Working on a Parallel Map Operator for Heron
Map is a higher-order function that transforms a list into a new list by applying a given function to each item in the list. I want to have a version of map built into Heron, which can be automatically parallelized by the compiler.
The map function is terribly useful, and can be easily implemented in many modern languages using generators.
In C# we could write the map function as follows:
IEnumerable<T> Map(IEnumerable<T> xs, Func<T, T> f) {
foreach (T x in xs)
yield return f(x);
}
Given generators it might seem that map is redundant. The reason I like a map operator built directly into a language is that there are opportunities for the compiler to perform list optimizations if the function is known to be stateless and free of side-effects. The most important optimization being: parallelization.
So I've decided to provide a map operator as part of the concrete syntax of Heron as follows:
var ys = mapeach (x in xs) to x * 2;
The only thing that concerns me is the fact that I have no guaranteed way to assure that programmers don't try to squeeze effects into a mapeach. On a single threaded version, no problems. However once a compiler decides to parallelize the map (e.g. because the input list is sufficiently big) it could cause some problems for users. Only time will tell if programmers can get used to this method of programming.
I'd love to hear your thoughts on the syntax and any ways to make this easier and safer to use.

