Haskell's Children
owenlynch.org
owenlynch.org
Idris as a descendant of Haskell I have no argument with.
Rust however, is more correctly viewed as (primarily) a descendant of ML languages, which all possess the properties the articles lists as derived from Haskell. Perhaps it is more correct to call Haskell and Rust cousins, both deriving from the same lineage of languages.
I would also not call Julia a child of Haskell. As it is more correctly viewed as an interpretation of Matlab as a Lisp, I'd therefore call Julia a descendant of Lisp. That Haskell single handedly popularized the utility of category theory in computing and that a category theory library is written in Julia does not make Julia a child of Haskell any more than it would have made Ocaml a child of Haskell (there used to be a category theory library written in Ocaml).
The popularization of functional programming, Domain specific languages as algebras, reactive (not FRP), property based testing, parser combinators, model view update for UIs, design principles secretly based on monads are all but a few of the ways Haskell has influenced programming. Which is the core point of the essay and with which I am in full agreement.
This thread somewhat has the air of an eulogy but it goes without saying that just because something is not the most popular in fashion does not mean it is dying.
Why would that be more correct? Would Rust have these features if they were not popularized by Haskell? I don't know the answer, but it seems like discussing which language some feature was taken, when more than one contemporary language has it, is a kind of meaningless pedantry.
Though I'm not really sure that it matters.
Though the original Rust compiler was written in OCaml, so depending on your viewpoint it could also be its child. I'm not sure if Rust lifted any features or syntax from OCaml specifically. I think the "O" in OCaml was not influential ...
----
edit: I found a list of influences, and it does credit Haskell with typeclasses, but SML/OCaml and C++ are credited first:
https://doc.rust-lang.org/reference/influences.html
I always thought of Rust as a mashup of OCaml and C++, but Haskell is in there too.
Some History of Functional Programming Languages (David Turner) https://www.cs.kent.ac.uk/people/staff/dat/tfp12/tfp12.pdf
Miranda was open sourced earlier this year! I remember downloading the tarball and looking at its source code.
https://old.reddit.com/r/programming/comments/fbtmar/miranda...
https://www.cs.kent.ac.uk/people/staff/dat/miranda/
https://www.cs.kent.ac.uk/people/staff/dat/miranda/downloads...
I would be interested to learn of features in Rust that came from ML and that aren’t features of Haskell - not counting strictness :-) I can think of a non-example: one of ML's distinguishing features is its functional (in the FP sense) module system; Haskell and Rust have more conventional module systems.
You could maybe argue that a Rust enum definition is more like an ML type definition than a Haskell data definition because the constructor parameters are tuples rather than curried, but it isn’t clear to me whether this is actual influence from ML, or Rust aiming to be more C-ish.
As I understand the history of algebraic data types and pattern matching, they first appeared in Hope, and from there they were added to ML, and on the separate branch of lazy functional languages they were inherited by Hope’s successors Miranda, Orwell, and Haskell.
> Rust that came from ML and that aren’t features of Haskell
> As I understand the history of algebraic data types and pattern matching, they first appeared in Hope, and from there they were added to ML, and later adopted by Hope’s successors
They appeared first in a practical form in Hope but I'm taken to understand that these ideas could already be found in Programs and their Proofs: An Algebraic Approach and the core ideas are clearly present in this 1968 paper by Burstall: http://www.cse.chalmers.se/edu/year/2010/course/DAT140_Types...
Creators of HOPE, VAX ML (which also had limited patterns and case analysis) and LCF/ML all influenced each other and played prominent roles in the design Standard ML, which would in turn strongly influence future functional languages, including Haskell and Ocaml. It's tricky to untangle the history but I don't think it's controversial to say SML popularized functional programming as prototypically understood today (before Haskell evangelized it further).
> Rust’s two most prominent type system features are lifetimes and traits
Other than the initial prototype being written in Ocaml, I'd say Rust's pragmatic feel: eager evaluation, immutability by default but a hatch to mutability and the implementation of the features listed in the posted article all make it feel more like an ML. I agree, though, that this is a subjective take.
Hmm...
> When I first started using Rust, I really missed monads.
Option and Result are Monads. Surely he misses do-notation.
> This all being said, I think it is worth looking at the features that are prominent in Haskell that ended up going to Rust ...[list of features that Rust got from SML/ocaml]
Here are the officially stated Rust influences: https://doc.rust-lang.org/reference/influences.html
To say that Rust is a child of Haskell is really overstating the case.
Well, functions and lists are also monads. I guess what the author means is that the language recognises the similarity of those disparate structures.
The article really rubbed me the wrong way with this tidbit
> (we don’t talk about Go…)
So since the author is speaking of "premier system’s language" we choose to ignore the world's premier cloud language? Interesting.
Whether or not these advances make a difference (positive or negative) in production is different question, but I'd rather not use "how much is a language used", as an inherent quality metric for a programming language. Otherwise, I fear we might all end up doing old-school Java, Javascript, Visual Basic and ABAP.
