Hacker News Reader

As someone who thinks that more advanced type systems and proof-carrying-code are some of the biggest advances in theoretical computer science and practical programming in the last few decades, but also is a C# compiler developer, I have also noticed this. Haskell is really interesting, but C# developers get shit done.

I'm not sure what the cause is, but it definitely gnaws at me.

> C# developers get shit done.

Yes, and that says a lot about the final quality too, no? We could say the same about Perl/PHP/Java/C++ and then we're sitting in the same horrible morass we live in today, because people settle for minimally bad code produced as fast as possible.

Good, cheap, fast. Pick 2. Almost invariably: Cheap+fast are the picked results.

> and then we're sitting in the same horrible morass we live in today

But you see, this is the claim that requires substantial evidence. Let's assume we're in a mess. If Haskell is one way out of it, as some people claim, why won't they show us the way? They've had more than 20 years to do it. There are enough Haskell developers out there to give us this pesky evidence we need. And yet we've seen absolutely none. There are way more impressive BASIC programs out there than Haskell programs. I think that there are two reasons for that: 1) People involved in PL research are less interested in programs that aren't compilers, and 2) languages like Haskell make certain tradeoffs that their designers fail to see, and as a result their advantages fall way short of the claims they make.

I don't even understand your question.

My internet is filled with half-baked webapps with bad errors and riddled with stupid security vulnerabilities, many of which can be defined to not occur in an adequately designed tech stack. Desktop programs work through dint of manpower, crufty testing environments, and users-as-QA.

What exactly are you looking for? Web app frameworks? they exist. database connectors? they exist. mathematical proofs relating to correctness? they exist, in abundance. data structures arcane and common? they exist. regex libs? they exist. Sophisticated programs running sophisticated systems? They exist, and you won't hear about them except as a rumor on message boards...

If you mean, "is there going to be a big marketing campaign to make me and my manager feel comfortable"... I doubt it.

> Sophisticated programs running sophisticated systems? They exist

Not really. Well, far less than in all those horrible, unusable languages. All I'm saying is, if Haskell makes writing correct code so easy, where is it? There should be tons of it now. It should be practically everywhere. In fact, it should be financially stupid to do anything in any language other than Haskell. So where are all the companies making a fortune by betting on Haskell?

You're arguing from perfection.

The reality is that most companies choose "fast+cheap" and usually incur significant technical debt which is paid off through throwing bodies at the problem. It is usually considered better to hire 3 people who are mediocre than 1 person who is very skilled. This is extremely well addressed in the software engineering literature going back to the late 60s.

It's also the case that software engineering is very conservative: "Don't change ANYTHING" is the default state of the industry with regards to practices.

I don't buy this either. If this were true you'd still see, for lack of a better term, the "redis" of Haskell. You can't argue that antirez skimped on development or that he isn't talented enough.

Where is the redis of Haskell? The ffmpeg of ML? There's something else at play here.

I think the simple answer is that a lot of problems people consider useful are intellectually not terribly interesting. Among other things, if it's been done before, the cool factor drops off a good deal.

e.g., I run a tutorial site for common lisp. One request has been for 'example of web framework'. At some point, I'll do it (it's useful), but my heart at the moment is in implementing an extensible indexing system for n-dimensional tuples. It's a good deal more esoteric and a lot more fun than screwing with http/html.

Not Haskell, but its relative (cousins of some sort I hear), OCaml, is used at Jane Street [1]. They seem to be making money.

[1] https://www.janestreet.com/technology/

Maybe it already is. That doesn't mean it would happen. People do not genuinely optimize for what they want, they tend to optimize for what they want given what they know today and their internal biases.

I'm not willing to claim that Haskell is perfect, but I am willing to claim it is better. I'm also willing to claim that if it had anything resembling the community, say, C# has (to pick a random example) then it'd solve most of the problems you're highlighting here.

And it's getting that way. The number of programmers proficient in it and projects succeeding using it grow year over year.

But if you want to argue from a perfect world then you need to note reality will look different and search for trajectories that aim to approach your perfect world... not just current states which are far from it.

