[lmm]

> Would such a language be viable? Or is it absolutely necessary to catch all possible type errors. In the past, tools like lint have been considered useful, although lint is not at all the kind of type system that I envision, namely a system that is in practice catching all the errors that ML catches -- it just doesn't (and cannot) guarantee it catches them.

It's entirely possible - I believe Dylan implements something on these lines. IMO it's harder to reason about than a consistent type system with "escape hatches" like unsafePerformIO, and about equivalent in usability.

> The restrictive nature of static type systems today is legendary, but I wonder sometimes whether people realize how restrictive they are. If you can't make your idea work with functors or typeclasses, chances are good there's no way to get it to compile and you have to write functions in longhand.

I have yet to find a piece of "good code" that I couldn't implement in a typesafe way. I've occasionally found code that worked but couldn't be made typesafe (e.g. the "big global Map<String, Object>" antipattern), but it's always been code that I would have wanted to rewrite for readability and sanity anyway.