Lambda the Ultimate

You probably had to throw out closures, tuples, datatypes. Not much left of ML then.

Excuse my ignorance, but why couldn't have GC on a microcontroller? A simple GC is really tiny, and since you probably control all the memory on a microcontroller anyway, it should be no problem to integrate one.

...a ML compiler for simple hardware without building a decent optimising compiler.

That's my thought.

For example, you'll need to be able to perform certain analysis (what is plural, btw) on the program to tell that some values are linear (as in linear logic - google for it) and could be kept on stack or malloc'ed. Or perform fully fledged region inference (and refuse to compile programs that cannot infere finitie regions).

I plan to implement an GC (to get a real ML)
but I want to understand what is posible without
it because some functions (interupt-handler for example;
the GC itself...)
should not dynamicaly allocate memory.


linear values... I'll google for it.


The target of my language will be an stack-based vm,
a mixture of (static)forth/ocamlrun/tiny-smalltalk(perhaps).


Throwing out closures especially currying will help,
I hope to pass tuples on the stack...

Supposedly: http://www.it-c.dk/research/mlkit/

Never used it personally, though.

you could do something like jhc?

Take a look at Pre-Scheme.

"Pre-Scheme: A Scheme Dialect for Systems Programming"

by Richard A. Kelsey.

http://citeseer.ist.psu.edu/kelsey97prescheme.html

Abstract: Pre-Scheme is a statically typed dialect of Scheme that gives the programmer the efficiency and lowlevel machine access of C while retaining many of the desirable features of Scheme. The PreScheme compiler makes use of type inference, partial evaluation and Scheme and Lisp compiler technology to compile the problematic features of Scheme, such as closures, into C code without significant run-time overhead. Use of such features in Pre-Scheme programs is restricted to those cases that can be compiled into efficient code. Type reconstruction is done using a modified Hindley/Milner algorithm that allows overloaded user-defined functions. All top-level forms in Pre-Scheme programs are evaluated at compile time, which gives the user additional control over the compiler's partial evaluation of the program. Pre-Scheme has been implemented and used to write a byte-code interpreter and associated support code for a complete Scheme implementation.

/Jens Axel Søgaard

Its mentioned in regards to MLKit, but still worth mentioning on its own. Regions are a way of statically (and typefully) allowing the programmer to explicitly do dynamic memory allocation in a safe manner. I'd suggest checking out the various papers by folks such as Talpin, Morrisett, and several others. There is also a chapter in ATAPL on the subject.

Hope that helps