mul/docs/GOALS.md

goals

learning!

the primary goal with mul is for me to work through implementing a compiler, something i have long wanted to do, but never actually succeeded in.

start simple and grow

while mul's implementation is still getting started, i want to implement features atomically, one at a time, in an end to end manner.

for example, we'll start with a simple system-f/bidirectional system, implement a parser, typechecker, interpreter, and code generator, then gradually layer features onto that, supporting them at every "level" of the compiler.

orthogonal and discoverable

avoid overly redundant or unrelated functionality. make the core set of functionality well documented and easy to inspect.

inspiration:

• fish shell design

robust static type system

without going to the extent of dependent types, mul's type system should be expressive and flexible.

experiment with capabilities and/or effects

one specific feature of the type system will be a focus on safety and security through control of side effects. two interesting avenues to explore for these aspects will be capabilities and effect handling.

capabilities:

• Lambda Capabilities
• What Are Capabilities?
• Spritely - What is CapTP, and what does it enable?
• Cap'n Proto

effects:

• Unison's Introduction to Abilities: A Mental Model
• Algebraic Effects for the Rest of Us
• The Koka Programming Language
• Eff's An Introduction to Algebraic Effects and Handlers
• Exotic Programming Ideas: Effect Systems

open question: how should this design interact with a trait system? can/should effects subsume traits?

application-friendly ergonomics without sacrificing performance

mul will take cues from visions for a "smaller" rust that is able to focus on a different set of problems by not specializing in systems programming.

inspiration:

• Notes on a smaller Rust
• Revisiting a 'smaller Rust'
• Rust's Ugly Syntax

in short, we want to take parts of rust's syntax and semantics that work well and are amenable to both machine and human analysis, while leaving behind the components that are designed around control of memory layout + representation, explicit mentions of ownership and borrowing, and other more complicated concepts.

while not prioritizing it in the early implementation, mul will try to learn from rust (and its progeny) to sacrifice as little performance as possible while still offering a substantially simpler interface.

incremental, interactive, intelligent

parsing, typechecking, evaluating, compiling, etc should be usable across a batch compiler, interactive REPL, and IDE tooling. program information should be preserved where possible.

inspiration:

• Query-based compiler architectures