My answer would be lazy evaluation by default. It's extremely unusual, and I've never seen a convincing enough justification for it, and it gives rise to performance bugs (space leaks) that can be fiendishly difficult to track down and fix and are disastrous in production. With the arrival of Idris, I think we can pretty conclusively say this was a mistake and that the Miranda branch of the PL family tree is likely to prove to be an evolutionary dead-end.
My perspective on this is that lazy be default made sticking to purity much more compelling as if you just dropped print statements in you weren't sure exactly when they get evaluated. This lead to important developments like IO (they started with user input just being a lazy list! Very possible to accidentally block trying to read too much), which might not have happened otherwise.
But now, I do feel like lazy is best kept for core data-structures and places where you know you can use it, and a strict by default variant of Haskell would be better, now that we have learned most of the lessens from lazyness.
"My perspective on this is that lazy be default made sticking to purity much more compelling as if you just dropped print statements in you weren't sure exactly when they get evaluated."
And that's basically a death sentence to the feature because it reduces it to a silver lining of desperate post-mortem optimism. As far as my understanding goes, lazy evaluation was originally included in the language because it was perceived to be a powerful optimization technique enabled by pure functional programming. You reduce the amount of work a program is doing dynamically at runtime while not trading off on readability or modularity. All win. Except there were subtle trade-offs that have made themselves clear over the years and like you said, most would agree that lazy-by-default is too extreme of a feature and doesn't really pay its share of the rent at the end of the day.
My main gripe with lazy evaluation is that it's implicit behavior. It's to evaluation what garbage collection is to memory or dynamic types are to types. It hides something from the programmer that is useful to know more often than not.
> My main gripe with lazy evaluation is that it's implicit behavior. It's to evaluation what garbage collection is to memory
Are you saying you would like to return to the languages with explicit memory management? Or why is, generally speaking, GC considered good and LE considered bad?
> Are you saying you would like to return to the languages with explicit memory management?
Not really, but that does seem to be the logical conclusion at the end of the day. Haskell makes a really good case for the use of a garbage collector via pure functional programming, but I'm open to more fine-grain programmer-driven mechanisms for handling memory. I need to try out Rust's borrowing system on a meaningfully large project before I can start saying anything conclusive.
Most programmers, and especially the hipsterish HN crowd, simply cannot be trusted to write correct file manipulation code. In general, anything they produce will be riddled with race conditions and erroneous assumptions (e.g., that rename works cross-device, that close cannot fail) and that breaks in rare but possibly catastrophic circumstances.
The solution is copy-on-write file systems such as ZFS and Btrfs that ensure neither data nor meta-data are ever altered in-place and the reuse of correct file manipulation code (written by adults) rather than rolling your own, either from a library or something higher-level like SQLite.
My answer would be lazy evaluation by default. It's extremely unusual, and I've never seen a convincing enough justification for it, and it gives rise to performance bugs (space leaks) that can be fiendishly difficult to track down and fix and are disastrous in production. With the arrival of Idris, I think we can pretty conclusively say this was a mistake and that the Miranda branch of the PL family tree is likely to prove to be an evolutionary dead-end.