Define quality. Because what I've realized with the PL-elitist crowd is this almost always boils down to "the ability to be expressive" which is fine but it certainly isn't a complete metric, and, to borrow a common refrain, we've had expressive languages in the 70's as well, we've had s-expressions for a while now. What about other metrics, such as "ability for mass amounts of programmers to program the computer to do the correct thing?" and "ability for programmer to maintain said program?" -- you know, real world quality. My point is, people's usual quality metrics are often incomplete or narrowly defined.
In either case, if Go is so primitive, then why didn't the 70s produce a language like Go? Why did developers who created languages, operating systems, and systems software in the 70s not make Go until the 2000s (you do know who created Go, right?)
I just don't buy this argument, especially having used those "advanced" type systems for many years, sorry.
But all that says is that the magic of Go isn't in the type system. The couldn't produce-it-until-the-2000s isn't in the type system. The thing that makes Go more used in the real world than Haskell isn't the sophistication of the type system.
I don't see much to argue with in those claims.
The "PL-elitist" crowd looks at programming languages from a computer science perspective, while the "PL-pragmatist" crowd (for lack of a better term) looks at it from a workplace perspective. At least that's how I see it.
In other words, one group writes tooling and libraries; the other writes business applications using said tooling and libraries.
One paradigm invites complex, highly expressive code as to minimize the chance of unforeseen errors (even if just by having a smaller codebase), and as to provide a clean and powerful interface that "does more with less".
The other invites simple, highly maintainable code as to minimize the training overhead of new employees and the chance that their lack of rigor might end up accidentally introducing or reintroducing bugs. Furthermore, a simpler language means that new teams can hop onto a pre-existing project faster, because they don't have to scrutinize each LOC to the same extent.
I might be wrong on any or all of this, but that's how I perceive it.
My second point was that the popularity of Go (or any language) does not mean that it is inherently a good language, where good can mean, productivity, defect rate, .... Adoption of a language depends on many things, I believe quality of the tools, availability of libraries, and good old marketing (the Google aura around Go has helped in adoption) are often more important. The effort to make an initial, crappy, proof-of-concept is in my experience often decisive for the choice of language. The cost of long term maintenance (where a good type system might bring value) is rarely considered.
For Go, it also hasn't hurt that some of the core developers where being paid by Google to work on it.
The metric of the "ability for mass amounts of programmers to program the computer", is indeed a very relevant one, and I think Go shines there. When reading Go code, I typically find it easy to understand what the goal of the code is. However, "to do the correct thing", is often less of a success: there are off-by-one error and corner cases that have bugs, and reimplementations of the same basic logic. If you have a lot of developers available, you simply have them fix these issues, and your Go project becomes a success.
My point is, that in this scenario, the language itself is of minor importance: success is determined by the fact that you have access to a large pool of relatively cheap labor of reasonable quality (Google or VC funded startups are perfect examples). A 20 or even 50% productivity gain by having a "better" language would simply not matter for the outcome, certainly not if it takes your labor longer to get up to speed.
So, in that context, compared to the alternatives, the quality of Go is relatively high. But I do think that, if during the design of Go they had taken a couple of basic concepts from ML like sum types, polymorphism, and a decent module system, they would have ended up with a language that would make the current Go look unproductive and poor quality, even in that same context. It wouldn't make a difference to Google, but it would make the life of many a developer a little bit more productive and joyful.
And yes, I do know who created Go, but arguments by authority carry little weight.
I have the opposite experience. At a micro-level you might be able to tell what a line of Go code is doing, like manipulating this variable into that state etc, but it's such a tedious language that drowns the reader in code that it can be very hard to see the forest for the trees.
As an example of a deliberately simple language that mostly easy to read, I would perhaps cite Erlang.
In any case, I agree wholeheartedly that Go would benefit from sum types. (Though given the weak type system otherwise, you couldn't even implement a useful `Maybe` in Go. You'd really want parametric polymorphism to make sum types shine.) A way to express immutability would also have been welcome, especially given that they pride themselves on concurrent programming.
That may offer consolation to those who think that certain languages are the best, and who wonder why those "best" languages are used so little. But I don't think it's true. I don't think programmers are stupid, ignorant, or sheep. I think if a tool offers them advantages, they'll use it.
Perhaps I should qualify: significant advantages. There is a cost to switching. The switched-to language has to be enough better to pay back the cost.
I remember what Sun did with Java back in the day. No, I do not agree that Go has been massively marketed.
I will admit, however, that Go was created and supported by Google, and that mattered. It mattered, not in the publicity, but in the tools and libraries (and maybe even in the tutorials).
So your point is because Java has been marketed more then Go that Go is not massively marketed? If you take the set of all languages and would make a sorted list of how many hours and dollars where spend to market it then I would be very, very surprised to not see Go in the top 5.
C++ has vendors behind it. Those vendors make actual dollars selling C++ (or at least tools that work on C++.) When Microsoft markets Dev Studio (with C++ support), is that marketing C++? How about C#?
In contrast, what does Google do to market Go? Put up a website, and put out notices of the next version? When have you ever seen an advertisement for Go? For a tool that supports Go? How about for C++, C#, and Java?
