Safety critical systems are generally real tine so Haskell is off the table. I'll also spend a lot of tine manually verifying the code, and using alot of external analysis and verification tools, so restricted C++ is fine in that case. Now, if you told me that the system was safety critical and nit real time, and the system wasn't important enough to merit lots of manual verification, then Haskell would be a great choice because its static type system is better than nothing.
Why do you think Haskell is fine as long as the system "isn't important enough to merit lots of manual verification"? That sounds puzzling to me. I'd say one thing complements the other: automatic and manual verification seems the best option.
Is lazy evaluation your biggest issue with Haskell? How about languages like ML or OCaml, which are arguably safer than "restricted C++" and do not default to lazy evaluation?
Or is GC your main problem with these languages? This would rule out most dynamically typed languages as well.
Anyways, you might want to read the stack overflow article on this subject:
http://stackoverflow.com/questions/243387/which-languages-ar...
That's a pretty non-standard view. It's as unhelpful as static typing proponents claiming dynamically typed languages such as Python are "unityped". As unhelpful as Simon Peyton Jones jokingly claiming that Haskell is the world's best imperative language :) Any of those claims may be technically true, but it doesn't help us understand, choose or use those languages. Other similarly unhelpful claims include the old "these languages are all Turing-complete anyway, so it doesn't which one you use."
So I disagree with you, philosophically, but since I acknowledge you're a very knowledgeable guy, let me ask you about your opinion:
- Would you say there is no advantage to using a statically typed language like Haskell/ML over a dynamically typed language?
- Or if you think there is an advantage, do you think the costs outweigh the benefits?
- If you were asked to develop a safety critical system, and given the choice of using Haskell (or ML if you dislike lazy evaluation by default, or OCaml if you prefer a more hybrid language) and Python (or a similar dynamic language of your choice), and every other analysis tool you can think of, static or run-time, which language would you pick? Assume you cannot pick anything else, you're given time to become proficient on the language you pick, and you cannot refuse the assignment. I know, this is a fantasy scenario, but indulge me.