Contents
1 Introduction
So far, I’ve usually written posts here when I feel I have something atomic and at least somewhat unique to say, but this one will be a bit more personal, retrospective, meandering and kinda pointless. I’d like to talk about something that’s quite dear to my heart, lisp.
But first of all, I’d like to update my dear readers of this website on where I’ve been! It’s been just about a year since my last post, and longer since the last redesign of the site design (if memory serves me right, we’re on the third or fourth iteration at the moment). My photos site is updated every so often, but I rarely take many photos unless I’m travelling – I plan to change this in the near future!
I spent the summer 2024 taking the last classes I’d kept until the end, working on an SDR time-of-flight estimation project, playing lots of badminton with Sidong (now ECE PhD at GaTech) and of course – as we did for 2 years – watching anime and yapping way too much with Stephen (now CS PhD at CMU). Excluding the year I spent interning at NTT, I was at Georgia Tech for 2 years. I’d rather not comment about the state of the admin, CS classes, or housing, but I really did come to like Atlanta and the people I spent my time with there. I changed considerably from 2021-2024, broadly attributable to two experiences. One was the year I lived in Japan, which gave me a bit a of a bigger picture about life and deeply influenced my sense of aesthetics; but the other was living in close proximity with Daniel and Stephen – who influenced everything from my sense of humor to my thoughts on programming languages (ostensibly, the topic of this post). My friends at GT were honestly one of a kind and had such strong, bright and interesting personalities, I’m so glad I got to know them :)
I was always vaguely amused by cool programming languages, but I
started seriously programming lisp in my first year of undergrad to
prove to my neovim-using friends that emacs was way cooler. I say
“amused” because really I’m just interested in alternate ways of
thinking about computing, since I don’t know much about PLT or
compilers. I had briefly interacted with Haskell in high school, but
hadn’t ever written anything in it. Daniel and I learned Haskell again
together in my first year, doing the monad challenges and a bunch of
advent of code problems, although my main use for Haskell was for
configuring Xmonad and Hakyll. Daniel was (and
is) very good at functional programming, and these moments are fond
memories to me. I remember particularly well getting badly stuck deriving
liftM2
while he cruised along ahead. I highly recommend that
challenge set if you’re interested in learning Haskell. Haskell was the first time I was exposed to FP proper,
and all the abstractions built up with types and higher order functions –
things like functors/applicative/monads, arrows, zippers, etc. What
caught my interest was the way that Haskell programmers continue
making abstractions until the actual problem to be solved appears
syntactically trivial – and relatedly, why the code of some Haskell projects
are practically DSLs. But this post isn’t about Haskell, and I don’t think
any of my friends from then had any love for lisp, but Haskell
forever changed the way I wrote lisp (and all other languages too).
1.1 Some rambling
One of the many magical things about lisp is that it’s the only language that can claim to be truly multi-paradigm – because it has minimal syntax. The more syntactical rules are baked into the language (like, for instance the way Python deals with generators), the more the language guides people into a certain idiomatic way of programming. Lisp on the other hand, is only a few macro expansions away from raw abstract syntax tree. This is what makes lisp so powerful in my eyes; it can implement a huge variety of ideas elegantly. Lisps can have GC or no GC, lazy eval or strict eval, object oriented systems, imperative loop facilities, and even static types. All this power, is of course due to macros, which rewrite code at compile-time.
For instance, the commonly used when
macro from many languages
adds a new syntactical construct:
define-macro (when cond exp . rest)
(if ,cond
`(begin ,exp . ,rest))) (
Which lets you run any number of forms when a condition is true. This means you can write
(when (at-war?)display "Lauching missiles!")
(display "Attack at dawn!")) (
instead of
if (at-war?)
(begin
(display "Lauching missiles!")