Advertisement is a subset of marketing. Back in the day when Go was "new" You saw daily posts on many programmer focused communities about it. Google bankrolled the entire development of Go. Blogposts of many googlers talking about the language. Google sponsored Go events. The positive image of Google itself at the time of Go initial introduction.
You try to weasel me towards C# and Java when those are not two language I even have mentioned. But since you want to hear about. Yes both of these were also heavily marketed.
Or, if the big company doesn't do anything more than is done for other languages (Rust, say), what difference does it make? Is Rust marketed in the same way Go is? Per your definitions, I would say yes, even though the Mozilla Foundation isn't a big company.
But then, is Haskell marketed in the same way? I see lots of posts on it, at least here. Lots of blogposts on it. But there's no big company, or even foundation, behind it. Is that marketing?
Why do you define a set of actions as marketing when people who work for Google do it, but not when others do it?
There are other languages I can take out of the bag like Mesa/Cedar, Interlisp-D, Smalltalk-80, although they are already on the 80's borderline, which then also brings Ada, Standard ML and Object Pascal into the picture.
Go have limited capabilities type system. C have limited capabilities type system. Both are quite successful. If talking about C I didn't shown "intellectual superiority", maybe it would be incorrect label if I where to talk about Go?
It's just that in Haskell you have to write fewer tests to be sure you did it right, because for many kinds of mistakes you could have made, the compiler can yell at you.
The other advantage of nicer types is that they serve as part of the documentation. Exactly because the types can constraint the behaviour of implementations more than in Go. So the reader can make more valid assumptions from the type alone.
Hoogle uses those types to give you a truly extra-ordinary way to search the Haskell libraries via types. That approach would be almost useless in Go, and you have to fall back to searching via keywords etc only.
EDIT: also, strictly speaking, Rust doesn't use Hindley-Milner, and we don't have parametric polymorphism.
struct Person<T> {
age: T,
}
There's also more minor things like https://github.com/rust-lang/rfcs/pull/1210#issuecomment-179...
If you ask all the developers out there to describe parametric and ad-hoc polymorphism I think a vast majority would give the example of a type parameter (e.g., Java generics or C++ templates) for parametric polymorphism and Java interfaces or Haskell's classes for ad-hoc polymorphism. I can even quote directly from Pierce (Types and Programming Languages):
> Parametric polymorphism, the topic of this chapter, allows a single piece of code to be typed "generically," using variables in place of actual types, and then instantiated with particular types as needed. Parametric definitions are uniform: all of their instances behave the same.
I think Rust and the aforementioned languages fit this definition. Outside of a specific compiler issue, claiming otherwise seems to only confuse the issue, especially for those just casually reading and not familiar with programming language theory.
Which doesn't mean it's terrible to get it wrong, just want to be clear about what Rust does and does not have. It is important because these kinds of definitions either hold or they don't; "sorta kinda mostly parametric" isn't the way folks tend to think about this. Which makes sense, because they're interested in proofs and formal definitions.
Yes, Pierce is great! But the issue is:
> all of their instances behave the same.
This is not true in Rust, as shown in my comment and the other replies. We have accepted APIs that break this property, and we have other language features on the way that break this property.
In practice it is often useful to drop our demands for rigour by a bit. But any reasonable definition of parametric polymorphism _has_ to exclude C++ templates.
C++ templates are much closer to the definition of ad-hoc polymorphism.
My practical less-than-rigorous rule-of-thumb for parametric polymorphism amounts to something like: whenever Wadler's Theorems for Free paper applies. https://ttic.uchicago.edu/~dreyer/course/papers/wadler.pdf
Theorems for Free means that eg the type of the identity function (forall a . a -> a) guarantees that even if you supply it a bool, it can't magically turn into an implementation for `not`, or multiply all integers by 2.
fn foo<T>() -> usize {
std::mem::size_of::<T>()
}
foo::<u8>(); // 1
foo::<u16>(); // 2The _behavior_ of this function is the same for all types, the _output_ is different. That is, for all types, the function body is the same. Maybe there is a more abstract definition of parametric polymorphism you are using, but as I said above, this seems pedantic.
fn foo<T>() -> usize {
let x = std::mem::size_of::<T>();
if x % 2 == 0 {
panic!();
}
}
foo::<u8>(); // 1
foo::<u16>(); // panic
> this seems pedantic
The first code example is merely the simplest demonstration, in the wild I would expect lots of `size_of` in generic contexts to result in type-dependent behavior somehow.
I'm not saying this is necessarily a very bad thing, nor do I have strong opinions on the usefulness of strict parametricity (which AFAIK Haskell doesn't have either). But in discussions relevant to parametricity, it's useful to know the ways a given language can subvert it (and Rust will further encourage it to be subverted, once the specialization feature is developed).
In practice Haskell seems to have pretty strong views on enforcing parametric polymorphism, doesn't it?
Haskell gives you ad-hoc polymorphism via typeclasses and there are also existential types and GADTs etc, if you need those. But once you declare something to abide by parametric polymorphism, you are expected to keep your end of the bargain.
(Yes, you could violate the pact via some unsafePerformIO trickery, but that's considered bad form.)
