[evincarofautumn]

Yeah, everyone’s got their eye on Rust as far as low-level expressive programming goes. Nimrod[1] also looks quite good:

> Nimrod is a statically typed, imperative programming language that tries to give the programmer ultimate power without compromises on runtime efficiency.

In my spare time, I’m working on a statically typed concatenative language called Kitten[2] with similar goals.

[1]: http://nimrod-code.org/

[2]: http://github.com/evincarofautumn/kitten

[yoklov]

Speaking as a C++ programmer, Rust appeals to me a great deal over Nimrod because you can turn off (or just not use) Rust's GC and still have memory-safe code.

I believe you can turn off Nimrods GC as well, but you lose any guarantee of memory safety when you do. Not to mention -- aren't all pointers in Nimrod reference counted? That's going to take a fairly significant performance toll due to cache effects alone.

[rbehrends]

Nimrod borrowed Modula-3's idea of having both traced and untraced pointers (using ref and ptr as keywords, respectively). As long as you only use ptr references, no overhead for reference counting (or other forms of garbage collection) is generated.

Of course, in order to use untraced references, you also have to use manual memory management (though you can use the macro/metaprogramming system to reduce the pain somewhat).

For traced references, Nimrod uses deferred reference counting; i.e. reference counts are only updated when they are stored on the heap or in a global variable (similar to write barriers for generational/incremental garbage collectors). If a reference count reaches zero, the reference is stored in a zero count table. At intervals, a separate pass checks if any references in the zero count table are still referenced from the stack (and does cycle detection where needed).

Deferred reference counting avoids the biggest problem with naive reference counting, which is that just assigning a pointer to a local variable (to inspect the contents of an object, say), can trigger up to two reference count updates. Conversely, with deferred reference counting, a read-only traversal of a data structure will not perform any reference count changes at all. Similarly, purely temporary allocations will not require any reference count changes, either.

[pcwalton]

> That's going to take a fairly significant performance toll due to cache effects alone.

It's not so bad because in Nimrod they are non-atomically reference counted and deferred reference counting is used to avoid touching the RC for stack references, at the cost of some minor latency for stack scanning. (Disclaimer: I work on Rust and my Nimrod info is possibly wrong and/or out of date.)

[flogic]

Even for non performance critical code, Rust is looking like it'll be a winner.

[Dewie]

Does Kitten also try to not make (too many) compromises on runtime efficiency?

[evincarofautumn]

It’s very much a work in progress, but yes, as far as the language design goes. Right now there’s a horrifically inefficient AST-walking interpreter, essentially the simplest thing that can possibly work. That gives us integration tests, a standard library, and a REPL. We are working on the design and implementation of a native runtime as free time allows.

The language looks and feels fairly high level, like a hybrid imperative-functional language. But in reality, it’s closer to (stack machine) assembly. There is a relatively straightforward mapping to x86 or ARM, so it’s easy to look at some code and mentally compile it at -O0 the way you might in C. Each Kitten instruction usually corresponds to one or two x86 instructions.

Evaluation is strict, because excessive laziness can negatively impact predictability and performance, and you can get many of the benefits of laziness in different ways. All types have immutable value semantics; sharing is essentially an optimisation, and can be implemented with a tracing GC or a reference-counting one because the language is cycle-free. And function definitions are often extremely short, so they benefit from aggressive inlining.

The general feeling is that we want absolutely as lightweight a runtime as possible—no type tagging, value boxing, runtime dispatch, reflection, garbage collection, &c. unless the programmer asks for them. At the same time, we want to avoid infecting code with low-level details such as pointers, lifetimes, and ownership. Correctness first, followed shortly by performance.

Kitten is attacking the same problems as Rust and Nimrod, but from a very different direction. And of course we are only a small handful of people working in our spare time, and I write most of the code. It would be wonderful to have some more people on board.