Interview with Zig language creator Andrew Kelley [video]

I work in HFT, and one of the key concerns when writing low-latency code is "is this code allocating memory, and if so, how can I stop it?" Zig is the perfect language for this use case as none of the standard library implicitly allocates, rather for anything that allocates, the caller must pass in an allocator. The stdlib also provides a handy arena allocator, which is often the best choice.

This is a huge advantage over C++ and Rust, because it makes it much harder for e.g. the intern to write code that repeatedly creates a vector or dynamically allocated string in a loop. Or to use something like std::unordered_map or std::deque that allocates wantonly.

There is some ongoing work towards custom allocators for containers in Rust std. [1]

Right now you could also go no_std (you still get the core library, which does not contain any allocating data structures) and use custom containers with a passed in allocator.

Zig is definitely a cool language, and it will be interesting if they can come up with good solutions to memory management!

But sentences like these in the documentation [2] would make me prefer Rust for most low level domains (for now):

> It is the Zig programmer's responsibility to ensure that a pointer is not accessed when the memory pointed to is no longer available. Note that a slice is a form of pointer, in that it references other memory. ...

> ... the documentation for the function should explain who "owns" the pointer

[1] https://github.com/rust-lang/wg-allocators

[2] https://ziglang.org/documentation/master/#toc-Lifetime-and-O...

> But sentences like these in the documentation [2] would make me prefer Rust for most low level domains

While such use-after-free issues are not prevented at compile time, the plan is to ultimately have safe Zig catch them (and panic) at runtime, i.e. safe Zig shouldn't have undefined behaviour. Because this is done with runtime checks, the expectation is that those checks will be turned off (selectively, perhaps) in production, after testing has satisfied you. In that case, the guarantees aren't as strong as Rusts, but those guarantees come at a significant cost -- to language complexity and compilation time -- that can also have a negative effect on correctness. So while Zig's approach to safety/correctness is certainly very different from Rust's, I don't think it is necessarily weaker (perhaps it could even be stronger, but the question is hard to settle).

> but those guarantees come at a significant cost -- to language complexity and compilation time

Aren't Rust's compilation time woes more due to the amount of IR the front/middle-end give to LLVM? I was under the impression that the type system-related stuff isn't that expensive.

The borrow-checking and ownership mechanism is cheap, and it's almost never a significant part of the big compilation time encountered in Rust.

What's not cheap, and is responsible for long compilation times (the order is arbitrary, the relative weights are highly dependent of the code-base):

- size of the code generation units (the whole crate vs individual files in C)

- procedural macros

- generics & traits

- interaction between generics & LLVM IR generation (a lot of bloat is created, to be removed by LLVM later)

- LLVM itself

Most of those are being worked on, but in the end it's mostly a social problem: as Rust users are used to long compile time, many of them don't especially take care of it, and most gains in the compiler are often outweighed by people writing slower code. It's already possible to write Rust code that compiles quickly, if you pay attention. The culture is evolving though, and more and more library authors are now mindful of compilation time of their crate (and the tooling to diagnose it is also improving).

Key takeaway: Memory safety isn't what makes Rust compile slowly, “zero-cost abstractions” is.