bufferfish

bufferfish is utility library for working with binary network messages between Rust and TypeScript, such as over WebSockets. It provides a simple API for encoding and decoding data into binary arrays, as well as generating TypeScript definitions and decoding functions from your Rust code.

This library has an unstable API and may be missing some basic functionality. I can't recommend using it in production, although I am using it for my own production project.

Table of Contents

• bufferfish
• Table of Contents
• Getting Started
  • Examples
    • Using Generated Decoding Functions (JavaScript)
    • Manually Decoding a Bufferfish (JavaScript)
  • TypeScript Code Generation
    • Codegen Example
• Encodable / Decodable Types
  • Notes
• Security
  • Contributing
  • License

Getting Started

Use cargo add bufferfish to add the Rust library to your project.

Use npm install bufferfish to add the TypeScript library to your project.

Examples

use bufferfish::{Encode};
use futures_util::SinkExt;
use tokio::net::{TcpListener, TcpStream};
use tokio_tungstenite::{accept_async, tungstenite::Message};

#[derive(Encode)]
#[repr(u16)]
enum PacketId {
    Join,
}

// We need to make sure we can convert our enum to a u16, as that is the type
// `bufferfish` uses to identify packets. You can use the `num_enum` crate and
// derive `IntoPrimitive` and `FromPrimitive` to remove this step completely.
impl From<PacketId> for u16 {
    fn from(id: PacketId) -> u16 {
        match id {
            PacketId::Join => 0,
        }
    }
}

// We annotate our packet with the #[Encode] macro to enable automatic
// encoding and decoding to or from a `Bufferfish`.
//
// Additionally, we use the #[bufferfish] attribute to specify the packet ID.
#[derive(Encode)]
#[bufferfish(PacketId::Join)]
struct JoinPacket {
    id: u32
    username: String,
}

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let listener = TcpListener::bind("127.0.0.1:3000").await?;

    while let Ok((stream, _)) = listener.accept().await {
        tokio::spawn(async move {
            if let Err(e) = process(stream).await {
                eprintln!("Error processing connection: {}", e);
            }
        });
    }

    Ok(())
}

async fn process(steam: TcpStream) -> Result<(), Box<dyn std::error::Error>> {
    let mut ws = accept_async(steam).await?;

    let packet = JoinPacket {
        id: 1,
        username: "Rob".to_string(),
    };
    let bf = packet.to_bufferfish()?;

    ws.send(Message::Binary(bf.into())).await?;

    Ok(())
}

Using Generated Decoding Functions (JavaScript)

const ws = new WebSocket("ws://127.0.0.1:3000")
ws.binaryType = "arraybuffer"

ws.onmessage = (event) => {
  const bf = new Bufferfish(event.data)
  const packetId = bf.readUint16()

    if (packetId === PacketId.Join) {
        const packet = decodeJoinPacket(bf)

        console.log(packet) // { id: 1, username: "Rob" }
    }
}

Manually Decoding a Bufferfish (JavaScript)

const ws = new WebSocket("ws://127.0.0.1:3000")
ws.binaryType = "arraybuffer"

ws.onmessage = (event) => {
    const bf = new Bufferfish(event.data)
    const packetId = bf.readUint16()

    if (packetId === PacketId.Join) {
        const id = bf.readUint32()
        const username = bf.readString()

        console.log({
            id,
            username,
        }) // { id: 1, username: "Rob" }
    }
}

TypeScript Code Generation

bufferfish provides a generate function that can be used in build.rs (or used in a CLI script, called by server at launch, etc) to generate TypeScript definitions and functions from your Rust code, meaning your Rust server becomes the source of truth for all network messages, and reducing manually interacting with bufferfish on the client.

// build.rs
fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("cargo:rerun-if-changed=build.rs");

    bufferfish::generate("src", "../client/src/generated/Packet.ts")?;

    Ok(())
}

Codegen Example

use bufferfish::Encode;

pub enum PacketId {
    Join = 0,
    Leave,
    Unknown = 255,
}

#[derive(Encode)]
#[bufferfish(PacketId::Join)]
pub struct JoinPacket {
    pub id: u8,
    pub username: String,
}

#[derive(Encode)]
#[bufferfish(PacketId::Leave)]
pub struct LeavePacket;

/* AUTOGENERATED BUFFERFISH FILE, DO NOT EDIT */
import { Bufferfish } from 'bufferfish';

export enum PacketId {
    Join = 0,
    Leave = 1,
    Unknown = 255,
}

export interface JoinPacket {
    id: number
    username: string
}

export const decodeJoinPacket = (bf: Bufferfish): JoinPacket => {
    return {
        id: bf.readUint8() as number
        username: bf.readString() as string
    }
}

Encodable / Decodable Types

u8 | number u16 | number u32 | number i8 | number i16 | number i32 | number bool | boolean String | string Vec<T> where T: Encodable | Array<T> T where T: Encodable | object or primitive

*The reverse is true for decoding.

Notes

• I recommend using the num_enum crate for deriving IntoPrimitive and FromPrimitve on enums you wish to Encode. This removes a lot of boilerplate.
• Enums in TypeScript are often mentioned as a "bad" feature, and this is generally true when considering typical web development use-cases. In the case of a list of "op codes", mapping to dev-friendly names, however, they are actually really useful. Modern bundlers - like esbuild - can actually inline them, meaning we just get integer literals in the final output..

Security

bufferfish functions ensure inputs are valid as a "best effort". Internal buffers are constructed with a maximum capacity (default of 1024 bytes), and will fail to construct if an input would cause the internal buffer to cross that threshold.

When reading data, you will always get the correct return type - however, you are still subject to corrupted data if the input was incorrect but technically valid. For example, if you call read_u8 on a buffer that contains a u16 at the cursor position, you will get a u8 back, as the buffer has no way to know that it was originally encoded as a u16. It is valid data, but will very likely be an unexpected value.

This kind of problem should be dealt with before operating on the buffer.

Decoding an oversized bufferfish via the Decode trait will just ignore / discard the additional data, as it is only going to read specific byte lengths generated by the Encodable impl.

Decoding an undersized bufferfish will return a BufferfishError::FailedWrite.

Contributing

bufferfish is open to contributions, however it should be noted that the library was created for my own game projects, and I am not interested in making it widely general-purpose. If you have a feature request or bug fix that you think would be useful to others, feel free to open an issue or PR either way.

License

bufferfish source code is dual-licensed under either

• MIT License
• Apache License, Version 2.0

at your option.