literary 1.8.1

Literate package development with Jupyter

Literary

TL;DR

Literary is a Python tool to make Jupyter (IPython) notebooks behave like pure-Python packages. This allows pure-Python packages to be generated from notebooks, and notebooks to be imported at runtime.

This package is an exploration of the literate programming idea pioneered by Donald Knuth and implemented in the nbdev package. Although nbdev looks to be a very mature and comprehensive tool, it is quite opinionated. This package is an investigation into what a smaller nbdev might look like.

Philosophy 📖

1. Low mental overhead
   Realistically, most Python programmers that wish to write packages need to have some familiarity with the Python package development model, including the conventional structure of a package. For this reason, I feel that it is important to design literary such that these skills translate directly to designing libraries with notebooks
2. Minimal downstream impact
   Users of literary packages should not realise that they are consuming notebook-generated code at runtime. This means that a pure-Python package needs to be generated from the notebooks, and it must use the conventional import model. For this reason, literary should only exist as a development dependency of the package.

Differences with nbdev

• Use of cell tags instead of comments or magics to dictate exports
• Use of nbconvert machinery to build the pure-Python lib package
• Use of import hooks to import other notebooks
  • Maintains a similar programming model to conventional module development
  • Reduces the need to modify notebook contents during conversion
• Minimal runtime overhead
  • Features like patch are removed from the generated module (& imported notebook source) using AST transformations
• Currently no documentation generation
  • Loosely, the plan is to use existing notebook-book tooling to re-use the existing Jupyter ecosystem

Differences with Knuth

Knuth introduced the tangle and weave programs to produce separate documentation and source code for compilation. Literary differs in treating the notebook as the "ground truth" for documentation + testing, and generating smaller source code for packaging.