(display "Attack at dawn!"))) (
Since the first one gets compiled to exactly the second one. Obivously
this is a trivial (but useful!) example, but macros become more complex when code is
generated programmatically rather than just by mechanical
substitution into a template. The loop
macro from common lisp is a
good example of another commonly used macro with a very complex
implementation. In common lisp, a for loop can be written with the
loop macro:
loop for i from 0 to 10 do (pprint i)) (
The loop macro has a quite intricate natural-language syntax.
Of course, macros aren’t unique to lisp by any means, but the degree of freedom and ease of definition of macros in lisp is due exactly to the homoiconicity; code is data, so manipulating code is just as easy as manipulating data! In other languages, like rust, c++, defining macros is limited, dangerous, difficult, and thus rarely done. In lisp, defining and using macros is natural. I will point out a notable exception to the “other languages” scope: Julia, which is remarkably close to the syntactic freedom one has with lisp – the lisp influence is markedly clear in many aspects of that language: metaprogramming, multiple dispatch, flexible typing, repl interactivity, etc.
Anyway, I become involved with the meow project in my first year of
college as well, and that marked my full immersion into elisp. My
emacs config became filled with functional commands like those from
dash
and weird macros. This was really the point of no return for
me, and I started writing much more lisp – but ran into
the classic issue: you’re spoiled for choice. There are tons of
different lisp dialects, some with tons of different
implementations. I finally settled on janet for scripting needs, it
was small, fast, and had a convenient standard library. But after
touring clojure, emacs lisp, common lisp, and scheme, I kept feeling
like I was missing some feature from some language, or preferring some
naming scheme that I was used to. This sparked matsurika, a fork of
janet with additional functions and macros wrapped up in the
binary. Essentially, every helper or abstraction I wanted to write
while actually solving a problem I’d generalize as best as possible
and put it into the interpreter instead of that particular
script. It’s very powerful to be able to modify your interpreter as
you go; and when you can do this freely knowing nobody will ever read
your scripts, and nobody will ever use your fork, it leads to
extremely concise and clean programs with a lot of hidden complexity.
Some of the macros added to matsurika include:
$
, which runs a shell command and returns the output – very useful to hack quickly, by using unix commands when convenient. For example, getting python files in the current directory:(filter |(s-suffix? ".py" $) ($ ls))
cli
, a concise way to define a main function make the args accessibles+
, a string concatenation facility with constants in scope:(s+ qt "hello" s "world" qt nl)
to print"hello world"\n
whichawk
, which runs lisp forms on every line of a string/file that matches a PEG (CFG-like grammar) – ported from TXR lisp’s awk macro->>
, chains a sequence of computations by threading a value as the last argument of each form…(->> 5 (+ 1) (- 5) (* 100))
evals to -100. However, it is sometimes convenient to change the arg order for only one computation in the chain. In my version of the threaded macro, prefixing a form with*
reverses the order of the args. So,(->> 5 (+ 1) *(- 5) (* 100))
evals to 100. I am a big fan of these threading macros, I used them first in Clojure, but find myself wanting them everywhere. My favorite macro library is swiss-arrows, which invents some new kinds of arrows with… rather odd names. I ported several of them to matsurika and use them surprisingly often. Actually, this is a good example of the contrast between a lisp-enabled “abstraction by rewriting” approach and a traditional fp “abstraction by higher order functions”.->>
can be easily interpreted asfoldl const
over a list of partial functions, or composing partial functions. This has effectively been turned into a new convenient syntactical construct with a macro. Similarly, the “Nil-shortcutting diamond wand” (??) from swiss-arrows, which ends the chain early if any intermediate value is nil, is equivalent to chainingMaybe
computations with>>=
in Haskell.
