MicroHs binaries are ~100× smaller and ~5–10× slower for this workload; for many data-wrangling tasks that’s a great swap
Under which conditions is that a great swap? A 5x increase in processing times is absolutely huge, and even for moderate data volumes could make a data processing pipeline completely non-viable.
This is what I'm thinking. There are still use cases I would say where small binaries really matter. But then you are really choosing the wrong tool for the job with haskell, and I say this as a haskell stan. I expect an optimized C binary is much, much smaller still.
Under which conditions is that a great swap? A 5x increase in processing times is absolutely huge, and even for moderate data volumes could make a data processing pipeline completely non-viable.