Quick and painless wrapping C code into Python.
- Free software: MIT license
- Documentation: https://cslug.readthedocs.io.
- Source: https://github.com/bwoodsend/cslug
- Releases: https://github.com/bwoodsend/cslug/releases
The cslug package provides a thin layer on top of the built-in ctypes library, making it easier to load functions and structures from C into Python.
Mixing C with Python is nothing new - there are plenty of other ways. A nice comparison of the various methods can be found here. cslug aims to be the simplest although it certainly isn't the most flexible.
Using ctypes driven wrapping has both advantages and disadvantages over Python extension modules and tools that write them (such as Cython).
- C code can be just plain high school level C. Even a hello world Python extension module is some 40 lines of incomprehensible macros.
- Binaries are not linked against Python and are therefore not tied to a specific Python version. A Python extension module needs to be recompiled for every minor version of Python (3.6, 3.7, 3.8, 3.9) and for every platform (Windows, macOS, Linux) whereas a cslug binary need only be compiled for every platform.
- You can use virtually any C compiler. Python extension modules must be built with clang on macOS and MSVC on Windows.
- File sizes of binaries are very small.
1000 lines of C code equates to about 20KB of binary on Linux.
Python extension modules are typically several times larger and
a bare-bones Cython-ised
import numpy
extension is several MBs.
- The surrounding Python code is less automated. A Python extension module looks and feels like a native Python module out the box whereas a small wrapper function is generally required for ctypes.
- You can't use native Python types such as
list
ordict
within C code. Using such types will generally reduce performance down to near pure Python levels anyway so this is a small loss in practice. - You can't use C++.
Before you commit yourself to any non Pure-Python you should bear in mind that:
- You'll need to ship wheels for every platform you wish to support. Otherwise, users of your code will have to install a C compiler to run it. This means that you either need access to all platforms, or you will have to setup Continuous Integration to build you package on a cloud server. Linux users can get around this by using Vagrant.
- Linux wheels must be built on a manylinux Docker image in order to be compatible with all distributions of Linux.
- Unless your users have the relevant security thrice disabled, uninstalled, blocked and scraped off the hard drive, recent macOS will block or delete any binary file you produce unless you either pay for a codesign license or your software becomes famous enough to be whitelisted for you by Apple. Windows users face a similar, albeit lesser, problem with Microsoft Defender.
The following OS/compiler combinations are fully supported and tested regularly.
Compiler | Linux | Windows | macOS | FreeBSD | Cygwin/msys2 |
---|---|---|---|---|---|
gcc | ✓ | ✓ | ✓ | ✓ | ✓ |
clang | ✓ | ✗ | ✓ | ✓ | ✗ |
MSVC | ✗ | ✗ | ✗ | ✗ | ✗ |
TinyCC | ✓ | ✓ | ✗ | ✗ | ✗ |
cslug requires a C compiler to compile C code. Its favourite compiler is gcc. Linux distributions typically come with it preinstalled. If you are on another OS or just don't have it then you should get it with mingw-w64. Windows users are recommended to use MinGW-Builds.
Check that you have it set up by running the following in a terminal:
gcc -v
Note
gcc is a build time dependency only. If you provide wheels for a package that contain binaries built with cslug, then your users will not need a compiler; only if they try to build your package from source.
By default, cslug will use gcc if it can find it. On macOS or FreeBSD it
will switch to clang if gcc is unavailable.
To use any other supported compiler, cslug respects the CC
environment
variable.
Set it to the name or full path of your alternative compiler.
Install cslug itself with the usual:
pip install cslug
Whilst cslug is still in its 0.x versions, breaking changes may occur on
minor version increments.
Please don't assume forward compatibility - pick a version you like and
pin it in a requirements.txt
.
Inspect the changelog for anything that may break your code.
Check out our quickstart page on readthedocs to get started.
Hall of fame for contributions to cslug.
Huge shout out to JetBrains for PyCharm and for providing their full range of products free to open source developers. (The ability to run Python from inside a docker image with completion, debugging, and all the other bells and whistles has been a big help to this project.)
This initial creation of this package was sped up considerably by Cookiecutter and a fork of the audreyr/cookiecutter-pypackage project template.