If you functions routinely violate those limits as a matter of course, those guarantees are useless.
I'm all for abusing notation and terminology a bit when it makes sense in practice, but loosening our definitions too much risks making them useless, too.
In practice in Haskell, I often only need a helper function for eg integers, but when the implementation allows, I will give the function the most parametric-polymorphic type that fits, because that makes the readers job easier:
Just like an invocation of `filter` is easier to read than a for-loop, because `filter` is strictly less powerful than the loop.
(In addition, the more general type serves to give the compiler a hint, so it can yell at me in case I accidentally do an operation on the integer that I didn't mean to.)
> However, in 2020, the premier system’s language is surely Rust.
Yesterday I was speaking with a friend who runs a small company doing contracted embedded programming (the embedded OS is often Linux but often not) and he'd never even heard of Rust. Maybe it's a little lower level than what's meant by "systems" here but still, the article's statement is just delusional.
I can go to Green Hills and get a contract that guarantees their compiler will be available and supported for my target platform for the next 10+ years. Rust is still a moving target with a single implementation, and no (present) way to get the guarantees needed for the embedded domains I've been involved in. I'd like to see Rust experimented with there, but it won't be the "go to" language for quite some time.
Absolutely. I've said it before, I will say it again, Rust will not be taken seriously as a systems language until it is specified. As I understand it years ago I was laughed at for this, but it is underway. I am grateful for this.
I forgot to preface my statements with I Love Rust, or, moreso, I actually love OCaml, and I love OCaml inspired languages, and I would love to write an OS in it (save ReactOS).
That being said, the desire to label it as the "premier systems language" (and no fault of the author, many before have tried to say this) is entirely not grounded in any sort of reality.
You can find monads everywhere if you look hard enough. But Haskell actually lets you use those discoveries for massive code reuse, while Rust forces you to write stuff like Result::and_then and Result::map over and over again.
> To say that Rust is a child of Haskell is really overstating the case.
Right, most of these are common influences. I'd say Rust's main contributions are (a) the ownership system, and (b) painting a thin coat of C++ syntax over ML features to make them more palatable, and neither of these seem all that inspired by Haskell.
The feature of Haskell that lets you avoid writing 'and_then" over and over is do-notation.
Beyond that, it's not clear that a monad trait would work, even if we did. It may, but it also may not. We'll see (or not) I guess.
You mean you couldn't express it in Rust, even with HKTs? Or you mean that it wouldn't be useful?
There have been some people working on these problems, take https://varkor.github.io/blog/2018/11/10/monadic-do-notation... for example. This is why I say "not clear" and not "impossible" or something. We just don't know.
(Of course you still want some form of nicer notation than the raw one, but that's beside the point.)
Out of the remaining monads that are in common use (Result, Option, Future), it doesn't seem particularly worthwhile to try cram them all into the same API, when you could instead try to design better specialized APIs (as has been done with Future)
In practice applicative functors are even more prevalent than their subset, monads.
If you can express your algorithm using the weaker applicative combinators only, there's more opportunity for optimisation and parallelism. (And the code is easier to understand for your readers, because there are fewer interactions for them to worry about, too.)
I just implemented the Haxl paper in Java as a weekend project (for implementing GraphQL APIs), so it totally does ;)
I think you're being a bit short-sighted here.
I fully appreciate that it might be extremely difficult to accommodate these HKTs, etc. in a language with the core values of Rust. That's absolutely fine, because that's probably several research-level projects in and of itself. OTOH, we shouldn't discount the value of such things.
It's absolutely fine and reasonable to choose e.g. efficiency over other things, but it's not the only value.
Little abbreviations might be widely used, but when people talk about code reuse I'm hoping to hear about ways to share larger chunks of code, not having shorter idioms.
In Rust, the functions Result::and_then, Result::map, Option::and_then, Option::map, etc. are pretty much all saying the exact same thing, but they need to be reimplemented over and over again. With Haskell, Result and Option (and [], QuickCheck.Gen, Data.Binary.Get, etc.) just need to implement return and (>>=), and they get a ridiculous amount of useful functions for free (mapM_, foldM_, forever, ap, etc.). That's a "large chunk of code" saved in my opinion.
https://hackage.haskell.org/package/base-4.14.0.0/docs/Contr...
And, of course, they also get some very nice sugar in the form of do notation.
Having popular interfaces like monad, applicative functors, monoids etc also suggests common ways to structure any API you are working on.
If you design an API from scratch and have to implement everything again and again, you might be tempted into a pretty arbitrary breakdown of the problem space.
Whereas in Haskell personal laziness drives you to look hard to see if you can express your API in terms of foldable and traversable or monads etc so that you get lots of functions for free.
Those common interfaces also automatically suggest gaps in your API that otherwise you would have only noticed after using your API for a while (or by being a very experienced programmer).
With the caveat that some languages support an encoding of HKTs. You can encode HKTs in OCaml, Kotlin, or TypeScript for example, and there are FP libraries taking advantage of that. I believe the encoding was inspired by this paper:
https://www.cl.cam.ac.uk/~jdy22/papers/lightweight-higher-ki...
