 +1831,356 @@ public abstract class CodedOutputStream extends ByteOutput {
   }
 
   /**
+   * A {@link CodedOutputStream} that writes directly to a direct {@link ByteBuffer} using {@code
+   * sun.misc.Unsafe}.
+   */
+  private static final class UnsafeDirectNioEncoder extends CodedOutputStream {
+    private final ByteBuffer originalBuffer;
+    private final ByteBuffer buffer;
+    private final long address;
+    private final long initialPosition;
+    private final long limit;
+    private final long oneVarintLimit;
+    private long position;
+
+    UnsafeDirectNioEncoder(ByteBuffer buffer) {
+      this.originalBuffer = buffer;
+      this.buffer = buffer.duplicate().order(ByteOrder.LITTLE_ENDIAN);
+      address = UnsafeUtil.addressOffset(buffer);
+      initialPosition = address + buffer.position();
+      limit = address + buffer.limit();
+      oneVarintLimit = limit - MAX_VARINT_SIZE;
+      position = initialPosition;
+    }
+
+    static boolean isSupported() {
+      return UnsafeUtil.hasUnsafeByteBufferOperations();
+    }
+
+    @Override
+    public void writeTag(int fieldNumber, int wireType) throws IOException {
+      writeUInt32NoTag(WireFormat.makeTag(fieldNumber, wireType));
+    }
+
+    @Override
+    public void writeInt32(int fieldNumber, int value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_VARINT);
+      writeInt32NoTag(value);
+    }
+
+    @Override
+    public void writeUInt32(int fieldNumber, int value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_VARINT);
+      writeUInt32NoTag(value);
+    }
+
+    @Override
+    public void writeFixed32(int fieldNumber, int value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_FIXED32);
+      writeFixed32NoTag(value);
+    }
+
+    @Override
+    public void writeUInt64(int fieldNumber, long value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_VARINT);
+      writeUInt64NoTag(value);
+    }
+
+    @Override
+    public void writeFixed64(int fieldNumber, long value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_FIXED64);
+      writeFixed64NoTag(value);
+    }
+
+    @Override
+    public void writeBool(int fieldNumber, boolean value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_VARINT);
+      write((byte) (value ? 1 : 0));
+    }
+
+    @Override
+    public void writeString(int fieldNumber, String value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_LENGTH_DELIMITED);
+      writeStringNoTag(value);
+    }
+
+    @Override
+    public void writeBytes(int fieldNumber, ByteString value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_LENGTH_DELIMITED);
+      writeBytesNoTag(value);
+    }
+
+    @Override
+    public void writeByteArray(int fieldNumber, byte[] value) throws IOException {
+      writeByteArray(fieldNumber, value, 0, value.length);
+    }
+
+    @Override
+    public void writeByteArray(int fieldNumber, byte[] value, int offset, int length)
+        throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_LENGTH_DELIMITED);
+      writeByteArrayNoTag(value, offset, length);
+    }
+
+    @Override
+    public void writeByteBuffer(int fieldNumber, ByteBuffer value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_LENGTH_DELIMITED);
+      writeUInt32NoTag(value.capacity());
+      writeRawBytes(value);
+    }
+
+    @Override
+    public void writeMessage(int fieldNumber, MessageLite value) throws IOException {
+      writeTag(fieldNumber, WireFormat.WIRETYPE_LENGTH_DELIMITED);
+      writeMessageNoTag(value);
+    }
+
+    @Override
+    public void writeMessageSetExtension(int fieldNumber, MessageLite value) throws IOException {
+      writeTag(WireFormat.MESSAGE_SET_ITEM, WireFormat.WIRETYPE_START_GROUP);
+      writeUInt32(WireFormat.MESSAGE_SET_TYPE_ID, fieldNumber);
+      writeMessage(WireFormat.MESSAGE_SET_MESSAGE, value);
+      writeTag(WireFormat.MESSAGE_SET_ITEM, WireFormat.WIRETYPE_END_GROUP);
+    }
+
+    @Override
+    public void writeRawMessageSetExtension(int fieldNumber, ByteString value) throws IOException {
+      writeTag(WireFormat.MESSAGE_SET_ITEM, WireFormat.WIRETYPE_START_GROUP);
+      writeUInt32(WireFormat.MESSAGE_SET_TYPE_ID, fieldNumber);
+      writeBytes(WireFormat.MESSAGE_SET_MESSAGE, value);
+      writeTag(WireFormat.MESSAGE_SET_ITEM, WireFormat.WIRETYPE_END_GROUP);
+    }
+
+    @Override
+    public void writeMessageNoTag(MessageLite value) throws IOException {
+      writeUInt32NoTag(value.getSerializedSize());
+      value.writeTo(this);
+    }
+
+    @Override
+    public void write(byte value) throws IOException {
+      if (position >= limit) {
+        throw new OutOfSpaceException(
+            String.format("Pos: %d, limit: %d, len: %d", position, limit, 1));
+      }
+      UnsafeUtil.putByte(position++, value);
+    }
+
+    @Override
+    public void writeBytesNoTag(ByteString value) throws IOException {
+      writeUInt32NoTag(value.size());
+      value.writeTo(this);
+    }
+
+    @Override
+    public void writeByteArrayNoTag(byte[] value, int offset, int length) throws IOException {
+      writeUInt32NoTag(length);
+      write(value, offset, length);
+    }
+
+    @Override
+    public void writeRawBytes(ByteBuffer value) throws IOException {
+      if (value.hasArray()) {
+        write(value.array(), value.arrayOffset(), value.capacity());
+      } else {
+        ByteBuffer duplicated = value.duplicate();
+        duplicated.clear();
+        write(duplicated);
+      }
+    }
+
+    @Override
+    public void writeInt32NoTag(int value) throws IOException {
+      if (value >= 0) {
+        writeUInt32NoTag(value);
+      } else {
+        // Must sign-extend.
+        writeUInt64NoTag(value);
+      }
+    }
+
+    @Override
+    public void writeUInt32NoTag(int value) throws IOException {
+      if (position <= oneVarintLimit) {
+        // Optimization to avoid bounds checks on each iteration.
+        while (true) {
+          if ((value & ~0x7F) == 0) {
+            UnsafeUtil.putByte(position++, (byte) value);
+            return;
+          } else {
+            UnsafeUtil.putByte(position++, (byte) ((value & 0x7F) | 0x80));
+            value >>>= 7;
+          }
+        }
+      } else {
+        while (position < limit) {
+          if ((value & ~0x7F) == 0) {
+            UnsafeUtil.putByte(position++, (byte) value);
+            return;
+          } else {
+            UnsafeUtil.putByte(position++, (byte) ((value & 0x7F) | 0x80));
+            value >>>= 7;
+          }
+        }
+        throw new OutOfSpaceException(
+            String.format("Pos: %d, limit: %d, len: %d", position, limit, 1));
+      }
+    }
+
+    @Override
+    public void writeFixed32NoTag(int value) throws IOException {
+      buffer.putInt(bufferPos(position), value);
+      position += FIXED_32_SIZE;
+    }
+
+    @Override
+    public void writeUInt64NoTag(long value) throws IOException {
+      if (position <= oneVarintLimit) {
+        // Optimization to avoid bounds checks on each iteration.
+        while (true) {
+          if ((value & ~0x7FL) == 0) {
+            UnsafeUtil.putByte(position++, (byte) value);
+            return;
+          } else {
+            UnsafeUtil.putByte(position++, (byte) (((int) value & 0x7F) | 0x80));
+            value >>>= 7;
+          }
+        }
+      } else {
+        while (position < limit) {
+          if ((value & ~0x7FL) == 0) {
+            UnsafeUtil.putByte(position++, (byte) value);
+            return;
+          } else {
+            UnsafeUtil.putByte(position++, (byte) (((int) value & 0x7F) | 0x80));
+            value >>>= 7;
+          }
+        }
+        throw new OutOfSpaceException(
+            String.format("Pos: %d, limit: %d, len: %d", position, limit, 1));
+      }
+    }
+
+    @Override
+    public void writeFixed64NoTag(long value) throws IOException {
+      buffer.putLong(bufferPos(position), value);
+      position += FIXED_64_SIZE;
+    }
+
+    @Override
+    public void write(byte[] value, int offset, int length) throws IOException {
+      if (value == null
+          || offset < 0
+          || length < 0
+          || (value.length - length) < offset
+          || (limit - length) < position) {
+        if (value == null) {
+          throw new NullPointerException("value");
+        }
+        throw new OutOfSpaceException(
+            String.format("Pos: %d, limit: %d, len: %d", position, limit, length));
+      }
+
+      UnsafeUtil.copyMemory(
+          value, UnsafeUtil.getArrayBaseOffset() + offset, null, position, length);
+      position += length;
+    }
+
+    @Override
+    public void writeLazy(byte[] value, int offset, int length) throws IOException {
+      write(value, offset, length);
+    }
+
+    @Override
+    public void write(ByteBuffer value) throws IOException {
+      try {
+        int length = value.remaining();
+        repositionBuffer(position);
+        buffer.put(value);
+        position += length;
+      } catch (BufferOverflowException e) {
+        throw new OutOfSpaceException(e);
+      }
+    }
+
+    @Override
+    public void writeLazy(ByteBuffer value) throws IOException {
+      write(value);
+    }
+
+    @Override
+    public void writeStringNoTag(String value) throws IOException {
+      long prevPos = position;
+      try {
+        // UTF-8 byte length of the string is at least its UTF-16 code unit length (value.length()),
+        // and at most 3 times of it. We take advantage of this in both branches below.
+        int maxEncodedSize = value.length() * Utf8.MAX_BYTES_PER_CHAR;
+        int maxLengthVarIntSize = computeUInt32SizeNoTag(maxEncodedSize);
+        int minLengthVarIntSize = computeUInt32SizeNoTag(value.length());
+        if (minLengthVarIntSize == maxLengthVarIntSize) {
+          // Save the current position and increment past the length field. We'll come back
+          // and write the length field after the encoding is complete.
+          int stringStart = bufferPos(position) + minLengthVarIntSize;
+          buffer.position(stringStart);
+
+          // Encode the string.
+          Utf8.encodeUtf8(value, buffer);
+
+          // Write the length and advance the position.
+          int length = buffer.position() - stringStart;
+          writeUInt32NoTag(length);
+          position += length;
+        } else {
+          // Calculate and write the encoded length.
+          int length = Utf8.encodedLength(value);
+          writeUInt32NoTag(length);
+
+          // Write the string and advance the position.
+          repositionBuffer(position);
+          Utf8.encodeUtf8(value, buffer);
+          position += length;
+        }
+      } catch (UnpairedSurrogateException e) {
+        // Roll back the change and convert to an IOException.
+        position = prevPos;
+        repositionBuffer(position);
+
+        // TODO(nathanmittler): We should throw an IOException here instead.
+        inefficientWriteStringNoTag(value, e);
+      } catch (IllegalArgumentException e) {
+        // Thrown by buffer.position() if out of range.
+        throw new OutOfSpaceException(e);
+      } catch (IndexOutOfBoundsException e) {
+        throw new OutOfSpaceException(e);
+      }
+    }
+
+    @Override
+    public void flush() {
+      // Update the position of the original buffer.
+      originalBuffer.position(bufferPos(position));
+    }
+
+    @Override
+    public int spaceLeft() {
+      return (int) (limit - position);
+    }
+
+    @Override
+    public int getTotalBytesWritten() {
+      return (int) (position - initialPosition);
+    }
+
+    private void repositionBuffer(long pos) {
+      buffer.position(bufferPos(pos));
+    }
+
+    private int bufferPos(long pos) {
+      return (int) (pos - address);
+    }
+  }
+
+  /**
    * Abstract base class for buffered encoders.
    */
   private abstract static class AbstractBufferedEncoder extends CodedOutputStream {
diff --git a/java/core/src/main/java/com/google/protobuf/Descriptors.java b/java/core/src/main/java/com/google/protobuf/Descriptors.java
index 1c34c24f..cab7118a 100644
--- a/java/core/src/main/java/com/google/protobuf/Descriptors.java
+++ b/java/core/src/main/java/com/google/protobuf/Descriptors.java
@@ -1212,33 +1212,20 @@ public final class Descriptors {
       private final Object defaultDefault;
     }
 
