heatshrink-rs

A no_std Rust implementation of Heatshrink compression and decompression for embedded systems.

Heatshrink is an LZSS-based algorithm designed for low-memory environments. It operates with bounded, incremental CPU use, making it suitable for hard real-time systems.

Features

• no_std, no allocator required — runs on bare-metal targets.
• Configurable parameters — window size W and lookahead L are const generics; any valid (W, L) combination can be used at zero runtime cost.
• Idiomatic Rust API — sink, poll, and finish return Result types; no C-style return codes.
• Optional search index — enable the heatshrink-use-index feature to significantly speed up compression at the cost of extra memory.
• embedded-io adapters — enable the embedded-io feature to use the encoder and decoder as Read/Write streams in embedded-io pipelines.
• ISC licence — free to use, including for commercial purposes.

Parameters

Both the encoder and decoder are parameterised by const generics:

BUF and WIN are redundant parameters required by a current Rust stable limitation: arithmetic expressions are not yet allowed in const-generic array sizes. Always set BUF = 2 << W and WIN = 1 << W.

The convenience type aliases [DefaultEncoder] and [DefaultDecoder] use W=8, L=4 to match the original C library defaults.

Quick Start

Convenience functions (single-call, buffer-to-buffer)

use heatshrink::{encoder, decoder};

let input = b"hello heatshrink";
let mut compressed   = [0u8; 64];
let mut decompressed = [0u8; 64];

let encoded = encoder::encode(input, &mut compressed).unwrap();
let decoded = decoder::decode(encoded, &mut decompressed).unwrap();

assert_eq!(input, decoded);

Streaming API

Use the streaming API when processing data in chunks or when working under tight memory constraints.

use heatshrink::{DefaultEncoder, DefaultDecoder, Poll, SinkError, Finish};

// ── Encoding ──────────────────────────────────────────────────────────────────
let input = b"hello heatshrink - streaming example";
let mut compressed = [0u8; 128];
let mut enc = DefaultEncoder::new();
let mut total_in  = 0;
let mut total_out = 0;

loop {
    if total_in < input.len() {
        match enc.sink(&input[total_in..]) {
            Ok(n)                  => total_in += n,
            Err(SinkError::Full)   => {}   // drain with poll() first
            Err(SinkError::Misuse) => panic!("sink after finish"),
        }
    }
    if total_in == input.len() {
        enc.finish();
    }
    match enc.poll(&mut compressed[total_out..]) {
        Ok(Poll::More(n))  => total_out += n,
        Ok(Poll::Empty(n)) => {
            total_out += n;
            if total_in == input.len() { break; }
        }
        Err(_) => panic!("empty output buffer"),
    }
}

// ── Decoding ──────────────────────────────────────────────────────────────────
let mut decompressed = [0u8; 128];
let mut dec = DefaultDecoder::new();
let mut dec_in  = 0;
let mut dec_out = 0;

loop {
    if dec_in < total_out {
        match dec.sink(&compressed[dec_in..total_out]) {
            Ok(n)                  => dec_in += n,
            Err(SinkError::Full)   => {}
            Err(SinkError::Misuse) => panic!(),
        }
    }
    match dec.poll(&mut decompressed[dec_out..]) {
        Ok(Poll::More(n))  => dec_out += n,
        Ok(Poll::Empty(n)) => {
            dec_out += n;
            if dec_in == total_out { break; }
        }
        Err(_) => panic!("empty output buffer"),
    }
}

assert_eq!(&input[..], &decompressed[..dec_out]);

Custom parameters

use heatshrink::{encoder::HeatshrinkEncoder, decoder::HeatshrinkDecoder};

// W=11, L=6 — BUF = 2<<11 = 4096, WIN = 1<<11 = 2048
type MyEncoder = HeatshrinkEncoder<11, 6, 4096>;
type MyDecoder = HeatshrinkDecoder<11, 6, 32, 2048>;

embedded-io adapters

Enable the embedded-io feature to use the encoder and decoder as embedded_io::Read and embedded_io::Write streams:

[dependencies]
heatshrink-lib = { version = "...", features = ["embedded-io"] }

Four adapters are available in the heatshrink::io module:

Call finish() on the Write adapters when all data has been written.

# #[cfg(feature = "embedded-io")] {
use heatshrink::io::{EncoderWriter, DecoderWriter, SliceSink};
use heatshrink::{DefaultEncoder, DefaultDecoder};
use embedded_io::Write as _;

let input = b"hello heatshrink embedded-io";
let mut compressed   = [0u8; 64];
let mut decompressed = [0u8; 64];

// Compress
let mut sink = SliceSink::new(&mut compressed);
let mut enc: EncoderWriter<_, DefaultEncoder> = EncoderWriter::new(&mut sink);
enc.write_all(input).unwrap();
enc.finish().unwrap();
let n_enc = sink.len();

// Decompress
let mut sink2 = SliceSink::new(&mut decompressed);
let mut dec: DecoderWriter<_, DefaultDecoder> = DecoderWriter::new(&mut sink2);
dec.write_all(&compressed[..n_enc]).unwrap();
dec.finish().unwrap();

assert_eq!(input, sink2.written());
# }

Cargo features

Memory usage

With the default parameters W=8, L=4 (measured with core::mem::size_of):

The search index adds 2 << W × 2 bytes to the encoder: for W=8 that is 512 × 2 = 1024 bytes. Larger window sizes grow the index proportionally — at W=14 the index alone occupies 32 KB.

Performance

The choice of W and L parameters affects compression ratio, compression speed, and decompression speed. The graphs below illustrate these trade-offs for a representative dataset.

Benchmarks were run on a Core i5-8350U @ 1.7 GHz using accelerometer compressed data.

Migrating from 0.4.x

Version 0.4.x had hardcoded parameters (W=8, L=4) and C-style return codes. The current API differs in the following ways:

• Parameters are now const generics. Use DefaultEncoder / DefaultDecoder to keep the previous behaviour, or choose any (W, L) pair.
• sink() returns Result<usize, SinkError> instead of HSsinkRes.
• poll() returns Result<Poll, PollError> instead of HSpollRes.
• finish() returns Finish instead of HSfinishRes.
• encode() / decode() return Result<&[u8], CodecError> instead of Result<&[u8], HSError>.

More information

• Blog post: Heatshrink overview
• Blog post: Lightweight indexing
• Original C library