repo-review

This is a framework for building checks designed to check to see if a repository follows guidelines. By itself, it does nothing - it requires at least one plugin to be installed..

With one or more plugins, it will produce a list of results - green checkmarks mean this rule is followed, red x’s mean the rule is not. A yellow warning sign means that the check was skipped because a previous required check failed.

sp-repo-review provides checks based on the Scientific-Python Development Guide at scientific-python/cookie.

Running repo-review

Repo-review supports running multiple ways:

• From the command line on a local folder
• From the command line on a remote repository on GitHub (gh:org/repo@branch)
• From WebAssembly in Pyodide (example in docs/index.html)

If the root of a package is not the repository root, pass --package-dir a/b/c.

Configuration

Repo-review supports configuration via pyproject.toml:

[tool.repo-review]
select = ["A", "B", "C100"]
ignore = ["A100"]

If --select or --ignore are given on the command line, they will override the pyproject.toml config.

Development of repo-review and plugins

This repository is intended to be fun and easy to develop - it requires and uses Python 3.10, and uses a lot of the new features in 3.9 and 3.10. It's maybe not entirely conventional, but it enables very simple plugin development. It works locally, remotely, and in WebAssembly (using Pyodide).

There are a few key designs that are very useful and make this possible. First, all paths are handled as Traversables. This allows a simple Traversable implementation based on open_url to provide a web interface for use in the webapp. It also would allow zipfile.Path to work just as well, too - no need to extract.

Checks can request fixtures (like pytest) as arguments. Check files can add new fixtures as needed. Fixtures are are specified with entry points, and take any other fixture as arguments as well - the package fixture represents the root of the package you are checking, and is the basis for all other fixtures. Checks are specified via an entrypoint that returns a dict of checks; this also can accept fixtures, allowing dynamic check listings.

Check files do not depend on the main library, and can be extended (similar to Flake8). You register new check files via entry-points - so extending this is with custom checks or custom fixtures is easy and trivial. There's no need to subclass or do anything with the base library - no dependency required.

Checks are as simple as possible so they are easy to write. A check is a class with the name (1-2 letters + number) and a docstring (the check message). It should define a set of requires with any checks it depends on (by name), and have a check classmethod. The docstring of this method is the failure message, and supports substitution. Arguments to this method are fixtures, and package is the built-in one providing the package directory as a Traversable. Any other fixtures are available by name. A new fixture is given the package Traversable, and can produce anything; fixtures are topologically sorted, pre-computed and cached.

The runner will topologically sort the checks, and checks that do not run will get a None result and the check method will not run. The front-end (Rich powered CLI or Pyodide webapp) will render the markdown-formatted check docstring only if the result is False.

Links

This project inspired Try-PyHF, an interface for a High Energy Physics package in Scikit-HEP.

This project inspired abSENSE, an web interface to abSENSE.

This was developed for Scikit-HEP before moving to Scientific-Python.