Linus introduced a race condition and a regression with a AI generated code, then handed off fixing the problem to someone else because he was going on vacation.
Fiasco might have been an exaggeration, but it’s still pretty lame.
This was caught during review, so nothing was “introduced” quite yet. And there’s no indication that Linus wouldn’t have produced a race condition (a notoriously more easy to miss kind of bug) without the use of AI tools.
It’s only being exaggerated so much because he used AI.
You are correct to point out the distinction between race conditions and data races. Rust only rules out data races.
I don’t know whether the race in question was a data race or not.
I would still wager that Rust code generally has fewer race conditions than C, given that mutable aliasing can be used as symbolic prevention of race conditions, i.e. your &mut T doesn’t necessarily need to hold data; it can represent some domain concept.
aren’t language specific
There are actually languages that guarantee deterministic concurrent execution, and this can rule out larger classes of race conditions than just data races. See the Bloom language and CALM theorem.
That is something I am surprisingly okay with. Quickly figuring out if something is feasible is good, and doing it in a back of the napkin style is the norm, even without LLMs. Of course then you need to do it properly, without falling into the trap of reimplementing the slop.
nothing to be nervous about, its literally software engineering and project management 101, programmer writes code reviewer catches bugs, nothing goes to production. this happens all the time, has happened in the linux kernel way before LLMs were a thing
People are welcome to stop spreading the falsehoods. In every thread where this is said (including this one), the truth is pointed out but those spreading it couldn’t care less. Just scroll. Others are pointing out the details here just like every other place.
Linus introduced a race condition and a regression with a AI generated code, then handed off fixing the problem to someone else because he was going on vacation.
Fiasco might have been an exaggeration, but it’s still pretty lame.
https://lkml.iu.edu/2610.1/13698.html
This was caught during review, so nothing was “introduced” quite yet. And there’s no indication that Linus wouldn’t have produced a race condition (a notoriously more easy to miss kind of bug) without the use of AI tools.
It’s only being exaggerated so much because he used AI.
I could attribute the race condition just as much to their use of the C language as to AI.
Race conditions I’m pretty sure aren’t language specific. Rust can have them too.
You are correct to point out the distinction between race conditions and data races. Rust only rules out data races.
I don’t know whether the race in question was a data race or not.
I would still wager that Rust code generally has fewer race conditions than C, given that mutable aliasing can be used as symbolic prevention of race conditions, i.e. your
&mut Tdoesn’t necessarily need to hold data; it can represent some domain concept.There are actually languages that guarantee deterministic concurrent execution, and this can rule out larger classes of race conditions than just data races. See the Bloom language and CALM theorem.
Yes, but the tools provided by Rust:
unsafe {}block to be used.Yea he basically said “This seems to work but is slop, lead Dev, make it real.”
That is something I am surprisingly okay with. Quickly figuring out if something is feasible is good, and doing it in a back of the napkin style is the norm, even without LLMs. Of course then you need to do it properly, without falling into the trap of reimplementing the slop.
Race conditions are not unusual, and it was caught during review, so there’s really no problem here.
Great
This doesn’t make me nervous at all…
nothing to be nervous about, its literally software engineering and project management 101, programmer writes code reviewer catches bugs, nothing goes to production. this happens all the time, has happened in the linux kernel way before LLMs were a thing
It shouldn’t, because it’s a lie.
:/
This is why dictionaries list more than 1 definition and you shouldn’t stop at the first one:
https://www.dictionary.com/browse/lame
People who write this dictionary should listen to the public more maybe. Lame doesn’t mean what you wrote for at least 25 years.
Not what happened. I’m not digging it up but this story has been debunked repeatedly
(you’re welcome to just point folks to a debunking)
People are welcome to stop spreading the falsehoods. In every thread where this is said (including this one), the truth is pointed out but those spreading it couldn’t care less. Just scroll. Others are pointing out the details here just like every other place.
Yeah there’s 240 comments here. I’m not shoveling through that.
Literally you got hit with an linux mailing list saying this is what happened?
Those words don’t make sense