com.google.protobuf.MessageSetSchema Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of protobuf-kotlin Show documentation
Show all versions of protobuf-kotlin Show documentation
Kotlin core Protocol Buffers library. Protocol Buffers are a way of encoding structured data in an
efficient yet extensible format.
// Protocol Buffers - Google's data interchange format
// Copyright 2008 Google Inc. All rights reserved.
//
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file or at
// https://developers.google.com/open-source/licenses/bsd
package com.google.protobuf;
import java.io.IOException;
import java.util.Iterator;
import java.util.Map.Entry;
/** Schema used for proto2 messages using message_set_wireformat. */
@CheckReturnValue
final class MessageSetSchema implements Schema {
private final MessageLite defaultInstance;
private final UnknownFieldSchema, ?> unknownFieldSchema;
private final boolean hasExtensions;
private final ExtensionSchema> extensionSchema;
private MessageSetSchema(
UnknownFieldSchema, ?> unknownFieldSchema,
ExtensionSchema> extensionSchema,
MessageLite defaultInstance) {
this.unknownFieldSchema = unknownFieldSchema;
this.hasExtensions = extensionSchema.hasExtensions(defaultInstance);
this.extensionSchema = extensionSchema;
this.defaultInstance = defaultInstance;
}
static MessageSetSchema newSchema(
UnknownFieldSchema, ?> unknownFieldSchema,
ExtensionSchema> extensionSchema,
MessageLite defaultInstance) {
return new MessageSetSchema(unknownFieldSchema, extensionSchema, defaultInstance);
}
@SuppressWarnings("unchecked")
@Override
public T newInstance() {
// TODO decide if we're keeping support for Full in schema classes and handle this
// better.
if (defaultInstance instanceof GeneratedMessageLite) {
return (T) ((GeneratedMessageLite, ?>) defaultInstance).newMutableInstance();
} else {
return (T) defaultInstance.newBuilderForType().buildPartial();
}
}
@Override
public boolean equals(T message, T other) {
Object messageUnknown = unknownFieldSchema.getFromMessage(message);
Object otherUnknown = unknownFieldSchema.getFromMessage(other);
if (!messageUnknown.equals(otherUnknown)) {
return false;
}
if (hasExtensions) {
FieldSet> messageExtensions = extensionSchema.getExtensions(message);
FieldSet> otherExtensions = extensionSchema.getExtensions(other);
return messageExtensions.equals(otherExtensions);
}
return true;
}
@Override
public int hashCode(T message) {
int hashCode = unknownFieldSchema.getFromMessage(message).hashCode();
if (hasExtensions) {
FieldSet> extensions = extensionSchema.getExtensions(message);
hashCode = (hashCode * 53) + extensions.hashCode();
}
return hashCode;
}
@Override
public void mergeFrom(T message, T other) {
SchemaUtil.mergeUnknownFields(unknownFieldSchema, message, other);
if (hasExtensions) {
SchemaUtil.mergeExtensions(extensionSchema, message, other);
}
}
@SuppressWarnings("unchecked")
@Override
public void writeTo(T message, Writer writer) throws IOException {
FieldSet> extensions = extensionSchema.getExtensions(message);
Iterator> iterator = extensions.iterator();
while (iterator.hasNext()) {
Entry, ?> extension = (Entry, ?>) iterator.next();
FieldSet.FieldDescriptorLite> fd = (FieldSet.FieldDescriptorLite>) extension.getKey();
if (fd.getLiteJavaType() != WireFormat.JavaType.MESSAGE || fd.isRepeated() || fd.isPacked()) {
throw new IllegalStateException("Found invalid MessageSet item.");
}
if (extension instanceof LazyField.LazyEntry) {
writer.writeMessageSetItem(
fd.getNumber(), ((LazyField.LazyEntry) extension).getField().toByteString());
} else {
writer.writeMessageSetItem(fd.getNumber(), extension.getValue());
}
}
writeUnknownFieldsHelper(unknownFieldSchema, message, writer);
}
/**
* A helper method for wildcard capture of {@code unknownFieldSchema}. See:
* https://docs.oracle.com/javase/tutorial/java/generics/capture.html
*/
private void writeUnknownFieldsHelper(
UnknownFieldSchema unknownFieldSchema, T message, Writer writer) throws IOException {
unknownFieldSchema.writeAsMessageSetTo(unknownFieldSchema.getFromMessage(message), writer);
}
@SuppressWarnings("ReferenceEquality")
@Override
public void mergeFrom(
T message, byte[] data, int position, int limit, ArrayDecoders.Registers registers)
throws IOException {
// TODO decide if we're keeping support for Full in schema classes and handle this
// better.