It kind of sucks, but it does let you work with a Monad type class, and it's better than not doing it at all.
I'm not sure if such an encoding is possible in Rust. I've been told that F# has a hard time with it, not sure why (possibly because of reified generics). In TypeScript and Kotlin it works because the runtime and the type systems are relaxed and you can just force downcasts if you're in trouble.
Just to add some tangential color here: even if you don't intend to use this encoding in production, it can be a fruitful intermediate step in refactoring! I just recently ran into a situation in Java where I had some tightly-coupled concerns for which HKTs were the obvious way out. Since Java doesn't have HKTs, I applied the HKT encoding to get the hard part out of the way, and then discovered a simplification that brought me back into the realm of idiomatic Java. I would not have found the final solution had I not pathed through HKTs -- I actually thought it would be impossible to do this decoupling properly at first!
More color: the paper itself describes this approach as an application of defunctionalization. I use defunctionalization a lot, too, to write an obvious recursive solution and then mechanically transform it into an iterative one. So it's nice to know how to apply it at the type level, too ;)
> Option and Result are Monads. Surely he misses do-notation.
Surely you are joking if you are trying to say someone writing haskell would say rust has monads in any practical sense. Sure, all languages have/can have A (or two) monad.
Rust does not have higher kinded types, which are required to have abstractions like the Functor and Monad type classes in Haskell.
Option and Result are monadic, but they must implement their own, separate methods and can not share a common interface. Which takes away a lot of expressive power.
There is a accepted proposal for something similar, known as GAT [1] (generic associated types). But that is blocked on other work and might never actually be implemented.
In a theoretical sense, sure, but in a practical sense, my understanding is that GATs already exist in chalk, and so it's still a matter of putting in the work, not a question of "is this possible or not."
You would know better of course, but it was my impression that there is no firm commitment to stabilizing GAT, even once the compiler fully switches to Chalk, due the potential disruptiveness and complication of the language.
(For those unaware: Chalk is a rewrite of the trait logic in rustc, based on logic programming [1] )
I think async trait methods could be stabilized without exposing GAT to the user.
Of course I would very, very much like to have them either way, they would make certain abstractions much easier to build.
Certainly, but not in the sense of being uncertain about whether or not GAT are worthwhile, only in the sense of not wanting to make promises that may not be able to be kept. The Rust developers want GAT--or something analogous to GAT--and much work has been put in towards that goal (work that will have been useful even without GAT), and at this juncture there's no foreseen theoretical impediments; however, that's not to say that some unforeseen impediment won't yet arise. Such a thing happened with specialization, which has been in the works for much longer (current status: https://github.com/rust-lang/rust/pull/68970), and the lesson has been to not incite expectation until you are extremely certain the promise will be able to be kept. As a mere half-informed bystander it seems that GAT not encountering the problems that specialization did, but don't take this as any sort of promise. :)
Rust is already a complicated language. Those two features would increase the complexity of the type system quite significantly.
Possibly beyond what is healthy for Rust adoption.
That's part of it. Option and Result are concrete instances of Monads. He misses the Monad abstraction.
When you write code that operates on something like a monad, without caring about what the monad is, you want code that has a type like <T1>[<T2>]. In haskell syntax, it's m a. <T1> (or m) in this case is a type that takes another type as an argument to complete a concrete type. AFAIK, rust doesn't have type variables that take type variables (or types) as arguments like that.
Soon you can even deploy haskell web apps in 15 seconds: https://twitter.com/larsparsfromage/status/13066539171876495...
Growing between 5% and 10% weekly. Many new people are starting their journey into haskell with IHP :-) While the community might be changing, it's definitly not dying.
E.g. on Sep 1. we had 870 GitHub stars and 321 Twitter Followers. Now (23 Sep) we're at 970 GitHub stars and 487 followers.
IHP has been launched around 3 months ago. So we're measuring it since then.
I am a huge Haskell fan though. The 'problem' with Haskell is that, because of its roots in academic research the language will not chase success at any price (read hacks); and tends to eventually find solutions to its problems even if it means not being popular for a long time. I like that. Watch out for stuff like linear types[1] coming to GHC soon.
That's the PR line, but only describes about 5% of the problem with haskell.
But haskell's not relatively unpopular because it was late to add things like linear types. It's relatively unpopular because it doesn't care much about user experience.
Consider: it's getting linear types before an efficient string type in `base`. The first goal of this language is not good engineering practices and getting things right. The first goal of this language is "cool type stuff", with good engineering maybe goal 10 or something (is O(n^2) `nub` ever going to be deprecated I wonder?). And that's fine, we need a language for cool type stuff, let's just be honest about it.
Context: I've been Haskelling for 5 years, I've written a fair amount of FOSS in it, and it's my favorite language. I just want to help spread an honest impression of it.
IMO Haskell remains as the go-to language for teaching functional programming and for experimenting with PL concepts. Long may it remain.
I like using Haskell for writing compilers, but I would probably not like to use it for numerics code. While I could replace Haskell with Idris, I'm still a bit lukewarm on the cost-benefit of heavy use of dependent types for many programs (although Idris has other nice advantages unrelated to the type system).
