This page describes exactly what Java code the protocol buffer compiler generates for any given protocol definition. You should read the language guide before reading this document.
The protocol buffer compiler produces Java output when invoked with the --java_out= command-line flag. The parameter to the --java_out= option is the directory where you want the compiler to write your Java output. The compiler creates a single .java for each .proto file input. This file contains a single outer class definition containing several nested classes and static fields based on the declarations in the .proto file.
The outer class's name is chosen as follows: If the .proto file contains a line like the following:
option java_outer_classname = "Foo";
Then the outer class name will be Foo. Otherwise, the outer class name is determined by converting the .proto file base name to camel case. For example, foo_bar.proto will become FooBar.
The Java package name is chosen as described under Packages, below.
The output file is chosen by concatenating the parameter to --java_out=, the package name (with .s replaced with /s), and the .java file name.
So, for example, let's say you invoke the compiler as follows:
protoc --proto_path=src --java_out=build/gen src/foo.proto
If foo.proto's java package is com.example and its outer classname is FooProtos, then the protocol buffer compiler will generate the file build/gen/com/example/FooProtos.java. The protocol buffer compiler will automatically create the build/gen/com and build/gen/com/example directories if needed. However, it will not create build/gen or build; they must already exist. You can specify multiple .proto files in a single invocation; all output files will be generated at once.
The generated class is placed in a Java package based on the java_package option. If the option is omitted, the package declaration is used instead.
For example, if the .proto file contains:
package foo.bar;
Then the resulting Java class will be placed in Java package foo.bar. However, if the .proto file also contains a java_package option, like so:
package foo.bar; option java_package = "com.example.foo.bar";
Then the class is placed in the com.example.foo.bar package instead. The java_package option is provided because normal .proto package declarations are not expected to start with a backwards domain name.
Given a simple message declaration:
|
message Foo {}
The protocol buffer compiler generates a class called Foo, which implements the Message interface. The class is declared final; no further subclassing is allowed. Foo extends GeneratedMessage, but this should be considered an implementation detail. By default, Foo overrides many methods of GeneratedMessage with specialized versions for maximum speed. However, if the .proto file contains the line:
option optimize_for = CODE_SIZE;
then Foo will override only the minimum set of methods necessary to function and rely on GeneratedMessage's reflection-based implementations of the rest. This significantly reduces the size of the generated code, but also reduces performance. Alternatively, if the .proto file contains:
option optimize_for = LITE_RUNTIME;
then Foo will include fast implementations of all methods, but will implement the MessageLite interface, which only contains a subset of the methods of Message. In particular, it does not support descriptors or reflection. However, in this mode, the generated code only needs to link against libprotobuf-lite.jar instead of libprotobuf.jar. The "lite" library is much smaller than the full library, and is more appropriate for resource-constrained systems such as mobile phones.
In addition to the methods defined by the Message interface, the Foo class defines the following static methods:
static Foo getDefaultInstance(): Returns a singleton instance of Foo, which is identical to what you'd get if you called Foo.newBuilder().build() (so all singular fields are unset and all repeated fields are empty). Note that the default instance of a message can be used as a factory by calling its newBuilderForType() method.
static Descriptor getDescriptor(): Returns the type's descriptor. This contains information about the type, including what fields it has and what their types are. This can be used with the reflection methods of the Message, such as getField().
static Foo parseFrom(...): Parses a message of type Foo from the given source and returns it. There is one parseFrom method corresponding to each variant of mergeFrom() in the Message.Builder interface. Note that parseFrom() never throws UninitializedMessageException; it throws InvalidProtocolBufferException if the parsed message is missing required fields. This makes it subtly different from calling Foo.newBuilder().mergeFrom(...).build().
Foo.Builder newBuilder(): Creates a new builder (described below).
Foo.Builder newBuilder(Foo prototype): Creates a new builder with all fields initialized to the same values that they have in prototype. Since embedded message and string objects are immutable, they are shared between the original and the copy.
Message objects – such as instances of the Foo class described above – are immutable, just like a Java String. To construct a message object, you need to use a builder. Each message class has its own builder class – so in our Foo example, the protocol buffer compiler generates a nested class Foo.Builder which can be used to build a Foo. Foo.Builder implements the Message.Builder interface. It extends the GeneratedMessage.Builder class, but, again, this should be considered an implementation detail. Like Foo, Foo.Builder may rely on generic method implementations in GeneratedMessage.Builder or, when the optimize_for option is used, generated custom code that is much faster.
Foo.Builder does not define any static methods. Its interface is exactly as defined by the Message.Builder interface, with the exception that return types are more specific: methods of Foo.Builder that modify the builder return type Foo.Builder, and build() returns type Foo.
