Things Zig comptime won't do

The partial evaluation/specialization is accomplished in D using a template. The example from the link:

    fn f(comptime x: u32, y: u32) u32 {
        if (x == 0) return y + 1;
        if (x == 1) return y * 2;
        return y;
    }

and in D:

    uint f(uint x)(uint y) {
        if (x == 0) return y + 1;
        if (x == 1) return y * 2;
        return y;
    }

The two parameter lists make it a function template, the first set of parameters are the template parameters, which are compile time. The second set are the runtime parameters. The compile time parameters can also be types, and aliased symbols.

Here is, I think, an interesting example of the kind of thing TFA is talking about. In case you’re not already familiar, there’s an issue that game devs sometimes struggle with, where, in C/C++, an array of structs (AoS) has a nice syntactic representation in the language and is easy to work with/avoid leaks, but a struct of arrays (SoA) has a more compact layout in memory and better performance.

Zig has a library to that allows you to have an AoS that is laid out in memory like a SoA: https://zig.news/kristoff/struct-of-arrays-soa-in-zig-easy-i... . If you read the implementation (https://github.com/ziglang/zig/blob/master/lib/std/multi_arr...) the SoA is an elaborately specialized type, parameterized on a struct type that it introspects at compile time.

It’s neat because one might reach for macros for this sort of the thing (and I’d expect the implementation to be quite complex, if it’s even possible) but the details of Zig’s comptime—you can inspect the fields of the type parameter struct, and the SoA can be highly flexible about its own fields—mean that you don’t need a macro system, and the Zig implementation is actually simpler than a macro approach probably would be.

Using a different type vs. a different syntax can be an important usability consideration, particularly since D also has templates and other features, where Zig provides only the comptime type for all of them. Homogeneity can also be a nice usability win, though there are downsides as well.

D doesn't have a macro system, either, so I don't understand what you mean.

Zig's use of comptime in a function argument makes it a template :-/

I bet if you use such a function with different comptime arguments, compile it, and dump the assembler you'll see that function appearing multiple times, each with somewhat different code generated for it.

> Zig's use of comptime in a function argument makes it a template :-/

That you can draw an isomorphism between two things does not mean they are ergonomically identical.

When we're responding to quite valid points about other languages having essentially the same features as Zig with subjective claims about ergonomics, the idea that Zig comptime is "revolutionary" is looking awfully flimsy. I agree with Walter: Zig isn't doing anything novel. Picking some features while leaving others out is something that every language does; if doing that is enough to make a language "revolutionary", then every language is revolutionary. The reality is a lot simpler and more boring: for Zig enthusiasts, the set of features that Zig has appeals to them. Just like enthusiasts of every programming language.

I'm sorry, but not being able to see that a design that uses a touchscreen to eliminate the keyboard is novel despite the touchscreen itself having been used elsewhere alongside a keyboard, shows a misunderstanding of what design is.

Show me the language that used a general purpose compile-time mechanisms to avoid specialised features such as generics/templates, interfaces/typeclasses, macros, and conditional compilation before Zig, then I'll say that language was revolutionary.

I also find it hard to believe that you can't see how replacing all these features with a single one (that isn't AST macros) is novel. I'm not saying you have to think it's a good idea -- that's a matter of personal taste (at least until we can collect more objective data) -- but it's clearly novel.

I don't know all the languages in the world and it's possible there was a language that did that before Zig, but none of the languages mentioned here did. Of course, it's possible that no other language did that because it's stupid, but that doesn't mean it's not novel (especially as the outcome does not appear stupid on the face of it).

>for Zig enthusiasts, the set of features that Zig has appeals to them. Just like enthusiasts of every programming language.

I find it rather amusing that it's a Java and a Rust enthusiast who are extolling Zig approach here! I am not particularly well read with respect to programming languages, but I don't recall many languages which define generic pair as

    fn Pair(A: type, B: type) type {
        return struct { fst: A, snd: B };
    }

The only one that comes to mind is 1ML, and I'd argue that it is also revolutionary.

Well, if you strip away the curly braces and return statement, that's just a regular type definition. Modeling generic types as functions from types to types is just System F, which goes back to 1975. Turing-complete type-level programming is common in tons of languages, from TypeScript to Scala to Haskell.

I think the innovation here is imperative type-level programming--languages that support type-level programming are typically functional languages, or functional languages at the type level. Certainly interesting, but not revolutionary IMO.