Elm is a more special-purpose child, but seems to have had an impact on JS practices, as a 'gateway' to functional programming. PureScript is also interesting but hasn't had as big an effect.
I agree with the article that Idris is really nice; although I'd lump Agda in with it as a dependently-typed child of Haskell, and contrast them both against Coq (rather than lumping Agda and Coq together).
I've also played around with Curry and found it quite interesting (functional logic programming), although it's essentially a research language.
I personally believe, agda is an excellent programming language. It's stdlib is good, concepts and construct are very well-thought out, it's very safe and works surprisingly well. I've been programming in agda for years and almost never had issue implementing anything even though it's not Turing complete. Using sized types you can prove most sane things to terminate anyway. And if you really need infinite loops (say, you're implementing an interpreter of a Turing complete language) then you have many options: (1) disable termination checking for a single function (2) use Haskell to implement a single function (3) use Rust, C, C++ etc to implement a function then with GHC ffi link it to agda.
Maybe I'm just being near sighted, but I'm really surprised more people don't use Agda or Idris in production. The number of times the type system saved my ass is uncountable. You can literally write unittests in your type system it fekkin runs your custom unittests while type checking. How is this not revolutionary? I don't understand.
My main gripe with dependently-typed languages is the amount of wheel-reinventing that's involved; e.g. I might have crafted a "ListOfIncreasing comparisonFunction" type, which is perfectly suited to the problem I need to solve, but then I find myself implementing a map function, a filter function, an append function, a length function, etc. Then I find myself needing a "length of appending commutes" lemma, and an "empty is left identity of append" lemma, and an "empty is right identity of append" lemma, and so on.
It's really nice when it works, as a nice self-contained little module; but any more can get exhausting. I haven't tried integrating Agda with Haskell, but that sounds like a nice tradeoff: do all of the "boilerplate" in Haskell (string plumbing, etc.) to get a "Maybe InputsForMyAgda" value, pass it into whatever tricky algorithm we've managed to implement thanks to Agda keeping us in check, then send the result back into Haskell for throwing into whatever external monstrosity we're interfacing with.
I don't know why you had problem with compiling... All you have to do is run `agda --compile src/Main.agda` and it just works...
> My main gripe with dependently-typed languages is the amount of wheel-reinventing that's involved; e.g. I might have crafted a "ListOfIncreasing comparisonFunction" type, which is perfectly suited to the problem I need to solve, but then I find myself implementing a map function, a filter function, an append function, a length function, etc. Then I find myself needing a "length of appending commutes" lemma, and an "empty is left identity of append" lemma, and an "empty is right identity of append" lemma, and so on.
This is true to some extent, but this has nothing to do with Agda. If you're working in an extremely strict type system like agda's where all types need to be provably constructable, you cannot just start coding. You need to have a good idea what your program will look like. E.g. if you're writing a parser (as I often do because writing parsers in agda is such fun) you need to know some mathematical structure behind parsers.
Take a look at this page: https://agda.readthedocs.io/en/v2.6.1.1/language/sized-types...
Here they build Kleene star, which expresses infinite computation. However, you need to construct it in a (coinductive) way such that it doesn't infinitely loop. You cannot just code something like this. In that page they use a paper to base their functions. You need to know the math behind this to have an idea what will work. But you also absolutely don't need to be a mathematician either. And if you have no clue where to start, having multiple iterations of your program usually results in success. E.g. sometimes I start coding agda, then realize my types aren't good enough, restart, rinse and repeat. In my experience Haskell is much better for programs like this. Once you write your program in Haskell and have an idea where things are going, start formalizing some functions in agda, and then try to rewrite everything in agda. Writing programs in agda doesn't mean it's correct, of course, your business logic can still be wrong. The power of agda that comes from dependent typing, is that you can encode tests and proofs of your invariants into the type system. This means if you really want X to be always correct you can prove X or of it's not easy, you can write a whole bunch of unittests that'll be checked at typecheck time.
Regarding FFI, it's actually very easy, read here: https://agda.readthedocs.io/en/v2.6.1.1/language/foreign-fun...
You can add arbitrary haskell code in your agda programs. If you compile with `agda --no-main` agda doesn't call GHC so you can manually call GHC to link your program. This means you can do everything you can do with GHC to your compiled Haskell code in MAlonzo/
My point was that most Agda code I've written, or seen others write, is used to prove theorems; the result is rarely run. Even when implementing "real" algorithmic programs, it's usually just to prove that they satisfy some property, rather than actually using them to do anything ;)
> This is true to some extent, but this has nothing to do with Agda.
It's not inherent to Agda, but it's true in the sense that I could do "stringly typed programming" in Agda (i.e. using big, imprecise types which require a bunch of redundant checking and the compiler won't help me with), but there wouldn't be much point (I might as well do that in, say, Python and benefit from the vast amount of libraries and tooling).
> you can write a whole bunch of unittests that'll be checked at typecheck time.
