00:59:33 The nfs rpc_lockd_flags and rpc_statd_flags was moved out of the ${name}_precmd() function and that broke things creating this regression. 01:26:43 I posted a patch to fix the 15.1-RELEASE nfs flags regression: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=296233#c2 01:27:01 There is also an easy workaround to implement until the problem is fixed in the next patch release. 02:44:14 jfc bugs.freebsd turned up their anubis 02:44:25 my fucking lptop fan kicked on 02:50:14 rtprio: Because of Anubis? Lol 02:57:03 Wow that is pretty slow on my laptop too. It didn't heat up and turn my fan on but it was like 20-30 seconds to get through it. 03:02:01 Okay. Now I gotta test it. 03:03:10 Whoa! 03:03:25 This has to be a mistake. Why did they turn the level up that high? Haha 03:06:31 That was definitely a good 15 seconds. 03:08:49 No fans or anything (but my laptop is no slouch). But even over 3-4 seconds is a LONG time. 03:09:58 "Don't sell yourself short, ek. Your laptop is an incredible slouch!" 04:45:23 hello folks 14:38:48 ek: yeah, usually difficulty is 2, this is 6, it took 30 seconds or more 15:25:04 oh god, Anubis in front of FreeBSD sites, yeah that is such a thing that makes me put on my shoes to go outside instead 15:30:13 www is dead. Switch to gemini over HF radio and never look back. 15:43:57 boru: isn't encryption banned on the radio 15:44:40 i mean, i'm not sure we've seen any reasonable alternatives to Anubis proposed 15:45:09 LXGHTNXNG: doesn't seem to stop meshtastic and friends. 15:46:52 https://git.gammaspectra.live/git/go-away fortasse? 15:47:22 boru: it's at least true specifically for radio amateurs, where they are operating under the conditions of their licence, which conditions are the main conditions much of the general public would be permitted to operate on HF under 15:47:48 I'm aware, I have licence. 15:48:04 "to obscure the meaning ..." unless you're controlling satellites or RC vehicles 15:48:13 at least in the US 15:48:54 what's wrong with anubis, it keeps many websites up and running in the modern era of AI harvesting 15:50:00 its annoying but what can you do, its just the nature of 2026 15:52:30 i will say i've not "failed" it yet, just wasted a second or two each site. and i can't say that about the centralized alternatives 15:53:26 Anubis is incompatible with browsers like Dillo 15:53:44 My understanding is that go-away is only incompatible in an absurd paranoia mode; most scrape bots don't pretend to be Dillo, yet. 15:58:17 Anubis is also hypocritical, being written with the same AI bots that are slamming your sites and making you need it 15:58:29 my understanding - I may be wrong - is that go-away is not 16:01:00 uh, it wasn't written by a bot 16:01:18 i don't believe. in fact i recall the author is in this channel from time to time 16:12:42 I have met the Anubis author Xe in person and I will vouch that she is a real person. You can watch on YT her lightning talk on it presented at BSDCAN 2025. 16:17:10 LXGHTNXNG: not sure where you're getting that AI & Anubis theory from. I know of nothing to substantiate it. 16:18:39 Anubis was written by a human. Xe's use of LLMs was for a specific implementation and generating piles of test cases, which, considering the product is _to reject LLMs_ seems like a uniquely appropriate use case. 16:19:15 hi 16:19:22 Speaking of :) 16:20:03 heh, kilostars 16:20:46 One of the kinda clever things I use locally hosted language models for is seeing if documentation is understandable 16:21:26 and by locally hosted i mean the box that is in my office that i can look at and only uses 130W power max running a very small model (Qwen 3.7 36b-a3b) 16:21:50 (though it usually idles at like 5 watts) 16:22:03 god i need more ram, i'm probably gonna have to buy another one of those boxes :( 16:22:19 hard part is figuring out a hostname 16:22:31 current box is zohar because it kinda looks like the zohar from Xenogears 16:24:05 ok, what box is it that only takes 130w 16:24:11 DGX Spark 16:24:13 and will it cost less than the memory to fill it 16:24:17 yes 16:24:19 god yes 16:24:37 128Gi of ram that will work in my tower (because AM5 is stupid) is like $7500 CAD 16:24:53 Starting at roughly $4,000 to $4,700, .. . oof 16:24:55 DGX Spark is $6700 CAD and comes with a 4Ti ssd which would be an additional $2000 CAD 16:25:18 plus taxable business expense so girl math makes it free 16:26:20 LXGHTNXNG: one cannot know how to combat a thing without understanding how that thing works in intimate detail 16:26:26 believe what you want though 16:26:42 if you think i'm evil, there is not likely much i can say to sway your opinion otherwise 16:27:09 my steam