> People do not genuinely optimize for what they want, they tend to optimize for what they want given what they know today and their internal biases.

Absolutely. Which means that there are substantial switching cost, which a slight marginal improvement can't overcome. But if benefits are so immense, they should be able to. I think Haskell's supporters tend to overstate its advantages (and it certainly has plenty of those), and discount its disadvantages (and it's got plenty of those, too). All in all, Haskell is an extremely interesting and very good language, but it most certainly does not solve all or even most of the challenges of modern software development. It's good, it's very interesting, but it's not the second coming. I, for one, would take the JVMs monitoring and hot code swapping capabilities over Haskell's type safety, because those solve a bigger pain point for me.

I think Haskell has, of late, entered into the echo chamber a bit and been dressed up as much more than it is. I believe Haskell has a cleanliness of semantics which makes it rich, but I don't believe that it has the best RTS at all. My understanding (via Scala, an experiment) is that it's difficult to embed these semantics into the JVM.

Ultimately, choose the tools with tradeoffs beneficial to your goals, obviously. Engineering of any kind is nothing if not about understanding tradeoffs.

That all said, I do think there are upgrade paths for both technologies—JVM languages with better semantics and "ML-alike" implementations with better runtimes. I'd be more than happy to use either.

"is there going to be a big marketing campaign to make me and my manager feel comfortable"... I doubt it.

And this is why better languages fail: because language weenies have atypical minds, and can't comprehend why something marketed inappropriately to the mainstream never catches on as something mainstream.

This is also precisely why Java succeeded. This is also why LightTable finally got across what Smalltalkers had been raving about. It's really not only about who has the best tech or the best message. It's who has the best tech that can make itself understood.

Show me a BASIC program as impressive as any of:

- Elm's Time Travelling Debugger: http://debug.elm-lang.org/

- Pandoc: https://github.com/jgm/pandoc

- XMonad: http://xmonad.org/

- (GHC, should really count)

It's interesting. There's so much more C# code out there, but I wonder about the lifetime of that code, and when it will be swept under the rug (i.e. if javascript eats the enterprise).

Haskell seems to disseminate ideas rather than code. It even disseminated LINQ into C#.

Perhaps this makes the lack of volume of production Haskell code less stinging because it has contributed to languages in other ways?

There's kind of an upside to this as well, because Haskell isn't entrenched in industry, it's still free to make changes and grow the language (i.e. Applicative becomes superclass of Monad), although this gap may be starting to slowly close with the recent popularity.

More often than not, it is: Good, existent, fast. Pick 2.

> Haskell is really interesting, but C# developers get shit done. > I'm not sure what the cause is, but it definitely gnaws at me.

Stuff that people are willing to pay for, which can be a very important concept for many (if not most) of us.

> I'm not sure what the cause is, but it definitely gnaws at me.

My theory is that it is because that is what we teach. If you go to school you learn Java (which is basically C#) and if you learn a language on your own most people recommend Java/C# because it will get you a job.

By extension we end up with the vast majority of coders knowing Java/C#, thus most code gets written that way.

I think many experienced programmers forget how hard it was to write good code in the first place, whatever abstraction they started to use. I "learned" Java and C++ very quickly. I'm sure I was banging out code and stopped looking up most things within weeks. However, it took me years to actually understand how to write good code in those languages. I had to learn the patterns, I had to learn where the short-cuts and sharp edges were. I had to learn what features to avoid and which ones to embrace. It took a long time, but now it's all habit and instinct. I don't even have to think about it.

Asking someone to switch to using a new paradigm—like moving from OOP to FP—is asking them to relearn how to structure a program. All those habits and instincts no longer apply and new ones need to be learned. It takes years to learn how to write good high-quality FP programs. I know I'm still learning and I first picked up Haskell as a primary hobby four years ago. It probably took me a year before I really understood what a monad was and could apply that abstraction, similar to how long it took me to really understand and apply things like visitor patterns.

I believe that if you took an entirely untrained person you could get them to be at least as productive if you taught them with FP versus OOP from the start.

