The Layman's Rust Thread

[lost_in_c]

Yeah but to attract people you have to give them something fun to do. Rust is apparently less fun than C. I don't think how increasing the amount of Rust code in a project gets in more volunteer developers.

It doesn't. It increases the efficiency of autogenerated software that might be used to develop Linux, and incentivizes people to come in who have no interest in writing code, but some in writing things in prompts.

What it does is help the process of divesting itself from developers. Obviate the need for more volunteer developers. Or help obviate.

 

[cracauer@]

It doesn't. It increases the efficiency of autogenerated software that might be used to develop Linux, and incentivizes people to come in who have no interest in writing code, but some in writing things in prompts.

What it does is help the process of divesting itself from developers. Obviate the need for more volunteer developers. Or help obviate.

I don't think that is going to happen for any of the OS kernels.

You have a point in that if somebody does vibe coding then Rust is probably the best language(*) to target since it has the most compile-time checks.

(* not counting some functional languages with very advanced type systems which also guide toward correctness at compile time)

 

[msplsh]

Then maybe rewatch the video. I'm sure anybody else that does with any amount of good faith will see how I arrive at that summary.

I encourage everyone to watch the video so they can ignore this thread.

 

[msplsh]

Yeah but to attract people you have to give them something fun to do. Rust is apparently less fun than C. I don't think how increasing the amount of Rust code in a project gets in more volunteer developers.

IDK, they seemed to have fun doing this. I think that's what's going on.

 

[lost_in_c]

I don't think that is going to happen for any of the OS kernels.

You have a point in that if somebody does vibe coding then Rust is probably the best language(*) to target since it has the most compile-time checks.

(* not counting some functional languages with very advanced type systems which also guide toward correctness at compile time)

If I have a point in that Rust is a good language for autogenerated software to use, much better than C (while it is not necessarily much better than C for humans), and it is also true that the pool of active developers is shrinking, and that the proportion of those that is interested in writing for Linux is also shrinking, and that the available jobs for developers is shrinking even faster, and (this is whether you accept it or not, I have no figures, but I'm pretty confident) it is a fact that many people already submit substantial portions of code to the Linux kernel that is purely machine generated, then it's not really a question of whether it will happen with any of the OSs, but of when it started happening.

 

[cracauer@]

IDK, they seemed to have fun doing this. I think that's what's going on.

For people who are already very experienced. On the level of writing drivers for a GPU for which no driver writing documentation exists.

I am more concerned with people who have talent and are starting out. Will they come in with existing Rust knowledge when they are at a level to contribute to an OS kernel?

 

[cracauer@]

If I have a point in that Rust is a good language for autogenerated software to use, much better than C (while it is not necessarily much better than C for humans), and it is also true that the pool of active developers is shrinking, and that the proportion of those that is interested in writing for Linux is also shrinking, and that the available jobs for developers is shrinking even faster, and (this is whether you accept it or not, I have no figures, but I'm pretty confident) it is a fact that many people already submit substantial portions of code to the Linux kernel that is purely machine generated, then it's not really a question of whether it will happen with any of the OSs, but of when it started happening.

Yeah, but again, the question is what generated that code in the kernel.

It is well known that AMD submits code for their GPU drivers that is automatically generated by their/out of their chip design software.

That is entirely different from using a LLM. I don't think that the Linux kernel has LLM code.

 

[lost_in_c]

I know what I'm about to ask isn't exactly what you meant, but answer it anyway:

It is well known [...] code [...] is automatically generated

That is entirely different from using a LLM.

How?

 

[cracauer@]

I know what I'm about to ask isn't exactly what you meant, but answer it anyway:

How?

?

Of course AI-generated code is different from code generated by a human-written script. The LLM makes stuff up for starters.

 

[lost_in_c]

That's not the question. The question is, if you are using autogenerated software to get code, in what way is that not autogenerated software autogenerating code?

 

[lost_in_c]

I don't think that the Linux kernel has LLM code.

That's as may be. The whole point of this thread, though, is that if there was, then:

You have a point in that if somebody does vibe coding then Rust is probably the best language(*) to target

This rather becomes relevant.

 

[cracauer@]

That's not the question. The question is, if you are using autogenerated software to get code, in what way is that not autogenerated software autogenerating code?