UnknownFieldSetLite unknownFields = ((GeneratedMessageLite) message).unknownFields;
if (unknownFields == UnknownFieldSetLite.getDefaultInstance()) {
unknownFields = UnknownFieldSetLite.newInstance();
((GeneratedMessageLite) message).unknownFields = unknownFields;
}
final FieldSet extensions =
((GeneratedMessageLite.ExtendableMessage, ?>) message).ensureExtensionsAreMutable();
GeneratedMessageLite.GeneratedExtension, ?> extension = null;
while (position < limit) {
position = ArrayDecoders.decodeVarint32(data, position, registers);
final int startTag = registers.int1;
if (startTag != WireFormat.MESSAGE_SET_ITEM_TAG) {
if (WireFormat.getTagWireType(startTag) == WireFormat.WIRETYPE_LENGTH_DELIMITED) {
extension =
(GeneratedMessageLite.GeneratedExtension, ?>) extensionSchema.findExtensionByNumber(
registers.extensionRegistry, defaultInstance,
WireFormat.getTagFieldNumber(startTag));
if (extension != null) {
position =
ArrayDecoders.decodeMessageField(
Protobuf.getInstance().schemaFor(
extension.getMessageDefaultInstance().getClass()),
data, position, limit, registers);
extensions.setField(extension.descriptor, registers.object1);
} else {
position =
ArrayDecoders.decodeUnknownField(
startTag, data, position, limit, unknownFields, registers);
}
} else {
position = ArrayDecoders.skipField(startTag, data, position, limit, registers);
}
continue;
}
int typeId = 0;
ByteString rawBytes = null;
while (position < limit) {
position = ArrayDecoders.decodeVarint32(data, position, registers);
final int tag = registers.int1;
final int number = WireFormat.getTagFieldNumber(tag);
final int wireType = WireFormat.getTagWireType(tag);
switch (number) {
case WireFormat.MESSAGE_SET_TYPE_ID:
if (wireType == WireFormat.WIRETYPE_VARINT) {
position = ArrayDecoders.decodeVarint32(data, position, registers);
typeId = registers.int1;
// TODO decide if we're keeping support for Full in schema classes and
// handle this better.
extension =
(GeneratedMessageLite.GeneratedExtension, ?>)
extensionSchema.findExtensionByNumber(
registers.extensionRegistry, defaultInstance, typeId);
continue;
}
break;
case WireFormat.MESSAGE_SET_MESSAGE:
if (extension != null) {
position = ArrayDecoders.decodeMessageField(
Protobuf.getInstance().schemaFor(
extension.getMessageDefaultInstance().getClass()),
data, position, limit, registers);
extensions.setField(extension.descriptor, registers.object1);
continue;
} else {
if (wireType == WireFormat.WIRETYPE_LENGTH_DELIMITED) {
position = ArrayDecoders.decodeBytes(data, position, registers);
rawBytes = (ByteString) registers.object1;
continue;
}
break;
}
default:
break;
}
if (tag == WireFormat.MESSAGE_SET_ITEM_END_TAG) {
break;
}
position = ArrayDecoders.skipField(tag, data, position, limit, registers);
}
if (rawBytes != null) {
unknownFields.storeField(
WireFormat.makeTag(typeId, WireFormat.WIRETYPE_LENGTH_DELIMITED), rawBytes);
}
}
if (position != limit) {
throw InvalidProtocolBufferException.parseFailure();
}
}
@Override
public void mergeFrom(T message, Reader reader, ExtensionRegistryLite extensionRegistry)
throws IOException {
mergeFromHelper(unknownFieldSchema, extensionSchema, message, reader, extensionRegistry);
}
/**
* A helper method for wildcard capture of {@code unknownFieldSchema}. See:
* https://docs.oracle.com/javase/tutorial/java/generics/capture.html
*/
private > void mergeFromHelper(
UnknownFieldSchema unknownFieldSchema,
ExtensionSchema extensionSchema,
T message,
Reader reader,
ExtensionRegistryLite extensionRegistry)
throws IOException {
UB unknownFields = unknownFieldSchema.getBuilderFromMessage(message);
FieldSet extensions = extensionSchema.getMutableExtensions(message);
try {
while (true) {
final int number = reader.getFieldNumber();
if (number == Reader.READ_DONE) {
return;
}
if (parseMessageSetItemOrUnknownField(
reader,
extensionRegistry,
extensionSchema,
extensions,
unknownFieldSchema,
unknownFields)) {
continue;
}
// Done reading.