I might be misunderstanding something, but this is how it works in D:

    struct Pair(A, B) { A fst; B snd; }

    Pair!(int, float) p; // declaration of p as instance of Pair

It's just a struct with the addition of type parameters.

But Zig's comptime only approximates the features you mentioned; it doesn't fully implement them. Which is what the original article is saying. To use your analogy, using a touchscreen to eliminate a keyboard isn't very impressive if your touchscreen keyboard is missing keys.

If you say that incomplete implementations count, then I could argue that the C preprocessor subsumes generics/templates, interfaces/typeclasses†, macros, and conditional compilation.

†Exercise for the reader: build a generics system in the C preprocessor that #error's out if the wrong type is passed using the trick in [1].

[1]: https://stackoverflow.com/a/45450646

The thing is, this is not type-level programming, this is term-level programming. That there's no separate language of types is the feature. Functional/imperative is orthogonal. You can imagine functional Zig which writes

    Pair :: type -> type -> type
    let Pair a b = product a b 

This is one half of the innovation, dependent-types lite.

The second half is how every other major feature is expressed _directly_ via comptime/partial evaluation, not even syntax sugar is necessary. Generic, macros, and conditional compilation are the three big ones.

> This is one half of the innovation, dependent-types lite.

But that's not dependent types. Dependent types are types that depend on values. If all the arguments to a function are either types or values, then you don't have dependent types: you have kind polymorphism, as implemented for example in GHC extensions [1].

> The second half is how every other major feature is expressed _directly_ via comptime/partial evaluation, not even syntax sugar is necessary. Generic, macros, and conditional compilation are the three big ones.

I'd argue that not having syntactic sugar is pretty minor, but reasonable people can differ I suppose.

[1]: https://ghc.gitlab.haskell.org/ghc/doc/users_guide/exts/poly...

> Dependent types are types that depend on values.

Like this?

    fn f(comptime x: bool) if (x) u32 else bool {
        return if (x) 0 else false;
    }

> But Zig's comptime only approximates the features you mentioned; it doesn't fully implement them

That's like saying that a touchscreen device without a keyboard only approximates a keyboard but doesn't fully implement one. The important thing is that the feature performs the duty of those other features.

> If you say that incomplete implementations count, then I could argue that the C preprocessor subsumes generics/templates, interfaces/typeclasses†, macros, and conditional compilation.

There are two problems with this, even if we assumed that the power of C's preprocessor is completely equivalent to Zig's comptime:

First, C's preprocessor is a distinct meta-language; one major point of Zig's comptime is that the metalanguage is the same language as the object language.

Second, it's unsurprising that macros -- whether they're more sophisticated or less -- can do the role of all those other features. As I wrote in my original comment (https://news.ycombinator.com/item?id=43745438) one of the exciting things about Zig is that a feature that isn't macros (and is strictly weaker than macros, as it's referentially transparent) can replace them for the most part, while enjoying a greater ease of understanding.

I remember that one of my first impressions of Zig was that it evoked the magic of Lisp (at least that was my gut feeling), but in a completely different way, one that doesn't involve AST manipulation, and doesn't suffer from many of the problems that make List macros problematic (i.e. creating DSLs with their own rules). I'm not saying it may not have other problems, but that is very novel.

I hadn't seen any such fresh designs in well over a decade. Now, it could be that I simply don't know enough languages, but you also haven't named other languages that work on this design principle, so I think my excitement was warranted. I'll let you know if I think that's not only a fresh and exciting design but also a good one in ten years.

BTW, I have no problem with you finding Zig's comptime unappealing to your tastes or even believing it suffers from fundamental issues that may prove problematic in practice (although personally I think that, when considering both pros and cons of this design versus the alternatives, there's some promise here). I just don't understand how you can say that the design isn't novel while not naming one other language with a similar core design: a mechanism for partial evaluation of the object language (with access to additional reflective operations) that replace those other features I mentioned (by performing their duty, if not exactly their mode of operation).

For example, I've looked at Terra, but it makes a distinction between the meta language and the object (or "runtime") language.

No, dependent types depend on runtime values.

>I'm not saying it may not have other problems, but that is very novel.

Just to explicitly acknowledge this, it inherits the C++ problem that you don't get type errors inside a function until you call the function and, when that happens, its not always immediately obvious whether the problem is in the caller or in the callee.

