Skip to content

Faster C implementation of the bitstruct Python library

License

Notifications You must be signed in to change notification settings

qchateau/cbitstruct

Repository files navigation

License Build Status

About

cbitstruct is a C implementation of the bitstruct library. Credits to Erik Moqvist for the original bitstruct library available on Github and pip.

The goal is to provide the same API an idealy to be able to interchange import bitstruct and import cbitstruct as bitstruct.

Obvious increased performance comes with limitations described below.

Installation

pip3 install cbitstruct

Documentation

Please refer to the bitstruct documentation as the aim of this library is to provide the same API with increased performance.

If you are not used to bitstruct, you should seriously consider using it first, before migrating to this library if you require higher performance.

Limitations

Limitation Will it be lifted ?
All types except padding are limited to 64 bits Maybe for 'raw' and 'text' types
May not work on big-endian architectures Maybe
Exceptions differ from bitstruct Probably not
CPython only Probably not
Error messages are unclear Will never be as clear as bitstruct
Python >= 3.5 No

Some limitations are there because I did not get the time or motivation to lift them up. Some other are deeply rooted into this library and may never be lifted.

Performance

Comparing to bitstruct

The script available in tests/test_perf.py measures performance comparing to the bitstruct library.

Here are the result "on my machine" (Ubuntu in Virtualbox on a laptop):

byteswap list of int      | x  8.779 (   8.638us ->   0.984us)
byteswap str              | x 17.466 (   9.158us ->   0.524us)
calcsize                  | x139.330 (  61.060us ->   0.438us)
compiled pack             | x 47.389 (  35.968us ->   0.759us)
compiled pack_dict        | x 27.184 (  34.588us ->   1.272us)
compiled pack_into        | x 32.037 (  38.650us ->   1.206us)
compiled pack_into_dict   | x 27.343 (  37.718us ->   1.379us)
compiled unpack           | x 33.928 (  31.278us ->   0.922us)
compiled unpack_dict      | x 21.627 (  31.597us ->   1.461us)
compiled unpack_from      | x 30.622 (  29.977us ->   0.979us)
compiled unpack_from_dict | x 20.479 (  30.936us ->   1.511us)
pack                      | x 77.003 ( 103.030us ->   1.338us)
pack_dict                 | x 53.254 ( 103.255us ->   1.939us)
pack_into                 | x 82.829 ( 119.373us ->   1.441us)
pack_into_dict            | x 52.173 ( 108.135us ->   2.073us)
unpack                    | x 78.459 (  91.896us ->   1.171us)
unpack_dict               | x 40.287 (  89.300us ->   2.217us)
unpack_from               | x 77.027 (  91.202us ->   1.184us)
unpack_from_dict          | x 39.467 (  88.043us ->   2.231us)

Disclaimer: these results may and will vary largely depending on the number of elements and types you pack/unpack. This script is provided as-is, and I will gladly accept an improved script providing more reliable results.

The dict API

The dict API is marginally slower than the traditional one. As the packing/unpacking performance is quite high, the overhead of performing dictionary lookups and hashing significantly increas pack and unpacking duration.