Skip to content

orlandos-nl/IkigaJSON

Repository files navigation

IkigaJSON is a really fast JSON parser. IkigaJSON is competitive to the modern Foundation JSON in benchmarks, and outperforms older versions of Foundation JSON by a large margin.

Aside from being more performant, IkigaJSON has a much lower and more stable memory footprint, too! By design, IkigaJSON scales better than Foundation on larger JSON payloads. All while providing an easy to use API with Codable support.

Join our Discord for any questions and friendly banter.

Please note that Swift libraries need to be built in RELEASE compilation mode in order to judge performnce. If you're testing performance on a DEBUG build, you'll find severe mis-optimisations by the compiler that cannot reasonably be fixed in libraries. When building Swift code on DEBUG compilation, it can be 10-20x slower than equivalent code on RELEASE.

Server-Side Swift

The above performance statement was tested on Foundation for macOS and iOS. If you're using Swift on Linux with Swift 5.5, your performance is slightly better if you use the new Foundation for Linux. Swift 5.5 does not improve Foundation's JSON performance on macOS or iOS. IkigaJSON performs increasingly better than Linux’ Foundation JSON the bigger your JSON payload gets.

Adding the dependency

The 1.x versions are reliant on SwiftNIO 1.x, and for SwiftNIO 2.x support use the 2.x versions of IkigaJSON.

// SwiftNIO 1.x
.package(url: "https://github.com/orlandos-nl/IkigaJSON.git", from: "1.0.0"),
// Or, for SwiftNIO 2
.package(url: "https://github.com/orlandos-nl/IkigaJSON.git", from: "2.0.0"),

Usage

import IkigaJSON

struct User: Codable {
    let id: Int
    let name: String
}

let data: Data = ...
var decoder = IkigaJSONDecoder()
let user = try decoder.decode(User.self, from: data)

In Hummingbird 2

Conform Ikiga to Hummingbird's protocols like so:

extension IkigaJSONEncoder: HBResponseEncoder {
    public func encode(_ value: some Encodable, from request: HBRequest, context: some HBBaseRequestContext) throws -> HBResponse {
        // Capacity should roughly cover the amount of data you regularly expect to encode
        // However, the buffer will grow if needed
        var buffer = context.allocator.buffer(capacity: 2048)
        try self.encodeAndWrite(value, into: &buffer)
        return HBResponse(
            status: .ok, 
            headers: [
                .contentType: "application/json; charset=utf-8",
            ], 
            body: .init(byteBuffer: buffer)
        )
    }
}

extension IkigaJSONDecoder: HBRequestDecoder {
    public func decode<T>(_ type: T.Type, from request: HBRequest, context: some HBBaseRequestContext) async throws -> T where T : Decodable {
        let data = try await request.body.collate(maxSize: context.maxUploadSize)
        return try self.decode(T.self, from: data)
    }
}

In Vapor 4

Conform Ikiga to Vapor 4's protocols like so:

extension IkigaJSONEncoder: ContentEncoder {
    public func encode<E: Encodable>(
        _ encodable: E,
        to body: inout ByteBuffer,
        headers: inout HTTPHeaders
    ) throws {
        headers.contentType = .json
        try self.encodeAndWrite(encodable, into: &body)
    }

    public func encode<E>(_ encodable: E, to body: inout ByteBuffer, headers: inout HTTPHeaders, userInfo: [CodingUserInfoKey : Sendable]) throws where E : Encodable {
        var encoder = self
        encoder.userInfo = userInfo
        headers.contentType = .json
        try encoder.encodeAndWrite(encodable, into: &body)
    }

    public func encode<E>(_ encodable: E, to body: inout ByteBuffer, headers: inout HTTPHeaders, userInfo: [CodingUserInfoKey : Any]) throws where E : Encodable {
        var encoder = self
        encoder.userInfo = userInfo
        headers.contentType = .json
        try encoder.encodeAndWrite(encodable, into: &body)
    }
}

extension IkigaJSONDecoder: ContentDecoder {
    public func decode<D: Decodable>(
        _ decodable: D.Type,
        from body: ByteBuffer,
        headers: HTTPHeaders
    ) throws -> D {
        return try self.decode(D.self, from: body)
    }
    
    public func decode<D>(_ decodable: D.Type, from body: ByteBuffer, headers: HTTPHeaders, userInfo: [CodingUserInfoKey : Sendable]) throws -> D where D : Decodable {
        let decoder = IkigaJSONDecoder(settings: settings)
        decoder.settings.userInfo = userInfo
        return try decoder.decode(D.self, from: body)
    }

    public func decode<D>(_ decodable: D.Type, from body: ByteBuffer, headers: HTTPHeaders, userInfo: [CodingUserInfoKey : Any]) throws -> D where D : Decodable {
        let decoder = IkigaJSONDecoder(settings: settings)
        decoder.settings.userInfo = userInfo
        return try decoder.decode(D.self, from: body)
    }
}

Register the encoder/decoder to Vapor like so:

var decoder = IkigaJSONDecoder()
decoder.settings.dateDecodingStrategy = .iso8601
ContentConfiguration.global.use(decoder: decoder, for: .json)

var encoder = IkigaJSONEncoder()
encoder.settings.dateEncodingStrategy = .iso8601
ContentConfiguration.global.use(encoder: encoder, for: .json)

Raw JSON

IkigaJSON supports raw JSON types (JSONObject and JSONArray) like many other libraries do, alongside the codable API described above. The critical difference is that IkigaJSON edits the JSON inline, so there's no additional conversion overhead from Swift type to JSON.

var user = JSONObject()
user["username"] = "Joannis"
user["roles"] = ["admin", "moderator", "user"] as JSONArray
user["programmer"] = true

print(user.string)

print(user["username"].string)
// OR
print(user["username"] as? String)

SwiftNIO support

The encoders and decoders support SwiftNIO.

var user = try JSONObject(buffer: byteBuffer)
print(user["username"].string)

We also have added the ability to use the IkigaJSONEncoder and IkigaJSONDecoder with JSON.

let user = try decoder.decode([User].self, from: byteBuffer)
var buffer: ByteBuffer = ...

try encoder.encodeAndWrite(user, into: &buffer)

The above method can be used to stream multiple entities from a source like a database over the socket asynchronously. This can greatly reduce memory usage.

Performance

By design you can build on top of any data storage as long as it exposes a pointer API. This way, IkigaJSON doesn't (need to) copy any data from your buffer keeping it lightweight. The entire parser can function with only 1 memory allocation and allows for reusing the Decoder to reuse the memory allocation.

This allocation (called the JSONDescription) acts as a filter over the original dataset, indicating to IkigaJSON where keys, values and objects start/end. Therefore IkigaJSON can do really fast inline mutations, and provide objects such as JSONObject/JSONDescription that are extremely performant at reading individual values. This also allows IkigaJSON to decode from its own helper types such as JSONObject and JSONArray, since it doesn't need to regenerate a JSONDescription and has the original buffer at hand.

Support

  • All decoding strategies that Foundation supports
  • Unicode
  • Codable
  • Escaping
  • Performance 🚀
  • Date/Data encoding strategies
  • Raw JSON APIs (non-codable)
  • Codable decoding from JSONObject and JSONArray
  • \u escaped unicode characters

Media

Architecture