Like most developers I suspect, I enjoy having a side project to work on during vacations. Almost always it’s something completely different from the day job; something I’m interested in, or never done before, or allows me to learn and explore something new. That’s the major attraction for me to software development: there’s always something new to learn. It’s also why my attention is span is like that of a labrador retriever’s in a field full of squirrels. My subconscious can never commit to which squirrel to chase for the next five minutes, let alone day.
This Christmas vacation is no different. Although I’m spending it with my immediate family, we’re all introverts (save Dad), so there’s plenty of down time for side projects. During these times, I usually work on a project from a set of domains that I, for whatever reason, find most interesting. And one of those domains happens to be compilers. I’m not sure why I settled on that this time, as opposed to, say, an RSS sync engine. Maybe my subconscious likes alliteration.
I’m not a compiler expert by any stretch of the imagination, but I’ve written a few of small ones before. They’re like a data structures winter wonderland. (Or maybe like a fat, slow moving squirrel, my subconscious helpfully points out.) I’ve read through the LLVM Kaleidoscope tutorial many times before and even used LLVM itself as a backend for a (half finished) Pascal compiler (codenamed “Pasco” for “Pascal & Company” or after the noted guitar amplifier photojournalist). So as I was reading through the tutorial recently, I thought: what can I do with this? It’s actually a great step-by-step tutorial for building a small, but interesting language, regardless of the chosen compiler toolset. But what toolset? Erlang seemed the most interesting, because my conscious self hates alliteration and/or my subconscious.
I taught myself Erlang a couple of years ago, I think. It was an easy dip into the functional languages pool, just to see what that was all about. Erlang has strict evaluation, a simple syntax, and a lack of pretentious academics (also known as monads), which made learning it easier than, say, Haskell. I love the actor model of concurrency and BEAM (the Erlang VM), but always wondered if there were a way to put a more powerful front end on it. (And apparently I’m not the only one who’s wondered this, given the existence of Elixir and other BEAM targeted languages.)
So that’s what I did. I used the Kaleidoscope tutorial to gradually implement the Kaleidoscope language in Erlang, targeting BEAM. Since the language didn’t map perfectly to the new runtime, I renamed the compiler and language to kalerl, which can be found on Github. (kalerl is pronounced “Kal-Earl”, like Superman’s redneck cousin. My subconscious thinks that’s hilarious.) I learned a lot from this exercise, mostly about generating Erlang parse trees, which are called “abstract forms” for some reason. I also had a lot of fun, a) because of learning of new stuff and b) because there’s something strangely satisfying about a tool that takes a programmer’s textual ramblings and spits out an executable binary. There’s magic going on inside of it, and I made that magic.
I’m contemplating writing a series of blog posts about the project (my subconscious likes love and acceptance, and has deluded itself into thinking it is achievable through these means). The difficulty is deciding how to break it up into digestible chunks, and figuring out what is interesting that someone (other than my subconscious) would consider worthwhile. Then I need to find the time and energy to do it, which is severely lacking in my current phase of life. But here’s to optimism, blog posts, and fat, slow moving squirrels.
(P.S. My subconscious also likes parenthetical asides and unedited prose, but I’d wager that’s obvious by now.)