A pure Julia implementation of the Zstandard (RFC 8878) compression algorithm. No C dependencies required.

• Compression and decompression of Zstandard frames
• Streaming decompression via ZstdDecompressorStream (implements IO interface)
• Huffman and FSE entropy coding
• Repeat offset encoding
• Multi-block support for data larger than 128KB
• Dictionary decompression
• Skippable frames
• Optional content checksums (XXH64)

• CodecZstd.jl -- wraps the C libzstd library via TranscodingStreams.jl, providing a mature and battle-tested streaming interface.
• ChunkCodecsLibZstd.jl -- wraps the C libzstd library via ChunkCodecs.jl, providing chunk-based encoding and decoding.

For production use, these C-backed packages are likely more reliable and performant. Zstandard.jl is useful when a pure Julia implementation is preferred, such as for environments where C dependencies are unavailable or for educational purposes.

using Pkg
Pkg.add(url="https://github.com/JuliaIO/Zstandard.jl")

Note: This package depends on a fork of XXHashNative.jl that adds XXH64 support. The [sources] section in Project.toml points to this fork until Moelf/XXHashNative.jl#28 is merged.

using Zstandard

# Compress
compressed = compress("Hello, world!")
compressed = compress(read("input.bin"))

# Decompress
data = decompress(compressed)

# Streaming decompression
open("file.zst") do io
    stream = ZstdDecompressorStream(io)
    decompressed = read(stream)
end

# Optional content checksum
compressed = compress(data, checksum=true)

Benchmarked against libzstd (via CodecZstd.jl) on an AMD Ryzen system, Julia 1.11:

Compression ratios are within 6% of libzstd across all tested datasets.

Compression is faster than libzstd on small text and incompressible data. Decompression is within 2-3x of libzstd for compressible data.

julia --project=. -e 'using Pkg; Pkg.test()'