I’ve written a number of scripts that use reasonably often in
janet/matsurika, and in general it’s been fun. However, I don’t think
that choosing janet was the right choice in retrospect. This is for a
number of reasons. First, janet (and by extension, clojure) is already
too opinionated, and is not a good base to mold to your tastes. I like some of those opinions (for
instance, the PEGs, the table syntax) but don’t like some
others. Second, the maintainance cost of having to hack on the janet
source code, combined with the fact that since its forked i will need
to periodically rebase to get the latest changes (with manual merge
conflicts), turned out to be nontrivial. Finally, needing your own
deranged binary to run your scripts is a bit awkward. One of Janet’s
biggest differentiators, and indeed a project goal is that it’s small,
written in C, has no dependencies, and is embeddable. My goals include
only “small.” Everything new in matsurika could easily have
implemented as a library providing new macros and functions. In the
near future, I plan to implement this for either r5rs scheme or
racket. Racket seems particularly appealing, since it has explicit
support for other lisp dialects using the #lang
keyword. I do enjoy
clojure as a language, but for mostly superficial reasons: it tends to
encourage a stateless pure FP style, and the standard library is pretty
good (batteries included). However, I’m not a huge fan of some of the
modern clojureisms like the square brackets; one might argue that
clojure is not a lisp at all because the code is not linked lists and
there are no cons cells; one of the minimal specifications for a lisp
according to the original paper by John McCarthy.
2 Other
I graduated college in August 2024 – and I’m now doing my Masters at the Institute of Science Tokyo (formerly Tokyo Tech). I love it here! I’m working on using using diffusion models with wireless data at Nlab. I’m living a lot slower than I did the last time I was here; there’s not much local tourism left to do, and it feels normal rather than magical as it once did. I bike a lot around the city, and it’s become one of my favorite hobbies.
I’ve been using emacs for a long time now, but lately I’ve been keeping my on lem; I think it’s a matter of time before I switch (probably after I port the core of meow to CL). I no longer believe as strongly as I once did in the future of emacs, but I do still feel that my current keybinding scheme on meow’s editing model is really close to optimal for me. I’m sure that emacs and its religious users will continue to hack away underground long after the nukes fall and wipe out surface life, but there are fundamental flaws that need to be fixed:
- emacs-lisp is really not that good
- decades of cruft has led to bad performance
- the epic long lines problem
- single-threaded
- the epic GC hanging problem
- relatively slow start up time
With a project of emacs age and popularity obviously there have been a number of attempts to hack it: GNU Emacs is itself a reimplementation for one (1984), Lucid emacs (late 80s), emacs-ng, remacs, commercial-emacs, etc that I’m probably forgetting. I like lem mainly because it makes the step of finally ditching emacs lisp for common lisp. It’s much better suited for developing editor packages, and cl compilers are more performant. I think of lem to emacs as perhaps neovim to vim; a tight, modern reimplementation that doesn’t forget the culture and soul of the original project.
No matter how much I wax about lisp, I write mainly python on a day to day basis. That’s why the “soul” (as I like to call it) of the language/ecosystem and the experience of writing in lisp is so important to me; it’s my reprieve. I’ve wasted more time than most readers could possibly imagine trying to convince people that lisp is the best programming language ever (true), literally goated (also true), alien tech from the future (so timeless!), divinely inspired (it’s said God came to JMC in his sleep) etc but it really doesn’t matter. What matters is that writing lisp is truly fun! It’s a joy to iterate and organically build up a solution, testing as you go in the repl, precisely manipulating the code with sophisticated tools (structural editing! paredit!). Lisp dialect tooling (especially CL, Schemes, Clojure) is blissful to use – the very first language servers were for lisps! The monkey-wrench move-fast-break-things attitude encouraged by dynamically typed lisp combined with the patterns of interactivity, self-documentation, and hot-swappability is to me, at the very core of hacker culture. If you spend enough time with lisp, the parentheses fade out with the stars and you’re left to admire the raw, pulsating heart of computation.
As Stallman puts it:
The most powerful programming language is Lisp. If you don’t know Lisp (or its variant, Scheme), you don’t know what it means for a programming language to be powerful and elegant. Once you learn Lisp, you will see what is lacking in most other languages.
Y F = F (Y F)