L
Remember the days when we used to build systems out of pluggable components talking to each other over generic interfaces? I don't see what would make XRAID specific to XFS. XRAID works because it relies on its callers sticking to a no-overwrite contract. Couldn't it therefore behave like a zoned block device via DM instead of having to be integrated into a specific filesystem? Zoned devices are already contractually no-overwrite and already provide the multiple write streams that XRAID wants.…
L
> If such breaks are "rude", then how do you call a customer dropping a supplier? Depends on the situation. Usually fine. > Or vendor lock-in? Yep, definitely rude. > Or MS laying off 60,000 employees in one go? That's actually a misinterpretation: they didn't lay off 60k employees, they offered a voluntary exit package with a monetary bonus to 60k employees. That's actually a really nice way to do things, vs the surprise "10k people: you don't have a job". The latter, though: that's also rude.…
L
> Folks that intentionally choose the GPL consider those "other uses" to be actively harmful and *want* them prohibited. Well, of course, I totally agree! But that doesn't seem particularly relevant as a reply, given that this subthread was about folks who _did not_ choose the GPL, but rather a BSD or MIT license. I even added the parenthetical "(under this definition)", in order to try to avoid getting such a response. Unsuccessfully.
L
I don't think that "safe SubC" idea can work because of two misunderstandings of why Safe Rust is Safe. First and most important: Safety is cultural. Rust's safety culture uses Rust's safety technology to deliver safety, but you can't just apply that technology and magically get safety, you need the culture. You're damn sure not getting the culture from C. I was a C programmer for decades. Secondly and maybe more understandably to people focused on technology, ALL of Rust follows the same rules…
L
OpenWRT is actually pretty broken. One pet peeve of mine is to enable DVB drivers in the kernel. I mean what do you do with tvheadend and a dvb stick if yoy don't enable the kernel drivers. So when enabling these simple features, it's more than easy to make it unbuildable by turning some config on. The build process is horribly error prone. I've more than 20 years of experience with kernel menuconfig, and at least 10-15 years with busybox and uboot. Openwrt is just broken. It took them almost…
L
I started working with embedded Linux in 2004. With OpenEmbedded because that was what OpenZaurus developers were using. Years later, I met with Buildroot in CirrOS project. And a few other places. There are Buildroot, OpenEmbedded and OpenWrt -- the three embedded Linux build systems which are still in use after 20+ years. I wish Buildroot to be with us for the next 25 years!
L
That's the kind of "high-level ergonomics" that I like from the Rust language. Getting rid of whole classes of bugs with a good compiler and tools. I'm less fond of the succinct syntax and prefer the verbosity and explicitness of C. See for example what I wrote about type inference or having a safe SubC language thoughts.
L
Assuming Glasswing is what Open Source Scanner uses, the reports I get for iproute and DPDK have been mostly on target. It does tend to identify things that are really niche corner cases, but that is what I would expect. Are they real bugs? yes Are they worth fixing? yes Does it prioritize correctly? it over reacts like a security researcher. Overall, I impressed but it does add to huge project overload.
L
You could just cross-compile...
L
It is interesting to me that LLMs does so much poorly in one area than another. It doesn't seem that surprising to me. Patches are smaller and should be easier to reason about. They're also likely to be way buggier than established code, so there are just more real bugs to find. I bet you'd have the same general trend if you ran the LLM bug finding on established projects. They'll generally do better on small projects than big ones, and they'll definitely do better on projects that have…
L
I was today years old when I learned of the existence of Dell Unix. https://gunkies.org/wiki/Dell_UNIX
L
Sort of, but it's surprisingly good at it: it finds things that humans don't find, and it's harder to fool with linting (that is, even if the author fixes all the signs that tell human reviewers this code isn't done carefully, it'll still notice the less superficial mistakes). The text of the comment is useful for confirming that the problem you found is the same one it found, even if you can't understand it before you re-examine the code. For example, if there's a PR where the new code assumes…
L
So, basically automated "code-smell" detection, but with a confusing report with too much confidence, rather than the "this looks suspicious, it probably isn't right" that would be more appropriate?
L
This reminds me a lot of Greg KH's talk at Kernel Recipes 2026 - Security in the LLM age. I watched the video of this talk online, and greatly appreciated Greg's perspective. I wish more people could be as balanced as Greg about this stuff.