Generated code is fine if the generator is controlled by a competent human. Heck, they built an entire programming language around generics (C++).

 

[lost_in_c]

Generated code is fine if the generator is controlled by a competent human. Heck, they built an entire programming language around generics (C++).

What exactly do you mean by fine? Fien according to who?

 

[cracauer@]

What exactly do you mean by fine? Fien according to who?

For starters - Linus, in the case of the AMD graphics drivers.

Google also uses Python scripts to write C++ code in quite a few places.

(they did when I was there, C++ made some progress since then, so maybe they don't have to anymore)

If you want to put it that way every modern C++ program is generated from code. As is any Lisp program using compile-time computing to compute code to compile.

 

[lost_in_c]

Well, in this thread, we really are only referring to autogenerated code for which Rust would be optimal.

And, by the standards you have just laid out, if they had an autogenerated software farm autogenerating Rust code for the Linux kernel, and Linus reviewed it all before commiting to the same extent that he reviews the AMD code, that would be fine. I'm still not sure what you mean by fine. Morally acceptable? Financially viable? Technically sound? Not sure.

 

[lost_in_c]

More specifically, code autogenerated by software that is autogenerated. Basically because I believe that is the only autogenertd code for which Rust would be optimal, in the context of using it in a project where humans also participate,a t least for starters.

 

[cracauer@]

Technically sound? Not sure.

I find code generated from LLMs to be not technically sound enough. And existing code that has been changed by LLMs even less so. You will have a hard time to come up with a prompt that changes existing code in one way without redoing lots of other details. No way you pass a commit review with that.

You can be pretty sure that nobody has reviewed all the code generated for the AMD GPU driver. There's a lot of it. Would it be better off if it was generating Rust? Maybe, it would catch some fatfingering in the generating script.

 

[lost_in_c]

You can be pretty sure that nobody has reviewed all the code generated for the AMD GPU driver. There's a lot of it.

Yes, well, certainly this is apropos.

Would it be better off if it was generating Rust? Maybe, it would catch some fatfingering in the generating script.

For the same reason I think it would be inapropriate for a purely autogenerated kernel, I don't: it would be inefficient.

I find code generated from LLMs to be not technically sound enough. And existing code that has been changed by LLMs even less so. You will have a hard time to come up with a prompt that changes existing code in one way without redoing lots of other details. No way you pass a commit review with that.

But I take it then that you take it for granted that code that was autogenerated, from autogenerated software, to whatever extent it may have later been modified by a person or not, is in the kernal in substancial amounts?

 

[cracauer@]

But I take it then that you take it for granted that code that was autogenerated, from autogenerated software, to whatever extent it may have later been modified by a person or not, is in the kernal in substancial amounts?

I never talked in this thread under the premise that the generator is autogenerated.

(except the Python/C++ example if you count C++ as autogenerated in itself)

 

[lost_in_c]

I never talked in this thread under the premise that the generator is autogenerated.

(except the Python/C++ example if you count C++ as autogenerated in itself)

Referring to this:

I find code generated from LLMs

I'm assuming you know enough about transformers and such to know that what they call "LLM"s are autogenerated.

 

[cracauer@]

Now you lost me. Code generated by an LLM is code generated by generated code? I don't classify it as that.

Code generated by an LLM is generated by a programmer who has severe weaknesses telling good information from bad.

 

[lost_in_c]

Now you lost me. Code generated by an LLM is code generated by generated code? I don't classify it as that.

If a so called "LLM" is generated code, and it is generating code, then by Thor, it is generated code generating code.

 

[lost_in_c]

Actually that is inaccurate, I got carried away in that religious frenzy. It is generted software (usually straight binaries) generating code.

 

[lost_in_c]

By the way, the reason I react to the term "Large Language Model" is that the technology that that term actually alludes to can do much more than generate language models. That is just a small subset of the capabilities of that technology.

 

[cracauer@]

By the way, the reason I react to the term "Large Language Model" is that the technology that that term actually alludes to can do much more than generate language models. That is just a small subset of the capabilities of that technology.

Yes, it can generate more than text and code. But the "language" part of the LLM designation comes from the input. The ability to describe in human language what you want it to do.