-    // TODO(xiaofeng): Implement it consistently across different languages. See b/24751348.
-    private static String fieldNameToLowerCamelCase(String name) {
+    // This method should match exactly with the ToJsonName() function in C++
+    // descriptor.cc.
+    private static String fieldNameToJsonName(String name) {
       StringBuilder result = new StringBuilder(name.length());
       boolean isNextUpperCase = false;
       for (int i = 0; i < name.length(); i++) {
         Character ch = name.charAt(i);
-        if (Character.isLowerCase(ch)) {
-          if (isNextUpperCase) {
-            result.append(Character.toUpperCase(ch));
-          } else {
-            result.append(ch);
-          }
-          isNextUpperCase = false;
-        } else if (Character.isUpperCase(ch)) {
-          if (i == 0) {
-            // Force first letter to lower-case.
-            result.append(Character.toLowerCase(ch));
-          } else {
-            // Capital letters after the first are left as-is.
-            result.append(ch);
-          }
-          isNextUpperCase = false;
-        } else if (Character.isDigit(ch)) {
-          result.append(ch);
+        if (ch == '_') {
+          isNextUpperCase = true;
+        } else if (isNextUpperCase) {
+          result.append(Character.toUpperCase(ch));
           isNextUpperCase = false;
         } else {
-          isNextUpperCase = true;
+          result.append(ch);
         }
       }
       return result.toString();
@@ -1257,7 +1244,7 @@ public final class Descriptors {
       if (proto.hasJsonName()) {
         jsonName = proto.getJsonName();
       } else {
-        jsonName = fieldNameToLowerCamelCase(proto.getName());
+        jsonName = fieldNameToJsonName(proto.getName());
       }
 