I do this whenever I'm writing Agda or Idris, and it's so satisfying. The 'test framework' is just the type system, e.g.
test_parse_empty : Eq (parse "[]") Nil
test_parse_empty = refl
_ : (f x y) == expected; _ = refl
I also put my tests in the same file that define them by adding a private module
module X
record X : ...
private
module test where
_ : (f x y) == expected; _ = refl
The benefits look really promising e.g. proving forall x -> f x == g x -> f == g. I really want to use this theorem in some of my agda projects, but I never needed it desperately enough that I had to drop everything and do this. My usual workaround is implement slowButEasyToUnderstand function. Do a whole bunch of proofs with slowButEasyToUnderstand. Implement fastButComplicated function. Prove forall x slowButEasyToUnderstand x == fastButComplicated x
I think one immediate application I can think of with cubical is, you can implement mergeSort : List X -> List X and then pass this around everywhere. You don't need a Sorter record or whatever since when you implement quickSort, you will be able to prove quickSort == mergeSort, so everything will typecheck. I don't know which will end up getting compiled though.
Wut? Rust doesn't have the purity and the capability to isolate side-effects that Haskell has. Rust is popular because of its ownership model that got into the mainstream. We still don't know what are the costs of it.
The only similarity between them is that both are statically typed, expression based languages (ex. Scala, Ruby, Wolfram, LISP) and moved the task of proving stuff from analysis tools to the programmer. But that doesn't mean that Haskell "is the science" behind anything. For what it's worth, Rust is closer to the ML (OCaml) family than Haskell, focusing on performance rather than purity. Though that is also debatable since it doesn't feature a GC and comes with zero-cost abstractions.
Rust has its very own merits. I'm eager to see how it will evolve and what hidden costs will be uncovered.
Sounds interesting, do you have some links on this topic?
Swift
https://swift.org/blog/swift-5-exclusivity/
https://github.com/apple/swift/blob/master/docs/OwnershipMan...
D
https://dlang.org/blog/2019/07/15/ownership-and-borrowing-in...
Haskell
https://gitlab.haskell.org/ghc/ghc/-/wikis/linear-types
Chapel
https://chapel-lang.org/docs/technotes/lifetimeChecking.html
ParaSail
https://adacore.github.io/ParaSail/
Ada/SPARK 2014 proposal
https://www.adacore.com/uploads/techPapers/Safe-Dynamic-Memo...
Pony
https://tutorial.ponylang.io/reference-capabilities/referenc...
Verona (very early stages)
https://www.microsoft.com/en-us/research/project/project-ver...
When I learned those, Haskell was still Miranda.
I really don't know why people think Haskell is the pinnacle of everything in PL design.
If you were a Java/C/C# programmer in the 2000s, you missed out on:
* lambda abstractions
* Generics (prior to 2004)
* Higher-kinded Generics
* Primitives in Generic collections
* no nulls
* do-notation
* ML type system
* global, decidable type inference
* Pattern matching
* Functions as first-class objects
* Currying
* And most importantly *side-effect-free* functions
Over the last decade I've realised that those statements aren't really truthful. Many functional languages allow null. Many functional languages won't enforce side-effect-free functions for you. So it was never really about FP, it was about Haskell. Anyway, that's how they got conflated in my mind.
So does Java in 2020, but I was writing about:
>> it feels really weird that now is fashionable to assert that Haskell === FP
When Googling "functional programming", you get back pages of definitions emphasising "pure functions", "no side-effects", "mathematical functions", "avoiding shared state", "referential transparency". So what Haskell provides is what the blogosphere has been labeling as "functional programming" for a decade. Those definitions could be wrong, but they're the definitions that are floating around out there.
Could you please elaborate on that? Is it related to the new features, like lambdas, pattern matching with switch and so on?
Haskell ticks all these boxes.
SML has most of those things, except the last: it was effectively frozen in 2001. This is discussed in section 9.3 of the HOPL IV History of Standard ML https://dl.acm.org/doi/10.1145/3386336
OCaml lacks a standard and multiple implementations, and seems to me to be doing better than SML in the other areas.
IMHO it is the first successfully implemented research language, though. I think that Haskell is the first language that, at the same time, had substantial commercial and research applications[1]. As such, it showed that the CS of PL is indeed of practical relevance.
Rust on the other hand, mostly applies this scientific research. And successfully so. Hence I call it the "engineering" part.
[1] To a certain degree this is also true for Lisp and SML, of course. But I think SML fell out of fashion quite quickly both in academia and industry. Lisp is, well, lisp. People that start using it seldom search for another language afterwards.
http://batteries.forge.ocamlcore.org/doc.preview:batteries-b...
But the sort of code you would write in Haskell gets very noisy without typeclasses, and it's just not an obviously good way to program in a language that's intended to be impure and strict.
For me, Haskell is on the list of languages (right next to Smalltalk and Forth) that I would like to learn in the future just to experience problem solving from that vantage point, but it's taken me so long to work through SICP that I wonder if I have the time to meaningfully experience all these languages.
I have fond memory of first working through SICP in my teens. But having recently revisited the book, I realised it's mostly nostalgia.
There are better books about programming than SICP nowadays.
> What books would you recommend instead?
