Skip to main content

Library for data compression using Nintendo's Yaz0 algorithm

Project description


Library for data compression using the Yaz0 algorithm

syaz0 is a native module for Python 3.6+ that provides fast data compression and decompression using Nintendo's Yaz0 algorithm.


Decompression performance is on par with existing Yaz0 decoders.

As of late December 2019, syaz0 is able to compress files much faster than existing Yaz0 encoders. Files that are representative of Breath of the Wild assets were compressed 20x to 30x faster than with existing public tools for an equivalent or better compression ratio, and 70-80x faster (with a slightly worse ratio) in extreme cases.

At the default compression level, file sizes are typically within 1% of Nintendo's.

For detailed benchmarks, see the results files in the test directory.


syaz0 can be installed with pip3 install syaz0. Binary builds are provided for Windows 64 bits (only). On all other platforms, building from source is required. Skip to the end of the README for more information.


Returns a syaz0.Header corresponding to the Yaz0 file header with the fields magic, uncompressed_size, data_alignment, reserved.


Decompresses Yaz0-compressed data from a bytes-like object data. Returns a bytes object containing the uncompressed data.


Decompresses Yaz0-compressed data from a bytes (not bytes-like) object data. Returns a bytes object containing the uncompressed data.

Unlike syaz0.decompress, this function assumes that the input data is well-formed. In exchange for slightly improved performance, no sanity checks are performed. Warning: Do not use on untrusted data.

syaz0.compress(data, data_alignment=0, level=7)

Compresses a bytes-like object data. Returns a bytes-like object containing the Yaz0-compressed data.

data_alignment is a hint for decoders to allocate buffers with the required data alignment. Defaults to 0, which indicates that no particular alignment is required.

level is the compression level (6-9). 6 is fastest and 9 is slowest. Higher compression levels result in better compression. 7 is a good compromise between compression ratio and performance.

Project information

syaz0 was written with two goals in mind: to improve performance for Yaz0 compression — which is excruciatingly slow if one also desires decent compression ratios — and to let me learn a bit more about compression.

After doing more research on compression algorithms and finding out just how similar Yaz0 and LZ77-style compression algorithms are, I tried to implement some common tricks to help improve the extremely poor compression performance due to the sliding window search.

But the implementation was still extremely slow compared to gzip. And probably badly implemented.

It turns out that many more tricks were needed for fast compression.

After stumbling upon zlib-ng and seeing how well-optimised it is and all the intrinsics I decided it was best not to reinvent the wheel. Thus syaz0 uses a copy of zlib-ng for all the heavy lifting (match searching). The following modifications were made:

  • The window size was reduced to 4K (2^12) to match Yaz0.
  • The compress function and the stream structures were changed to take a callback that is invoked every time a distance/length pair or a literal is emitted. (I'm not proud, but it works.)
  • MAX_MATCH was not increased. zlib assumes it is equal to 258 in too many places and increasing it actually gives worse compression ratios.

Building from source

Building syaz0 from source requires:

  • CMake 3.10+
  • A compiler that supports C++17
  • Everything needed to build zlib-ng
  • pybind11 2.4+ (including CMake config files)
  • setuptools

When no binary build is available, pip will automatically build from source during the install process.

To build from source manually, run python3 bdist_wheel.


This software is licensed under the terms of the GNU General Public License, version 2 or later.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

syaz0-1.0.1.tar.gz (704.2 kB view hashes)

Uploaded Source

Built Distributions

syaz0-1.0.1-cp38-cp38-win_amd64.whl (129.1 kB view hashes)

Uploaded CPython 3.8 Windows x86-64

syaz0-1.0.1-cp38-cp38-manylinux2014_x86_64.whl (147.7 kB view hashes)

Uploaded CPython 3.8

syaz0-1.0.1-cp37-cp37m-win_amd64.whl (129.1 kB view hashes)

Uploaded CPython 3.7m Windows x86-64

syaz0-1.0.1-cp37-cp37m-manylinux2014_x86_64.whl (150.0 kB view hashes)

Uploaded CPython 3.7m

syaz0-1.0.1-cp36-cp36m-win_amd64.whl (129.0 kB view hashes)

Uploaded CPython 3.6m Windows x86-64

syaz0-1.0.1-cp36-cp36m-manylinux2014_x86_64.whl (150.0 kB view hashes)

Uploaded CPython 3.6m

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page