Python

Python applications that use supported build systems like Meson, CMake, or Autotools can be built using the standard method. However, many Python applications use custom install scripts or are expected to be installed through Setuptools and pip.

For these cases, flatpak-builder provides the simple buildsystem. Rather than automating the build process, simple accepts a build-commands array of strings, which are executed in sequence.

For example, the following YAML makes building the popular requests module rather straightforward:

name: requests
buildsystem: simple
build-commands:
  - pip3 install --prefix=/app --no-deps .
sources:
  - type: archive
    url: https://files.pythonhosted.org/packages/source/r/requests/requests-2.18.4.tar.gz
    sha256: 9c443e7324ba5b85070c4a818ade28bfabedf16ea10206da1132edaa6dda237e

Here, build-commands is an array containing the commands required to build and install the module. As can be seen, in this case pip is run to do this. Here, the --prefix=/app option is important, because otherwise pip would try to install the module under /usr/ which, because /usr/ is mounted read-only inside the sandbox, would fail.

Note that --no-deps is only used for the purpose of the example - since the requests module has its own dependencies, the build would fail. If multiple dependencies are required, it is better to install them using the method in the next section, instead.

Building multiple Python dependencies

Even though the example above installs, it won’t actually work. This is because the requests module has a number of dependencies that haven’t been installed:

  • certifi
  • chardet
  • idna
  • urllib3

Four dependencies aren’t very many, and could be installed using the simple method described above. However, anything more complex than this would quickly become tedious.

For these cases, flatpak-pip-generator can be used to generate the necessary manifest JSON. This is a Python script that takes a package name and uses pip to identify its dependencies, along with their tarball URLs and hashes.

Using flatpak-pip-generator is as simple as running:

$ python3 flatpak-pip-generator requests

Or if you have a requirements.txt file you can generate modules for every listed package:

$ python3 flatpak-pip-generator --requirements-file=requirements.txt

This will output a file called python3-requests.json, containing the necessary manifest JSON, which can then be included in your application’s manifest file. Even if your manifest uses YAML, you can still include JSON like this:

modules:
  - python3-requests.json
  # (other modules go here)