iq-scm / stitch

Write reproducible reports in Markdown

Stitch

A knitr- RMarkdown-like library, in Python.

Note: You might want to consider Jan Schulz's knitpy instead. It's probably more mature at this point. However, I wanted to see if there was a simpler way of doing things.

The high-level goal of this type of library (knitr/RMarkdown, knitpy, and stitch) is to make writing reproducible reports easier.

Documentation is available here.

Examples

See the project's examples page for a side-by-side comparison of input markdown and stitched HTML.

More complex examples are linked to from there as well.

Install

stitch supports Python 3.5 and above. At the moment stitch can be installed from pip via

pip install knotr

I know, it's confusing. I've filed a claim for stitch on PyPI, but I think the people working that support queue are over-worked. Once that gets processed, I'll put it up on conda-forge as well. If you need a mnemonic, it's "I want knitr, but not the one written in R." Also I wanted to confuse R users. And knots are kind of like a buggy version of knits.

stitch requires pandoc>=1.18. This can be installed using your system package manager, or pypandoc.

Design

The goal was to keep stitch itself extremely simple by reusing existing libraries. A high level overview of our tasks is

1. Command-line Interface
2. Parse markdown file
3. Execute code chunks, capturing the output
4. Collate execution output into the document
5. Render to final output

Fortunately the building blocks are all there.

We reuse

• pandoc via pypandoc for parsing markdown and rendering the final output
• jupyter for language kernels, executing code, and collecting the output
• Use pandocfilters to collate the execution output into the document

So all stitch has to do is to provide a command-line interface, scan the document for code chunks, manage some kernels, hand the code to the kernels, pass the output to an appropriate pandocfilter.

The biggest departure from knitpy is the use of pandoc's JSON AST. This is what you get from pandoc -t json input.md

This saves us from having do any kind of custom parsing of the markdown. The only drawback so far is somewhat inscrutable Haskell exceptions if stitch happens to produce a bad document.

Documentation

Stitch's documentation has an odd build process, so standard tools like readthedocs weren't flexible enough. To make the docs, install stitch and all the extra dependencies. Clone https://github.com/pystitch/pystitch.github.io

Checkout the src branch.

Run make html.