       if (proto.hasType()) {
diff --git a/java/core/src/main/java/com/google/protobuf/GeneratedMessageV3.java b/java/core/src/main/java/com/google/protobuf/GeneratedMessageV3.java
index 5dfe7ff7..5c0b9967 100644
--- a/java/core/src/main/java/com/google/protobuf/GeneratedMessageV3.java
+++ b/java/core/src/main/java/com/google/protobuf/GeneratedMessageV3.java
@@ -36,7 +36,6 @@ import com.google.protobuf.Descriptors.EnumValueDescriptor;
 import com.google.protobuf.Descriptors.FieldDescriptor;
 import com.google.protobuf.Descriptors.FileDescriptor;
 import com.google.protobuf.Descriptors.OneofDescriptor;
-import com.google.protobuf.GeneratedMessage.GeneratedExtension;
 
 import java.io.IOException;
 import java.io.InputStream;
@@ -45,6 +44,7 @@ import java.io.Serializable;
 import java.lang.reflect.InvocationTargetException;
 import java.lang.reflect.Method;
 import java.util.ArrayList;
+import java.util.Arrays;
 import java.util.Collections;
 import java.util.Iterator;
 import java.util.List;
@@ -1609,6 +1609,43 @@ public abstract class GeneratedMessageV3 extends AbstractMessage
     FieldDescriptor getDescriptor();
   }
 
+  /** For use by generated code only. */
+  public static <ContainingType extends Message, Type>
+      GeneratedExtension<ContainingType, Type>
+      newMessageScopedGeneratedExtension(final Message scope,
+                                         final int descriptorIndex,
+                                         final Class singularType,
+                                         final Message defaultInstance) {
+    // For extensions scoped within a Message, we use the Message to resolve
+    // the outer class's descriptor, from which the extension descriptor is
+    // obtained.
+    return new GeneratedExtension<ContainingType, Type>(
+        new CachedDescriptorRetriever() {
+          @Override
+          public FieldDescriptor loadDescriptor() {
+            return scope.getDescriptorForType().getExtensions().get(descriptorIndex);
+          }
+        },
+        singularType,
+        defaultInstance,
+        Extension.ExtensionType.IMMUTABLE);
+  }
+
+  /** For use by generated code only. */
+  public static <ContainingType extends Message, Type>
+     GeneratedExtension<ContainingType, Type>
+     newFileScopedGeneratedExtension(final Class singularType,
+                                     final Message defaultInstance) {
+    // For extensions scoped within a file, we rely on the outer class's
+    // static initializer to call internalInit() on the extension when the
+    // descriptor is available.
+    return new GeneratedExtension<ContainingType, Type>(
+        null,  // ExtensionDescriptorRetriever is initialized in internalInit();
+        singularType,
+        defaultInstance,
+        Extension.ExtensionType.IMMUTABLE);
+  }
+
   private abstract static class CachedDescriptorRetriever
       implements ExtensionDescriptorRetriever {
     private volatile FieldDescriptor descriptor;
@@ -1627,6 +1664,301 @@ public abstract class GeneratedMessageV3 extends AbstractMessage
     }
   }
 
