That one’s interesting to me because it speaks to both sides of the debate. Yes, safe Rust is good for catching those kinds of errors at compile time. But also, it should theoretically be very easy for a compiler to avoid those problems in Zig, too, if you are using the language the way it wants to be used.

Which, from what I’ve experienced so far, does seem to take a whole lot more effort if you’re using a coding agent. I spend an incredible amount of time making sure mine doesn’t bloat our codebase with Java-flavored Python, and all the noise and defects and performance problems that it brings.

> using the language the way it wants to be used.

This sounds like just the coding conventions dependency they're trying to avoid.

Yup. But that also gets to the “they’re both right” angle. Zig’s convention may not be how the Bun developers wanted to work, but it’s a good choice for the kinds of projects Zig is designed for. Which overlaps a bit with, but isn't the same as, Rust’s intended use cases. For example, I might rather go with the Zig philosophy if I’ve got hard memory constraints and want to avoid dynamic memory allocation.

Which, ironically, is something I’d love for my browser’s JavaScript runtime to do. But I can also respect others not wanting that.

IMO it's the same problem with Rust. LLMs are trained on vast quantities of code that violate Rust soundness in safe code.

LLMs are prediction engines: you can move the needle (heavily) on the predictions from lazy to correct by construction, but the defaults even on high end models like Opus are always riddled with shortcuts.

LLMs can produce coherent & safe Rust 1.92, LLMs can produce coherent Zig 0.16, but that's dependent on in context learning & instruction following.

> LLMs are trained on vast quantities of code that violate Rust soundness in safe code.

Why is your reasoning behind this? Most Rust code in the wild does not use unsafe. Most projects don't have a single line of unsafe.

> that's dependent on in context learning & instruction following

Not just that. For Rust in particular, since unsafe code is uncompilable code (unless explicitly marked "unsafe"), the LLM is essentially forced to continue iterating even if the code already works but is unsafe, until it finds an implementation that's both correct (per the spec) and safe (per the compiler). That's the difference from Zig, where "make the code safe" is essentially part of the spec, and so may be missed by the LLM and not be discovered by a human until there's an observable runtime issue, or CVE.

> But also, it should theoretically be very easy for a compiler to avoid those problems in Zig, too, if you are using the language the way it wants to be used.

agreed. in my side project im tinkering with a memory safety checker in zig that intercepts a compiler artifact, and it works better with idiomatic zig, problematic code is when you try to write c-isms in zig -- so i basicallt tell the checker to reject c-isms that create ambiguity for safety checking.

The way I'm thinking of it is that they ported Bun from Zig to Unsafe Rust, and are still working on now porting it from Unsafe Rust to Rust (not 100% safe, but a reasonable amount).

I'm optimistic for that refactoring. But if one in 30 lines sprinkled all over everywhere is unsafe then you haven't actually done much to improve safety. When you have a large percent of fully safe functions and modules, that's when you're making a difference. Counting the lines is not a very helpful metric.

> one in 30 lines

> Counting the lines is not a very helpful metric.

Seems self-contradictory.

And they did say they're working to reduce the unsafe code over time. That huge percentage of memory bugs that they were able to better triage and fix due to the port was also them improving safety.

The other part of that line is important. Measuring lines is better than nothing but the distribution is much more important. If they're scattered all over that's a bigger downside than the line reduction is an upside.

If you want good metrics, measure functions and data structures without unsafe. Not lines.

> If you want good metrics, measure functions and data structures without unsafe.

I doubt it, as it's possible a substantial amount of those unsafe areas can't even be removed, given the project's dependent on C/C++ libraries. Without proper calibration those metrics won't tell the full story, and with calibration the incentive for the story will decrease over time with the unsafe refactoring. I doubt the distribution of the unsafe code really matters that much; what matters is proper handling, wherever it is, so stability and security isn't compromised.