Methods that modify the contents of a builder – including field setters – always return a reference to the builder (i.e. they "return this;"). This allows multiple method calls to be chained together in one line. For example:
builder.mergeFrom(obj).setFoo(1).setBar("abc").clearBaz();
A message can be declared inside another message. For example:
message Foo {
message Bar {
}
}
In this case, the compiler simply generates Bar as an inner class nested inside Foo.
In addition to the methods described in the previous section, the protocol buffer compiler generates a set of accessor methods for each field defined within the message in the .proto file. The methods that read the field value are defined both in the message class and its corresponding builder; the methods that modify the value are only defined in the builder only.
Note that method names always use camel-case naming, even if the field name in the .proto file uses lower-case with underscores (as it should). The case-conversion works as follows:
Thus, the field foo_bar_baz becomes fooBarBaz. If prefixed with get, it would be getFooBarBaz.
As well as accessor methods, the compiler generates an integer constant for each field containing its field number. The constant name is the field name converted to upper-case followed by _FIELD_NUMBER. For example, given the field optional int32 foo_bar = 5;, the compiler will generate the constant public static final int FOO_BAR_FIELD_NUMBER = 5;.
For either of these field definitions:
optional int32 foo = 1; required int32 foo = 1;
The compiler will generate the following accessor methods in both the message class and its builder:
bool hasFoo(): Returns true if the field is set.
int getFoo(): Returns the current value of the field. If the field is not set, returns the default value.
The compiler will generate the following methods only in the message's builder:
Builder setFoo(int value): Sets the value of the field. After calling this, hasFoo() will return true and getFoo() will return value.
Builder clearFoo(): Clears the value of the field. After calling this, hasFoo() will return false and getFoo() will return the default value.
For other simple field types, the corresponding Java type is chosen according to the scalar value types table. For message and enum types, the value type is replaced with the message or enum class.
For message types, setFoo() also accepts an instance of the message's builder type as the parameter. This is just a shortcut which is equivalent to calling .build() on the builder and passing the result to the method.
For this field definition:
repeated int32 foo = 1;
The compiler will generate the following accessor methods in both the message class and its builder:
int getFooCount(): Returns the number of elements currently in the field.
int getFoo(int index): Returns the element at the given zero-based index.
List<Integer> getFooList(): Returns the entire field as an immutable list.
The compiler will generate the following methods only in the message's builder:
Builder setFoo(int index, int value): Sets the value of the element at the given zero-based index.
Builder addFoo(int value): Appends a new element to the field with the given value.
Builder addAllFoo(List<Integer> value): Appends all elements in the given list to the field.
Builder clearFoo(): Removes all elements from the field. After calling this, fooSize() will return zero.
For other simple field types, the corresponding Java type is chosen according to the scalar value types table. For message and enum types, the type is the message or enum class.
For message types, setFoo() and addFoo() also accept an instance of the message's builder type as the parameter. This is just a shortcut which is equivalent to calling .build() on the builder and passing the result to the method.
Given an enum definition like:
enum Foo {
VALUE_A = 1;
VALUE_B = 5;
VALUE_C = 1234;
}
The protocol buffer compiler will generate a Java enum type called Foo with the same set of values. Additionally, the values of this enum type have the following special methods:
int getNumber(): Returns the object's numeric value as defined in the .proto file.
EnumValueDescriptor getValueDescriptor(): Returns the value's descriptor, which contains information about the value's name, number, and type.
EnumDescriptor getDescriptorForType(): Returns the enum type's descriptor, which contains e.g. information about each defined value.
Additionally, the Foo enum type contains the following static methods:
static Foo valueOf(int value): Returns the enum object corresponding to the given numeric value.
static Foo valueOf(EnumValueDescriptor descriptor): Returns the enum object corresponding to the given value descriptor. May be faster than valueOf(int).
EnumDescriptor getDescriptor(): Returns the enum type's descriptor, which contains e.g. information about each defined value. (This differs from getDescriptorForType() only in that it is a static method.)
Note that the .proto language allows multiple enum symbols to have the same numeric value. Symbols with the same numeric value are synonyms. For example:
enum Foo {
BAR = 1;
BAZ = 1;
}
In this case, BAZ is a synonym for BAR. In Java, BAZ will be defined as a static final field like so:
static final Foo BAZ = BAR;
Thus, BAR and BAZ compare equal, and BAZ should never appear in switch statements. The compiler always chooses the first symbol defined with a given numeric value to be the "canonical" version of that symbol; all subsequent symbols with the same number are just aliases.
An enum can be defined nested within a message type. The compiler generates the Java enum definition nested within that message type's class.
Given a message with an extension range:
message Foo {
extensions 100 to 199;
}
The protocol buffer compiler will make Foo extend GeneratedMessage.ExtendableMessage instead of the usual GeneratedMessage. Similarly, Foo's builder will extend GeneratedMessage.ExtendableBuilder. You should never refer to these base types by name (GeneratedMessage is considered an implementation detail). However, these superclasses define a number of additional methods that you can use to manipulate extensions.
