Hacker News Reader

Comment by parenthesis | Hacker News Reader

Haskell is that rare beast: a language that Lispers may envy.

Isn't this problematic? Lispers fling around lots of superlatives, as if their favorite language will never be dethroned from its position atop the language kingdom. And they always felt justified about this.

Shouldn't we confront them about this?

I do think there's a science of programming languages which can inform us how to best design the tools we use. Isn't it just possible that these Hindley-Milner languages are the next step in the progression that LISP started?

Look at the transition from Newtonian to Einsteinian mechanics. They were both great at the time of release, but one is clearly a step forward. The former is usually embeddable within the latter, but edge cases in the former don't quite work the same in the latter. And some cases in the former are just downright illegal in the latter. And unfortunately, the latter is a lot more complex for the power users. But, the latter is more expressive, teaches us more about the world... and enables completely new possibilities like nuclear fusion/fission.

This is like the difference between dynamic and static typing. Dynamic typing is certainly embeddable within Haskell. Just define a data type Dyn which is the sum of types Int, String, Float, [Dyn], and Dyn -> Dyn. You will see how much LISP is possible without much added syntactic cruft. But Haskell users don't need and want to operate in this way, because they can leverage the benefits of static typing.

Remember, these things don't come from whims of clever people or corporations. LISP and Haskell both come from research of very fundamental ideas. his is much more like real science than social science. And we should keep our eyes open to the future of this field, instead of proclaiming that we're finally done with our "100 year language".

There isn't some unified continuum of power in languages, though. There are some problems for which a H-M-ish type system is a tremendous advantage, and there are others for which it gets in the way. Same with lazy evaluation - it can make a problem much simpler or much harder to reason about. Sometimes using a language built from the ground up to support distributed and concurrent programming (e.g. Erlang), unification and backtracking (Prolog), or constraint programming is the right tool for the job. Sometimes being able to work close to the Unix kernel or run on an embedded system without an OS is more important.

Languages are a means to an end -- they help manage complexity while solving a problem. There's value in being able to extend a language's semantics to support a superset of several major language families, but unless handled very carefully, it can turn the host language into a sprawling mess in the process.

That's why we need a language with a scalable type system that supports both static and dynamic typing. (Actually VB had that years ago with Variant data types, but the rest of the type system was horrible.)

I suspect it might be fruitful to do static analysis for constraints, with type identity as just one attribute. While inferring that X is an int is useful, inferring that it's an int which is always positive and less than 256 would allow a lot of other optimizations. Even if an inference engine can't completely prove something is always a (string * int list) pair, it would still be useful to know that it's (string * (either int or double) list), and the list cannot be empty. Etc. Type declarations or inference are a bit all-or-nothing, and I think being able to read through the properties that the compiler could infer (or at least confirm) would help find bugs, suggest optimizations, etc.

If I ever get past the first dozen projects on my list, I'd like to write a compiler for a dialect of Prolog designed with constraint analysis in mind. (I also need to read more about what's already been tried, first - this is just me being curious about how far constraint analysis could go and wondering out loud.) It would be tricky, but more feasible with Prolog-like semantics than in, say, C.

Isn't it just possible that these Hindley-Milner languages are the next step in the progression that LISP started?

It's possible they've discovered interesting ideas, but they're not on the line of development Lisp started. They grow more out of the Algol tradition.

I don't think anyone has ever proclaimed "we're finally done with our 100 year language." That would be extremely unlikely. The question is more which present languages are on the path to it.

    Dynamic typing is certainly embeddable within Haskell. Just define a data type Dyn which is the sum of types Int, String, Float, [Dyn], and Dyn -> Dyn. You will see how much LISP is possible without much added syntactic cruft. But Haskell users don't need and want to operate in this way, because they can leverage the benefits of static typing.

An enumerated type (as you describe) is very simple to implement, but true dynamic typing is a bit more complicated, and there's some constructs from dynamically-typed languages which can't be expressed within the bounds of Haskell's type system.

The Data.Dynamic[1] module is a pretty good implementation of dynamic typing in Haskell, but it's still a bit awkward to use compared to a language which supports dynamic typing natively (eg Python).

[1] http://www.haskell.org/ghc/docs/latest/html/libraries/base-4...

> Isn't it just possible that these Hindley-Milner languages are the next step in the progression that LISP started?

Why do these preclude lisp? It seems to me you're drawing a false dichotomy.

Well, I'm arguing from the Bob Harper standpoint that untyped lambda calculus is simply a mode of use of typed lambda calculus. That is, untyped lambda calculus is typed lambda calculus with exactly one type.

Now, neither LISP nor Haskell are just lambda calculus... but I look at Haskell as a bigger language that subsumes most of LISP.

But, Haskell has no answer to LISP macros.

> But, Haskell has no answer to LISP macros.

Lazy evaluation. Template Haskell. Quasiquotations (EDSLs). The GHC API. (Liskell.)

I believe between them anything Lisp macros can do they can do, though it may not be so easy. (Strangely, we seem to have little need of them, but I will leave it to the reader as to whether this is a Blub situation or regular Haskell is just that good.)

