David Thomas wrote:
With such a big speedup and not a lot of downside
That is far from proven, and frankly, almost certainly untrue. But again, it isn't the point here. There *are* case where you do need nubOrd, and those cases are common enough for it to make sense to get it into the libraries.
I'd really rather it wind up central enough to be used in preference to nub wherever possible
You are welcome to your own preference. I, and many other experienced Haskellers I know, prefer nub in many cases. Let's just make both easily available please.
(including in base!).
I wouldn't mind having ordNub in base. But let's at least get it in somewhere convenient and easily usable. If you focus the energy on trying to get it into base, this proposal will likely die the same death as all previous attempts.
Perhaps they can duplicate the minimum functionality necessary from Data.Set, internally behind the scenes?
Why should the GHC team agree to the maintenance burden of that? They have enough of a challenge keeping up with the work they already have. But you know what, give it a try. Just don't get stuck on that point if you meet with resistance.
Alternatively, maybe we can move nub out of base?
No. nub is part of the Haskell standard, so it's required to be there. Even if you disagree with my view that it should be there on its own merits. -Yitz