In particular Foo and Foo.Builder will inherit the methods hasExtension(), getExtension(), and getExtensionCount(). Additionally, Foo.Builder will inherit methods setExtension() and clearExtension(). Each of these methods takes, as its first parameter, an extension identifier (described below), which identifies an extension field. The remaining parameters and the return value are exactly the same as those for the corresponding accessor methods that would be generated for a normal (non-extension) field of the same type as the extension identifier.
Given an extension definition:
extend Foo {
optional int32 bar = 123;
}
The protocol buffer compiler generates an "extension identifier" called bar, which you can use with Foo's extension accessors to access this extension, like so:
Foo foo =
Foo.newBuilder()
.setExtension(bar, 1)
.build();
assert foo.hasExtension(bar);
assert foo.getExtension(bar) == 1;
(The exact implementation of extension identifiers is complicated and involves magical use of generics – however, you don't need to worry about how extension identifiers work to use them.)
Note that bar would be declared as a static field of the outer class for the .proto file, as described above; we have omitted the outer class name in the example.
Extensions can be declared nested inside of another type. For example, a common pattern is to do something like this:
message Baz {
extend Foo {
optional Baz foo_ext = 124;
}
}
In this case, the extension identifier foo_ext is declared nested inside Baz. It can be used as follows:
Baz baz = createMyBaz();
Foo foo =
Foo.newBuilder()
.setExtension(Baz.fooExt, baz)
.build();
Given a service definition:
service Foo {
rpc Bar(FooRequest) returns(FooResponse);
}
The protocol buffer compiler will generate an abstract class Foo to represent this service. Foo will have an abstract method for each method defined in the service definition. In this case, the method Bar is defined as:
abstract void bar(RpcController controller, FooRequest request, RpcCallback<FooResponse> done);
The parameters are equivalent to the parameters of Service.CallMethod(), except that the method argument is implied and request and done specify their exact type.
Foo subclasses the Service interface. The protocol buffer compiler automatically generates implementations of the methods of Service as follows:
getDescriptorForType: Returns the service's ServiceDescriptor.
callMethod: Determines which method is being called based on the provided method descriptor and calls it directly, down-casting the request message and callback to the correct types.
getRequestPrototype and getRequestPrototype: Returns the default instance of the request or response of the correct type for the given method.
The following static method is also generated:
static ServiceDescriptor getDescriptor(): Returns the type's descriptor, which contains information about what methods this service has and what their input and output types are.
Foo will also contain a nested interface Foo.Interface. This is a pure interface that again contains methods corresponding to each method in your service definition. However, this interface does not extend the Service interface. This is a problem because RPC server implementations are usually written to use abstract Service objects, not your particular service. To solve this problem, if you have an object impl implementing Foo.Interface, you can call Foo.newReflectiveService(impl) to construct an instance of Foo that simply delegates to impl, and implements Service.
To recap, when implementing your own service, you have two options:
Foo and implement its methods as appropriate, then hand instances of your subclass directly to the RPC server implementation. This is usually easiest, but some consider it less "pure".
Foo.Interface and use Foo.newReflectiveService(Foo.Interface) to construct a Service wrapping it, then pass the wrapper to your RPC implementation.
The protocol buffer compiler also generates a "stub" implementation of every service interface, which is used by clients wishing to send requests to servers implementing the service. For the Foo service (above), the stub implementation Foo.Stub will be defined as a nested class.
Foo.Stub is a subclass of Foo which also implements the following methods:
Foo.Stub(RpcChannel channel): Constructs a new stub which sends requests on the given channel.
RpcChannel getChannel(): Returns this stub's channel, as passed to the constructor.
The stub additionally implements each of the service's methods as a wrapper around the channel. Calling one of the methods simply calls channel.callMethod().
The Protocol Buffer library does not include an RPC implementation. However, it includes all of the tools you need to hook up a generated service class to any arbitrary RPC implementation of your choice. You need only provide implementations of RpcChannel and RpcController.
The RPC classes described above all have non-blocking semantics: when you call a method, you provide a callback object which will be invoked once the method completes. Often it is easier (though possibly less scalable) to write code using blocking semantics, where the method simply doesn't return until it is done. To accomodate this, the protocol buffer compiler also generates blocking versions of your service class. Foo.BlockingInterface is equivalent to Foo.Interface except that each method simply returns the result rather than call a callback. So, for example, bar is defined as:
abstract FooResponse bar(RpcController controller, FooRequest request) throws ServiceException;
Analogous to non-blocking services, Foo.newReflectiveBlockingService(Foo.BlockingInterface) returns a BlockingService wrapping some Foo.BlockingInterface. Finally, Foo.BlockingStub returns a stub implementation of Foo.BlockingInterface that sends requests to a particular BlockingRpcChannel.