Skip to content
/ pbj Public

A performance optimized Google Protocol Buffers code generator, parser, and Gradle module.

License

Notifications You must be signed in to change notification settings

hashgraph/pbj

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 

pbj Protobuf Library

An alternative Google Protocol Buffers code generator, parser, and Gradle module. The project has two design goals:

  • Nice Java Record Objects - parse the proto files into nice clean Java record objects. With as clean as API as possible.
  • Performance Optimized - be as fast or faster than the standard Google ProtoC generated code
  • Minimal Garbage Generated - produce the minimum amount of garbage Java objects as possible

These design goals often compete with each other so this project tries to strike the right balance for use in the Hedera Node project. The hope is that balance might well be useful in many other projects. There is still plenty of work to achieve these goals and will probably always we improvements that can be made but this is what the project is striving for.

There is 3 gradle projects, the root project that builds two libraries:

  • PBJ Core pbj-core which has 2 sub projects
    • Compiler Gradle Plugin pbj-compiler which produces hedera.pbj-compiler.gradle.plugin-VERSION.jar
    • Runtime Library pbj-runtime which produces pnj-runtime-VERSION.jar
  • Integration Tests pbj-integration-tests which uses PBJ with test .proto files to generate code and run generated unit tests

Generated Model Objects

Each protobuf message is generated into a Java Record model object. With static instances for two codecs and a default value. There is an inner Builder class generated for fluid style building of the model object. There is also a copyBuilder() method that lets you easily create a derivative of this immutable record with some changes.

Protobuf Message Example

message Fraction {
    int64 numerator = 1;
    int64 denominator = 2;
}

Example generated model Java record

public record Fraction(
       long numerator,
       long denominator
){
    public static final Codec<Fraction> PROTOBUF = new com.hedera.hapi.node.base.codec.FractionProtoCodec();
    public static final JsonCodec<Fraction> JSON = new com.hedera.hapi.node.base.codec.FractionJsonCodec();
    public static final Fraction DEFAULT = newBuilder().build();

    public Builder copyBuilder() {...}
    public static Builder newBuilder() {...}
    
    public static final class Builder {...}
}

Generated Codecs

Codecs are generated for each model class, one for Protobuf binary encoding and one for Protobuf JSON encoding. There are static instances accessible via MyModelOject.PROTOBUF and MyModelOject.JSON.

Generated Unit Tests

For each generated model object there is a unit test generate that tests the protobuf and JSON codecs against the protoc generated code to make sure they are 100% byte for byte binary compatible.

pbj Data IO Library

com.hedera.pbj.runtime.io

As well as protobuf code generation, PBJ also provides a data IO library that provides an alternative to the default java.io & java.nio libraries. The PBJ Data IO library has some key design decisions, of which some are alternatives to the standard Java IO libraries:

  • The starting point was to provide an API that could unify Streams and Buffers. So a single parser could work on either a InputStream or a ByteBuffer. This was needed as we have both uses cases in the Hedera Node project and did not want to maintain two code paths.
  • End Of File(EOF) in streams is handled with a EOFException rather than by returning -1 from read bytes.
  • Provide an immutable alternative to byte[]. With the generated Record objects being immutable, it left any byte[] fields as dangerous mutable exceptions. This is provided by the Bytes class. It tries hard to avoid leakage of the underlying byte[] array while trying to balance that with performance.

The key data IO Interfaces

  • SequentialData Represents sequential data which may either be buffered or streamed. Provides all the position, limit and capacity management API but no reading or writing. It only allows reading or writing in a forwards direction.
  • RandomAccessData Provides random access reading and writing at offsets.
  • BufferedSequentialData Extends SequentialData and RandomAccessData to make position movable and allowing random access reading and writing.
  • ReadableSequentialData Extends SequentialData to provide reading API.
  • WritableSequentialData Extends SequentialData to provide writing API.

The key data IO Classes

  • ReadableStreamingData A ReadableSequentialData that reads from a InputStream.
  • WritableStreamingData A WritableSequentialData that writes to a OutputStream.
  • BufferedData A buffer that wraps a byte[] or ByteBuffer and implements BufferedSequentialData, ReadableSequentialData, WritableSequentialData and RandomAccessData.
  • RandomAccessSequenceAdapter A RandomAccessData that wraps a RandomAccessData and provides a ReadableSequentialData.

Build Libraries

Running gradle build in pbj-core directory will build the 2 libraries into the local maven repository for compiler and runtime.

Run Integration Tests

Running gradle build in pbj-integration-tests will check out the latest proto source files from the hedera-protobufs repository and generate code using pbj-compiler then run all the 100k+ generated unit tests which takes a few minutes. The hedera protobufs provide a good coverage of protobuf features and insure all the generated code is tested for hedera node use case.

These tests should be run before committing any new code for PBJ.

About

A performance optimized Google Protocol Buffers code generator, parser, and Gradle module.

Topics

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Packages

No packages published

Languages