There is also the saying that Haskell is the world's best imperative language, and that is not without reason, since you can re-use its usual power of abstraction while writing imperative code too.
It's awkward, it has a bad ecosystem, it has just as many issues with hidden state, etc...
The only reason you would want to write imperative Haskell over a modern imperative language is the typesystem, but even that can be an issue sometimes, and I'd risk saying that the Rust or C++ typesystems while being very slightly less expressive are better at scale overall due to their flexibility imo.
Another big one is that it really sucks to debug. Seriously, it's brutally slow compared to what you can do with a good debugger in other languages.
If you care about performance, Haskell can be significantly worse than others, both lower level compiled languages (lazy evaluation, large amounts of trash to be collected by the GC, cache trashing, etc...), and higher level interpreted languages, due to the lack of native accommodations to bind into fast code, or even into GPUs.
There are a lot of disadvantages to Haskell if you're going to do mostly imperative code.
I don’t have time to go into details but nearly every statement you’ve made here is in my opinion wrong, other than the difficulty in debugging (and there are very good reasons why this is difficult).
I’ e never found it hard to write fast Haskell, because I put in the effort to learn how to do it. People forget they also put in the time to learn how to write fast C++, and Java; it’s a very large portion of any software engineering curriculum, it’s literally what algorithms is all about. People find writing fast Haskell hard not because it is fundamentally hard, but because they were never taught how.
> due to the lack of native accommodations to bind into fast code, or even into GPUs.
I have no idea what you mean by this, GHC provides a very good FFI, and there are libraries for binding to Rust, R, Objective-C and others. We also have Accelerate for data processing on GPUs, which includes runtime compilation of native code for both GPUs and CPUs.