Very interesting to know these. And don’t get me wrong, I’m not surprised some people need performance. I’m just saying given the requirements and options, pandoc really doesn’t fit your need.

Other than pandoc, I think compiling Haskell with stack is much easier. I’m no Haskell expert but this is what I heard. I agree cabal dependency hell is frustrating.

Yes, “Haskell is used in science”. It just doesn’t really “has a presence”. Generally, people talked about using C/C++/UPC (and CUDA), Fortran, even Python (PyCUDA), and then the new Julia for HPC (High Performance Computing). But I never heard of people talking about deploying a scientific HPC application written in Haskell. What makes me sad is that actually Haskell is faster than Python. But scientist chose Python over Haskell because of e.g. the ease to learn it, distribute, and interfacing with existing Fortran and C/C++ code base, and all these are difficult to be accomplished in Haskell. One problem many scientists have is that they are not professional programmers, and they don’t care about spending time learning a language (except when needed in their research). So many of them write whatever code that’s getting their research done in the least amount of time. For computer time dominant projects, they use Fortran/C/C++; for programmers’ time dominant projects, they use Python. But they (almost) never choose Haskell because it is hard to learn (lots of programmers time to pick it up, but develop time might actually be much less. But they don’t invest in the future but want to code right now) and slow (lots of computer time needed). So it’s in the middle of nowhere.

I also heard part of the reason Haskell didn’t take off in the way say Python does is a matter of PR — they just are not good at selling their product.

And then perhaps another reason Haskell is not a good fit for science is that Haskell focus too much on correctness. But in scientific computing, most of the time we actually didn’t really care if it is “correct”, but close enough and fast.

I wish I can use Haskell in any part of my research such that it gives me excuse to spend more time to learn such a beautiful language (and to contribute more to pandoc).

P.S. as I’m typing the above, I found that Intel has finally released the long-promised Haskell compiler! In the past, they released studies showing that in some situation, the Haskell code beats the C code by 8% in speed. Hopefully this could change the adoption of Haskell in scientific programming. Reference: IntelLabs/flrc: Haskell Research Compiler, Intel Haskell Research Compiler | Hacker News