• Public

  • Public
  • Network
  • Groups
  • Featured
  • Popular
  • People

The current outcome is extremely frustrating. While the blind spots here were real, we were driven by the requirements of the academic research community to spend nearly three years finding a way to do complete inference over mutability. That was an enormous effort, and it delayed our recognition that we were sitting on the wrong kind of underlying type system entirely. While I continue to think that there is some value in mutability inference, I think it's a shame that a fairly insignificant wart in the original inference mechanism managed to prevent larger-scale success in the overall project for what amount to political reasons. If not for that distraction, I think we would probably have learned enough about the I/O and the instance coherency issues to have moved to a different type system while we still had a group to do it with, and we would have a working and useful language today. The distractions of academia aside, it is fair to ask why we weren't building small "concept test" programs as a sanity check of our design. There are a number answers, none very satisfactory: Research languages can adopt simplifications on primitive types (notably integers) that systems languages cannot. That's what pushed us into type classes in the first place, we new that polymorphism over unboxed types hadn't seen a lot of attention in the literature, and we knew that mutability inference had never been done. We had limited manpower, so we chose to focus on those issues first...