> I'm not sure what the cause is, but it definitely gnaws at me.

Here's my hypothesis: programming language are made for people to use, and people spending time thinking about designing Watson, don't want to spend it thinking about expressing their program using lambda calculus. The only group for whom the stuff they develop coincides with language concepts are those writing compilers. So they get confused because their domain is the programming language, so they don't feel like they're spending their mind power on two different things. They can't imagine that someone trying to solve a tricky scheduling problem wouldn't want to spend time thinking about types.

The other group that can adopt languages like Haskell are those whose domain doesn't require too much thought: namely CRUD applications. These guys want to spend time thinking about type-driven-design. It might actually make their code quality better. They, too, can't understand why people solving a hard concurrent-data-structure problem wouldn't want to spend time thinking about expressing their ideas elegantly through types.

Both of these groups have one thing in common: expendable energy to spend thinking about programming language concepts. So to them, it seems easy; and it probably is, if you're willing to put some effort into it, which is precisely what many developers don't want to do.

Or, in other words, advanced languages have some cognitive overhead that people working on extraordinarily complex applications can't spare?

Maybe. Then again, I've been using C++ again recently and I feel like the cognitive overhead there is so huge that I can barely understand my simple programs. Nonetheless, Microsoft pretty much runs on C++ (and C#).

I've started recently working in and properly learning C++ (most because I hoped the OO model would suit me over C).

Sometimes, I get so frustrated by how weird the syntax can get (the templates always confuse me to the point I have to stop and slowly read the code before I get its meaning)!

It's not so much that they can't but that they won't. And the same people don't use C++'s confusing features (crazy template stuff) either. But the problem isn't just the cognitive overhead inherent to the language, but that of the switching cost. As I said in another comment, while Haskell certainly has some advantages, they don't easily overcome the switching cost. People writing compilers happily pay the cognitive price for the switch because that's their core business, while those doing CRUD apps can certainly afford the price.

It's not a good thing - for production software - for a language to be too much fun to use. It means that developers spend all their time using the language and not enough time solving problems. The suckiness of Java/C# tends to encourage people to solve their problem and move on quickly to the next problem, because it's honestly not much fun trying to extend the language or build abstractions. Haskell & Lisp programmers, however, can spend hours constructing utility libraries, DSLs, and custom monads to solve their problem elegantly, and as a result they have extremely elegant source code that nobody but themselves will ever read, a library of utilities that they will probably not use again, and a solved problem that took 4 hours while the hacky Java solution took an hour.

Then I guess that Emacs users never get any work done? ;) (I wouldn't know, but I would be surprised if that was the case)

The average emacs user I know doesn't actually know or write Elisp, they just grab preconfigured .el files off the Internet or corporate intranet and add the appropriate snippets to their .emacs.

Elisp gurus, in my experience, don't get all that much work done, other than writing lots of editor extensions. This is great if their job is writing editor extensions, less great if their job is something else. It's the same as with compiler writers: they're great at making other people more productive, but when it comes to writing end-user software, not so much. In general, of course - there are exceptions.

Nonsense. A high-level language developer uses the higher concepts to get more work done, in better ways, and less time. Hacks always strike back, sooner or later.

I disagree with these premises.

I find Ruby, C[0] and Go really fun. The reason I get shit done is because they are so fun! I can't wait to do something else in them!

Also, the elegant utility library code that took four hours to write can—and likely will—be reused, and save you hours upon hours.

Meanwhile, the Java hacker will always have +1 hour of coding, because he never learns.

[0]: I realise calling C fun is weird, but I enjoy writing C code, and it's my guilty pleasure, I guess.

You get the hacky Lisp version in 15 minutes.

this is a thought-provoking article on the subject: http://prog21.dadgum.com/21.html

> Yet more useful working code has probably been written in Java and C than all other languages combined since the invention of the computer

Wikipedia claims that there are over 200 billion (with a 'B') lines of COBOL in production. That further supports your point, of course.

