tendermint/binary

tendermint/binary

The binary submodule encodes primary types and structs into bytes.

Primary types

uint*, int*, string, time, byteslice and byteslice-slice types can be encoded and decoded with the following methods:

The following writes o uint64 to w io.Writer, and increments n and/or sets err

WriteUint64(o uint64, w io.Writer, n *int64, err *error)

// Typical usage:
buf, n, err := new(bytes.Buffer), new(int64), new(error)
WriteUint64(uint64(x), buf, n, err)
if *err != nil {
    panic(err)
}

The following reads a uint64 from r io.Reader, and increments n and/or sets err

var o = ReadUint64(r io.Reader, n *int64, err *error)

Similar methods for uint32, uint16, uint8, int64, int32, int16, int8 exist. Protobuf variable length encoding is done with uint and int types:

WriteUvarint(o uint, w io.Writer, n *int64, err *error)
var o = ReadUvarint(r io.Reader, n *int64, err *error)

Byteslices can be written with:

WriteByteSlice(bz []byte, w io.Writer, n *int64, err *error)

Byteslices (and all slices such as byteslice-slices) are prepended with uvarint encoded length, so ReadByteSlice() knows how many bytes to read.

Note that there is no type information encoded -- the caller is assumed to know what types to decode.

Struct Types

Struct types can be automatically encoded with reflection. Unlike json-encoding, no field name or type information is encoded. Field values are simply encoded in order.

type Foo struct {
    MyString        string
    MyUint32        uint32
    myPrivateBytes  []byte
}

foo := Foo{"my string", math.MaxUint32, []byte("my private bytes")}

buf, n, err := new(bytes.Buffer), new(int64), new(error)
WriteBinary(foo, buf, n, err)

// fmt.Printf("%X", buf.Bytes()) gives:
// 096D7920737472696E67FFFFFFFF
// 09:                           uvarint encoded length of string "my string"
//   6D7920737472696E67:         bytes of string "my string"
//                     FFFFFFFF: bytes for MaxUint32 
// Note that the unexported "myPrivateBytes" isn't encoded.

foo2 := ReadBinary(Foo{}, buf, n, err).(Foo)

// Or, to decode onto a pointer:
foo2 := ReadBinary(&Foo{}, buf, n, err).(*Foo)

WriteBinary and ReadBinary can encode/decode structs recursively. However, interface field values are a bit more complicated.

type Greeter interface {
	Greet() string
}

type Dog struct{}
func (d Dog) Greet() string { return "Woof!" }

type Cat struct{}
func (c Cat) Greet() string { return "Meow!" }

type Foo struct {
	Greeter
}

foo := Foo{Dog{}}

buf, n, err := new(bytes.Buffer), new(int64), new(error)
WriteBinary(foo, buf, n, err)

// This errors with "Unknown field type interface"
foo2 := ReadBinary(Foo{}, buf, n, err)

In the above example, ReadBinary() fails because the Greeter field for Foo{} is ambiguous -- it could be either a Dog{} or a Cat{}, like a union structure. In this case, the user must define a custom encoder/decoder as follows:

const (
	GreeterTypeDog = byte(0x01)
	GreeterTypeCat = byte(0x02)
)

func GreeterEncoder(o interface{}, w io.Writer, n *int64, err *error) {
	switch o.(type) {
	case Dog:
		WriteByte(GreeterTypeDog, w, n, err)
		WriteBinary(o, w, n, err)
	case Cat:
		WriteByte(GreeterTypeCat, w, n, err)
		WriteBinary(o, w, n, err)
	default:
		*err = errors.New(fmt.Sprintf("Unknown greeter type %v", o))
	}
}

func GreeterDecoder(r *bytes.Reader, n *int64, err *error) interface{} {
	// We must peek the type byte because ReadBinary() expects it.
	switch t := PeekByte(r, n, err); t {
	case GreeterTypeDog:
		return ReadBinary(Dog{}, r, n, err)
	case GreeterTypeCat:
		return ReadBinary(Cat{}, r, n, err)
	default:
		*err = errors.New(fmt.Sprintf("Unknown greeter type byte %X", t))
		return nil
	}
}

// This tells the reflection system to use the custom encoder/decoder functions
// for encoding/decoding interface struct-field types.
var _ = RegisterType(&TypeInfo{
	Type: reflect.TypeOf((*Greeter)(nil)).Elem(),
	Encoder: GreeterEncoder,
	Decoder: GreeterDecoder,
})

Sometimes you want to always prefix a globally unique type byte while encoding, whether or not the declared type is an interface or concrete type. In this case, you can declare a "TypeByte() byte" function on the struct (as a value receiver, not a pointer receiver!), and you can skip the declaration of a custom decoder.

type Dog struct{}
func (d Dog) TypeByte() byte { return GreeterTypeDog }
func (d Dog) Greet() string { return "Woof!" }

type Cat struct{}
func (c Cat) TypeByte() byte { return GreeterTypeCat }
func (c Cat) Greet() string { return "Meow!" }

var _ = RegisterType(&TypeInfo{
	Type: reflect.TypeOf((*Greeter)(nil)).Elem(),
	Decoder: GreeterDecoder,
})