My theory, as someone admittedly inexpert in both Lisp and Haskell but having used both on occasion, is that both are so ridiculously powerful in different ways compared to "normal" languages that very few people ever even reach the point of being comfortable Blub programmers in either, never mind hitting the wall and wondering "what's next?" [0]

Really, how many programmers are out there who simultaneously 1) spend enough time with either language to master it 2) are smart enough to not only realize the language is limiting them, but to invent a new language to surpass those limits 3) aren't heavily tied to their current language 4) have enough spare time to bootstrap a new language with all the associated scaffolding (libraries, &c.) needed for anyone to want to try it?

[0] Feel free to substitute Scheme, an ML dialect, or other related languages into that sentence.

For something going beyond (but including) Haskell, see Curry [1]. Curry marries functional and logic programming, and compiles down to Haskell in recent versions.

[1] http://en.wikipedia.org/wiki/Curry_(programming_language)

The "may not be so easy" part is the trick.

But that's a tradeoff you make. If you have anything more than the most rudimentary of basic syntax, macros become hard. Whereas most of the things that Haskell brings to the table can be incorporated into any other functional language.

I, as a Haskeller, envy Lispers as well. Metaprogramming is so incredibly easy in Lisp. Oftentimes I can find good solutions in Haskell for things that people would normally do using metaprogramming, but still. And almost always the types really help me, but every now and then they get in the way.

All I really want in a language is something similar to Haskell, with Lisp-like metaprogramming, easy support for Erlang-style concurrency, and syntax that doesn't immediately terrify the Blub programmers. Is that so much to ask?

Metaprogramming and syntax don't go that well together. If you're going to manipulate programs effectively, you're going to be operating on source trees, and it's easiest to do that when source code and source tree are of the same form.

Thank you for picking up on that half of my joke. Everyone else went for the type system part...

Not that it isn't easy to make a candied Lisp dialect with "friendly" syntax more appealing to non-Lisp programmers, but anyone who actually learns a Lisp quickly realizes that it just gets in the way...

If I were to personally try creating an "ideal language" to match my tongue-in-cheek plea, I'd probably start with a terse Lisp-like syntax and a HM-like type system and go from there.

Yes I think that is too much to ask. Metaprogramming, I think, is by definition incompatible with compile time type checking.

The "meta" in Metaprogramming means that the program is able to manipulate itself (recursively) at runtime. I believe it's theoretically impossible to combine that with AOT type checking and Haskell without its type system would not be Haskell at all.

[edit] Of course you could argue that the sort of thing C++ does with templates is metaprogramming as well, only at compile time. In that case I suggest that we need another term for that because it's totally unlikle what Lisp is so good at.

A large part of what constitutes lisp-style metaprogramming is macros, which are definitely a compile-time feature.

That's a very valid point. So I think we have these two types of metaprogramming, the runtime type and the compile time type. The latter is compatible with static typing but the former clearly is not.

If the compiler is part of the runtime (i.e. an interpreter that compiles expressions in order to evaluate them), and then the type checker is part of the compiler, even eval should be able to throw type errors. What, then, is the problem?

The problem is that the type error occurs at runtime, which is exactly what static type checking is supposed to prevent in the first place.

It depends on what you mean by "runtime." If you're creating a new datatype, is that really "at runtime," or are you just talking to the compiler interactively?

To put it another way: metaprogramming might have effects at runtime, but not the kind of effects that change depending on what the runtime does. Metaprogramming should be deterministic for your program to be considered to be "in production."

It's not a statically typed language like Haskell anymore?

you might want to google "staged programming" - it's what strongly typed languages do to get metaprogramming.

(edit: afaik it doesn't allow the full "recursive manipulation" you describe, but it does let you generate code with code)

Thank you so much! I've had informal ideas quite like this from working on my "HTML rewriting system" part of my (unreleased) Common Lisp web framework. I now have a very specific search term to acquire more formalized knowledge which should speed up some future developments a lot because I won't have to independently come up with solutions to as many of the problems now.

heh. no problem. if you're interested in this kind of thing check out http://lambda-the-ultimate.com from time to time. just skimming the conversations can keep you up to date...

If I create syntax with a program (code with code) is that metaprogramming? Doesn't c++ or any text generating code already do this?

Maybe it's not "native" and I have to run system level calls to force compiles and execution of created code.

Almost all programs do metaprogramming (interpreted code, scala on the jvm, etc).

What's the differentiatior for Lisps Metaprogramming (self contained languange structures?).

I'm pretty interested in simplification of programmer interfaces on iteratively more functional (but perhaps more complex) underbellies. I will certainly investigate.

My definition of metaprogramming is writing a program that manipulates itself. I'm not sure I would call any arbitrary code generator metaprogramming as this would cause the "self" in my definition suffer a severe identity crisis :-)

But yes, C++ templates as well as Lisp macros are compile time metaprogramming. All languages that have eval and/or allow function/method bindings to be replaced at runtime allow runtime metaprogramming.

