compress-utils 0.7.1

Simple & high-performance compression utilities for Python

compress-utils

Unified Python interface for six compression algorithms — Brotli, bzip2, LZ4, XZ/LZMA, zlib, and Zstandard — backed by a single high-performance C library. Same API for every algorithm.

Installation

pip install compress-utils

The wheel is self-contained: no system codec libraries required at runtime. Type stubs (.pyi) and a py.typed marker ship with the wheel, so mypy, pyright, and IDE autocomplete work out of the box.

Quick start

Functional API (most common)

import compress_utils as cu

data = b"the quick brown fox jumps over the lazy dog" * 100

compressed = cu.compress(data, "zstd", level=5)
restored   = cu.decompress(compressed, "zstd")
assert restored == data

Algorithm names accept both lowercase strings ("zstd", "brotli", …) and the typed enum:

cu.compress(data, cu.Algorithm.zstd, level=5)

Streaming API

For inputs that don't fit in memory or arrive incrementally:

from compress_utils import CompressStream, DecompressStream

cs = CompressStream("zstd", level=5)
chunks = [cs.compress(chunk) for chunk in iter_chunks(some_file)]
chunks.append(cs.finish())
compressed = b"".join(chunks)

ds = DecompressStream("zstd")
restored = ds.decompress(compressed) + ds.finish()

Supported algorithms

Algorithm	Spelling	Notes
Zstandard	"zstd"	Wire format: ZSTD frame with content size.
Brotli	"brotli"	Wire format: raw Brotli stream.
zlib	"zlib" (also "gzip")	Wire format: zlib wrapper (RFC 1950).
bzip2	"bz2" (also "bzip2")	bzip2 stream.
LZ4	"lz4"	LZ4 frame format (compatible with lz4 CLI / .lz4 files).
XZ/LZMA	"xz" (also "lzma")	XZ stream with CRC64.

cu.is_available(name_or_enum) returns True for algorithms compiled into the installed wheel.

Other helpful APIs

cu.version()                                  # "0.1.0"
cu.set_max_decompressed_size(256 * 1024**2)   # bound one-shot decompression
                                              # (default: 1 GiB; 0 = unbounded)

try:
    cu.decompress(garbage, "zstd")
except cu.CompressError as e:
    print(e)

Compression levels

Every algorithm accepts a level from 1 (fastest) to 10 (smallest). The Python binding maps this to each algorithm's native range automatically — you don't need to know that ZSTD goes 1–22 or zlib goes 1–9. Defaults to 5 if omitted.

Type checking

The package ships PEP 561 type information:

import compress_utils as cu
reveal_type(cu.compress)           # → (data: Buffer, algorithm: object, level: int = 5) -> bytes
reveal_type(cu.Algorithm.zstd)     # → Algorithm

Stubs are regenerated from the compiled module at build time (via pybind11-stubgen), so they cannot drift from the binding signatures.

Performance notes

The Python binding is a thin pybind11 wrapper over the C library. Streaming uses chunked buffers with an internal drain protocol — output is yielded in 64 KB pages, so streaming a multi-GB file does not hold the whole compressed result in memory.

For applications that round-trip many small payloads with the same algorithm, prefer the functional API over creating a new CompressStream per payload — internal codec state is short-lived and reused.

Project layout

This is one of four official bindings to the compress-utils C library:

• C — the canonical ABI in include/compress_utils.h.
• C++ — header-only cu:: namespace, bindings/cpp/.
• Python — this package.
• JS / TS (WASM) — npm install compress-utils, bindings/wasm/.

Go, Rust, Swift, and Java bindings are tracked in TODO.md.

License

MIT. See LICENSE.

---

Built by Nico Dupont.