Why? This doesn't agree with my experience using Java or Go at all. Both languages have ways to reuse objects that significantly decrease GC pressure for bottlenecks to achieve competitive performance. Also anyone running net services at scale frequently runs into performance issues that are much more thorny than misplaced quadratic functions (are you deserializing JSON in a hot loop? is your HTTP connection pool too large? is your hash table too large, leading to significant collisions?)
Rust or C++ is overkill when reusing some objects, moving data structures, or memoizing some serde is all you need. Reasoning about them in Haskell can also be quite difficult, especially if you build a lazy chain of operations that leads to executing an expensive operation frequently in a hot loop, or forget to preallocate HTTP connections in a pool, or something like that. It's a high cost to pay for a better type system and awkwardly unifying abstractions.
I agree with that, but the consequence of these things is that developers don't have to care about it themselves. Thus these languages are usually good enough that optimization is not a concern for most devs in these ecosystems.
> (are you deserializing JSON in a hot loop? is your HTTP connection pool too large? is your hash table too large, leading to significant collisions?)
None of these things are language-related, though.
> Reasoning about them in Haskell can also be quite difficult
We didn't have specific trouble when this kind of problem came up, maybe your mileage is different ? Being used to it, it's kind of hard to understand what specifically you find hard in this.
And I spent way less time learning Go than learning Haskell.
...And you've got what you paid for.
I've read GHC assembly and it seems to be intentionally trying to run poorly - eg it calls everything through unpredictable function pointers. No, this isn't CPUs' fault, as GHC people do try to be optimal, they are just failing. jhc produced good code but isn't maintained afaik.
> Yes, given the fact that it's praised to such a high degree, users might have extremely high expectations
May I suggest the problem is not with the language, then? As a fellow HNer that you can trust, I also have an extremely well-made bridge to sell you if you're interested.
> Because performance problems are particularly hard to solve in Haskell due to its high level of abstraction.
That doesn't match with my experience using the language. But again, how often do you have to solve performance issues that are more complex than a misplaced quadratic function in other languages? If it's frequent enough, wouldn't you consider using a language like rust or CPP rather than something like Java, Go or Python?