delta-pack

Binary serialization with delta compression for real-time state synchronization.

Delta-Pack is a cross-language serialization framework optimized for networked applications where you need to efficiently synchronize state between clients and servers. It provides both full encoding and delta encoding (only transmitting what changed).

Features

Compact binary format - Smaller than JSON, MessagePack, and often Protobuf
Delta compression - Encode only the differences between two states
Cross-language - Compatible with TypeScript and C# implementations
Zero-copy strings - String dictionary deduplication within each message
Serde integration - Generated types derive Serialize/Deserialize

Installation

Add to your Cargo.toml:

[dependencies]
delta-pack = "0.1"

Usage

Delta-Pack uses code generation from YAML schemas. Define your schema, generate Rust code, and use the generated types.

1. Define a Schema

# schema.yml
HairColor:
  - BLACK
  - BROWN
  - BLOND
  - RED

Address:
  street: string
  city: string
  zip: string

User:
  id: string
  name: string
  age: uint
  weight: float(precision=0.01)
  hairColor: HairColor
  address: Address?
  tags: string[]
  metadata: <string, string>

2. Generate Rust Code

npx delta-pack generate schema.yml -l rust > src/generated.rs

3. Use the Generated Types

use crate::generated::{User, Address, HairColor};

fn main() {
    // Create a user
    let user1 = User {
        id: "user-123".into(),
        name: "Alice".into(),
        age: 30,
        weight: 65.5,
        hair_color: HairColor::Brown,
        address: Some(Address {
            street: "123 Main St".into(),
            city: "Springfield".into(),
            zip: "12345".into(),
        }),
        tags: vec!["admin".into(), "verified".into()],
        metadata: [("level".into(), "5".into())].into(),
    };

    // Full encode/decode
    let bytes = user1.encode();
    let decoded = User::decode(&bytes);
    assert!(user1.equals(&decoded));

    // Delta encoding - only send what changed
    let user2 = User {
        age: 31,  // birthday!
        ..user1.clone()
    };

    let full_size = user2.encode().len();
    let diff = User::encode_diff(&user1, &user2);
    let diff_size = diff.len();

    println!("Full: {} bytes, Diff: {} bytes", full_size, diff_size);
    // Full: 58 bytes, Diff: 3 bytes

    // Apply the diff to reconstruct user2
    let reconstructed = User::decode_diff(&user1, &diff);
    assert!(user2.equals(&reconstructed));
}

Generated API

Every generated type provides these methods:

Method	Description
`encode(&self) -> Vec<u8>`	Serialize to binary
`decode(buf: &[u8]) -> Self`	Deserialize from binary
`encode_diff(a: &Self, b: &Self) -> Vec<u8>`	Encode only the differences from `a` to `b`
`decode_diff(a: &Self, diff: &[u8]) -> Self`	Apply a diff to `a` to produce `b`
`equals(&self, other: &Self) -> bool`	Deep equality (respects float precision)

Generated types also derive:

Clone, Debug - Standard traits
Default - All fields initialized to zero/empty values
Serialize, Deserialize - Serde support for JSON interop

Schema Types

Primitives

Schema	Rust Type	Notes
`string`	`String`	UTF-8, dictionary-compressed
`int`	`i64`	Signed, varint-encoded
`uint`	`u64`	Unsigned, varint-encoded
`int(min=0, max=100)`	`u64`	Bounded, more compact
`float`	`f32`	32-bit IEEE 754
`float(precision=0.01)`	`f32`	Quantized for smaller diffs
`boolean`	`bool`	1 bit, RLE-compressed

Containers

Schema	Rust Type
`T[]`	`Vec<T>`
`T?`	`Option<T>`
`<K, V>`	`HashMap<K, V>`

Named Types

# Enum (list of strings)
Direction:
  - up
  - down
  - left
  - right

# Object (key-value properties)
Player:
  name: string
  score: uint
  position: Position

# Union (list of type references)
Message:
  - ChatMessage
  - MoveMessage
  - AttackMessage

Self-References

Recursive types are supported and generate Box<T>:

TreeNode:
  value: int
  children: TreeNode[]  # Generates Vec<Box<TreeNode>>

Binary Format

[data section][RLE section][numRleBits: reverse varint]

Data section: Primitives encoded sequentially (strings with dictionary, varints, floats)
RLE section: Run-length encoded bits (booleans, optional flags, change indicators)
Reverse varint: Bit count stored at end for streaming decode

Performance

Benchmarks comparing encode throughput (higher is better):

Schema	DeltaPack	JSON	MessagePack
Primitives	34.1M ops/s	11.1M	10.1M
GameState	5.8M ops/s	673K	816K
User	5.3M ops/s	1.8M	2.0M

Run benchmarks:

cd rust/benchmarks
./build.sh  # Generate benchmark schemas
cargo run --release

Faster Decoding with mimalloc

Decode performance is allocation-bound. Using mimalloc instead of the system allocator improves decode throughput by ~30%:

# Cargo.toml
[dependencies]
mimalloc = "0.1"

// main.rs
#[global_allocator]
static GLOBAL: mimalloc::MiMalloc = mimalloc::MiMalloc;

Cross-Language Compatibility

Delta-Pack ensures binary compatibility across Rust, TypeScript, and C#:

Same schema produces interoperable binary format
Conformance tests verify encode/decode compatibility
Diff encoding uses sorted keys for deterministic output

Note: Full encodes may produce different byte sequences across languages due to HashMap iteration order, but decoded values are identical.

License

MIT

delta-pack 0.1.0