This is a forge-like environment that can be used to build wheels for mobile platforms. It is currently only tested for iOS, but in theory, it should also be usable for Android. Contributions to verify Android support, and to add more package recipes, are definitely encouraged.
Before using Mobile Forge, you'll need to compile Python for your build platform (e.g., your laptop), and for host platform (e.g., for iOS). It may be helpful to use a project like Python-Apple-support to manage this compilation process.
If you do use Python-Apple-support, this repo contains an activation script that will configure your environment so it's ready to use.
Set an environment variable declaring the location of your Python-Apple-support checkout:
$ export PYTHON_APPLE_SUPPORT=/path/to/Python-Apple-support
Clone this repository, and run the activate script for the Python version you want to use:
$ git clone https://github.com/beeware/mobile-forge.git $ cd mobile-forge $ source ./setup-iOS.sh 3.11
This will create a Python virtual environment, install mobile forge, and provide some hints at forge commands you can run.
If a virtual environment already exists, it will be activated, and the same hints displayed.
If you're not using Python-Apple-support, the setup process requires more manual steps:
Create and activate a virtual environment using the Python build platform, using the build platform Python compiled in step 1.
Clone this repository, and install it into your freshly created virtual environment:
(venv3.11) $ git clone https://github.com/beeware/mobile-forge.git (venv3.11) $ cd mobile-forge (venv3.11) $ pip install -e .
Ensure your
PATH
contains any tools that were necessary to compile the host CPython, and does not contain any macOS development libraries. Mobile-forge will clean thePATH
to remove known problematic paths (e.g., paths added by Homebrew, rbenv, npm, etc).Set environment variables that define the location of the Python executable for each of the host platforms you intend to target. For iOS, this means defining 3 environment variables:
(venv3.11) $ export MOBILE_FORGE_IPHONEOS_ARM64=... (venv3.11) $ export MOBILE_FORGE_IPHONESIMULATOR_ARM64=... (venv3.11) $ export MOBILE_FORGE_IPHONESIMULATOR_X86_64=...
Build a package. The
packages
folder contains recipes for packages.lru-dict
is a good first package to try:(venv3.11) $ forge iOS lru-dict
Or, to build a wheel for a single architecture:
(venv3.11) $ forge iphonesimulator:12.0:arm64 lru-dict
Once this command completes, there should be a wheel for each platform in the dist
folder. A log for each successful build will be in the logs
folder; a log for each
unsuccessful build (if there are any) will be in the errors
folder.
Mobile Forge is trying to support multiple packages, building on multiple Python versions, for multiple architectures; and some of those Python versions were released before the release of ARM64 macOS hardware. As a result, some versions of some packages have some quirks that must be taken into account.
Pandas uses a meta-package named oldest-supported-numpy
to ensure ABI compatibility
during compilation. However, this can install a different version of numpy, depending on
the platform. This is especially problematic for Python 3.9, because the minimum
supported version for Python 3.9 on ARM64 is different to the version that is installed
for x86_64. Mobile-forge produces a replacement oldest-supported-numpy
package, tagged
as version 2999.1.1, which ensures that consistent versions are available for build
purposes; however, this wheel should not be published.
Cryptography currently builds a very old version (3.4.8). This is the last version that could be built without a Rust compiler. The recipe works as is on all Python versions except Python 3.8 on ARM64, because there was no ARM64-compatible wheel published for cffi 1.15.1. However, if you run:
$ pip wheel -w dist --no-deps cffi==1.15.1
you can build a universal Python3.8 CFFI wheel for CFFI 1.15.1, which can be used to satisfy this build-time requirement.
To include these wheels in a test project, you can add the dist
folder as a links
source in your requires
definition in your Briefcase pyproject.toml
. For
example, the following will install the lru-dict
wheels you've just compiled:
requires = [ "--find-links", "/path/to/mobile-forge/dist", "lru-dict", ]
If there's a package that you want that doesn't have an existing recipe, you can add a recipe for that package.
Create a directory in recipes
. The name of the directory must be in PyPI normalized
form (PEP 503). Alternatively, you can create this directory somewhere else, and pass
its path when calling forge
.
Inside the recipe directory, add the following files.
A meta.yaml file. This supports a subset of Conda syntax, defined in meta-schema.yaml.
A test.py file (or test package), to run on a target installation. This should contain a pytest suite which imports the package and does some basic checks.
