GNU epsilon

Speaking of experimental software, after working on a private repo for a long time I’ve finally pushed the GNU epsilon sources to Savannah: http://savannah.gnu.org/projects/epsilon. The software is definitely not for the faint of heart in its current state, but fellow Lispers and language jocks might already have some fun with it.

bzr branch bzr://bzr.savannah.gnu.org/epsilon/trunk/

I hadn’t make any public announcement before. I created the public branch on Savannah right before my impromptu epsilon talk at the recent Düsseldorf GHM (http://www.gnu.org/ghm/2012/ddorf), showing it to the people who were physically present. Now you know it’s there as well: play with it if you want, and please let me know if you have any constructive comments.

I plan to use this space to speak informally about my practical epsilon hacks, so that people can see how things are done in practice. Or just me hearing myself speak if you will, which is useful anyway to gauge the presentation style. Think of this as a preliminary form of hands-on documentation, before the software is stable enough for me to start writing the real Texinfo documentation.

My forthcoming PhD thesis will also be useful for understanding epsilon, including its philosophical rationale. Unfortunately the document has not ended up as universally accessible as I would have liked. I suppose there isn’t much I can do: the system is simple but not trivial, and some formalism is ultimately necessary. I hope the non-mathematical part will be of interest to a broader public.

epsilon is a very, very small language. But it’s surprisingly hard to introduce informally because of its idiosyncratic nature, which needs many preliminary explanations.

My short-term to-do list for epsilon contains a lot of fun items:

• implement an s-expression frontend, so that I don’t depend on Guile any longer. I’m not sure if my old s-expression frontend written in ML is worth translating and adapting;
• implement ML-style pattern-matching;
• implement a faster runtime not relying on Guile SMOBs, with unboxed fixnums;
• implement a yet faster runtime with no boxedness tags;
• finish my quick-n-dirty bytecode compiler;
• implement a quick-n-dirty native compiler, by generating MIPS and x86_64 assembly from the bytecode form. I think that this naïve version might already be faster than slow bytecode engines, such as CPython;
• optimize free variable computation by avoiding the quadratic algorithm representing variable sets as lists;
• clean up my CPS transform; even better, trash it and just implement delimited continuations instead;
• implement library functions for fixed-point rationals, including I/O;
• implement buffered input and output — I can translate my ML prototype if needed;
• integrate my parallel mark-sweep BiBOP garbage collector;
• make the garbage collector generational, by adapting the semispace collector into a young generation and the parallel mark-sweep BiBOP thing into an old generation;
• general-purpose expression optimizer by term rewriting.

In some unspecified order.

Tags: english, epsilon, gnu, hacking, meta