L
Well, given the description of parity declustering is under the heading "Solutions are well-known", perhaps the author didn't consider it worthwhile to go further into a history lesson on parity declustering. And no, ZFS wasn't the first implementation of it.
L
There is a risk in Cargo from unused dependencies - Cargo has a build graph based on the declared dependencies in Cargo.toml, and chooses what build steps to run based on that graph. One thing a crate can contain is a build script - which is code that's built and run on the build host at the time of compilation, and can do anything that code running as part of the build system can do. In theory, an unused dependency can contain a build.rs that's got a security-relevant bug in it; that would…
L
I don't know exactly where to draw the line, but a slave-written patch is clearly on the wrong side of the line! Individual projects need to decide where they draw the line. There are a lot of projects with policies I don't agree with, but I prefer that to projects that don't exclude anyone and allow jerks or bad actors to ruin them.
L
I wrote about the WTO. It is true. That there exist /some/, multi-lateral trade agreements with much more _limited_ memberships/signatories, which do have some provisions for higher minimum labour standards is something I thought wouldn't have needed to be said. Though, CPTPP is not the example I'd have thought of - I would have pointed at the EU! I'm not aware of what kind of labour standards CPTPP requires myself. But good to know there's something. Still, there's nothing at a global level…
L
Sadly no comparison to https://openzfs.github.io/openzfs-docs/Basic%20Concepts/P... neither any mention or acknowledgement.
L
Per the information linked in the article, Polo uses libweston, i.e. Weston's compositor framework.
L
From a security point of view, was fetching an unused dependency a potential attack vector? (like changing global state even without calling exported functions or using global data). I don't know Rust enough to be sure but I don't think so. It only affect the package manager and not the compiler.
L
I kind of think that the obvious thing is to have something you can put in the compilation unit or on the function to say that the programmer is in the second camp. If you want to document something and also check the same thing, you should be able to have the compiler act like you wrote it out again in your function body without actually repeating yourself, but it shouldn't necessarily happen unless you direct it to.
L
> But it's not that big a deal because it's the dependencies in Cargo, the tool that will fetch external dependencies From a security point of view, was fetching an unused dependency a potential attack vector? (like changing global state even without calling exported functions or using global data). A little like the use of pkg-config in C. pkg-config is usable (mainly) thanks to packaging efforts done by distributions, so it's not an upstream-only thing (unlike Cargo, Npm, PyPI, etc).
L
I think the issue is that dereferencing a pointer can have side effects, if it gives a bus error or segfault or accesses MMIO or affects caches. If you think about those kinds of side effects, it's hard to have useful contracts that can be proven not to have any. On the other hand, it might be possible to say that passing a misaligned pointer to a function might trap or not unpredictably independent of contracts, and enforcing the contract might influence that, but you couldn't be sure that's…
L
Wayland support in Mutter predates the existence of wlroots by 4 years.
L
> But it can also swing the other way: trust in traditional media has been lost, especially in the US. I wasn't including "online media", I was only referring to _unknown and unsolicited_ media. Whether you are consuming Fox News or the New York Times, it's really not hard to verify that you are really consuming those and not some impersonation. That particular problem is basically solved. Once you have verified that what you see actually came from Fox News or the New York Times and you know…
L
Ah sorry, I had misinterpreted what you were saying as being about applying US law to the UK. I think what you're saying is that the development/reverse engineering work would be easier to defend in court in the US than the UK, is that correct?
L
> And I'm pretty sure that are hundreds of more or less "serious" things to be considered (un)ethical. I don't blame maintainers who do not want to play this game. Anyone who tries to play these "what you are doing is unethical because you exist in a society" games that is not practicing strict Janism is a complete hypocrite.
L
I find that, when it says there's a bug, there usually is; however, it's much less reliable as to what is actually wrong, and what it says is usually harder to understand the bug from than looking at the code and then checking whether the flaw you find can be described by what it said; and it's frequently wrong about what ought to be done to correct the problem, if it gives a suggestion. I find that my coworkers ignore the LLM code review comments because they don't make any sense, but pretty…
L
The big problem is that where do you draw the line? For example some might say that my patch is unethical because while creating that patch I ate a hamburger and butchering (and eating) a (holy) cow is unethical. Or the usage of vast natural resources to raise that cow is also unethical. And I'm pretty sure that are hundreds of more or less "serious" things to be considered (un)ethical. I don't blame maintainers who do not want to play this game.