Bitcode

A bitwise encoder/decoder similar to bincode.

Motivation

We noticed relatively low entropy in data serialized with bincode. This library attempts to shrink the serialized size without sacrificing too much speed (as would be the case with compression).

Bitcode does not attempt to have a stable format, so we are free to optimize it.

Comparison with bincode

Features (serde)

• Bitwise serialization
• Gamma encoded lengths and enum variant indices
• Implemented in 100% safe Rust

Additional features with #[derive(bitcode::Encode, bitcode::Decode)]

• Enums use the fewest possible bits, e.g. an enum with 4 variants uses 2 bits
• Specify frequency of enum variants with #[bitcode_hint(frequency = 123) to use Huffman coding
• Opt into Gamma encoded integers with #[bitcode_hint(gamma)]
• Specify expected range of integers with #[bitcode_hint(expected_range = "50..100"]
• Hint that floats are expected to be between 0 and 1 with #[bitcode_hint(expected_range = "0.0..1.0"] (makes f32 ~25 bits and f64 ~54 bits)
• Serialization with bitcode::encode will never return an error unless you:
• Fall back to serde on specific fields with #[bitcode(with_serde)]

Limitations

• Doesn't support streaming APIs
• Format is unstable between versions
• When using feature = "derive", structs/enums that contain themselves must be labeled with #[bitcode(recursive)] or you will get a compile error

Benchmarks vs. bincode and postcard

Speed (nanoseconds)

Aims to be as fast as bincode and postcard. See rust serialization benchmark for more benchmarks.

Size (bits)

Random Struct Benchmark

The following data structure was used for benchmarking.

struct Data {
    x: Option<f32>,
    y: Option<i8>,
    z: u16,
    s: String,
    e: DataEnum,
}

enum DataEnum {
    Bar,
    Baz(String),
    Foo(Option<u8>),
}

In the table below, Size (bytes) is the average size of a randomly generated Data struct. Zero Bytes are the percentage of bytes that are 0 in the output. If the result contains a large percentage of zero bytes, that is a sign that it could be compressed more.

A note on enums

When using serde to serialize enums. Enum variants are encoded such that variants declared earlier will occupy fewer bits. It is advantageous to sort variant declarations from most common to least common.

This limitation can be avoided by using bitcode's derive macros.

Testing

Fuzzing

cargo install cargo-fuzz
cargo fuzz run fuzz

32-bit

sudo apt install gcc-multilib
rustup target add i686-unknown-linux-gnu
cargo test --target i686-unknown-linux-gnu

Acknowledgement

Some test cases were derived from bincode (see comment in tests.rs).

License

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.