The Basil Programming Language
The goal of Basil is to try to have it all - high-level language features like macros, dynamic code evaluation, minimal type annotations - without any of the runtime costs normally associated with high-level languages - large runtime systems, slow performance, high memory use.
To accomplish this, Basil makes use of a novel system to compile its code. The main innovation is that the Basil compiler is actually an interpreter, able to interpret any Basil expression. When it gets to an expression it wants to save for runtime, it wraps it up and saves it to the end of the program. These runtime values propagate and result in an expression tree of all code reachable at runtime - notably without unnecessary operations between constants or dead code. Basil also statically type-checks this tree, so none of the code it ultimately compiles incurs a dynamic typing or boxing overhead.
The primary goal of Basil as a language, however, isn't just speed. Basil exists to help programmers express complex problems in natural ways. While keeping Basil totally compiled and statically-typed, we've managed to implement the following features to enable uniquely customizable syntax:
Simple, homoiconic code! Basil is a Lisp dialect at its core, and has extremely simple base syntax with no reserved words and few special symbols.
Extremely simple macros! Only one primitive for code interpolation/evaluation, as opposed to the three or so usually present in Lisp dialects.
User-defined infix functions and macros! Can have any arity (unary, binary, or more!) and custom precedence.
Mixfix function syntax! Functions can be defined with keywords in their argument lists, to help break up high-arity function calls into natural-language phrases.
The Basil compiler was implemented entirely within the language jam, almost entirely from scratch in C++. Only two libraries were used (both created before the jam):
Jasmine, a runtime x86_64 assembler and object format.
Collections, an alternative C++ standard library.
Everything else is totally original, created since August 10th.
The Basil compiler has no runtime dependencies other than libc, and the entire compiler and runtime fit comfortably in under 400 kB.
The code is first tree-walk interpreted by the compiler, using dynamically-typed Basil values to represent code. This is then lowered for runtime values to typed nodes, which are then lowered to a simple IR and then to x86_64 machine code. This code is then linked, loaded and executed just-in-time by the Basil environment.
Try It Out!
The best way to give Basil a try is to work through the interactive tour! It walks you through every core Basil feature, from the basic syntax to infix macros.
You can also try running the Basil REPL. You can reach it by exiting the tour on the project page, and then running
./basil in command line. The REPL mode features a number of commands to view compiler intermediates or generated assembly, so if you're curious how that all fits together give it a shot!
If you're curious how some part of the project works, or are looking for a list of all the language's features, take a look at our README for a somewhat informal language specification.
Support the Project
If you'd like to get in touch (with a question, suggestion, etc), the best way to reach us is on Twitter or Discord (our handles are elucent#1839 and minchy#2474).
Thanks so much for taking a look at our project! It was a lot of fun making it, and we're going to be continuing work on it past the jam - so stay tuned! :)
@xxpertHacker I deleted that part when I realized how similar to reference counting it was.
Right, there'd have to be some kind of branch. Except for situations where it could be deduced at compile-time what type of value was getting passed to the function.
What do you mean? In the example you give there, zero copies would need to be made. The function selected there would be passed a reference (the reference passed to
h), and whichever function that was would get ownership of it (it doesn't matter which one, it just matters that a function borrowed it).
Agreed, I'm interested to see how Basil handles it.
Right. Also, you can always implement static analysis in stages, i.e. start with a naive algorithm and slowly build up to a complex one.