+  /**
+   * Used in proto1 generated code only.
+   *
+   * After enabling bridge, we can define proto2 extensions (the extended type
+   * is a proto2 mutable message) in a proto1 .proto file. For these extensions
+   * we should generate proto2 GeneratedExtensions.
+   */
+  public static <ContainingType extends Message, Type>
+      GeneratedExtension<ContainingType, Type>
+      newMessageScopedGeneratedExtension(
+          final Message scope, final String name,
+          final Class singularType, final Message defaultInstance) {
+    // For extensions scoped within a Message, we use the Message to resolve
+    // the outer class's descriptor, from which the extension descriptor is
+    // obtained.
+    return new GeneratedExtension<ContainingType, Type>(
+        new CachedDescriptorRetriever() {
+          @Override
+          protected FieldDescriptor loadDescriptor() {
+            return scope.getDescriptorForType().findFieldByName(name);
+          }
+        },
+        singularType,
+        defaultInstance,
+        Extension.ExtensionType.MUTABLE);
+  }
+
+  /**
+   * Used in proto1 generated code only.
+   *
+   * After enabling bridge, we can define proto2 extensions (the extended type
+   * is a proto2 mutable message) in a proto1 .proto file. For these extensions
+   * we should generate proto2 GeneratedExtensions.
+   */
+  public static <ContainingType extends Message, Type>
+     GeneratedExtension<ContainingType, Type>
+     newFileScopedGeneratedExtension(
+         final Class singularType, final Message defaultInstance,
+         final String descriptorOuterClass, final String extensionName) {
+    // For extensions scoped within a file, we load the descriptor outer
+    // class and rely on it to get the FileDescriptor which then can be
+    // used to obtain the extension's FieldDescriptor.
+    return new GeneratedExtension<ContainingType, Type>(
+        new CachedDescriptorRetriever() {
+          @Override
+          protected FieldDescriptor loadDescriptor() {
+            try {
+              Class clazz = singularType.getClassLoader().loadClass(descriptorOuterClass);
+              FileDescriptor file = (FileDescriptor) clazz.getField("descriptor").get(null);
+              return file.findExtensionByName(extensionName);
+            } catch (Exception e) {
+              throw new RuntimeException(
+                  "Cannot load descriptors: "
+                      + descriptorOuterClass
+                      + " is not a valid descriptor class name",
+                  e);
+            }
+          }
+        },
+        singularType,
+        defaultInstance,
+        Extension.ExtensionType.MUTABLE);
+  }
+
+  /**
+   * Type used to represent generated extensions.  The protocol compiler
+   * generates a static singleton instance of this class for each extension.
+   *
+   * <p>For example, imagine you have the {@code .proto} file:
+   *
+   * <pre>
+   * option java_class = "MyProto";
+   *
+   * message Foo {
+   *   extensions 1000 to max;
+   * }
+   *
+   * extend Foo {
+   *   optional int32 bar;
+   * }
+   * </pre>
+   *
+   * <p>Then, {@code MyProto.Foo.bar} has type
+   * {@code GeneratedExtension<MyProto.Foo, Integer>}.
+   *
+   * <p>In general, users should ignore the details of this type, and simply use
+   * these static singletons as parameters to the extension accessors defined
+   * in {@link ExtendableMessage} and {@link ExtendableBuilder}.
+   */
+  public static class GeneratedExtension<
+      ContainingType extends Message, Type> extends
+      Extension<ContainingType, Type> {
+    // TODO(kenton):  Find ways to avoid using Java reflection within this
+    //   class.  Also try to avoid suppressing unchecked warnings.
+
+    // We can't always initialize the descriptor of a GeneratedExtension when
+    // we first construct it due to initialization order difficulties (namely,
+    // the descriptor may not have been constructed yet, since it is often
+    // constructed by the initializer of a separate module).
+    //
+    // In the case of nested extensions, we initialize the
+    // ExtensionDescriptorRetriever with an instance that uses the scoping
+    // Message's default instance to retrieve the extension's descriptor.
+    //
+    // In the case of non-nested extensions, we initialize the
+    // ExtensionDescriptorRetriever to null and rely on the outer class's static
+    // initializer to call internalInit() after the descriptor has been parsed.
+    GeneratedExtension(ExtensionDescriptorRetriever descriptorRetriever,
+        Class singularType,
+        Message messageDefaultInstance,
+        ExtensionType extensionType) {
+      if (Message.class.isAssignableFrom(singularType) &&
+          !singularType.isInstance(messageDefaultInstance)) {
+        throw new IllegalArgumentException(
+            "Bad messageDefaultInstance for " + singularType.getName());
+      }
+      this.descriptorRetriever = descriptorRetriever;
+      this.singularType = singularType;
+      this.messageDefaultInstance = messageDefaultInstance;
+
+      if (ProtocolMessageEnum.class.isAssignableFrom(singularType)) {
+        this.enumValueOf = getMethodOrDie(singularType, "valueOf",
+                                          EnumValueDescriptor.class);
+        this.enumGetValueDescriptor =
+            getMethodOrDie(singularType, "getValueDescriptor");
+      } else {
+        this.enumValueOf = null;
+        this.enumGetValueDescriptor = null;
+      }
+      this.extensionType = extensionType;
+    }
+
+    /** For use by generated code only. */
+    public void internalInit(final FieldDescriptor descriptor) {
+      if (descriptorRetriever != null) {
+        throw new IllegalStateException("Already initialized.");
+      }
+      descriptorRetriever =
+          new ExtensionDescriptorRetriever() {
+            @Override
+            public FieldDescriptor getDescriptor() {
+              return descriptor;
+            }
+          };
+    }
+
+    private ExtensionDescriptorRetriever descriptorRetriever;
+    private final Class singularType;
+    private final Message messageDefaultInstance;
+    private final Method enumValueOf;
+    private final Method enumGetValueDescriptor;
+    private final ExtensionType extensionType;
+
+    @Override
+    public FieldDescriptor getDescriptor() {
+      if (descriptorRetriever == null) {
+        throw new IllegalStateException(
+            "getDescriptor() called before internalInit()");
+      }
+      return descriptorRetriever.getDescriptor();
+    }
+
+    /**
+     * If the extension is an embedded message or group, returns the default
+     * instance of the message.
+     */
+    @Override
+    public Message getMessageDefaultInstance() {
+      return messageDefaultInstance;
+    }
+
+    @Override
+    protected ExtensionType getExtensionType() {
+      return extensionType;
+    }
+
+    /**
+     * Convert from the type used by the reflection accessors to the type used
+     * by native accessors.  E.g., for enums, the reflection accessors use
+     * EnumValueDescriptors but the native accessors use the generated enum
+     * type.
+     */
+    @Override
+    @SuppressWarnings("unchecked")
+    protected Object fromReflectionType(final Object value) {
+      FieldDescriptor descriptor = getDescriptor();
+      if (descriptor.isRepeated()) {
+        if (descriptor.getJavaType() == FieldDescriptor.JavaType.MESSAGE ||
+            descriptor.getJavaType() == FieldDescriptor.JavaType.ENUM) {
+          // Must convert the whole list.
+          final List result = new ArrayList();
+          for (final Object element : (List) value) {
+            result.add(singularFromReflectionType(element));
+          }
+          return result;
+        } else {
+          return value;
+        }
+      } else {
+        return singularFromReflectionType(value);
+      }
+    }
+
+    /**
+     * Like {@link #fromReflectionType(Object)}, but if the type is a repeated
+     * type, this converts a single element.
+     */
+    @Override
+    protected Object singularFromReflectionType(final Object value) {
+      FieldDescriptor descriptor = getDescriptor();
+      switch (descriptor.getJavaType()) {
+        case MESSAGE:
+          if (singularType.isInstance(value)) {
+            return value;
+          } else {
+            return messageDefaultInstance.newBuilderForType()
+                .mergeFrom((Message) value).build();
+          }
+        case ENUM:
+          return invokeOrDie(enumValueOf, null, (EnumValueDescriptor) value);
+        default:
+          return value;
+      }
+    }
+
+    /**
+     * Convert from the type used by the native accessors to the type used
+     * by reflection accessors.  E.g., for enums, the reflection accessors use
+     * EnumValueDescriptors but the native accessors use the generated enum
+     * type.
+     */
+    @Override
+    @SuppressWarnings("unchecked")
+    protected Object toReflectionType(final Object value) {
+      FieldDescriptor descriptor = getDescriptor();
+      if (descriptor.isRepeated()) {
+        if (descriptor.getJavaType() == FieldDescriptor.JavaType.ENUM) {
+          // Must convert the whole list.
+          final List result = new ArrayList();
+          for (final Object element : (List) value) {
+            result.add(singularToReflectionType(element));
+          }
+          return result;
+        } else {
+          return value;
+        }
+      } else {
+        return singularToReflectionType(value);
+      }
+    }
+
+    /**
+     * Like {@link #toReflectionType(Object)}, but if the type is a repeated
+     * type, this converts a single element.
+     */
+    @Override
+    protected Object singularToReflectionType(final Object value) {
+      FieldDescriptor descriptor = getDescriptor();
+      switch (descriptor.getJavaType()) {
+        case ENUM:
+          return invokeOrDie(enumGetValueDescriptor, value);
+        default:
+          return value;
+      }
+    }
+
+    @Override
+    public int getNumber() {
+      return getDescriptor().getNumber();
+    }
+
+    @Override
+    public WireFormat.FieldType getLiteType() {
+      return getDescriptor().getLiteType();
+    }
+
+    @Override
+    public boolean isRepeated() {
+      return getDescriptor().isRepeated();
+    }
+
+    @Override
+    @SuppressWarnings("unchecked")
+    public Type getDefaultValue() {
+      if (isRepeated()) {
+        return (Type) Collections.emptyList();
+      }
+      if (getDescriptor().getJavaType() == FieldDescriptor.JavaType.MESSAGE) {
+        return (Type) messageDefaultInstance;
+      }
+      return (Type) singularFromReflectionType(
+          getDescriptor().getDefaultValue());
+    }
+  }
+
   // =================================================================
 
