> Practically, this means that you can write your program in the ordinary lambda-calculus without any extra restrictions or annotations, and at the same time are allowed to control how resources are used in the program.
You can rewrite any Haskell program into the ordinary lambda-calculus, and this can be automated. The only possible problem might be that Haskell uses lazy evaluation, and I don't expect this to interfere with the control of resources but I could be wrong.
Haskell also allows you to share heap-allocated immutable data structures between threads, right? I think someone said this language doesn't support reference sharing across threads, but that could've been wrong
But as long as you're not storing closures in those shared data-structures, these structures will always be acyclic, and could be collected through simple reference counting. Perhaps that's an acceptable compromise.
cyclic, but without closure? unless closure is so broad as to capture essentially any haskell value. it's not going to be obvious when there is a cyclic reference (e.g. if you use `let x = repeat 1` instead of the definition above)