Haskell contributed materially in the seL4 effort for a verified microkernel: http://ssrg.nicta.com.au/projects/seL4/

The problem probably is that Haskell is still not expressive or useful enough compared to some of the more "primitive languages". Take for example high performance / numerical computing, in principle a well adapted functional language should have a lot of potential in that area (e.g. sisal), yet Libraries like repa or accelerate-cuda fall short as soon as you want to go beyond what they expose in high level functionality (for example dynamic stencils in the case of repa). Try to express matrix vector multiplication tailored to matrices arising for discontinuous galerkin of a certain pde (they are typically sparse but in a predictable way) in accelerate-cuda and compare to the ~500-1k straightforward lines of code to do the same with intel thread building blocks or cuda. A language like C++ will give you both the possibility to write very high level code at the expense of loss of control and low level code with very fine grained control.

There are lots of examples like this in Haskell, where you come to a point at which the abstractions provided to you don't work, mostly when the problem has more structure than you can hope to capture in the type system and some generic operations. In an imperative program it is much easier to "teach" or "tell" the computer what to do, when the compiler or language can not do it for you, whereas in a pure functional language you essentially have to give up at that point.

There are a few people who are doing that. One good example is the work that Anil Madhavapeddy's been doing on the Mirage OS: http://www.openmirage.org/ It's a really impressive use of OCaml.

Not sure why there are so few other good examples to point to though.

I'm rather taken with Tidal: https://github.com/yaxu/Tidal/

It basically allows you to procedurally generate thumping dance music by live coding in a Haskell DSL in Emacs.

Which, given Haskell's rather academic reputation, is pretty damn cool.

Maybe it turns out that the sociological side of programming trumps the technical side; that languages that forces or fosters a common vocabulary will always "win" over languages where you have so much power that you can easily diverge from the mainstream dialect, and there is not much culture of restraining this freedom.

On the other hand, these dialects are embedded DSLs; specialized languages created and hosted by a base language. People seem to balk less at the idea of DSLs compared to the idea of powerful languages.

I think that part of the problem is that more 'powerful' languages often more concept heavy than more 'pragmatic' languages and that makes them harder to learn for a lot of programmers.

Take Clojure for example. A lot of people including myself like this language a lot, but to be productive in it, you have to get used to the JVM, Lisp s-expressions, a heavily functional programming style with few side effects, a significant break from a normal approach to OOP, and Clojure's distinctive concurrency constructs which are more complex than Go's. It's not that any one of these things is hard, but you have to know all of them. Unless you have Lisp, or Haskell (or similar) experience, the only one you're likely to know is the JVM, and even then, maybe not.

This isn't like C++ or Java which are huge languages with a lot to wrap your head around but at core simply reuse concepts that most programmers are familiar with, (OOP, basic types) It's the very things that make the language interesting to begin with that makes it trickier to pick up.

Most of this holds for Haskell, Erlang, etc.

Then again, real code is written in these languages, but they don't tend to dominate.

Not to mention emacs and the REPL. Emacs was my biggest hurdle learning Common Lisp. Learning a language - OK. Learning an editor - OK. Learning both at the same time - Not OK. I don't think light table was around at the time.

I had the same problem when I learned CL. On the flip-side, I really like Emacs now so I'm glad I learned it.

The first programming course we expose students to at Edinburgh has them doing Haskell in Emacs on Scientific Linux. We like steep learning curves, it seems.

I think it's actually a third-order effect resulting from this: being harder to learn means that fewer people learn it, which means that there are fewer libraries and less community support available for the language. That means that even experienced, highly-skilled practitioners who could easily pick up the new language in a weekend are less productive in it.

I've written Haskell, I've written Scheme, I picked up Go in a couple weeks, and learning a new functional language shouldn't take me more a few weeks. Nevertheless, if I were starting a project now, I would probably pick some combination of Python/Java/C++. Why? Because their library support & ecosystem far outweighs any productivity boosts I could get with Go, Node.js, Clojure, Erlang, Haskell, etc. And yeah, I know you can use Java classes with Scala/Clojure, but there's an impedance mismatch mapping the concepts and existing library structure onto a functional language that doesn't exist with Jython.