   /** Calls Class.getMethod and throws a RuntimeException if it fails. */
@@ -2223,6 +2555,20 @@ public abstract class GeneratedMessageV3 extends AbstractMessage
             field.getNumber());
       }
 
+      private Message coerceType(Message value) {
+        if (value == null) {
+          return null;
+        }
+        if (mapEntryMessageDefaultInstance.getClass().isInstance(value)) {
+          return value;
+        }
+        // The value is not the exact right message type.  However, if it
+        // is an alternative implementation of the same type -- e.g. a
+        // DynamicMessage -- we should accept it.  In this case we can make
+        // a copy of the message.
+        return mapEntryMessageDefaultInstance.toBuilder().mergeFrom(value).build();
+      }
+
       @Override
       public Object get(GeneratedMessageV3 message) {
         List result = new ArrayList();
@@ -2278,15 +2624,15 @@ public abstract class GeneratedMessageV3 extends AbstractMessage
       public Object getRepeatedRaw(Builder builder, int index) {
         return getRepeated(builder, index);
       }
-
+      
       @Override
       public void setRepeated(Builder builder, int index, Object value) {
-        getMutableMapField(builder).getMutableList().set(index, (Message) value);
+        getMutableMapField(builder).getMutableList().set(index, coerceType((Message) value));
       }
 
       @Override
       public void addRepeated(Builder builder, Object value) {
-        getMutableMapField(builder).getMutableList().add((Message) value);
+        getMutableMapField(builder).getMutableList().add(coerceType((Message) value));
       }
 
       @Override
@@ -2713,4 +3059,131 @@ public abstract class GeneratedMessageV3 extends AbstractMessage
       output.writeBytesNoTag((ByteString) value);
     }
   }
+
+  protected static <V> void serializeIntegerMapTo(
+      CodedOutputStream out,
+      MapField<Integer, V> field,
+      MapEntry<Integer, V> defaultEntry,
+      int fieldNumber) throws IOException {
+    Map<Integer, V> m = field.getMap();
+    if (!out.isSerializationDeterministic()) {
+      serializeMapTo(out, m, defaultEntry, fieldNumber);
+      return;
+    }
+    // Sorting the unboxed keys and then look up the values during serialziation is 2x faster
+    // than sorting map entries with a custom comparator directly.
+    int[] keys = new int[m.size()];
+    int index = 0;
+    for (int k : m.keySet()) {
+      keys[index++] = k;
+    }
+    Arrays.sort(keys);
+    for (int key : keys) {
+      out.writeMessage(fieldNumber,
+          defaultEntry.newBuilderForType()
+              .setKey(key)
+              .setValue(m.get(key))
+              .build());
+    }
+  }
+
+  protected static <V> void serializeLongMapTo(
+      CodedOutputStream out,
+      MapField<Long, V> field,
+      MapEntry<Long, V> defaultEntry,
+      int fieldNumber)
+      throws IOException {
+    Map<Long, V> m = field.getMap();
+    if (!out.isSerializationDeterministic()) {
+      serializeMapTo(out, m, defaultEntry, fieldNumber);
+      return;
+    }
+
+    long[] keys = new long[m.size()];
+    int index = 0;
+    for (long k : m.keySet()) {
+      keys[index++] = k;
+    }
+    Arrays.sort(keys);
+    for (long key : keys) {
+      out.writeMessage(fieldNumber,
+          defaultEntry.newBuilderForType()
+              .setKey(key)
+              .setValue(m.get(key))
+              .build());
+    }
+  }
+
+  protected static <V> void serializeStringMapTo(
+      CodedOutputStream out,
+      MapField<String, V> field,
+      MapEntry<String, V> defaultEntry,
+      int fieldNumber)
+      throws IOException {
+    Map<String, V> m = field.getMap();
+    if (!out.isSerializationDeterministic()) {
+      serializeMapTo(out, m, defaultEntry, fieldNumber);
+      return;
+    }
+
+    // Sorting the String keys and then look up the values during serialziation is 25% faster than
+    // sorting map entries with a custom comparator directly.
+    String[] keys = new String[m.size()];
+    keys = m.keySet().toArray(keys);
+    Arrays.sort(keys);
+    for (String key : keys) {
+      out.writeMessage(fieldNumber,
+          defaultEntry.newBuilderForType()
+              .setKey(key)
+              .setValue(m.get(key))
+              .build());
+    }
+  }
+
+  protected static <V> void serializeBooleanMapTo(
+      CodedOutputStream out,
+      MapField<Boolean, V> field,
+      MapEntry<Boolean, V> defaultEntry,
+      int fieldNumber)
+      throws IOException {
+    Map<Boolean, V> m = field.getMap();
+    if (!out.isSerializationDeterministic()) {
+      serializeMapTo(out, m, defaultEntry, fieldNumber);
+      return;
+    }
+    maybeSerializeBooleanEntryTo(out, m, defaultEntry, fieldNumber, false);
+    maybeSerializeBooleanEntryTo(out, m, defaultEntry, fieldNumber, true);
+  }
+
+  private static <V> void maybeSerializeBooleanEntryTo(
+      CodedOutputStream out,
+      Map<Boolean, V> m,
+      MapEntry<Boolean, V> defaultEntry,
+      int fieldNumber,
+      boolean key)
+      throws IOException {
+    if (m.containsKey(key)) {
+      out.writeMessage(fieldNumber,
+          defaultEntry.newBuilderForType()
+              .setKey(key)
+              .setValue(m.get(key))
+              .build());
+    }
+  }
+
+  /** Serialize the map using the iteration order. */
+  private static <K, V> void serializeMapTo(
+      CodedOutputStream out,
+      Map<K, V> m,
+      MapEntry<K, V> defaultEntry,
+      int fieldNumber)
+      throws IOException {
+    for (Map.Entry<K, V> entry : m.entrySet()) {
+      out.writeMessage(fieldNumber,
+          defaultEntry.newBuilderForType()
+              .setKey(entry.getKey())
+              .setValue(entry.getValue())
+              .build());
+    }
+  }
 }
diff --git a/java/core/src/main/java/com/google/protobuf/Internal.java b/java/core/src/main/java/com/google/protobuf/Internal.java
index 3b4a0412..c234559c 100644
--- a/java/core/src/main/java/com/google/protobuf/Internal.java
+++ b/java/core/src/main/java/com/google/protobuf/Internal.java
@@ -60,6 +60,16 @@ public final class Internal {
   static final Charset ISO_8859_1 = Charset.forName("ISO-8859-1");
 
