The `Overwrite` trait and `Pin`

In July, boats presented a compelling vision in their post pinned places. With the Overwrite trait that I introduced in my previous post, however, I think we can get somewhere even more compelling, albeit at the cost of a tricky transition. As I will argue in this post, the Overwrite trait effectively becomes a better version of the existing Unpin trait, one that effects not only pinned references but also regular &mut references.

26 September 2024

Making overwrite opt-in #crazyideas

What would you say if I told you that it was possible to (a) eliminate a lot of “inter-method borrow conflicts” without introducing something like view types and (b) make pinning easier even than boats’s pinned places proposal, all without needing pinned fields or even a pinned keyword? You’d probably say “Sounds great… what’s the catch?” The catch it requires us to change Rust’s fundamental assumption that, given x: &mut T, you can always overwrite *x by doing *x = /* new value */, for any type T: Sized.

26 June 2024

More thoughts on claiming

This is the first of what I think will be several follow-up posts to “Claiming, auto and otherwise”. This post is focused on clarifying and tweaking the design I laid out previously in response to some of the feedback I’ve gotten. In future posts I want to lay out some of the alternative designs I’ve heard. TL;DR: People like it If there’s any one thing I can take away from what I’ve heard, is that people really like the idea of making working with reference counted or cheaply cloneable data more ergonomic than it is today.

21 June 2024

Claiming, auto and otherwise

This blog post proposes adding a third trait, Claim, that would live alongside Copy and Clone. The goal of this trait is to improve Rust’s existing split, where types are categorized as either Copy (for “plain old data”¹ that is safe to memcpy) and Clone (for types that require executing custom code or which have destructors). This split has served Rust fairly well but also has some shortcomings that we’ve seen over time, including maintenance hazards, performance footguns, and (at times quite significant) ergonomic pain and user confusion.

2 June 2024

The borrow checker within

This post lays out a 4-part roadmap for the borrow checker that I call “the borrow checker within”. These changes are meant to help Rust become a better version of itself, enabling patterns of code which feel like they fit within Rust’s spirit, but run afoul of the letter of its law. I feel fairly comfortable with the design for each of these items, though work remains to scope out the details.

2 May 2024

Unwind considered harmful?

I’ve been thinking a wild thought lately: we should deprecate panic=unwind. Most production users I know either already run with panic=abort or use unwinding in a very limited fashion, basically just to run to cleanup, not to truly recover. Removing unwinding from most case meanwhile has a number of benefits, allowing us to extend the type system in interesting and potentially very impactful ways. It also removes a common source of subtle bugs.

see all posts →