I've seen a number of projects (often in JavaScript or Python) fail because of poor effect management—but if you don't have much experience with functional programming, you'd see it as yet another project failing because of bad development practices or lack of discipline.
With experience, it became pretty clear that what we needed were not iron-willed, self-disciplined developers but just better tools, of which effect management (ie monads) and functional programming are some of the most powerful.
OCaml less so, but the Jane Street monoculture means that, if it's not itchy for a trading firm, it's not gotta get scratched. Trading's an odd industry, tech-wise. There are aspects of the OCaml ecosystem that eliminate any of my own desire to use the language that I would have seen as neutral or even strongly positive when I worked at a trading firm. (At least during working hours.)
So, yeah, I do think that one of the big things that's missing from typed FP right now is a strong language that has similar semantics to an ML or (preferably) Haskell, but with a much more pragmatic focus. F# is really nice if .NET is an option for you, but I'd sure like to see something that is statically compiled and can produce C-friendly binaries, so that it can play nice with a wider variety of pre-existing codebases.
As for compilation time, Go compilers use very basic optimizations as such they tend to compile faster, but hardly impressive for those that ever used TP, C++ Builder, Delphi, Modula-2 and similar languages.
Can you give some examples? Really curious.
I'm guessing 8-bit strings is a minor positive, on the basis that you really don't need to deal with non-ASCII text in that kind of setting, and there's probably some performance benefit to not supporting it. Even if you stick to characters from 7-bit ASCII in your own data, if the strings are technically UTF-8 or UTF-16, then the fact that it's a variable-length encoding means that more string operations will involve costly branch instructions. And if UTF-32, your strings will be 4x as big as they need to be.
Not working on non-Unix is neutral. There's no benefit to it, but nothing was going to be running on Windows, anyway, so there'd have been no harm in it, either.
Is it really so. I am only asking. A mathematical model allows deeper insight on structure (backed by hundreds of years of prior mathematical thought), and we then sugar it appropriately for users. Just like civil engineers who learn Strength of Materials etc mathematically before they design the bridges we drive over daily.
Tooling is one of my gripes also. There was a time decades ago when I couldn't go beyond compiling the simple hello world example. This has gotten much better after many years. Same with Haskell. Is this due to difficulties in getting the necessary financial support.