How much of the wheel to reinvent?

Sticking with the tried and tested

I like to produce my own material (notes, custom worksheets, question compilations, etc) and for years my tool of choice has been $\text{L}\text{A}\text{T}\text{E}\text{X}$ in TeXnicCenter.

It did a decent job for my purpose and sensibilities then with just enough GUI elements, syntax highlighting, one-click compilation to PDF and error logs. And it became my workhorse for the longest time. An unchanging workhorse, given its last release in 2013, and that I have used it regularly until at least 2020.

Discovering a more modern approach

I tried venturing into web development in 2020 and eventually discovered VS Code. Having been working with text-editing experience not that different from Notepad++ from the early/mid 2000s, the text editing experience in VS Code felt so revolutionary. How have I been working the past decade plus without these tools?!

And with the discovery of the LaTeX-Workshop extension along with personal customizations via snippets I enjoyed a huge productivity boost.

But yet the increased productivity only left me wanting more: $\text{L}\text{A}\text{T}\text{E}\text{X}$ produces such beautiful output documents, but the language and syntax feels a bit clunky at times. Moreover, I was starting to experiment with code generation ideas to further enhance my previous workflow of alternating between calculation tools (calculators, CASes and the trusty pen and paper or chalk and board) and markup/typesetting tools to produce content.

Before I knew it, what started as a primarily hobby project to make a math app/website became much deeper experiments in developing tooling to support the creation of mathematical content.

Reinventing the wheel, V0

As the major tools that I know of (late 2000s versions of MATLAB, Mathematica, SymPy and Markdown) all differed significantly from my personal goals, I decided to build my own, with very little external dependencies.

The prototype Mathlified framework builds on top of

• JavaScript/Typescript for control flow and general purpose programming
• HTML output via KaTeX
• PDF generation via $\text{L}\text{A}\text{T}\text{E}\text{X}$

These are then coordinated via the following custom libraries

• Mathlify CAS to handle mathematical calculations and manipulations
• Mathlifier as a KaTeX wrapper to simplify the generation of mathematical markup within JavaScript template strings
• Mathlifier2 to product LaTeX strings instead of KaTeX HTML output
• Vite-plugin-sveltekit-tex to coordinate everything on top of Vite's and SvelteKit's infrastructure. A JS/TS file is then all we need to have a Svelte webapp/webpage up and running, along with the generation of $\text{L}\text{A}\text{T}\text{E}\text{X}$ markup and PDF.

Changing into more modern wheel hubs

MathJax and KaTeX into Temml

I was blown away when I first learned about MathJax from Prof Chan during a complex analysis lecture. Up to that point, displaying beautiful and professional mathematics (ie not psuedo-markup like x^2 or images) means providing download links to pdf files.

I soon switched to KaTeX for faster performance (though MathJax has since caught up), but the project has started to feel a bit stagnant and bloated in recent years. With MathML getting better support I've now decided on Temml as the way forward to generate web-facing mathematics.

Markdown to djot

Meanwhile, there are the other common text markup features (eg emphasis, paragraphing and section headings). $\text{L}\text{A}\text{T}\text{E}\text{X}$ and HTML tend to be a bit too verbose for content-heavy writing. Markdown is awesome for most applications, but trying to create more dynamic math-heavy content around it has led me to discover edge cases that are difficult to solve.

I have initially settled on a 'markdown-lite' custom syntax, but I discovered Djot and realized they solve many of the features I will want (in addition to the usual paragraphing and style markups, the ability to add custom blocks and attributes is very promising).

So, instead of totally reinventing the wheel with custom syntax and parser, we can piggyback on top of Djot. Mathematics markup is the main part our framework is focused on so we will provide custom functionality but Djot can handle the rest (in particular the intermediate AST representation) in between markup and final output.

In a sense, it feels like we started out reinventing the wheel (with our custom syntax), but then discovered someone else had a better wheel and decided to modify theirs.

Typst

I recently rediscovered Typst and am very impressed with their project, especially the showcase on the site. The seamless mix of general programming control flow along with markup for content is what I have been working towards in Mathlified.

Typst vs Mathlified

The biggest design difference for me is that Typst is markup-first, with escapes to general programming when necessary, while Mathlified starts with programming (the underlying JavaScript) and adds the components (CAS, markup handling) on top of it.

If I had access to Typst a few years ago it may well have developed into my main tool in creating mathematical content. And if it performs well there will have been less impetus to develop something like Mathlified.

But as it stands, I think there are sufficient differences in the two projects for me to continue plugging along on Mathlified.

First, Mathlified can be thought of as web-first, translating more naturally to HTML. Out framework then transform our input into $\text{L}\text{A}\text{T}\text{E}\text{X}$ before compilation to a PDF. Typst's custom rust-based compiler seems really promising and circumvent some of my problems with $\text{L}\text{A}\text{T}\text{E}\text{X}$ while retaining many of the pros. Unfortunately, using Typst will then necessitate a new rethink about how to handle web/mobile.