return;
}
} finally {
unknownFieldSchema.setBuilderToMessage(message, unknownFields);
}
}
@Override
public void makeImmutable(T message) {
unknownFieldSchema.makeImmutable(message);
extensionSchema.makeImmutable(message);
}
private >
boolean parseMessageSetItemOrUnknownField(
Reader reader,
ExtensionRegistryLite extensionRegistry,
ExtensionSchema extensionSchema,
FieldSet extensions,
UnknownFieldSchema unknownFieldSchema,
UB unknownFields)
throws IOException {
int startTag = reader.getTag();
if (startTag != WireFormat.MESSAGE_SET_ITEM_TAG) {
if (WireFormat.getTagWireType(startTag) == WireFormat.WIRETYPE_LENGTH_DELIMITED) {
Object extension =
extensionSchema.findExtensionByNumber(
extensionRegistry, defaultInstance, WireFormat.getTagFieldNumber(startTag));
if (extension != null) {
extensionSchema.parseLengthPrefixedMessageSetItem(
reader, extension, extensionRegistry, extensions);
return true;
} else {
return unknownFieldSchema.mergeOneFieldFrom(unknownFields, reader, /* currentDepth= */ 0);
}
} else {
return reader.skipField();
}
}
// The wire format for MessageSet is:
// message MessageSet {
// repeated group Item = 1 {
// required uint32 typeId = 2;
// required bytes message = 3;
// }
// }
// "typeId" is the extension's field number. The extension can only be
// a message type, where "message" contains the encoded bytes of that
// message.
//
// In practice, we will probably never see a MessageSet item in which
// the message appears before the type ID, or where either field does not
// appear exactly once. However, in theory such cases are valid, so we
// should be prepared to accept them.
int typeId = 0;
ByteString rawBytes = null; // If we encounter "message" before "typeId"
Object extension = null;
// Read bytes from input, if we get it's type first then parse it eagerly,
// otherwise we store the raw bytes in a local variable.
loop:
while (true) {
final int number = reader.getFieldNumber();
if (number == Reader.READ_DONE) {
break;
}
final int tag = reader.getTag();
if (tag == WireFormat.MESSAGE_SET_TYPE_ID_TAG) {
typeId = reader.readUInt32();
extension =
extensionSchema.findExtensionByNumber(extensionRegistry, defaultInstance, typeId);
continue;
} else if (tag == WireFormat.MESSAGE_SET_MESSAGE_TAG) {
if (extension != null) {
extensionSchema.parseLengthPrefixedMessageSetItem(
reader, extension, extensionRegistry, extensions);
continue;
}
// We haven't seen a type ID yet or we want parse message lazily.
rawBytes = reader.readBytes();
continue;
} else {
if (!reader.skipField()) {
break loop; // End of group
}
}
}
if (reader.getTag() != WireFormat.MESSAGE_SET_ITEM_END_TAG) {
throw InvalidProtocolBufferException.invalidEndTag();
}
// If there are any rawBytes left, it means the message content appears before the type ID.
if (rawBytes != null) {
if (extension != null) { // We known the type
// TODO: Instead of reading into a temporary ByteString, maybe there is a way
// to read directly from Reader to the submessage?
extensionSchema.parseMessageSetItem(rawBytes, extension, extensionRegistry, extensions);
} else {
unknownFieldSchema.addLengthDelimited(unknownFields, typeId, rawBytes);
}
}
return true;
}
@Override
public final boolean isInitialized(T message) {
FieldSet> extensions = extensionSchema.getExtensions(message);
return extensions.isInitialized();
}
@Override
public int getSerializedSize(T message) {
int size = 0;
size += getUnknownFieldsSerializedSize(unknownFieldSchema, message);
if (hasExtensions) {
size += extensionSchema.getExtensions(message).getMessageSetSerializedSize();
}
return size;
}
private int getUnknownFieldsSerializedSize(
UnknownFieldSchema schema, T message) {
UT unknowns = schema.getFromMessage(message);
return schema.getSerializedSizeAsMessageSet(unknowns);
}
}