rig takes more power than this :| 16:27:27 if you're also thinking that because my job makes me use generative AI tools and i haven't quit out of protest, i don't currently accept complaints about my employment situation that are not also included with an equivalent or better job offer attached 16:27:44 well, i'll add it to the list of things to buy when i find a duffel bag full of cash 16:27:51 (note: **offer**, not **opportunity**) 16:28:01 yeah rtprio it's kinda nuts 16:28:22 my tower's 4080 uses more power idling in ffxiv than the entire box does doing AI model finetuning 16:29:40 unforced slop use is unforced slop use 16:33:59 go unforce your opinions somewhere else 16:35:05 Are opinions not allowed on IRC now? 16:35:26 Apparently not. 16:37:45 also LXGHTNXNG Anubis does work with Dillo if the administrator configures the meta-refresh challenge mode 16:37:50 if **the administrator** 16:37:56 thus, their problem, not mine 16:38:05 i mean, tbf, this isn't really the forum for AI rants; #freebsd-social would be a bit more appropriate 16:38:42 This discussion tags onto one about the HTTPS service at freebsd.org. So it started ontopic and then kinda just fell off the motorway. 16:39:18 yes, you're welcome to create a port of whatever Anubis alternative works and try to convince clusteradm@ to use it, but it has to be pretty effective 16:39:37 kevans: but they did turn up the difficulty lately, yeah? 16:40:08 it solves real problems that created complaints from our users, so it's not like it just got deployed because it's the going trend- it happens to be very easy to deploy 16:40:10 I've been working on a database of known datacentre IP addresses after trying and failing that way 16:40:13 no idea re: difficulty 16:40:42 (I've also been recovering from some pretty major surgery, which is why development has been kinda stalled) 16:41:04 I think it did crank up the difficulty. Modest laptops such as my Thinkpad X270 take 20-30 seconds to pass the challenge on bugs.freebsd.org now. 16:41:20 yeah, the worst offenders have started using headless chrome 16:41:58 i've been trying for at least half a year to try and identify headless chrome better up to and including using a ralph wiggum loop to try and identify it better 16:42:00 IMHO, stonewalling them isn't the way to go. That just becomes an arms race. If they fail the anubis gate, send them off to a tarpit and let them run up their electricity bill. 16:43:51 And poison their data set. 16:44:13 anubis invoking a script with failed client details would be kind of cool, but maybe also problematic with incurred load for each bot that failed? I guess that tapers off, though 16:44:41 (if you're just promptly tossing them into your firewall rules) 16:44:53 small problem there 16:45:10 if the client fails because the JS isn't running and the HTML isn't being interpreted by a browser 16:45:17 there's no way to know if it actually failed or not 16:45:40 fair point 16:46:14 i reckon you risk just blocking legitimate users with shit luck in their browsers often enough to be annoying 16:46:19 Xe: i didn't realize you can just chain all sparks together 16:47:07 kevans: i mean the false "block" rate is already kinda too high because people will anger at me and sometimes email my boss calling me a threat to children because i built anubis 16:48:04 Xe: yikes. I've only really had problems with what I assume must be poorly configured deployments 16:48:41 it's a hard balance to meet 16:48:59 i'm working on a lighter weight proof of work challenge using webassembly 16:49:06 which i expect will get people angering at me 16:49:20 but i also am compiling the webassembly to JS a-la the birth and death of javascript 16:52:18 the problem is that people that disable wasm also usually disable the JS JIT (the thing that makes javascript fast) so 16:52:23 /shrug 16:52:27 ain't no winning in detroit 16:54:59 leads to finding fun bugs though 16:55:04 like this one in LLVM: https://github.com/llvm/llvm-project/issues/204883 16:55:17 (first legit compiler bug of my career) 17:00:26 lol 17:01:28 the reason for that ended up being blog-worthy: https://xeiaso.net/notes/2026/anubis-wasm-vendor-binary/ 17:08:25 oh heck i need to test the wasm stuff on freebsd 17:08:37 it should be fine, but i don't know if wazero's jit works there or not 17:11:25 Certainly worth testing! 17:12:14 i mean, i'm not sure we've seen any reasonable alternatives 17:12:35 ^^ This right here. Anubis works fantastically for me (and many, many others). 17:13:29 I'm not sure why bugs.* has raised the level to what they have, but surely there must be a reason. Even if it was accidental or just testing something. 