   /**
+   * Throws an appropriate {@link NullPointerException} if the given objects is {@code null}.
+   */
+  static <T> T checkNotNull(T obj, String message) {
+    if (obj == null) {
+      throw new NullPointerException(message);
+    }
+    return obj;
+  }
+
+  /**
    * Helper called by generated code to construct default values for string
    * fields.
    * <p>
diff --git a/java/core/src/main/java/com/google/protobuf/InvalidProtocolBufferException.java b/java/core/src/main/java/com/google/protobuf/InvalidProtocolBufferException.java
index 85ce7b24..55e33d21 100644
--- a/java/core/src/main/java/com/google/protobuf/InvalidProtocolBufferException.java
+++ b/java/core/src/main/java/com/google/protobuf/InvalidProtocolBufferException.java
@@ -107,11 +107,23 @@ public class InvalidProtocolBufferException extends IOException {
       "Protocol message end-group tag did not match expected tag.");
   }
 
-  static InvalidProtocolBufferException invalidWireType() {
-    return new InvalidProtocolBufferException(
+  static InvalidWireTypeException invalidWireType() {
+    return new InvalidWireTypeException(
       "Protocol message tag had invalid wire type.");
   }
 
+  /**
+   * Exception indicating that and unexpected wire type was encountered for a field.
+   */
+  @ExperimentalApi
+  public static class InvalidWireTypeException extends InvalidProtocolBufferException {
+    private static final long serialVersionUID = 3283890091615336259L;
+
+    public InvalidWireTypeException(String description) {
+      super(description);
+    }
+  }
+
   static InvalidProtocolBufferException recursionLimitExceeded() {
     return new InvalidProtocolBufferException(
       "Protocol message had too many levels of nesting.  May be malicious.  " +
diff --git a/java/core/src/main/java/com/google/protobuf/MapEntry.java b/java/core/src/main/java/com/google/protobuf/MapEntry.java
index 117cd911..4f0351f4 100644
--- a/java/core/src/main/java/com/google/protobuf/MapEntry.java
+++ b/java/core/src/main/java/com/google/protobuf/MapEntry.java
@@ -334,6 +334,15 @@ public final class MapEntry<K, V> extends AbstractMessage {
       } else {
         if (field.getType() == FieldDescriptor.Type.ENUM) {
           value = ((EnumValueDescriptor) value).getNumber();
+        } else if (field.getType() == FieldDescriptor.Type.MESSAGE) {
+          if (value != null && !metadata.defaultValue.getClass().isInstance(value)) {
+            // The value is not the exact right message type.  However, if it
+            // is an alternative implementation of the same type -- e.g. a
+            // DynamicMessage -- we should accept it.  In this case we can make
+            // a copy of the message.
+            value =
+                ((Message) metadata.defaultValue).toBuilder().mergeFrom((Message) value).build();
+          }
         }
         setValue((V) value);
       }
diff --git a/java/core/src/main/java/com/google/protobuf/NioByteString.java b/java/core/src/main/java/com/google/protobuf/NioByteString.java
index 6163c7b1..76594809 100644
--- a/java/core/src/main/java/com/google/protobuf/NioByteString.java
+++ b/java/core/src/main/java/com/google/protobuf/NioByteString.java
@@ -30,6 +30,8 @@
 
 package com.google.protobuf;
 
+import static com.google.protobuf.Internal.checkNotNull;
+
 import java.io.IOException;
 import java.io.InputStream;
 import java.io.InvalidObjectException;
@@ -49,10 +51,9 @@ final class NioByteString extends ByteString.LeafByteString {
   private final ByteBuffer buffer;
 
   NioByteString(ByteBuffer buffer) {
-    if (buffer == null) {
-      throw new NullPointerException("buffer");
-    }
+    checkNotNull(buffer, "buffer");
 
+    // Use native byte order for fast fixed32/64 operations.
     this.buffer = buffer.slice().order(ByteOrder.nativeOrder());
   }
 
@@ -266,7 +267,7 @@ final class NioByteString extends ByteString.LeafByteString {
 
   @Override
   public CodedInputStream newCodedInput() {
-    return CodedInputStream.newInstance(buffer);
+    return CodedInputStream.newInstance(buffer, true);
   }
 
   /**
diff --git a/java/core/src/main/java/com/google/protobuf/RopeByteString.java b/java/core/src/main/java/com/google/protobuf/RopeByteString.java
index 3f3e9bd1..6fa555df 100644
--- a/java/core/src/main/java/com/google/protobuf/RopeByteString.java
+++ b/java/core/src/main/java/com/google/protobuf/RopeByteString.java
@@ -406,6 +406,7 @@ final class RopeByteString extends ByteString {
     right.writeTo(output);
   }
 