'Think Python' is a good introduction to programming in general. https://greenteapress.com/wp/think-python-2e/
'How To Design Programs' is another good introduction. https://htdp.org/2020-8-1/Book/part_preface.html
For functional programming in general, Okasaki's 'Purely Functional Datastructures' is a real treat.
Paul Graham's 'On Lisp' is worth reading. It's a bit on the pragmatic side. http://www.paulgraham.com/onlisp.html
Above all of them, the best book in general is 'The Pleasures of Counting'.
I think rust got this right, rustup and cargo are great tools that are easy to install and work with.
Cabal has fixed the problems from a few years ago, and Nix is a nice ecosystem to plug into, but these can get in the way at the beginning, at a point where the goal is to "create my first project, write my first program, build it, test it, run it".
(Then again, maybe you do have multiple examples and/or have strong evidence this system was indeed "Haskell done right", which is why I'm asking).
Haskell is surprisingly well suited for back-end services (e.g REST). Warp (an HTTP server), Servant (a way to define your backend API), and Mio [1] (the Haskell network IO implementation) are amazingly performant and scalable.
Of course, that does not prevent you from messing up on the business logic level. I suspect that they lacked awareness of performance implications of their choices, this is not something Haskell will save you from. On the contrary, parts of the Haskell community do tend to go to over-abstraction, and I believe this does have a negative effect on people writing production software.
- XMonad
- Pandoc
- Shellcheck
Everything else is just tools for using Haskell (Cabal etc.).
Seriously, compare:
- Haskell https://github.com/search?l=&o=desc&q=stars%3A%3E1000+langua...
- Rust https://github.com/search?l=&o=desc&q=stars%3A%3E1000+langua...
- Dart https://github.com/search?l=&o=desc&q=stars%3A%3E1000+langua...
Haskell has been around for a long time. It's not just "avoiding success at all costs," it's a nothing. It inspired good features like Option and Maybe in other languages, but it itself is on the same scale as Dart.
The first one that comes to mind is pattern matching. All your ML derivatives have had it for years, and it's an extremely powerful concept. Unfortunately, most people (myself included) had never heard of it until it hit more main stream languages like Rust and C#. Streams and the more wide spread use of anonymous functions is another great example of functional concepts making imperative languages much more expressive and powerful.
I don't like writing in pure functional languages very much, but I do enjoy a lot of the power the have and how we can make these concepts more accessible to traditional imperative languages.
I just have one thing to say.
In the real world (i.e., unlike in a tree), children have multiple parents.
I use Python / numpy heavily, and the giant hole numpy opens in my type annotations causes me grief daily. If there's something better out there, I want to try it, but it has to work. It has to work not just in the "I'm putting together a journal article and I need it to run" sense, but in the "I'm building a system that crunches some heavy engineering numbers all the time, and customers expect consistent results" sense. Can Julia be trusted?
The "time to first plot" (a consequence of the aggressive JIT that compiles large programs with heavy optimizations, as Julia doesn't have an interpreter outside of the debugger) still exists, but it is also improving at a fast rate.
[1] https://julialang.org/blog/2020/05/rr/
[2] https://juliacomputing.com/case-studies/
[3] https://github.com/JuliaPlots/Plots.jl/issues/2838#issuecomm...
JuliaCon2020: Julia Is Production Ready - https://bkamins.github.io/julialang/2020/08/07/production-re...
Discussion: https://news.ycombinator.com/item?id=24082281
There are comments suggesting that Julia is plenty capable, with libraries covering equivalent (or more) features provided by Python libraries like NumPy. YMMV.
>"The type system in Rust is obviously very influenced by the type system of Haskell, but they also implemented “ownership” which allows for the killer feature garbage-collection free automatic memory management."
I always assumed that garbage collection was simply a convenience offered by a particular language.
I was also under the impression that memory leaks were the class of problems that were addressed by garbage collection and that memory safety was a class of problems that was concerned with things like buffer overflows and corruption and addressed by runtime checks.
However after reading I feel like I have failed to understand the role of garbage collection in enforcing memory safety. Is memory management and memory safety not so cleanly divided?
Could someone explain or point me to a good resource for understanding how garbage collection is used to enforce memory safety?
Does that make sense?
- use after free
- memory leaks (failing to free)
- concurrent and uncontrolled access from multiple threads (race conditions)
Garbage collection helps with the first two. The system will not free memory until it is certain (by whatever method is being used) that the memory is no longer in use. But this comes with a run-time cost. Rust's innovation here is a more reliable way of tracking memory usage statically, and removes some of the run-time cost. Paired with its ownership/borrow mechanisms it also works to solve the third issue, which GC doesn't deal with at all.
I use PureScript everyday, and I love it. It's nearly identical to Haskell, with a few small changes that can be described on about a page
It's a fun language, and if you play with it and its ecosystem, it might give you ideas to import in your own environment.
Pick up a good book, and learn it, because it's fun. And no, you're not too dumb to learn it. Java and its ecosystem can be more complicated IMO, you don't realize it because you've put years in learning it.
Advice gratefully received!
An early version of Perl 6 was written in Haskell, which led to further Haskell influence on the language.