Linux is the name of a computer operating system kernel and a group of systems built on it, like Ubuntu or Mint. Linux is the basic infrastructure of the Internet. It’s important. A report ju…
You do realize that all code for the kernel is code reviewed and tested by so many people right? The patches are probably some of the most human reviewed code on the planet…
This feels like the time when every armchair expert became experts on immunology…
And I can’t find any evidence they’re actually involved with any development projects either. Literally ANY projects including stupid little pet ones every dev has
We are completely overwhelmed, of course. The glimmer of hope is that we secured sufficient LLM budget and access (thank you Meta!) to run reviews with multiple frontier models on each patch. This eliminates some hallucinations.
That’s not the author’s words, it’s a quote from the Linux mailing list…
The actual policy is that it all gets reviewed by a human I think when I was looking into it.
The bots are likely most useful to reject obviously bad patches which have obvious bugs or don’t follow guidelines. And identifying bad code is something which AI tends to be good at.
In our code at work, we use weakly typed languages which aren’t strict and on those it tends to be even better. We catch things like table keys that are typed wrong, etc. On C, things like pointers it probably does a good job with identifying as wrong
If it passes AI code review it still gets checked by human for things that are wrong and needs fixing. But if it fails, it probably needs fixing regardless.
A good language server / ide will also complain about incorrect types. That specific example is a deterministic process, no llm is required for it at all.
I find that very hard to believe, if you’re truly just talking about typing. If this is true you’re using an awful language server / linter / whatever.
Coding style is also a deterministic process, for which every widely used language has a linter or formatter or whatever. Also no LLMs needed.
As a professional software developer of 8 years, I honestly have no idea what you’re talking about
The thing about typing is that it is easy for a deterministic compiler to flag violations.
There are hundreds of other “best practices” that determinisitic compilers aren’t able to flag. Even the almighty Rust language is suceptible to race conditions, deadlocks, and many potential crash conditions.
I’m an old C programmer, and Rust pisses me off when it won’t let me do something that’s demonstrably 100% safe. However, a lot of the things that Rust and other general practice rules forbid or advise against, should really be avoided when they get past a very simple application. What was “proven safe” quickly becomes unsafe when a later maintainer who doesn’t understand the risks “extends” the code.
You do realize that all code for the kernel is code reviewed and tested by so many people right? The patches are probably some of the most human reviewed code on the planet…
This feels like the time when every armchair expert became experts on immunology…
The article is literally about how they have so much AI slop that the code is no longer reviewed by people it’s reviewed by a Meta chatbot…
Can you post the quote where it says that it’s untested and it goes straight to master
I’m lost.
Also, looking through the authors articles, frankly, the list looks like low quality human slop.
https://davidgerard.co.uk/other.html
And I can’t find any evidence they’re actually involved with any development projects either. Literally ANY projects including stupid little pet ones every dev has
I mean, having pet projects doesn’t require being “involved” with one nor requires making anything public…
Except I can’t see any references to this guy getting involved with any projects at all at a technical level.
If you check his articles, he’s definitely targeting a specific crowd based on what I can see too.
Saying “there is an AI bot that identifies if code is obviously bad” doesn’t really make an exciting story though…
The article is one angry person’s rant and you are further going beyond what it actually says.
That’s not the author’s words, it’s a quote from the Linux mailing list…
Yes and even that one quote from one person does not say what you did, so that’s what i meant :/
Yeah. Pretty much.
The actual policy is that it all gets reviewed by a human I think when I was looking into it.
The bots are likely most useful to reject obviously bad patches which have obvious bugs or don’t follow guidelines. And identifying bad code is something which AI tends to be good at.
In our code at work, we use weakly typed languages which aren’t strict and on those it tends to be even better. We catch things like table keys that are typed wrong, etc. On C, things like pointers it probably does a good job with identifying as wrong
If it passes AI code review it still gets checked by human for things that are wrong and needs fixing. But if it fails, it probably needs fixing regardless.
A good language server / ide will also complain about incorrect types. That specific example is a deterministic process, no llm is required for it at all.
We do use one. But, in practice, they can’t catch a lot of things which AI will
At least in our case too based on what’s available, it also can’t check stuff like coding style well (but our development relies on that less).
Some might though
I find that very hard to believe, if you’re truly just talking about typing. If this is true you’re using an awful language server / linter / whatever.
Coding style is also a deterministic process, for which every widely used language has a linter or formatter or whatever. Also no LLMs needed.
As a professional software developer of 8 years, I honestly have no idea what you’re talking about
The thing about typing is that it is easy for a deterministic compiler to flag violations.
There are hundreds of other “best practices” that determinisitic compilers aren’t able to flag. Even the almighty Rust language is suceptible to race conditions, deadlocks, and many potential crash conditions.
I’m an old C programmer, and Rust pisses me off when it won’t let me do something that’s demonstrably 100% safe. However, a lot of the things that Rust and other general practice rules forbid or advise against, should really be avoided when they get past a very simple application. What was “proven safe” quickly becomes unsafe when a later maintainer who doesn’t understand the risks “extends” the code.
This is Lua code. LuaLS is basically the language server available.
And sorry, I should have said “dynamically typed”. But, weak tables.
There’s a lot of things that don’t get caught, such as concatenating a boolean which causes a lua error in our environment. LuaLS won’t catch that.
Formatting isn’t a problem for us, but it does affect the kernel. But, the VSCode one we use does a lot of weird things honestly.
Oh, quit your bullshit.