Test-driven system administration with a little extra logic.

View the Project on GitHub larsyencken/marelle

Marelle ("hopscotch")

Test-driven system administration in SWI-Prolog, in the style of Babushka.

Marelle uses logic programming to describe system targets and rules by which these targets can be met. Prolog's built-in search mechanism makes writing and using these dependencies elegant. Anecdotally, writing deps for Marelle has the feel of teaching it about types of packages, rather than the feel of writing package templates.

Hopscotch for Seniors

Current status

Working and in active use.


Marelle has some features common to other configuration management frameworks:

It also has some interesting differences:

Installing marelle


Run the bootstrap script:

bash -c "`curl`"

This will install marelle as the current user, putting the executable in ~/.local/bin/marelle.

Manual version

  1. Get Prolog
    • On OS X, with Homebrew: brew install swi-prolog
    • On Ubuntu, with apt-get: sudo apt-get install swi-prolog-nox
  2. Get git
    • On OS X, with Homebrew: brew install git
    • On Ubuntu, with apt-get: sudo apt-get install git
  3. Clone and set up marelle
# clone the repo
mkdir -p ~/.local
git clone ~/.local/marelle

# set up an executable
mkdir -p ~/.local/bin
cat >~/.local/bin/marelle <<EOF
exec swipl -q -t main -s ~/.local/marelle/ "$@"
chmod a+x ~/.local/bin/marelle

Writing deps

Make a marelle-deps/ folder inside your project repo. Each package has two components, a met/2 goal which checks if the dependency is met, and an meet/2 goal with instructions on how to actually meet it if it's missing.

For example, suppose I want to write a dep for Python that works on recent Ubuntu flavours. I might write:

% python is a target to meet

% it's installed if it exists at /usr/bin/python
met(python, linux(_)) :- exists_file('/usr/bin/python').

% we can install by running apt-get in shell
meet(python, linux(_)) :-
    % could also use: install_apt('python-dev')
    bash('sudo apt-get install -y python-dev').

To install python on a machine, I'd now run marelle meet python.

To install pip, I might write:


% pip is installed if we can run it
met(pip, _) :- which(pip).

% on all flavours of linux, try to install the python-pip package
meet(pip, linux(_)) :- install_apt('python-pip').

% on all platforms, pip depends on python
depends(pip, _, [python]).

Note our our use of platform specifiers and the _ wildcard in their place. To see your current platform as described by marelle, run marelle platform. Examples include: osx, linux(precise) and linux(raring).

Running deps

See available deps

This runs every met/2 statement that's valid for your platform.

marelle scan

Install something

This will run the meet/2 clause for your package, provided a valid one exists for your current platform.

marelle meet python

See your platform

To find the right platform code to use in deps you're writing, run:

marelle platform

It reports the code for the platform you're currently on.

Where to put your deps

Like both Babushka and Babashka, Marelle looks for deps in ~/.marelle/deps and in a folder called marelle-deps in the current directory, if either exists. This allows you to set up a personal set of deps for your environment, as well as project-specific deps.


See my marelle-deps repo for working examples.