This page is a resume-like overview of my open-source work. It describes the purpose of each project and the extent of my contributions. Most projects also have an adequate README file, but these are intended for a typical user of the project and may require niche knowledge (e.g. of J) to understand.
All code designed and written by me.
BQN (BQN, mainly)
BQN is my second major programming language, driven by the frustration that no one seems yet to have built a “sensible” APL derivative: one that maintains the core array programming ideas but accounts for well-understood best practices in array theory and general programming language design. K is perhaps the nearest fit aside from the unmaintained A+, but it doesn’t use multidimensional arrays at all, and doesn’t make outer scopes visible from inner ones, leaving a lot to be desired by these criteria. BQN is not without problems but seems to do pretty well at the basics.
The BQN implementation uses the array-based compiler paradigm pioneered by Co-dfns, extending its scope from only the core compiler to scanning, parsing, and code generation as well. The documentation is also converted from markdown to HTML using similar techniques. Because so much of the language is self-hosted, it is very easy to port to new platforms, with the drawback that so many implementation layers leads to a very slow runtime (improvements to the compiler can probably fix this, with substantial effort).
To make learning BQN easier, I cloned Adám Brudzewsky’s APLcart, translating (some of) the APL examples to BQN. The website code is all Adám’s except for small modifications.
I is an experimental (avant-garde, even) programming language inspired by J. I began creating it after the stunning realization that my two favorite features of J—rank and forks—were actually the same concept! A paper describing this idea, a generalization of what Haskell calls the Functor typeclass, is shown in the I repository. It assumes no knowledge of any specific language. The language uses a number of other interesting ideas which are described in its documentation and guided examples.
Some features of the implementation include:
- Memory management implemented with reference counts.
- A soft-typing framework built into the language. Any type can be a combination of elementary types, in which case values of that type are type-value pairs.
- JIT compilation targetting x86 directly for pretty large portions of the language.
Headphone reviewer Crinacle had measurements of hundreds (like, a lot of hundreds) of in-ear monitors with no easy way to put two of them in the same graph. I volunteered to make a comparison tool to do just this, and it’s now a popular feature of an increasingly popular site, with audio enthusiasts using the screenshot tool to share nicely labelled comparisons on a daily basis. In addition to the README, this project has pretty extensive documentation which really should stand on its own but you’re obviously here to read about me so I’ll get in a few bullet points worth of bragging that don’t belong in the docs:
- That smoothing spline’s no easy feat. I wrote my own diagonal LDL decomposition and solver. The results are noticeably cleaner and more faithful than typical octave smoothing.
- Normalizing the headphones requires solving what volume would give you a specific target loudness. How to find this? Newton’s method. I took the derivative of ISO 226:2003.
Crinacle helped guide the design, and I also took input from Discord user space_wadet, who is a professional UI designer.
A port of my J music programming library to BQN. It’s powerful enough for real production work, but fairly slow because key features like filters, FFTs, and data encoding are implemented in BQN rather than a compiled language.
Synthesized music implemented in BQN. Nothing too polished: my more serious publications are on Bandcamp and as of 2021 have all been mixed in J.
This is a half-compiler, half-preprocessor which creates LaTeX using J code. It replaces J operations with symbolic counterparts to build a syntax tree, and then turns the tree into LaTeX code. A particular design goal is to make the output code look natural even when the input is idiomatic J. The converter uses knowledge about order of operations and mathematical convention to parenthesize only when necessary, and it is very rare to need to add or remove parentheses while using JtoLaTeX. Addons provide other utilities like matrix handling, uncertainty calculation, greek and other symbols, and set notation.
The base code and the more important addons are well documented in the
These are the utilities I use for music mixing and mastering, as well as some other tools for working with sound. Features include a tool for reading and writing wave files, a large collection of IIR filters, and sound synthesis tools.
Since this is more of a library than an application, documentation is provided only in the form of comments in the source files.
Various J scripts not large enough to merit an entire repository.
Projects that I have contributed to significantly. My contributions are noted.
Singeli (Java, BQN)
A domain-specific language designed for building fast array languages. Designed in collaboration with dzaima, who built an initial Java implementation. I’ve been working on one in BQN that fixes some deficiencies, in particular allowing custom operator definitions.
A school project that I and three other students designed for a class at UNC, with input from a small team at IBM. I am now the sole active maintainer. This does not mean I want to work in web dev.
Projects that I have changed, but only to a small extent.
J programming language (C)
I made some contributions to the J source code around 2016 and have since stopped (I am now using BQN rather than J for my own programming when possible).
- jsource: the official Github mirror of J’s source.
- unbox: a fork of the J source code. I prefer to work with it over the official J source since the build system is nicer, and my changes appear here first.
- openj/core: defunct. I keep it around because there are still some changes which I have not added to unbox.
A project which aims to unify research on the fast multipole method so that variations on the algorithm (in particular, different kernels) are easily comparable. I made several modifications to exafmm for my master’s thesis and am working on merging some of them back in.
The terminal I use. The only significant change from the code distributed by suckless is the addition of xft font rendering.