Tooling for Snailz

I built a synthetic data generator last year to use in a “software design for data scientists” tutorial that I never finished writing. Over the last week I rewrote it as a way of exploring some new tools:

• The command-line interface to Claude. I’m disgusted by the amorality of the AI industry, but am now convinced that the coding tools are here to stay: writing tests and refactoring code with Claude’s help was much faster than doing it by hand.

• uv for managing the packages and virtual environment and for running commands. It’s the first time I’ve used it as itself rather than running uv pip whatever; never going back.

• ruff and pyright for checking the code. ruff is a joy, but checking type annotations with pyright cost me a couple of hours. It’s not the tool’s fault, though: figuring out how to annotate the last 5% of the code led to me writing my first protocol and then throwing up my hands and replacing it with Any.

• pydantic for storing and validating records, including the data that snailz generates and the parameters used in generation. I started with dataclasses, but switched as soon as I found myself¹ adding methods that pydantic already had.

• doit to run commands instead of Make. I’m more comfortable with the latter, but I recognize that writing Makefiles is a dying art.

• mkdocs for documentation instead of Sphinx or pdoc². I find mkdocs easier to work with than Sphinx, and pdoc doesn’t support “extra” Markdown pages as well as mkdocs. (If the folks at Astral are looking for new product ideas, a better documentation generator would have at least one paying customer the day it launched…)

• pytest, pyfakefs, and faker for testing and click for building the command-line interface instead of argparse.