ReML

ReML extends Standard ML with explicit regions, effects, and effect constraints. It combines MLKit’s region inference with annotations that let you express where values are allocated and constrain how functions use memory. Standard ML programs are also ReML programs, so annotations can be introduced where you need more control.

Regions and effects

  • Explicit regions let you name regions and specify allocation in source code.
  • Region and effect parameters describe how functions use memory.
  • Effect constraints express relationships between effects, including mutation effects, and support reasoning about parallel computations.

ReML supports parallel threads. It currently runs without reference-tracing garbage collection; memory is managed through regions.

Getting started

ReML is included in the MLKit sources and distribution. Compile a source file with the reml executable:

reml example.sml
./run

For example, save this program as example.sml:

fun down `r (n : int) : int list`r =
    case n of
        0 => nil
      | _ => n :: down (n - 1)

val first =
    let with r
    in hd (down `r 5)
    end

val _ = print (Int.toString first ^ "\n")

The list is allocated in the local region r. Only its first element, an integer, leaves the scope; the region holding the list can be reclaimed when that scope ends. The program prints 5.

Learn more