IRC channel logs
2026-04-27.log
back to list of logs
<jab>Alicia: what are you up to today ? <Alicia>tinkering with some automation (unrelated) <jab>what kind of automation ? <Alicia>automatically sending some files through a web chat, using a tool that finds a sample on the screen to click on buttons and stuff <jab>hmmm...sounds like a fun little project. <SurrealHG>OK, I’ve read up on Hurd’s RPC… But I wonder: Does Hurd not have per-process namespaces? (A key feature of Plan 9.) Or have I missed that? <jab>SurrealHG: I don't actually know if the Hurd has a "per-process namespaces"...but that sounds like a super cool idea! <youpi>processes have their own notion of "/" <youpi>see the remap translator for instance <SurrealHG>youpi: Nice! So with the usual features like being able to run a process in an empty / with some translators added, or with a subset of the parent process’s /? <SurrealHG>youpi: Also, can I assume that since RPC servers provide everything and the rights on the ports a process has decide which ones it can use (and how), that “their own notion of /” means *everything*, like in Plan 9? So no need for separate namespaces for networking and other resources, yes? <SurrealHG>(Sorry, I am so new on actually looking into Hurd, that I literally only started today. I already read everything I could find in the wiki about this, so I would not bother you with already answered stuff, but I didn’t come across anything explicitly saying such a thing.) <Gooberpatrol66>is there any toolkit on hurd for using object capability mechanisms on a process like attenuation, proxying, etc.? <SurrealHG>In other words: Is everything a file… no a translator… no a superset called an RPC server that is not limited to file semantics? Or are there other things a process could use that would make it not a full sandbox? <Gooberpatrol66>processes use rpcs and file i/o is a higher level thing built on top of the rpcs <jab>SurrealHG: I don't know if this answers your question...but I am under the impression that regular users can only attach a translator to a filesystem node that they own. <jab>So I can attach translators to ~/joshua/ with ease. But I need root privledges to attach a translator to /root/ or most files in /etc (that are root owned). <SurrealHG>Gooberpatrol66: Yes, I read about the object capability approach, but I wan unclear about the mach kernel interface. I mean a program needs system calls to get RPC / port / file system access in the first place. But what else can a program do with the kernel interface? (I say that because Plan 9 intentionally pretty much *only* had file system <SurrealHG>syscalls, and so everything was done through files, meaning its name space was all-encompassing. So I would like to find out if Hurd’s kernel interface’s “namespace” handling is like that too, or if it is limited to RPCs / translators / the file system… <SurrealHG>jab: so Hurd still as a notion of a root user? Hmm… <youpi>SurrealHG: network cards show up as /dev/eth*, and tcp/ip shows up in /servers/socket, etc. so interposing / means interposing everything, yes. <youpi>the kernel actually doesn't know about file systems <SurrealHG>youpi: Okay, that is the best possible answer. :) <youpi>it just passes a port to / between parent and child processes <youpi>it doesn't even know what that port is, it's just convention between processes <SurrealHG>Man, back when I first read about Unix/Linux, in the 90s, why didn’t somebody scream about Hurd from the rooftops? I’m almost angry! <SurrealHG>(I knew about Hurd, but only that it exists. I didn’t even know about Plan 9) <jab>SurrealHG: I'd love to help your help in marketing the Hurd! <jab>our leader, Samuel, doesn't quite want logos for every translator, but for some it makes some sense to have cool logos. <jab>SurrealHG: also, the Hurd runs pretty well on real hardware. I am using it on a T420. We will hopefully be using ext3fs instead of ext2fs. <SurrealHG45>Given how it went with Linux, the best marketing is compatibility. MS always knew that, sadly. So the best marketing is news about the Hurd supporting modern hardware. I see there were already efforts in taking Linux drivers with a adaptor layer. But has anyone ever thought about starting a project to *automate* turning any current Linux code base <SurrealHG45>into a pure hardware-abstraction layer (HAL)? As in: Basically beheading Linux and turning its entire body of drivers and their interfaces into an RPC server that *only* offers driver functionality? <sam_>I recommend trying out the Hurd if you haven't already and researching the current approaches to hw compatibility <sam_>you will find answers to a bunch of these <sam_>(but yes, trying to use parts of Linux via a framework didn't scale) <sam_>i