I'd probably look for something very different from Arc -- something at least partly complementary. However, I don't know what that would be. What language would at least somewhat complement Arc but not be a low-level language like C?
I wouldn't recommend Ruby because I was not happy with the state of their docs and it didn't seem different enough from Perl or Python to warrant a major effort.
I wouldn't recommend Perl, because Perl is mostly for getting stuff done. If you want to get things done, use Perl, but if you want to learn a language for the intellectual gratification of learning something new, Perl might not be the best for that.
Java is mostly a verbose Python, so I wouldn't recommend that either.
Python works well enough. Not terribly interesting, imo.
Erlang seems pretty heavily focused on concurrency, so unless you need that I probably wouldn't specifically recommend Erlang.
I've heard Lua is simple and tiny. That might be interesting.
I've started haskell. I chose haskell because it appears there is a good balance between performance/documentation/libraries. I already know one language that can be slow, so I think learning a language that's blazing fast will balance the mix. And although Python/Ruby have better docs/libraries - they are much slower than even scheme/arc. Erlang docs are horrific. IMHO.
You came to pretty much the same conclusions I did. I'll probably try out Haskell too. It fits nicely with the kind of side-effect-free-even-to-the-point-of-monads-and-arrows programming I try to do in other languages anyway, and the main thing keeping me away from it has been a perception that it had a really slow implementation.
What I had ended up voting for in the poll though was Clojure. I figure it's lenient about types, it's very reliably cross-platform ('cause of all the business interest in maintaining Java), it has better documentation than Erlang, it can still communicate with C and Erlang (as pretty much any language can, apparently), and beyond that, it has a language-level focus on communicating with other JVM code, which should give you good access to lots of other well-worked-on codebases. As far as I can tell, once you've got access to C and JVM libraries, Python libraries are the next gold mine, but I don't know how to access those libraries except through Python somehow (like through system calls or something), so it didn't affect my decision. (Incidentally, for many of the same reasons I recommend Clojure, I'm hooked on Groovy, but I did consciously try not to let my own criteria make a difference.)
In any case, yeah, now that I see it as fast, Haskell looks great. :-p
I wonder if it's possible to build the beautiful arc syntax as a layer upon haskell. Probably not doable ~ macro's/expansion and all, but still - would be awesome.
Liskell sounds very close to what you're thinking about. Also, the "Write Yourself a Scheme in 48 Hours" Haskell tutorial might interest you. I think this is a pretty popular train of thought. ^_^
That's actually where I'm starting with Haskell. It turns out Haskell is a much more natural home for Blade (my pet lisplike) than any of the other languages I've tried: JavaScript, Groovy, Arc, and Blade itself (with an intent to port/bootstrap it). It's almost scary!
