compiler could detect "entry iterator pattern" in comprehensions

[jurgenvinju]

map[&K, &V] M = ...
{<k,v> | &K k <- M, &V v <- M[k]}

Here we see the typical generation of keys k from a map M, only to iterate directly again over the values v of every M[k]. The same result would be obtained if Rascal would iterate over key/value pairs from a map generator instead of over the keys:

{<k,v> | <&K k, &V v> <- M}

but that is not the semantics of the map generator...

If the compiler could detect the comprehension pattern it could directly call IMap.entryIterator and bind v and u in the same order as the comprehension does, but without having to construct all the intermediate result sets of M[k] (and looking them up in the first place).

The computational complexity of the current comprehension pattern is quadratic, while the entryIterator solution would be linear. Also the pressure on memory allocation would reduce drastically.

The current workaround is that programmers must call toRel from the Relation library. They should now about this, and also toRel has an overhead of constructing the intermediate relation which the compiler would not have since it would only allocate the entryIterator.