also recently combined the two :) <youpi>SurrealHG45: linux would be a beast for that, the netbsd codebase is way more longterm-proof, and that's the idea of rump <SurrealHG45>jab: The wiki seriously needs updating. It took me literal hours before I saw the first notion that development didn’t stop in 2016 and it still says in the wiki that it doesn’t support more than 1.7GB of RAM, can’t do USB, sound, SATA (let alone NVMe) or any graphics beyond AGP, and “hundrets of gigabytes” is used as if it was an <youpi>? the front page litteraly contains the recent qoth <SurrealHG45>sam_: Ah, okay. Didn’t it scale because people tried turning every module into a RPC server? Because having it all be ONE RPC server would solve that, no? <sam_>it doesn't scale because linux both changes a lot, and changes a lot *in a lot of different ways* <sam_>if you've ever maintained a patch to linux.git or an out of tree kernel module, you'll see how <sam_>it's just not a good fit for it at all <sam_>rump on the other hand is a perfect fit <jab>SurrealHG45: you are right in that the Hurd wiki at hurd.gnu.org is out of date. I personally use hurd.ion.nu That wiki is up to date. I also do a fair amount of the wiki updating. You are welcome to help me out! <youpi>hurd.gnu.org is not *that* out of date <SurrealHG69>sam_: Ah, that makes sense. Yeah, it really changes a lot. Reminds me of how Google “alledgedly” intentionally added “features” to Chrome so fast that no other browser maker could keep up, and everyone but Firefox died because of it. (For Linux I don’t imply intention. But I definitely criticize the lack of stable in-kernel interfaces. <SurrealHG69>There should be an industry standard driver interface that allows separation of drivers from kernels.) <SurrealHG69>sam_: I managed to avoid C for 36 years, so no, I haven’t tried maintaining a Linux patch. ;) <jab>youpi: not trying to throw stones, but I personally don't use hurd.gnu.org at the moment. I've completely replaced it with hurd.ion.nu . I know that one all of the web pages are current. :) I should probably try to figure out how to update hurd.gnu.org, since I gripe about it all the time. <sam_>in their defence it makes some maintenance easier to be able to do tree-wide changes, but it sucks from every other perspective <jab>I'm not trying to be critical of the FSF (who hosts hurd.gnu.org), but it's frustrating for new Hurd users that assume that site is the "defacto" source of hurd information. It's confusing to discover that some of those wiki pages are just out of date. <SurrealHG69>youpi: LOL, yes, you are right, it can be seen right on the home page. But in my defense, I didn’t look for news, but for a info on how Hurd works and then never went back to the home page. Also, a newbie would have NO idea what a “qoth” is. (Quoppa goth?) XD <sam_>"quarter of the hurd" I think (quarterly news) <youpi>jab: as already mentioned, updating hurd.gnu.org is *easy <youpi>the only thing is that I have to proofread the changes <jab>then, I have no excuse. I need to do it. :) <jab>assuming, you trust me to have enough knowledge to know what information gnu.org allows to be there. <SurrealHG69>jab: hurd.ion.nu doesn’t show up in searches, unfortunately. Not even if you literally type it into the search field. Needs more link farms! XD <youpi>so if something is blatantly out of date, it's really *only* a matter of pointing me at it <jab>youpi: yeah, but your time is better spent elsewhere. <youpi>but saying "then updated it all" won't work, I can't take the time to review everything in a go <youpi>jab: that's the whole point, it's really me who need to review <SurrealHG69>jab: hurd.ion.nu doesn’t show up in searches, unfortunately. Not even if you literally type it into the search field. Needs more link farms! XD <youpi>to make sure what goes there is accurate <SurrealHG69>jab: Also, I am utterly clueless. I would not dare to update the wiki. XD <SurrealHG69>youpi: Honestly, a single proofread with a way to make quick notes so you or someone else can later go through the notes would already be extremely useful. I, as a n00b, would have no problem if it simply said “this section is out of date” or “note: this has changed in thisandthatway. wiki update pending”. <SurrealHG69>I’m currently working on a novel type of graphical shell, so my only way to contribute to the Hurd, is by, when it is good enough to work, make it Hurd-first. (It needs SDL and OpenGL though. But it’s very efficient, so it should run smoothly even with zero GPU acceleration.). <SurrealHG69>If I suddenly became rich for some reason, I would absolutely spend a nice sum on hiring people to write drivers. ^^ I’ll notify you then. :P <jab>SurrealHG69: I'll ping you in 5 years when you're a billionare.