Building the module

Binary releases

If you’re on Windows, you can download the current binary release and ignore most of this page. Some older installers are available on Github, but you should really get the more recent installers that Christoph Gohlke provides on his website: http://www.lfd.uci.edu/~gohlke/pythonlibs/#pysfml

On other platforms, there may still be easier ways to build the module. For example, someone has written AUR scripts for Arch Linux users:

• https://aur.archlinux.org/packages.php?ID=50841
• https://aur.archlinux.org/packages.php?ID=50842

Getting SFML 2

The first thing you should do is get SFML 2 and make sure it works. If you want the latest version, please refer to the official tutorial: http://sfml-dev.org/tutorials/2.0/compile-with-cmake.php There is also a slightly older RC binary release that is mainly used by Windows users: http://www.sfml-dev.org/download.php#2.0-rc

Some platforms may make it easier to install it, for example Arch Linux users can get it from the AUR.

If you are on Windows, you will probably want to copy SFML’s headers and libraries directories to the corresponding directories of your compiler/IDE, and SFML’s DLLs to Windows’ DLL directory.

Building on Windows

If you don’t have a C++ compiler installed, I suggest using MinGW.

If you are using a recent version of MinGW, you may encounter this error when building the module:

error: unrecognized command line option '-mno-cygwin'

The problem is that the -mno-cygwin has been dropped in recent MinGW releases. A quick way to fix this is to remove the option from the distutils source. Find the distutils/cygwinccompiler.py in your Python installation (it should be something like C:\Python27\Lib\distutils\cygwinccompiler.py). Find the MinGW32CCompiler class and remove the -mno-cygwin options:

# class CygwinCCompiler
self.set_executables(compiler='gcc -mno-cygwin -O -Wall',
                     compiler_so='gcc -mno-cygwin -mdll -O -Wall',
                     compiler_cxx='g++ -mno-cygwin -O -Wall',
                     linker_exe='gcc -mno-cygwin',
                     linker_so='%s -mno-cygwin %s %s'
                                % (self.linker_dll, shared_option,
                                   entry_point))

If you are using Visual C++, please use the 2008 version. Python was built with this version, and it’s apparently difficult to use 2010 because it links to another C or C++ runtime.

Common build options

You can build the module with the setup.py script (or setup3k.py for Python 3). This section discusses some common options that you may need or find useful.

--inplace means that the module will be dropped in the current directory. I find this more practical, as it makes it easier to test the module once built.

--compiler=mingw32 obviously means that MinGW will be invoked instead of the default compiler. This is needed when you want to use GCC on Windows. This command will show you the list of compilers you can specify: python setup.py build_ext --help-compiler. Visual Studio is the default compiler and should work without using this option.

In the end, the command will look something like this:

python setup.py build_ext --inplace --compiler=mingw32

Building without Cython

Warning

Github has removed the downloads feature, so I don’t plan to package source releases anymore. Windows users can use the installers, and it should be easier for other users to build the module now that recent versions of Cython correctly build the module out of the box.

If you download a source release at the download page, you don’t need to install Cython, since the release already contains the files that Cython would generate.

Make sure that USE_CYTHON is set to False in setup.py (or setup3k.py, if you’re building for Python 3). You can then build the module by typing this command:

python setup.py build_ext

Building with Cython installed

Warning

With older versions of Cython, the binding won’t work correctly when built straight from the Git repo. If you want to build from the source, you’re encouraged to use the latest source release. See Building without Cython. If you really want to build from Git, you need to modify the generated sfml.cpp file. You need all these declarations:

__PYX_EXTERN_C DL_EXPORT(PyObject) *wrap_time_instance(sf::Time *);
__PYX_EXTERN_C DL_EXPORT(PyObject) *wrap_render_target_instance(sf::RenderTarget *);
__PYX_EXTERN_C void set_error_message(char*);
__PYX_EXTERN_C DL_EXPORT(PyObject) *wrap_chunk_instance(sf::SoundStream::Chunk*, int);

I don’t know if it’s the same exact problem everywhere, but on my system I get the first declaration and not the others. So you can look for the wrap_time_instance() declaration, and copy-paste the others after it.

This bug should be fixed in the next Cython release.

Warning

Several Ubuntu users reported that they can’t build the module because the Cython package is currently outdated. One solution is to install Cython manually, for example with easy_install cython.

If you downloaded the source straight from the Git repo or if you have modified the source, you’ll need to install Cython to build a module including the changes. Also, make sure that USE_CYTHON is set to True in setup.py.

When you’ve done so, you can build the module by typing this command:

python setup.py build_ext

If you get an error related with DL_IMPORT, refer to the end of the Building a Python 3 module section.

Building a Python 3 module

It’s possible to build a Python 3 module, but you may encounter a few minor problems.

First of all, on my machine, the Cython class used in setup3k.py to automate Cython invocation is only installed for Python 2. It’s probably possible to install it for Python 3, but it’s not complicated to invoke Cython manually:

cython --cplus sfml.pyx

The next step is to invoke the setup3k.py script to build the module. Since we called Cython already, make sure that USE_CYTHON is set to False in setup3k.py, then invoke this command:

python3 setup3k.py build_ext

(Note that you may have to type python instead of python3; typically, GNU/Linux systems provide this as a way to call a specific version of the interpreter, but I’m not sure that’s the case for all of them as well as Windows.)

(Also note that on GNU/Linux, the generated file won’t be called sfml.so but something like sfml.cpython-32mu.so. Apparently, on Windows it’s still sfml.pyd.)

The second problem used to be that you had to use bytes instead of Unicode e.g. when passing a filename or window title to SFML. This is now gone, except possibly in methods that I forgot to fix; make sure to report the issue if you encounter such a case. When you pass a Unicode object to these methods, they now encode it in UTF-8 before passing them to SFML. You can change the encoding by setting the default_encoding variable at any time.

Finally, compilation may fail because the src/sfml.h file generated by Cython uses the deprecated DL_IMPORT() macro. At the root of the project, there is a patch.py script that will remove the offending macros for you. The trick is that src/sfml.h will not exist at first; the setup script will create it, then try to compile it and fail. That’s when you need to use patch.py, and build the module again.