But Bob Harper really makes two points, and Simon Marlow's article only covers one of them. The fist point is that parallel functional languages makes it possible to write correct parallel programs easily, by not exposing any nondeterminism to the programmer. The second point is that for such a language it's particularly easy to describe what the running time of the program will be, by talking about the depth and work of a computation.
Of course, the reason Marlow doesn't make that point is that it's not true for Haskell! In ML, the work involved in computing f(g(1)) is the work of computing g plus the work of computing f. But in Haskell that's not true, since if f doesn't use it's argument, g will never run.
Of course, the reason Marlow doesn't make that point is that it's not true for Haskell! In ML, the work involved in computing f(g(1)) is the work of computing g plus the work of computing f. But in Haskell that's not true, since if f doesn't use it's argument, g will never run.