+
   @Override
   protected String toStringInternal(Charset charset) {
     return new String(toByteArray(), charset);
diff --git a/java/core/src/main/java/com/google/protobuf/TextFormat.java b/java/core/src/main/java/com/google/protobuf/TextFormat.java
index ff13675d..49708242 100644
--- a/java/core/src/main/java/com/google/protobuf/TextFormat.java
+++ b/java/core/src/main/java/com/google/protobuf/TextFormat.java
@@ -1576,14 +1576,22 @@ public final class TextFormat {
       // Support specifying repeated field values as a comma-separated list.
       // Ex."foo: [1, 2, 3]"
       if (field.isRepeated() && tokenizer.tryConsume("[")) {
-        while (true) {
-          consumeFieldValue(tokenizer, extensionRegistry, target, field, extension,
-              parseTreeBuilder, unknownFields);
-          if (tokenizer.tryConsume("]")) {
-            // End of list.
-            break;
+        if (!tokenizer.tryConsume("]")) {  // Allow "foo: []" to be treated as empty.
+          while (true) {
+            consumeFieldValue(
+                tokenizer,
+                extensionRegistry,
+                target,
+                field,
+                extension,
+                parseTreeBuilder,
+                unknownFields);
+            if (tokenizer.tryConsume("]")) {
+              // End of list.
+              break;
+            }
+            tokenizer.consume(",");
           }
-          tokenizer.consume(",");
         }
       } else {
         consumeFieldValue(tokenizer, extensionRegistry, target, field,
diff --git a/java/core/src/main/java/com/google/protobuf/TextFormatParseInfoTree.java b/java/core/src/main/java/com/google/protobuf/TextFormatParseInfoTree.java
index 5c43b2c3..0127ce92 100644
--- a/java/core/src/main/java/com/google/protobuf/TextFormatParseInfoTree.java
+++ b/java/core/src/main/java/com/google/protobuf/TextFormatParseInfoTree.java
@@ -45,7 +45,8 @@ import java.util.Map.Entry;
  *
  * <p>The locations of primary fields values are retrieved by {@code getLocation} or
  * {@code getLocations}.  The locations of sub message values are within nested
- * {@code TextFormatParseInfoTree}s and are retrieve by {@code getNestedTree} or {@code getNestedTrees}.
+ * {@code TextFormatParseInfoTree}s and are retrieve by {@code getNestedTree} or
+ * {@code getNestedTrees}.
  *
  * <p>The {@code TextFormatParseInfoTree} is created by a Builder.
  */
diff --git a/java/core/src/main/java/com/google/protobuf/UnsafeByteOperations.java b/java/core/src/main/java/com/google/protobuf/UnsafeByteOperations.java
index e72a6b4d..878c7758 100644
--- a/java/core/src/main/java/com/google/protobuf/UnsafeByteOperations.java
+++ b/java/core/src/main/java/com/google/protobuf/UnsafeByteOperations.java
@@ -67,16 +67,34 @@ public final class UnsafeByteOperations {
   /**
    * An unsafe operation that returns a {@link ByteString} that is backed by the provided buffer.
    *
+   * @param buffer the buffer to be wrapped
+   * @return a {@link ByteString} backed by the provided buffer
+   */
+  public static ByteString unsafeWrap(byte[] buffer) {
+    return ByteString.wrap(buffer);
+  }
+
+  /**
+   * An unsafe operation that returns a {@link ByteString} that is backed by a subregion of the
+   * provided buffer.
+   *
+   * @param buffer the buffer to be wrapped
+   * @param offset the offset of the wrapped region
+   * @param length the number of bytes of the wrapped region
+   * @return a {@link ByteString} backed by the provided buffer
+   */
+  public static ByteString unsafeWrap(byte[] buffer, int offset, int length) {
+    return ByteString.wrap(buffer, offset, length);
+  }
+
+  /**
+   * An unsafe operation that returns a {@link ByteString} that is backed by the provided buffer.
+   *
    * @param buffer the Java NIO buffer to be wrapped
    * @return a {@link ByteString} backed by the provided buffer
    */
   public static ByteString unsafeWrap(ByteBuffer buffer) {
-    if (buffer.hasArray()) {
-      final int offset = buffer.arrayOffset();
-      return ByteString.wrap(buffer.array(), offset + buffer.position(), buffer.remaining());
-    } else {
-      return new NioByteString(buffer);
-    }
+    return ByteString.wrap(buffer);
   }
 
   /**
@@ -98,4 +116,5 @@ public final class UnsafeByteOperations {
   public static void unsafeWriteTo(ByteString bytes, ByteOutput output) throws IOException {
     bytes.writeTo(output);
   }
+
 }
diff --git a/java/core/src/main/java/com/google/protobuf/UnsafeUtil.java b/java/core/src/main/java/com/google/protobuf/UnsafeUtil.java
index 6a4787d1..5f7bafd6 100644
--- a/java/core/src/main/java/com/google/protobuf/UnsafeUtil.java
+++ b/java/core/src/main/java/com/google/protobuf/UnsafeUtil.java
@@ -30,17 +30,14 @@
 
 package com.google.protobuf;
 
-import sun.misc.Unsafe;
-
 import java.lang.reflect.Field;
 import java.nio.Buffer;
 import java.nio.ByteBuffer;
 import java.security.AccessController;
 import java.security.PrivilegedExceptionAction;
+import sun.misc.Unsafe;
 
-/**
- * Utility class for working with unsafe operations.
- */
+/** Utility class for working with unsafe operations. */
 // TODO(nathanmittler): Add support for Android Memory/MemoryBlock
 final class UnsafeUtil {
   private static final sun.misc.Unsafe UNSAFE = getUnsafe();
@@ -50,8 +47,7 @@ final class UnsafeUtil {
   private static final long ARRAY_BASE_OFFSET = byteArrayBaseOffset();
   private static final long BUFFER_ADDRESS_OFFSET = fieldOffset(field(Buffer.class, "address"));
 
-  private UnsafeUtil() {
-  }
+  private UnsafeUtil() {}
 
   static boolean hasUnsafeArrayOperations() {
     return HAS_UNSAFE_ARRAY_OPERATIONS;
@@ -61,27 +57,83 @@ final class UnsafeUtil {
     return HAS_UNSAFE_BYTEBUFFER_OPERATIONS;
   }
 
+  static Object allocateInstance(Class<?> clazz) {
+    try {
+      return UNSAFE.allocateInstance(clazz);
+    } catch (InstantiationException e) {
+      throw new RuntimeException(e);
+    }
+  }
+
+  static long objectFieldOffset(Field field) {
+    return UNSAFE.objectFieldOffset(field);
+  }
+
   static long getArrayBaseOffset() {
     return ARRAY_BASE_OFFSET;
   }
 
-  static byte getByte(byte[] target, long offset) {
+  static byte getByte(Object target, long offset) {
     return UNSAFE.getByte(target, offset);
   }
 
