[TransAtlToonz]

I don't understand why rust bindings imply a freezing (or chilling) of the ABI—surely rust is bound by roughly the same constraints C is, being fully ABI-compatible in terms of consuming and being consumed. Is this commentary on how Rust is essentially, inherently more committed to backwards compatibility, or is this commentary on the fact that two languages will necessarily bring constraints that retard the ability to make breaking changes?

[AshamedCaptain]

Obviously the latter, which is already the point of contention that has started this entire discussion.

[TransAtlToonz]

Can you explain why you think this? I don't understand the reasoning and it's certainly not "obvious". There's certainly no technical reason implying this, so is this just resistance to learning rust? C'mon, kernel developers can surely learn new tricks. This just seems like a defeatist attitude.

EDIT: The process overhead seems straightforwardly worth it—rust can largely preserve semantics, offers the potential to increase confidence in code, and can encourage a new generation of contribution with a faster ramp-up to writing quality code. Notably nowhere here is a guarantee of better code quality, but presumably the existing quality-guaranteeing processes can translate fine to a roughly equivalently-capable language that offers more compile-time mechanisms for quality guarantees.

[AshamedCaptain]

You phrased it rather well -- "increased constraints will retard the ability to make breaking changes". You are adding a second layer of abstraction that brings very little generalization, but still doubles the mental load; there's no way it doesn't significant put additional pressure when making breaking changes. The natural reaction is that there will be less such breaking changes and interfaces will ossify. One can even argue this is what has already happened here.

In addition, depending on the skill of the "binding writer", the second set of interfaces may simply be actually easier to use (and generally true, since the rust bindings are actually designed instead of evolved organically). This is yet another mental barrier. There may not even be a point to evolving one interface, or the other. Which just further contributes to splitting the project into two worlds.

[kelnos]

> The natural reaction is that there will be less such breaking changes and interfaces will ossify. One can even argue this is what has already happened here.

I don't think I'd agree with that. Current kernel policy is that the C interfaces can evolve and change in whatever way they need to, and if that breaks Rust code, that's fine. Certainly some subsystem maintainers will want to be involved in helping fix that Rust code, or help provide direction on how the Rust side should evolve, but that's not required, and C maintainers can pick and choose when they do that, if at all.

Obviously if Rust is to become a first-class, fully-supported part of the kernel, that policy will eventually change. And yes, that will slow down changes to C interfaces. But I think suggesting that interfaces will ossify is an overreaction. The rate of change can slow to a still-acceptable level without stopping completely.

And frankly I think that when this time comes, maintainers who want to ignore Rust completely will be few and far between, and might be faced with a choice to either get on board or step down. That's difficult and uncomfortable, to be sure, but I think it's reasonable, if it comes to pass.

[TransAtlToonz]

> You are adding a second layer of abstraction that brings very little generalization

Presumably, this is an investment in replacing code written in C. There's no way around abstraction or overhead in such a venture.

> there's no way it doesn't significant put additional pressure when making breaking changes

This is the cost of investment.

> The natural reaction is that there will be less such breaking changes and interfaces will ossify.

A) "fewer", not "less". Breaking changes are countable.

B) A slower velocity of changes does not imply ossification. Furthermore, I'm not sure this is true—the benefits of formal verification of constraints surrounding memory-safety seems as if it would naturally lead to long-term higher velocity. Finally, I can't speak to the benefits of a freely-breakable kernel interface (I've never had to maintain a kernel for clients myself, thank god) but again, this seems like a worthwhile short-term investment for long-term gain.

> In addition, depending on the skill of the "binding writer" (and generally, since the rust bindings are actually designed instead of evolving organically), the second set of interfaces may simply be actually easier to use. There may not even be a point to evolving one interface, or the other. Which just further contributes to splitting the project.

Sure, this is possible. I present two questions, then: 1) what is lost with lesser popularity of the C interface with allegedly less stability, and 2) is the stability, popularity, and confidence in the new interface worth it? I think it might be, but I have no clue how to reason about the politics of the Linux ABI.

I have never written stable kernel code, so I don't have confident guidance myself. But I can say that if you put a kernel developer in front of me of genius ability, I would still trust and be more willing to engage with rust code. I cannot conceive of a C programmer skilled enough they would not benefit from the additional tooling and magnification of ability. There seems to be some attitude that if C is abandoned, something vital is lost. I submit that what is lost may not be of technical, but rather cultural (or, eek, egoist), value. Surely we can compensate for this if it is true.

EDIT, follow-up: if an unstable, less-used interface is desirable, surely this could be solved in the long term with two rust bindings.

EDIT2: in response to an aunt comment, I am surely abusing the term "ABI". I'm using it as a loose term for compatibility of interfaces at a linker-object level.

[dralley]

>Presumably, this is an investment in replacing code written in C. There's no way around abstraction or overhead in such a venture.

Nobody is proposing replacing code right now. Maybe that will happen eventually, but it's off limits for now.

R4L is about new drivers. Not even kernel subsystems, just drivers, and only new ones. IIRC there is a rule against having duplicate drivers for the same hardware. I suppose it's possible to rewrite a driver in-place, but I doubt anyone plans to do that.

[surajrmal]

There is a binder driver rewrite in rust. Companies who care are certainly rewriting drivers. If there is pushback in upstreaming them that will cause a lot of noise.

[biorach]

> A) "fewer", not "less". Breaking changes are countable.

this just makes you look pedantic and passive aggressive