GameJSON 0.0.1

Game JSON

A fast no dependency JSON library for C#, particularly aimed at games. This was built to satisfy the need that most JSON libraries cannot be sped up enough to (de)serialize files in a sufficient amount of time without incurring a framerate drop. Currently only deserializes strings and serializes to strings.

Structure

Comes in 2 parts

• GameJSON.ReflectionParsing - Reflection Library for serializing/deserializing objects to/from JSON. Contains JSON class.
• GameJSON.ManualParsing - A manual parsing library that ReflectionParsing is built on top of. If ReflectionParsing doesn't (de)serialize your JSON fast enough, using ManualParsing to manually (de)serialize will give you a massive performance boost. Contains JSONReader and JSONWriter class.

Reflection Parsing Goals

• All fields on an object will be serialized and deserialized
• Users can opt into serializing properties with SerializeProperty attribute
• Auto properties are serialized automatically (i.e. public int MyAutoProperty { get; set; })

Manual Parsing Goals

• Have as few additional allocations as possible
• Give the user as much power over serializing the JSON as they want to achieve great performance

Performance

These are the performance test results. As a summary, this behaves significantly faster than Newtonsoft in most cases except when loading using reflection with a hot cache. This is most likely because GameJSON does absolutely no caching. Most importantly though, GameJSON manually deserializes JSON at 5ms cold (i.e. first run with data type) and 2ms hot (i.e. second run with data type), whereas Newtonsoft manually deserializes in 95ms cold and 11ms hot. This means that files can be loaded with GameJSON in games without incurring a huge frame spike.

On Android, the differences between GameJSON and Newtonsoft/JSON.Net are vast, with GameJSON being faster in every category.

Tests

I created an object TestPositions with a string and a Vector3, and had each library serialize and deserialize these objects.

• Windows - i5-9600K with Mono backend
• Android - Galaxy S10 Lite with IL2CPP backend

1000 Test Positions Cold Platform

This captures the first run of serializing/deserializing 1000 TestPositions on the respective Platform

1000 Test Positions Hot Platform

This captures the second run of serializing/deserializing of 1000 TestPosition on the respective Platform

Results