Yeah, that one Zig can not do, hence "-lite".

That's still just a function of type ∀K∀L.K → L with a bound on K. From a type theory perspective, a comptime argument, when the function is used in such a way as to return a type, is not a value, even though it looks like one. Rather, true or false in this context is a type. (Yes, really. This is a good example of why Zig reusing the keyword "comptime" obscures the semantics.) If comptime true or comptime false were actually values, then you could put runtime values in there too.

The point is that comptime isn't dependent types at all. If your types can't depend on runtime values, they aren't dependent types. It's something more like kind polymorphism in GHC (except more dynamically typed), something which GHC explicitly calls out as not dependent types. (Also it's 12 years old [1]).

[1]: https://www.seas.upenn.edu/~sweirich/papers/fckinds.pdf

IIUC, it does have code generation—the ability to generate strings at compile-time and feed them back into the compiler.

The argument that the author of TFA is making is that Zig’s comptime is a very limited feature (which, they argue, is good. It restricts users from introducing architecture dependencies/cross-compilation bugs, is more amenable to optimization, etc), and yet it allows users to do most of the things that more general alternatives (such as code generation or a macro system) are often used for.

In other words, while Zig of course didn’t invent compile-time functions (see lisp macros), it is notable and useful from a PL perspective if Zig users are doing things that seem to require macros or code generation without actually having those features. D users, in contrast, do have code generation.

Or, alternatively, while many languages support metaprogramming of some kind, Zig’s metaprogramming language is at a unique maxima of safety (which macros and code generation lack) and utility (which e.g. Java/Go runtime reflection, which couldn’t do the AoS/SoA thing, lack)

Edit Ok, I think Zig comptime expressions are just like D templates, like you said. The syntax is nicer than C++ templates. Zig’s “No host leakage” (to guarantee cross-compile-ability) looks like the one possibly substantively different thing.

> Picking some features while leaving others out is something that every language does; if doing that is enough to make a language "revolutionary", then every language is revolutionary.

Picking a set of well motivated and orthogonal features that combine well in flexible ways is definitely enough to be revolutionary if that combination permits expressive programming in ways that used to be unwieldy, error-prone or redundant, eg. "redundant" in the sense that you have multiple ways of expressing the same thing in overlapping but possibly incompatible ways. It doesn't follow that every language must be revolutionary just because they pick features too, there are conditions to qualify.

For systems programming, I think Zig is revolutionary. I don't think any other language matches Zig's cross-compilation, cross-platform and metaprogramming story in such a simple package. And I don't even use Zig, I'm just a programming language theory enthusiast.

> I agree with Walter: Zig isn't doing anything novel.

"Novel" is relative. Anyone familiar with MetaOCaml wouldn't have seen Zig as particularly novel in a theoretical sense, as comptime is effectively a restricted multistage language. It's definitely revolutionary for an industry language though. I think D has too much baggage to qualify, even if many Zig expressions have translations into D.

> Zig’s “No host leakage” (to guarantee cross-compile-ability) looks like the one possibly substantively different thing.

That is a good idea, but could be problematic if one relies on size_t, which changes in size from 32 to 64 bit. D's CTFE adds checks for undefined behavior, such as shifting by more bits than are in the type being shifted. These checks are not done at runtime for performance reasons.

D's CTFE also does not allow calling the operating system, and only works on functions that are "pure".

> The important thing is that the feature performs the duty of those other features.

Zig's comptime doesn't do everything that Rust (or Java, or C#, or Swift, etc.) generics do, and I know you know this given your background in type theory. Zig doesn't allow for the inference and type-directed method resolution that Rust or the above languages do, because the "generics" that you create using Zig comptime aren't typechecked until they're instantiated. You can improve the error messages using "comptime if" or whatever Zig calls it (at the cost of a lot of ergonomics), but the compiler still can't reliably typecheck the bodies of generic functions before the compiler does the comptime evaluation.

Now I imagine you think that this feature doesn't matter, or at least doesn't matter enough to be worth the complexity it adds to the compiler. (I disagree, of course, because I find reliable IDE autocomplete and inline error messages to be enormously useful when writing generic Rust functions.) But that's the entire point: Zig comptime is not performing the duty of generics; it's approximating generics in a way that offers a tradeoff.

