Table Of Contents
Create a package for IOS¶
New in version 1.2.0.
From the 4th march 2015, the toolchain for iOS has been rewritten. The previous instructions doesn’t work anymore. If you still want to use the older version, try the old-toolchain tag in git. The new version supports i386, x86_64, armv7, arm64 = it works on emulator.
Currently, packages for iOS can only be generated with Python 2.7. Python 3.3+ support is on the way.
The overall process for creating a package for IOS can be explained in 4 steps:
- Compile python + modules for IOS
- Create an Xcode project and link your source code
This process has been tested with Xcode 6.1
You need to install some dependencies, like cython, autotools, etc. We encourage you to use Homebrew to install thoses dependencies:
brew install autoconf automake libtool pkg-config brew link libtool sudo easy_install pip sudo pip install cython==0.21.2
For more detail, see IOS Prerequisites. Just ensure that everything is ok before starting the second step!
Compile the distribution¶
Open a terminal, and type:
$ git clone git://github.com/kivy/kivy-ios $ cd kivy-ios $ ./toolchain.py build kivy
Most of the python distribution will be packed into a python27.zip. If you experience any issues or would like more detail on this process, please refer to Compiling for IOS.
Create an Xcode project¶
Before proceeding to the next step, ensure your application entry point is a file named main.py.
We provide a script that creates an initial Xcode project to start with. In the command line below, replace test with your project name. It must be a name without any spaces or illegal characters:
$ # ./toolchain.py create /path/to/your/appdir $ ./toolchain.py create /Users/tito/code/kivy/examples/demo/touchtracer
You must use a fully qualified path to your application directory.
Now you can open the Xcode project:
$ open touchtracer-ios/touchtracer.xcodeproj
Updating an Xcode project¶
Let’s say you want to add numpy to your project, but you didn’t have it compiled prior the XCode project creation. First, ensure to build it:
$ ./toolchain.py build numpy
Then, update your Xcode project:
$ ./toolchain.py update touchtracer-ios
All the libraries / frameworks necessary to run all the compiled recipes will be added to your Xcode project.
You can customize the build in many ways:
- Minimize the build/python/lib/python27.zip: this contains all the python modules. You can edit the zip file and remove all the files you’ll not use (reduce encodings, remove xml, email...)
- Change the icon, orientation, etc... According to the Apple policy :)
- Go to the settings panel > build, search for “strip” options, and triple-check that they are all set to NO. Stripping does not work with Python dynamic modules and will remove needed symbols.
- Indicate a launch image in portrait/landscape for iPad with and without retina display.
Launch Images are supported. By default, XCode want you to build an Image Sets. This is your responsability to fill all the images needed for the Sets, depending of your target. However, Kivy use SDL, and as soon as the application starts the SDL main, the launch image will disapear. To prevent that, you need to have 2 files named Default.png and Default-Landscape.png, and put them in the Resources folder in Xcode (not in your application folder)
Currently, the project has a few known issues (we’ll fix these in future versions):
- You can’t export your project outside the kivy-ios directory because the libraries included in the project are relative to it.
- Removing some libraries (like SDL_Mixer for audio) is currently not possible because the kivy project requires it.
- And more, just too technical to be written here.
Application quit abnormally!¶
By default, all the print statements to the console and files are ignored. If you have an issue when running your application, you can activate the log by commenting out this line in main.m:
Then you should see all the Kivy logging on the Xcode console.
How can Apple accept a python app ?¶
We managed to merge the app binary with all the libraries into a single binary, called libpython. This means all binary modules are loaded beforehand, so nothing is dynamically loaded.