A high performance json library for Golang

The code I write here is very similar to github.com/json-iterator/go, so you may ask why reinvent the wheel.

For sure that I benefit a lot from the jsoniter library, but i found some inconvenience for me to use it in some condition, for example:

• the iterator methods ReadString accepts null, there is no method which accepts exactly string. I have to do some extra check before calling.
• some behavior is not compatible with the standard library.
• I want a chained streamer

On the other hand, I also want to learn how the jsoniter works, so there is this repo.

Here are some of the differences:

• the iterator methods accept the exact type, for example ReadString accepts only string, not null
• the behavior is almost the same as the standard library (when an error returns, the behavior may differ from the standard library)
• the error of the iterator is returned instead of being saved inside iterator
• the decoder/encoder interface has additional options, like struct tag options

Some features of jsoniter are not implemented, and may be not implemented in the future neither. I choose only the ones I need to implement.

I tried implemented a version which is completely compatible with the standard library:

https://github.com/zerosnake0/jzon/tree/reflect

The benchmark shows that it's much faster than the standard library. However it is still much slower than the current version, which cannot be exactly the same as standard library (at least in my POV).

The major incompatibility is about the two following interfaces:

• json.Marshaler
• encoding.TextMarshaler

The method on pointer receiver may be called with an unaddressable value, for example:

type field struct {}

func (*field) MarshalJSON() ([]byte, error)

type st struct {
    F field
}

json.Marshal(st{}) // will not call field.MarshalJSON
jzon.Marshal(st{}) // will call field.MarshalJSON

So the user should be care when marshaling a value when method on pointer receiver is involved

You can check the tests for more detailed info about the difference

import "github.com/zerosnake0/jzon"

// Unmarshal
err := jzon.Unmarshal(b, &data)

// Marshal
b, err := jzon.Marshal(&data)

// Decoder
dec := jzon.NewDecoder(reader)
defer dec.Release()
err := dec.Decode(&data)

// Encoder
enc := jzon.NewEncoder(writer)
defer enc.Release()
err := enc.Encode(&data)

iter := jzon.NewIterator()
defer iter.Release()
iter.Reset(b)
jzon.ReadVal(&data)

var w io.Writer

streamer := jzon.NewStreamer()
defer streamer.Release()
streamer.Reset(w)
streamer.Value(&data)
streamer.Flush()

see decoder_test.go

type testIntDecoder struct{}

func (*testIntDecoder) Decode(ptr unsafe.Pointer, it *Iterator, opts *DecOpts) error {
    ...
}

dec := NewDecoderConfig(&DecoderOption{
    ValDecoders: map[reflect.Type]ValDecoder{
        reflect.TypeOf(int(0)): (*testIntDecoder)(nil),
    },
    CaseSensitive: true,
})

// standard library like
err := dec.Unmarshal(b, &data)

// iterator
iter := dec.NewIterator()
defer iter.Release()

see encoder_test.go

type testIntEncoder struct{}

func (*testIntEncoder) IsEmpty(ptr unsafe.Pointer) bool {
    ...
}

func (*testIntEncoder) Encode(ptr unsafe.Pointer, s *Streamer, opts *EncOpts) {
    ...
}

enc := NewEncoderConfig(&EncoderOption{
    ValEncoders: map[reflect.Type]ValEncoder{
        reflect.TypeOf(int(0)): (*testIntEncoder)(nil),
    },
})

// standard library like
b, err := enc.Marshal(&data)

// streamer
streamer := enc.NewStreamer()
defer streamer.Release()