Since static typing guarantees certain invariants it cannot be compatible with a program that violates those invariants at runtime. You could still decide to consider it metaprogramming when the program manipulates itself only within the limits of those guarantees, but when you look at what real world runtime metaprogramming is being used for (for instance in Rails) you will realise that these things (e.g method_missing) would not be possible within the limits of a statically typed language.

[edit] Lisp makes both compile time and runtime metaprogramming exceptionally easy due to its homoiconic nature.

It's certainly possible for a typed language like ML or Haskell to support metaprogramming. As other have noted, Haskell has Template Haskell. However, systems like MetaML and MetaOCaml support metaprogramming and give much stronger typing guarantees than Template Haskell. See http://www.metaocaml.org/examples/ for inspiration.

Sure, but there are theoretical limits to what runtime metprogramming can do whilst upholding the guarantees that a static type system provides.

For instance, in order to check that a particular function call conforms to the function signature, the function signature must at least exist. Type checking a call to a function for which not even the signature exists anwhere within the system is impossible.

The post above doesn't mention compile time type checking. Are you implying that to support one of the features he asks for that it is required?

One of Haskell's defining features is a very expressive and powerful static type system, so yeah, compile-time type inference/checking was pretty heavily implied by "similar to Haskell".

And yes, the fact that I was essentially asking for a language with both static typing and runtime self-modification was the reason for the tongue-in-cheek "is that so much to ask". Might as well ask for a program that can compile any legal perl program [0].

That said, I suspect there's a lot that could be done to allow certain subsets of metaprogramming techniques in a static-typed language; some sort of crazy meta-type system that lets the compiler prove that something will only produce code with a particular polymorphic type, maybe? I don't know.

[0] EDIT: My bad, it's actually parsing perl that's impossible, cf. http://www.perlmonks.org/?node_id=663393

Yes. A lot of the meta-programming is actually programming over the structure. For example, given a Java class, you could generate database code. In Java, this relies heavily on introspection, but you if you're interested in doing this in a type-safe way then you could look at Generic Programming (in Haskell).

Also, you can generate code (at runtime) which is type-safe, compile the code, etc. Everything you ask for is possible in Haskell, however, it is definitely more complicated than in Lisp.

Compiling code at runtime may improve performance, but it doesn't give the kind of guarantees that static typing gives you.

Static type checking proves that certain invariants hold at runtime and hence that certain defects are not possible. If a program can manipulate itself at runtime in arbitrary ways (not just reflect on itself in a read-only fashion), those proofs become invalid.

Yes I do imply that. camccann puts it very well, and as I said, what would Haskell be without its static type system?

There would still remain some cool parts. Like the syntax. And you could do something similar to typeclasses with duck typing.

Lisp-like metaprogramming is really tricky, because a lot of the language that you interact with is defined by macros. The Haskell that you see is a thick layer of syntactic sugar over a much smaller Haskell core.

Lisp-like metaprogramming is really tricky

Whoa. That doesn't match my experience at all. I'd say it's simple and practically effortless (compared to what you'd have to do elsewhere). It would be surprising if it weren't, given that the whole language is organized around code=data.

I think he meant Lisp-like metaprogramming in Haskell.

Ah. Thanks for the clarification.

You should check out the Qi language, though the developer who has historically contributed the most to it has departed some months ago. It has the ideas you are looking for (sans message-passing concurrency, I think.)

Haskell has message passing concurrency within a single (OS) process. It's not quite erlang-style concurrency, but it gives you a lot of it.

http://cvs.haskell.org/Hugs/pages/libraries/base/Control-Con...

If someone built an AMQP library for haskell, we'd probably be 90% of the way there.

All I want is a language which can tap directly in to my imagination and rapidly create machine functionality and simulations.

Couldn't you obtain most of that within a Lisp library?

Does Template Haskell affect this discussion at all? (That's not a rhetorical question - I haven't tried any Template Haskell, so I'm genuinely interested to know what it's capable of and how it compares to Lisp macros.)

TH does a lot of the same stuff. On a practical level, it isn't as good, though. The implementation leaves a lot to be desired. For example, I once tried to use TH to check XMonad configurations at compile-times (specifically, that their keymaps were sensible), but turns out TH cannot operate on anything defined in the same module!

See http://www.haskell.org/pipermail/xmonad/2008-November/006723... (if you don't like context, scroll down to 'I had to admit failure').

Can you give an example of a situation where Haskell type checking gets in your way? Not to take issue, but because I'm a collector. I've been told that it's not that easy to make a list of functions in Haskell unless they all have the same input and output types, which seems restrictive to me.

It's not "easy" to make a list, but instead maybe you would be able to work with a tuple (which is easy).

More practically, Haskell is going to ask you implicitly why you want a heterogenous list of functions. In all likelihood they're going to have a common interface at some point and you abstract around that so that the list is still "homogenous" (which is harder, but still pretty easy).

Well, polymorphic lists are one thing that are somewhat awkward. Other thing is that a small change in some part of program (say introducing an additional parameter, or having to use IO or state) can require massive changes in rest of the program, or a PhD in category-theory.

And a language that is more and more usable for non-pure math tasks.

Reply on news.ycombinator.com