Garden exists because I want a language that is interactively editable. This is impossible to add to an existing language.
Common Lisp and Smalltalk have a superpower: you can modify running programs. You can redefine functions and modify class definitions without restarting your program.
Garden aims to provide that superpower in a conventional language design. I'm trying to provide a very familiar syntax and workflow.
Garden prioritises developer experience over language functionality. The syntax and semantics are deliberately small, so a small team (or just me!) can build a delightful, credible language experience.
Types are a nice tool to have too.
Most gradual type systems are intended for incremental adoption in a dynamically typed language. Garden is new, but requires a typing model that works with code of the form:
function foo(): String { return ""; }
function bar(): Unit {
eval("function foo(): Int { return 1; }");
let x = foo() + 1;
}
Garden supports sandbox execution without third-party libraries. This is important for two workflows:
(1) The IDE can speculatively evaluate snippets of code safely. This is convenient and enables 'mutation testing' for code completion.
(2) Libraries can declare what capabilities they need, such as network or disk. This makes third-party libraries safer to install. This will be coarse permissions, similar to OpenBSD's Pledge.
Garden splits its semantics into (1) simple evaluation that is amenable to static analysis, and (2) reflective, dynamic evaluation that is friendly to developer experience.
This means that errors in userland are plain Result types,
but unhandled errors can be resumed in a Lisp/Smalltalk manner.