Tooling still has a way to go, but I don’t think that’s all. Haskell’s abstraction is just too much when you need to design something which is ultimately kind of straitforward, and most apps are (design app, game, utility app, web server, etc.)
I do think Haskell is more than a research language, it does have some use in practical, mainstream programming. In particular, Haskell is great at anything which is “input to output”, like compilers or document converters. Or in a program which is ultimately complex, and not just for optimization but “math” complexity, like math programs or simulators. In those programs the implicit laziness actually comes in handy.
But Haskell for a real-time video game or editor, and you’re just wrapping everything in an IO monad or some restricted stateful equivalent, and executing most of the instructions in sequence anyways.
I've done a couple small projects in Blueprint which is a purely functional programming language embedded in UnrealEngine 4 and I enjoyed it a lot. Blueprint notably does not expose any type classes, at least not when I tried it a couple years ago, but it felt very expressive for building complex game logic.
Disagree from me. I spent a year writing Haskell professionally, and from my own personal experience Haskell's tooling was overly complicated and difficult to use.
Maybe my impression is totally wrong or outdated, and it's surely not applicable to the whole community, so I'm sorry if this is a little out of touch now, but I sometimes got the sense that Haskell's community just doesn't care about the developer experience. They hold to this false dichotomy that if you make tools easy to use, then you can't have advanced features. Or simply, Cabal works just fine the way it is, why do we need Stack? Why would you need more than one GHC installation on a machine? Holy wars on the mailing lists re: Stack, Cabal, Haskell Platform. And if someone tells me to "just use Nix" one more time I will cry...
Rust feels like a breath of fresh air. The community seriously cares about their developer experience both for newbies and experienced professionals, and it really feels like a lot of work has gone into making the tools and documentation accessible and useful.
I don't think they are. Having used both, Rust's tooling is more accessible to beginners, which makes a big difference. I find it even easier to use than npm, while I find Haskell's tooling harder (on roughly the same level as OCaml with dune, or Maven).
> Haskell’s LSP is making rapid progress and there are many formatters and linters.
Haskell's LSP is good, however it's not as good as rust-analyzer, and has a habit of chewing through a lot of RAM.
[0] https://www.researchgate.net/publication/3246772_Planning_th...
Some more effort in packaging this ergonomically, and we'll see purity annotations in TS and explicit passage of state objects around.
I feel like "not tending to companies needs for profits" is a way to elevate Haskell and monadic FP abstractions with appeal to conspiracy/(postmodern) ideology. I don't mean this to be a pro-or-anti capitalism sentiment, but the excuse here is that "in a more just/less captalist/<other> world, Haskell would be a great tool, but our flawed world makes it not so" and I disagree with that. I just don't think Haskell is that good of a tool regardless (and I see the Haskell community these days appealing to this take a lot, trying to blame anyone but themselves.)
I think that people genuinely don't derive much value from Haskell and monadic FP. My experience, having written and maintained non-production-grade Haskell code, has been that the community is obsessed with toys and small examples, derives a lot of pleasure from clever abstractions which don't materially affect development (like Lenses), and most importantly doesn't really care about shipping software. There's exceptions like SPJ and Joey Hess who definitely ship useful Haskell, but the community doesn't care nearly enough about making it easy to actually produce software that a non-Haskell enthusiast would care about using.
Libraries are often created for a Masters project and abandoned. Even if a library works, debugging edge cases is never thought of. Almost nothing is written about testing in Haskell. No two libraries can agree on which monad stack to use, so a non-trivial amount of integration code involves matching and lifting into other monads (nor the performance implications of doing all of this matching and lifting). There are very few examples of organizational patterns for Haskell codebases for folks trying to create a project other than GHC itself, Pandoc, and perhaps XMonad. The community loves talking about abstractions and unifying concepts but doesn't really seem to have any practical tips about writing software. The string types in the language are confusing and poorly differentiated. The cabal and stack split is confusing and makes no sense to outsiders. The weird cargo-cult obsession in the community about category theory from people with little formal mathematical knowledge is bizarre. I can go on and on, but I don't want to dunk on the language endlessly (and I can certainly do this to any language I've used in love or anger :)
Fundamentally I don't think Haskell offers the programmer much more than what you'd get out of a language like Rust, Java, or Go (yes even Go) as the GP said. It's a language with a lot of interesting concepts that's fun for folks who like to solve algebra problems and is certainly a clean introduction to monadic FP. I'm looking forward to a successor that takes its concepts, cleans them up, and offers a way to write actual software with them. Then again, Multicore Ocaml is landing soon...
The reason for that is that people who do haskell for a living are dwarfed by the hobbyist community which makes them invisible, and they're spread thin on maintaining the ecosystem. GHC, HLS and all the other good things there are are the work of people on their free time, while other, more successful languages simply have companies dedicating time, money and contracts to tooling.
All of that means that it's not easy to move from the hobbyist haskeller group to the paid haskeller group. Most people usually get there by getting hired in a place where the stack, library choices, etc. are made by people who had done haskell professionally before and have solved the problem you cite for the team.
Once you're in, the benefits of Haskell are tangible, otherwise people would stop using it, and there wouldn't be a professional community. That is not to say that there are no problems, it's just that they are different problems, and people that stick to haskell for their job usually do it as a tradeoff.
The tech itself really works well, but it required either an org mature enough to do the things correctly, or a shop purely focused on haskell that would accept the complexity.
Also, sometimes the library ecosystem is a bit lacking. It wasn't an immovable roadblock, but we did write stuff that may have been available as a lib in other languages. This was a minor impediment for us, but it can get bad in other contexts.
That's very true and the case at my job. We've taught people with 0 Haskell experience to write useful production code in under a month.
> Even if a library works, debugging edge cases is never thought of.
Many libraries use sum types to express their api which naturally leads one to consider edge cases.
> Almost nothing is written about testing in Haskell.
Yes there is! Did you verify your claim with a quick search? Here's what one gave me:
- https://mmhaskell.com/testing/test-driven-development
- https://jaspervdj.be/posts/2015-03-13-practical-testing-in-h...
Heck, here's one from 2007:
https://www.jasani.org/posts/2007/unit-testing-with-hunit-in...
It feels like this web of lies about Haskeller's not caring about real world concerns is being spun, but even with minimal effort that is trivial to disprove.
> No two libraries can agree on which monad stack to use,
The mtl library is the most used monad stack by far and it's not typical to see others very often.
> so a non-trivial amount of integration code involves matching and lifting into other monads (nor the performance implications of doing all of this matching and lifting).
I feel this may betray a lack of understanding because lifting between monad stacks is expected.
I agree there are performance implications on theory, but if you get out of the performance ivory tower you'll find it doesn't happen much in the real world ;)
Oh but they do materially affect development in the same type of way LINQ does for C# programmers.
https://youtube.com/watch?v=QZy4Yml3LTY
Then for a great book that gives a great understanding see Chris Penner's amazing and fun "Optics by example:
About the last point, my tiny observations is that language too advanced don't get used as dependencies.. they get ignored until the crowd matures enough to get the ideas in their own ways and pull a gene transfer.
- lisp idioms (map, filter, destruct) => all dyn langs
- react / vue .. basically let over lambda and closure based programming (I even believe vue provide/inject is near equal to dynamic variables)
wait until someone there likes to thread state or rng seeds through 'chains of functions' and you'll see monads pop up