Bless you! Standard ML is the closest thing we have to a perfect language, in my opinion. My only wish would be a solution to the non-extensible overloading issue, but to fix it without also fixing the formal semantics around it feels like throwing the baby out with the bathwater, and is why I haven't gotten into Felix or Mythryl.
I mostly use Haskell myself. I find myself missing the laziness and purity when I use Standard ML, but I think the biggest obstacle for me is the shortage of libraries. I'm happy to use a language without a community, but compared to Haskell it feels a bit like a ghost town. Such a shame, it really is fantastic and would have been the perfect tool for building huge systems on.
Yes, the overloading issue requires changes to both the parsing rules and some fairly tricky formal semantics changes as well. I've talked with Dave MacQueen about it in the past, and it didn't sound like something he was terribly averse to, just that it would take a long time to really get _right_. And then it places a burden on the SML implementations, though at this point that's really SML/NJ, MLton, and possibly Poly/ML (though I don't know that team very well yet).
Unfortunately, the SML implementor community largely either moved on to other research projects after the implementation-y papers had been mined or left to work in finance (interestingly, it seems to be about 50/50, with the exception of a Googler or two). We don't really have anybody in research labs to do the kind of amazing full-time implementation, support, evangelism, and publication work that the Simons do so fabulously for Haskell!
That said, as a researcher, having many more SML users would require us (particularly Manticore) to Get Real in ways that are not on the path to publication. And considering I'm already only going to have 7-ish publications when I graduate, I see a probable future locked behind the doors of Big Finance for myself as well...
Apart from the obvious Jane Street example, quite a few. F# is used pretty widely at this point in the Chicago area. Many of the larger banks either have their own private dialect of ML or a strict, ML-like subset of Haskell (Standard Chartered).
But for most it's not their primary language; those roles are usually still filled by C on the server and some GUI-friendly language for the trader apps. It's mainly used less latency-sensitive work that is more financial modeling in nature.
Sorry for being vague with names; other than Jane Street and Standard Chartered, most of them will only talk about it confidentiality. Many take secrecy to the point of lunacy.
I mostly use Haskell myself. I find myself missing the laziness and purity when I use Standard ML, but I think the biggest obstacle for me is the shortage of libraries. I'm happy to use a language without a community, but compared to Haskell it feels a bit like a ghost town. Such a shame, it really is fantastic and would have been the perfect tool for building huge systems on.