-  static void putByte(byte[] target, long offset, byte value) {
+  static void putByte(Object target, long offset, byte value) {
     UNSAFE.putByte(target, offset, value);
   }
 
-  static void copyMemory(
-      byte[] src, long srcOffset, byte[] target, long targetOffset, long length) {
-    UNSAFE.copyMemory(src, srcOffset, target, targetOffset, length);
+  static int getInt(Object target, long offset) {
+    return UNSAFE.getInt(target, offset);
   }
 
-  static long getLong(byte[] target, long offset) {
+  static void putInt(Object target, long offset, int value) {
+    UNSAFE.putInt(target, offset, value);
+  }
+
+  static long getLong(Object target, long offset) {
     return UNSAFE.getLong(target, offset);
   }
 
+  static void putLong(Object target, long offset, long value) {
+    UNSAFE.putLong(target, offset, value);
+  }
+
+  static boolean getBoolean(Object target, long offset) {
+    return UNSAFE.getBoolean(target, offset);
+  }
+
+  static void putBoolean(Object target, long offset, boolean value) {
+    UNSAFE.putBoolean(target, offset, value);
+  }
+
+  static float getFloat(Object target, long offset) {
+    return UNSAFE.getFloat(target, offset);
+  }
+
+  static void putFloat(Object target, long offset, float value) {
+    UNSAFE.putFloat(target, offset, value);
+  }
+
+  static double getDouble(Object target, long offset) {
+    return UNSAFE.getDouble(target, offset);
+  }
+
+  static void putDouble(Object target, long offset, double value) {
+    UNSAFE.putDouble(target, offset, value);
+  }
+
+  static Object getObject(Object target, long offset) {
+    return UNSAFE.getObject(target, offset);
+  }
+
+  static void putObject(Object target, long offset, Object value) {
+    UNSAFE.putObject(target, offset, value);
+  }
+
+  static void copyMemory(
+      Object src, long srcOffset, Object target, long targetOffset, long length) {
+    UNSAFE.copyMemory(src, srcOffset, target, targetOffset, length);
+  }
+
   static byte getByte(long address) {
     return UNSAFE.getByte(address);
   }
@@ -90,14 +142,30 @@ final class UnsafeUtil {
     UNSAFE.putByte(address, value);
   }
 
+  static int getInt(long address) {
+    return UNSAFE.getInt(address);
+  }
+
+  static void putInt(long address, int value) {
+    UNSAFE.putInt(address, value);
+  }
+
   static long getLong(long address) {
     return UNSAFE.getLong(address);
   }
 
+  static void putLong(long address, long value) {
+    UNSAFE.putLong(address, value);
+  }
+
   static void copyMemory(long srcAddress, long targetAddress, long length) {
     UNSAFE.copyMemory(srcAddress, targetAddress, length);
   }
 
+  static void setMemory(long address, long numBytes, byte value) {
+    UNSAFE.setMemory(address, numBytes, value);
+  }
+
   /**
    * Gets the offset of the {@code address} field of the given direct {@link ByteBuffer}.
    */
@@ -136,18 +204,29 @@ final class UnsafeUtil {
     return unsafe;
   }
 
-  /**
-   * Indicates whether or not unsafe array operations are supported on this platform.
-   */
+  /** Indicates whether or not unsafe array operations are supported on this platform. */
   private static boolean supportsUnsafeArrayOperations() {
     boolean supported = false;
     if (UNSAFE != null) {
       try {
         Class<?> clazz = UNSAFE.getClass();
+        clazz.getMethod("objectFieldOffset", Field.class);
+        clazz.getMethod("allocateInstance", Class.class);
         clazz.getMethod("arrayBaseOffset", Class.class);
         clazz.getMethod("getByte", Object.class, long.class);
         clazz.getMethod("putByte", Object.class, long.class, byte.class);
+        clazz.getMethod("getBoolean", Object.class, long.class);
+        clazz.getMethod("putBoolean", Object.class, long.class, boolean.class);
+        clazz.getMethod("getInt", Object.class, long.class);
+        clazz.getMethod("putInt", Object.class, long.class, int.class);
         clazz.getMethod("getLong", Object.class, long.class);
+        clazz.getMethod("putLong", Object.class, long.class, long.class);
+        clazz.getMethod("getFloat", Object.class, long.class);
+        clazz.getMethod("putFloat", Object.class, long.class, float.class);
+        clazz.getMethod("getDouble", Object.class, long.class);
+        clazz.getMethod("putDouble", Object.class, long.class, double.class);
+        clazz.getMethod("getObject", Object.class, long.class);
+        clazz.getMethod("putObject", Object.class, long.class, Object.class);
         clazz.getMethod(
             "copyMemory", Object.class, long.class, Object.class, long.class, long.class);
         supported = true;
@@ -163,11 +242,17 @@ final class UnsafeUtil {
     if (UNSAFE != null) {
       try {
         Class<?> clazz = UNSAFE.getClass();
+        // Methods for getting direct buffer address.
         clazz.getMethod("objectFieldOffset", Field.class);
-        clazz.getMethod("getByte", long.class);
         clazz.getMethod("getLong", Object.class, long.class);
+
+        clazz.getMethod("getByte", long.class);
         clazz.getMethod("putByte", long.class, byte.class);
+        clazz.getMethod("getInt", long.class);
+        clazz.getMethod("putInt", long.class, int.class);
         clazz.getMethod("getLong", long.class);
+        clazz.getMethod("putLong", long.class, long.class);
+        clazz.getMethod("setMemory", long.class, long.class, byte.class);
         clazz.getMethod("copyMemory", long.class, long.class, long.class);
         supported = true;
       } catch (Throwable e) {
diff --git a/java/core/src/main/java/com/google/protobuf/WireFormat.java b/java/core/src/main/java/com/google/protobuf/WireFormat.java
index 6a58b081..0b0cdb7d 100644
--- a/java/core/src/main/java/com/google/protobuf/WireFormat.java
+++ b/java/core/src/main/java/com/google/protobuf/WireFormat.java
@@ -47,6 +47,10 @@ public final class WireFormat {
   // Do not allow instantiation.
   private WireFormat() {}
 
+  static final int FIXED_32_SIZE = 4;
+  static final int FIXED_64_SIZE = 8;
+  static final int MAX_VARINT_SIZE = 10;
+
   public static final int WIRETYPE_VARINT           = 0;
   public static final int WIRETYPE_FIXED64          = 1;
   public static final int WIRETYPE_LENGTH_DELIMITED = 2;