Masonry

A command line tool for composable project templating.

Masonry aims to reduce the need to write boiler plate code and setup files when starting or extending a project. It does so by allowing cookiecutter templates to be combined in a series of layers to build up a projects file structure.

Applying different combinations of these template building blocks then allow for a greater variety of project types to be more easily supported, compared to defining each template permutation separately.

Masonry also includes a cli application mason that makes applying and managing these template layers straight forward.

Installation

pip install masonry

Usage

1. Create a series of template layers to use for a particular project layout (see below and in the project-templates directory for examples).

2. Initialise a new project with its default starting template

   mason init project/template/path

3. Add an extra templating layer to the project

   mason add template_name

Creating and Using Template Layers for a Custom Project

The individual template layers are themselves cookiecutter templates. To combine several of these into a project to use with masonry:

• Each cookiecutter template should be in its own directory named after the templates purpose

• All these are then held in a parent directory, which is named after the project type all these layers relate to.

• Included in the project group directory is a metadata.json file that specifies the “default” template to use, as well as any dependencies between layers.

For example, imagine a situation that wished to combine the following three layers into a project:

1. One to create a base python package

2. Another to define the unittest structure with pytest

3. A third to add a build file for continuous integration.

Placing these all in the same directory called “python-project” would give the following structure:

path/to/python-project/
            ├── ci-build
            │   ├── {{cookiecutter.project_name}}
            │   └── cookiecutter.json
            │
            ├── package
            │   ├── {{cookiecutter.project_name}}
            │   └── cookiecutter.json
            │
            ├── pytest
            │   ├── {{cookiecutter.project_name}}
            │   └── cookiecutter.json
            │
            └── metadata.json

The metadata.json file would then specify the package template as the default layer, with the pytest template layer depending on the package template, and the ci-build template layer depending on the pytest template. The structure of the resulting JSON file is shown below:

{
    "default": "package",
    "dependencies": {
        "pytest": ["package"],
        "ci-build": ["pytest"]
    }
}

Now using mason we can create a new project following the package template with:

mason init path/to/python-project/

And at a later time, when we want to start adding tests and a build process to the project, we could run:

mason add ci-build

Note that even though we only specified “ci-build” above, mason is able to work out all needed template layer dependencies from the metadata.json and apply them in the right order. This means both the pytest and ci-build template layers will be applied in that order.

Projects as a Collection of Components

Splitting out the templates above may not seem to have gained you very much. After all, you could have just defined all these files for package + tests + CI in a single template structure. However, as you start to add different components to your projects under different scenarios, this modular approach becomes more beneficial.

For example, say you wanted to support different CI services such as circle CI, travis, and GitLab Runners on different projects; and on some projects you have a Makefile, and on others something more cross platform compatible like an invoke tasks.py file.

Accommodating all these options would either mean maintaining 6 different templates with a lot of repetition, or one large one with a lot of control flow logic in the jinja template.

Stomemason provides a middle ground of breaking up these components of the project into separate layers.

It also has the major benefit of being able to apply any additional layers after the initial project was created. So if you didn’t see the need to also create a conda package for your project till now? No problem, just apply the conda-package layer to the current project (assuming you have defined one of course!).

Additional Features

• Pre and post project creation hooks used by cookiecutter are supported.

• Cookiecutter variables are remembered and reused between template layers, meaning you only need to specify values for new variables.

• If project path is omitted, mason init allows you to interactively select one from a list of previously used projects.

• If the template name is omitted, mason add, allows you to interactively select one from a list of templates that can still be added to the project.

• Colourful UI and easy multiple choice selection thanks to the inquirer library.

Other Related Projects

• python boilerplate A web application for interactively filling out a template for a new project.

• yeoman A project scaffolding tool for web development written in javascript.

• mason A project in R similar to cookiecutter but with a colourful UI, and one of the inspirations for this project.

• usethis A project in the R community looking at similar ideas around project template composability, and being able to add them as needed on a project.