Optionally, one or more patch files in a folder named
patches
. These patches will be applied when the source code is unpacked for a given platform.For non-Python packages, a
build.sh
script. This is the script that will be executed in the build environment build the package. This script should invoke anyconfigure
,make
, or any other compilation steps needed to build the package. This script will be executed in an environment that defines the following environment variables:AR
- theAR
value used to compile the host Python, as determined fromsysconfig
CC
- theCC
value used to compile the host Python, as determined fromsysconfig
.CFLAGS
- theCFLAGS
value used to compile the host Python, as determined fromsysconfig
, augmented with the include paths for the SDK, andopt/include
in the host environment's site-packages.LDFLAGS
- theCFLAGS
value used to compile the host Python, as determined fromsysconfig
, augmented with the library paths for the SDK, andopt/lib
in the host environment's site-packages.CPU_COUNT
- The number of CPUs that are available, as determined bymultiprocessing.cpu_count()
HOST_TRIPLET
- the GCC compiler triplet for the host platform (e.g.,aarch64-apple-ios12.0-simulator
)BUILD_TRIPLET
- the GCC compiler triplet for the build platform (e.g.,aarch64-apple-darwin
)PREFIX
- a location where the compiled package can be installed in preparation for packaging.
This script should install the package into
$PREFIX
. Mobile Forge will package any content installed into$PREFIX
into a "wheel" that can be installed as a host requirement.
All Python projects are compiled using python -m build
, using a clean crossenv virtual environment for each platform of a
package. Any PEP518 build requirements will be included in both the host and build
environments.
If you're lucky, all you'll need to do is define a meta.yaml
that describes the
package name and version: e.g.,:
package: name: blis version: 0.4.1
If this doesn't result in a successful build, it will likely be for one of the following reasons:
The build process has a dependency on a system library. For example, Pillow has a dependency on
libjpeg
.libjpeg
isn't available on PyPI; but it is possible to build a "wheel" forlibjpeg
, so it can be specified as a requirement.A non-python "wheel" is constructed by compiling the package for your target platform, then installing it into a folder named
opt
. As a result of this "install", you'll usually end up with anopt/include
andopt/lib
folder; Mobile Forge will then wrap up thisopt
folder in a wheel, along with Python wheel metadata.When this "wheel" is specified as a host requirement, the "wheel" will be unpacked into the site packages folder of your cross-compilation host environment. This path the
include
andlib
paths will be automatically included in theCFLAGS
/LDFLAGS
environment variables when the Python build is executed.The build process has a dependency on external tooling. Mobile Forge will configure a C and C++ compiler using the same configuration that was used to compile the support libraries; however a package may require addition build tooling (e.g., a Fortran compiler) to complete the build. If this is the case, you'll need to find a version of the tool that can target mobile platforms, and work out how to modify the build process to apply any necessary compiler flags.
The build script has platform-specific logic. For example, if the
setup.py
file contain anif sys.platform == ...
clauses, it is unlikely that a mobile platform will trigger the right logic.
If you need to make any alterations to a project's source code for a build to succeed,
you can provide those patches by putting them in one or more files in a folder named
patches
in the recipe folder. These patches will be applied once the source code
has been unpacked.
If the project includes a configure script, you will likely need to provide a patch for config.sub. config.sub is the tools used by configure to identify the architecture and machine type; however, it doesn't currently recognize the host triples used by Apple. If you get the error:
checking host system type... Invalid configuration `arm64-apple-ios': machine `arm64-apple' not recognized configure: error: /bin/sh config/config.sub arm64-apple-ios failed
you will need to patch config.sub. There are several examples of patched config.sub scripts in the packages contained in this repository, and in the Python-Apple-support project; it is quite possible one of those patches can be used for the library you are trying to compile. The config.sub script has a datestamp at the top of the file; that can be used to identify which patch you will need.
Mobile Forge is part of the BeeWare suite. You can talk to the community through:
- @[email protected] on Mastodon
- Discord
- The Mobile Forge Github Discussions forum
We foster a welcoming and respectful community as described in our BeeWare Community Code of Conduct.
If you experience problems with Mobile Forge, log them on GitHub. If you want to contribute code, please fork the code and submit a pull request.
This project draws significantly on the implementation and knowledge developed in the Chaquopy package builder. Although this is largely a "clean room" reimplementation of that project, many details from that project have been used in the development of this one.