When I first looked at Zig comptime, it didn't evoke the "magic of Lisp" at all in me (and I do share an appreciation of simplicity in programming languages, though I feel like Scheme offers more of that than Lisp). Rather, my reaction was "oh, this is basically just what D does", having played with D a decent amount in years prior. Nothing I've seen in the intervening years has changed that impression. Zig's metaprogramming features are a spin on metaprogramming facilities that D thoroughly explored over a decade before Zig came on the scene.

Edit: Here's an experiment. Start with D and start removing features: GC, the class system, exceptions, etc. etc. Do you get to something that's more or less Zig modulo syntax? From what I can tell, you do. That's what I mean by "not revolutionary".

> Zig doesn't allow for the inference and type-directed method resolution that Rust or the above languages do

Well, but Zig also doesn't allow for overloads and always opts for explicitness regardless of comptime, so I would say that's consonant with the rest of the design.

> Now I imagine you think that this feature doesn't matter, or at least doesn't matter enough to be worth the complexity it adds to the compiler.

I don't care too much about the complexity of the compiler (except in how compilation times are affected), but I do care about the complexity of the language. And yes, there are obviously tradeoffs here, but they're not the same tradeoffs as C++ templates and I think it's refreshing. I can't yet tell how "good" the tradeoff is.

> Here's an experiment. Start with D and start removing features: GC, the class system, exceptions, etc. etc. Do you get to something that's more or less Zig modulo syntax?

I don't know D well enough to tell. I'd probably start by looking at how D would do this [1]: https://ziglang.org/documentation/master/#Case-Study-print-i...

For instance, the notion of a comptime variable (for which I couldn't find an analogue in D) is essential to the point that the "metalanguage" and the object language are pretty much the same language.

Interestingly, in Zig, the "metalanguage" is closer to being a superset of the object language whereas in other languages with compile-time phases, the metalanguage, if not distinct, is closer to being a subset. I think Terra is an interesting point of comparison, because there, while distinct, the metalanguage is also very rich.

[1] which, to me, gives the "magical Lisp feeling" except without macros.

> Picking some features while leaving others out is something that every language does; if doing that is enough to make a language "revolutionary", then every language is revolutionary.

You can say that about the design of any product. Yet, once in a while, we get revolutionary designs (even if every feature in isolation is not completely novel) when the choice of what to include and what to leave out is radically different from other products in the same category in a way that creates a unique experience.

> the notion of a comptime variable (for which I couldn't find an analogue in D)

A comptime variable in D would look like:

    enum v = foo(3);

Since an enum initialization is a ConstExpression, it's initialization must be evaluated at compile time.

A comptime function parameter in D looks like:

    int mars(int x)(int y) { ... }

where the first parameter list consists of compile time parameters, and the second the runtime parameters.

D does not have a switch-over-types statement, but the equivalent can be done with a sequence of static-if statements:

    static if (is(T == int)) { ... }
    else static if (is(T == float)) { ... }

Static If is always evaluated at compile time. The IsExpression does pattern matching on types.

A comptime variable in Zig isn't a constant whose value is computed at compile time (that would just be a Zig constant) but rather variable that's potentially mutable by comptime: https://ziglang.org/documentation/master/#Compile-Time-Varia...

This is one of the things that allow the "comptime language" to just be Zig, as in this example: https://ziglang.org/documentation/master/#Case-Study-print-i...

You can mutate variables at compile time in D. See the compile time Newton's method example: https://tour.dlang.org/tour/en/gems/compile-time-function-ev...

I don't think that's the same thing (rather, it's more like ordinary Zig variables in code that's evaluated at compile-time), as there's no arbitrary mixing of compile-time and runtime computation. Again, compare with https://ziglang.org/documentation/master/#Case-Study-print-i...

Anyway, I found this article that concludes that D's compile time evaluation is equivalent in power to Zig's, although it also doesn't cover how comptime variables can be used in Zig: https://renato.athaydes.com/posts/comptime-programming

However, as I've said many times, knowing about the theoretical power of partial evaluation, what excites me in Zig isn't what comptime can do (although I am impressed with the syntactic elegance of the mechanism) but how it is used to avoid adding other features.

A phone with a touchscreen is evolutionary; a phone without a keypad is revolutionary. The revolution is in the unique experience of using "just comptime" for many things.

It is, of course, a tradeoff, and whether or not that tradeoff is "good" remains to be seen, but I think this design is one of the most novel designs in programming languages in many, many years.