>>13
In R6RS I believe the only major issue would be the evaluation order of operands, binding forms, and iterators when side-effects are involved. For the sake of portability the compiler would signal an error when side-effects are used in these forms. One would then tend to use many begin, or let* forms, potentially with new special forms such as Shivers' loop or Clojure style "threading" macros to relieve some of the inconvenience of this. How could you conveniently handle IO though, and what about when modifying a global is actually useful to do in a subroutine? I'm certain FP people would have some good ideas on this topic...