Moontrade heavily utilizes WebAssembly (TinyGo currently) for running custom code/algorithms to reduce pricing streams into custom data models, execute trades, re-balance portfolios, import/export transformations, etc.
WAP is a flattened buffer format with zero parsing. Given the heavy utilization of numerical computation, numerical types are stored in little-endian binary representation to natively support WASM without needing to transform/parse while supporting all native primitive types of WASM. WAP supports other serialization formats which serialize and deserialize into the flat format.
Given the low-level nature of Wasm, there doesn't exist a uniform way to describe data schemas efficiently without the need to deserialze.
General purpose format. Even though WASM is a primary target, it has first class support to compile schemas to other languages and runtimes (TypeScript/JavaScript, Go, Rust, C#, Python, etc) utilizing the lowest level features of the respective target as much as possible for peak performance and correctness.
Schemas are represented in ".wap" files. It's a bit of a hybrid between protobuf and flatbuffer schemas.
MoonProto is all about throughput over size. MoonProto messages can be compressed with a high-performance algorithm like LZ4 which Moontrade uses extensively.
MoonProto is built primarily with streaming in mind. Moontrade utilizes various streaming technologies like Redis, Kafka, PubSub, etc.
MoonProto exclusively uses little-endian. Language implementations of MoonProto may provide support for big-endian format.
WAP can transparently serialize and deserialize utilizing JSON format while taking advantage of the efficient flat deserialized format.
WAP provides the ability to describe 2 names for the same field. Both must be unique within the record. The auto-generated JSON reader will resolve either name.
WAP can transparently serialize and deserialize utilizing Protocol Buffers format while taking advantage of the efficient flat deserialized format.
WAP supports TypeScript/JavaScript just dandy. Graphically charting numerical streams (financial charts) is extremely common in web browsers. Moontrade UI uses TypeScript HTML Canvas charting built on top of TradingView.
For maximum performance structs are ideal. However, some structures require variable length strings, lists, etc. WAP can support buffers up to 2GB in length.
WAP supports references within a single graph. Adding a reference to a type will turn the type into a flexible length type in order to resolve references in the graph.
WAP provides a custom streaming format.
Logs timestamped monotonic records
Time-Series records are timestamped monotonic records and have a time range / intervals.
FlatBuffers generally introduces an additional layer of indirection which can make property accesses slower with the added benefit of a more flexible evolutionary data structure. WAP still provides the ability to evolve your schemas, however WAP is geared for "native struct-like" access. In addition, it's important to point out that WAP was designed explicitly to meet Moontrade's product requirements / use cases. A good example of performance issues with FlatBuffers was evident inside Deno before they switched to their own direct (struct-like zero-copy) format. Similar performance gains can be expected for WAP.
In addition, the mutable access pattern for FlatBuffers is awkward.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent