Hacker News Reader

Comment by phplovesong | Hacker News Reader

I find Zig syntax noicy. I dont like the @TypeOf (at symbol) and pals, and the weird .{.x} syntax feels off.

Zig has some nice things going on but somehow code is really hard to read, admitting its a skill issue as im not that versed in zig.

View on HN

Zig is noisy and and the syntax is really not elegant. One reason I like odin's syntax, it's minimal and so well thought out.

Yes, Zig’s syntax is a bit noisier, but it enables things such as using if/for/while/switch in expressions, or using anonymous struct literals to emulate named and default parameters in functions.

The dot is just a placeholder for an inferred type, and IMHO that makes a lot of sense. E.g. you can either write this:

    const p = Point{ .x = 123, .y = 234 };

...or this:

    const p: Point = .{ .x = 123, .y = 234 };

When calling a function which expects a Point you can omit the verbose type:

    takePoint(.{ .x = 123, .y = 234 });

In Rust I need to explicitly write the type:

    takePoint(Point{ x: 123, y: 234);

...and in nested struct initializations the inferred form is very handy, e.g. Rust requires you to write this (not sure if I got the syntax right):

    const x = Rect{
        top_left: Point{ x: 123, y: 234 },
        bottom_right: Point{ x: 456, y: 456 },
    };

...but the compiler already knows that Rect consists of two nested Points, so what's the point of requiring the user to type that out? So in Zig it's just:

    const x = Rect{
        .top_left = .{ .x = 123, .y = 234 },
        .bottom_right = .{ .x = 456, .y = 456 },
    };

Requiring the explicit type on everything can get noisy really fast in Rust.

Of course the question is whether the leading dot in '.{' could be omitted, and personally I would be in favour of that. Apparently it simplifies the parser, but such implementation details should get in the way of convenience IMHO.

And then there's `.x = 123` vs `x: 123`. The Zig form is copied from C99, the Rust form from Javascript. Since I write both a lot of C99 and Typescript I don't either form (and both Zig and Rust are not even close to the flexibility and convenience of the C99 designated initialization syntax unfortunately).

Edit: fixed the Rust struct init syntax.

Zig is planning to get rid of explicit `T{}` syntax, in favor of only supporting inferred types.

https://github.com/ziglang/zig/issues/5038

So the explanation of a dot standing in for a type doesn't make sense in the long run.

Ah, I wasn't aware of that proposal. But yeah in that case I would also heavily prefer to "drop the dot" :)

IMHO Odin got it exactly right. For a variable with explicit type:

   a_variable : type = val;

...or for inferred type:

   a_variable := val;

...and the same for constants:

   a_const : type : val;
   a_const :: val;

...but I think that doesn't fit into Zig's parser design philosophy (e.g. requiring some sort of keyword upfront so that the parser knows the context it's in right from the start instead of delaying that decision to a later time).

Ahh, a perfect example of why Zig is uninteresting to me:

https://github.com/ziglang/zig/issues/5038#issuecomment-2441...

A language hostile to LSP/intellisense.

How is it hostile? The lsp can do type inference just the same as the compiler.

Yes. And by "do type inference just the same as the compiler", that includes having to do comptime execution. Plus with no interfaces/traits/concepts and comptime being mostly any-typed makes it difficult to have helpful intellisense.

That's... honestly really disappointing. I use explicit `T{}` because otherwise becomes too unreadable, too Assembly-like: I like knowing what types I'm using. It also provides a convenient thing to click on to inspect the type. I genuinely do not understand this headlong pursuit of conciseness to the detriment of readability.

Thanks for the explanation, but I don’t think you’ve sold me on .x

Think I’d rather do the Point{} syntax.

Because we've said x is a constant we're obliged to specify its type. For variables we're allowed to use inference and in most cases the type can be correctly inferred, but for constants or function signatures inference is deliberately prohibited.

    const x: Rect = ....

[Note that in Zig what you've written isn't a constant, Zig takes the same attitude as C and C++ of using const to indicate an immutable rather than a constant]

> Zig takes the same attitude as C and C++ of using const to indicate an immutable rather than a constant

I think it's a bit more complicated than that: AFAIK Zig consts without explicit type may be comptime_int or comptime_float, and those don't exist at runtime. Only consts with an explicit type annotation are 'runtime consts'.

Also see:

https://www.godbolt.org/z/esTr463bT

...still, I think Rust should allow to infer the type at least inside struct initialization, it would make designated init code like this a lot less noisy (Zig would suffer from the same problem if it hadn't the .{} syntax):

https://github.com/floooh/sokol-rust/blob/main/examples/texc...

...C99 is still the 'benchmark' when it comes to struct initialization:

https://github.com/floooh/sokol-samples/blob/29d5e9f4a56ae18...

Surely Rust can infer the type in your example? It just doesn't provide Zig's syntax to use the inferred type to manually initialize. If you wrote some_fn() here where some_fn's return type was genericised, Rust would ask for the appropriately typed some_fn not say it doesn't know the type.

> Surely Rust can infer the type in your example?

Well in Rust code like this:

    pass_action.colors[0] = sg::ColorAttachmentAction {
        load_action: sg::LoadAction::Clear,
        clear_value: sg::Color { r: 0.25, g: 0.5, b: 0.75, a: 1.0 },
        ..Default::default()
    };

...I cannot write:

    pass_action.colors[0] = {
        load_action: sg::LoadAction::Clear,
        clear_value: { r: 0.25, g: 0.5, b: 0.75, a: 1.0 },
        ..Default::default()
    };

...even though the Rust compiler has all the type information it needs (from the 'left-hand-side').

For comparison, in Zig it would look like this:

    pass_action.colors[0] = .{
        .load_action = .Clear,
        .clear_value = .{ .r=0.25, .g=0.5, .b=0.75, .a=1.0 },
    };

...Zig is still only halfway there compared to C99 (e.g. Zig doesn't allow designator chaining and is much less flexible for initializing nested arrays - in those areas it's closer to Rust than C).

Right but your call to Default::default() gives the game away that we do have inference. Default::default() is generic, if we didn't have inference we'd need to tell it which of the enormous number of implementations of that trait it should call.

Would you take syntax like clear_value: _ { r: 0.25, g: 0.5, g: 0.75, a: 1.0 } ?? Then we're saying that we know we need to pick a type here but the type can be inferred where our underscore was, just like when we

let words: Vec<_> = "A sentence broken by spaces".split_whitespace().collect();

Reply on news.ycombinator.com