17:13:56 It's not a terribly big deal, but I do understand some people's frustration with waiting 30+ seconds to continue moving forward. 17:14:04 it was getting pounded with LLM shit causing breakage 17:14:12 zi: Of course it was. 17:15:16 Before I put Anubis into place, one of my web servers was literally being pounded so hard it was no longer able to keep up. Fans, CPUs, RAM all at 100%. 17:15:17 ek: the thing people don't get is that it's either annoy people or take the service offline in many cases :/ 17:16:32 Xe: Sure. In my case, I'd rather annoy people for a few seconds as opposed to annoy them them their entire site visitation by have super slow responses. 17:16:40 Anubis works great! 17:16:45 I try <3 17:16:49 anubis does not take 30s for me 17:17:06 maybe 3-4s 17:17:09 it's luck-based 17:17:25 Never seen 30sec either, even on older devices... bugs.freebsd.org took about 3 or 4 sec 17:17:25 I know you do and I'm very thankful. So, thank you! 17:17:30 ah same time 17:17:53 yeah half of my problem is that i believe in the axiom of pay for better hardware or pay in your time 17:17:54 zi: This was yesterday on bugs.* and may have changed now after things slowed down. 17:18:00 money is replaceable but time is not 17:18:04 It wasn't 30+ for me (but it was 15-20). 17:18:07 so i tend to buy allyesconfig machines lol 17:19:25 Just checked. Difficulty dropped to 4 so definitely quicker. 17:19:49 yeah, the new wasm checks are much more granular 17:19:50 i didnt read all of the backscroll--is there a specific problem? 17:20:05 And, again, it doesn't really bother me. It's a one-time thing for that session. I go ahead and get my stuff on bugs done. 17:20:08 the current ones work difficulty being matching nibbles (4 bits), the new ones is bit-wise granularity 17:20:25 zi, The trigger for this was that bugs.freebsd.org cranked up the difficulty and it caused longer than normal delays. 17:20:30 i think the cool part about the new one is that i have the same .wasm code running on both the client and server 17:20:41 zi: Not a problem. Just someone noticed it was taking 30+ seconds to access bugs.* yesterday via the Anubis difficulty change. 17:20:50 rwp: ah ok. yeah--we had $things pounding the crap out of the instance to the point where it was unusable 17:20:59 it does add rust to the build closure of anubis, but i can't avoid that 17:21:13 zi, I understand. I fight that problem on other servers myself. It's a global problem. 17:21:13 It's a thing. 17:21:26 (doing the equivalent in Go would add at least 10Mi to download sizes for the solver wasm blob, which is unacceptable in my book) 17:21:52 Xe: That doesn't surprise me at all. 17:21:57 rwp: yea :/ 17:22:09 rwp: we do have plans for some $newer hw, but that's only gonna get us so far 17:22:33 the limiter is usually not the hardware, it's the scripting language bugs. was implemented in 17:23:01 Yep. My web server that being brought to it's knees was pretty beefy. Those $things are relentless. Anubis literally saved that web server. 17:23:08 thought about making a UI and API compatible version of bugzilla in Go/Rust with giga caching as a thing for exiting projects to migrate to, but i don't have enough time in the day :( 17:23:55 The free software conundrum! 17:24:33 if my husband was employed i could risk taking a chance on techaro 17:24:36 Dang. I had a question for you, Xe, and now I can't remember what it was. Typical. 17:25:10 That would be cool. 17:25:51 i've been working on ideas for the techaro seed round arc, but i may end up not going down that road 17:47:46 Xe: Not going down the techaro road at all or just not the arc road? 17:53:26 i'm the only income earner for this household, nothing techaro does has the word "agents" in it, and i really don't want to job hunt in this market 17:53:56 i'm just gonna go for moonlight bootstrapping i guess? idk 18:02:36 That's fair. And job hunting right now is definitely no picnic. The market is absolute junk (at least here in the USofA.) 18:23:41 It's worse in Canada lol 18:44:54 it's bad worldwide 18:46:58 I am running FreeBSD 15.0-CURRENT on my desktop, can you guys recommend anything better ? 18:47:00 bsdrobert: I would think so based on the reasoning. 18:47:20 mzar: 15.1-RELEASE? 18:47:57 what about upgrade to 16.0-CURRENT ? 18:48:28 sure, why not 18:48:55 15 was branched quite a while ago, so you're very out of date either way 18:49:01 Nothing wrong with that either as long as you don't encounter any issues. But, be prepared for the possibility of breakage. -CURRENT is not a stable release, but the latest. And code changes A LOT. 18:50:21 TBH I am running 16.0-CURRENT, but I am getting dyslectic with age and misstyped it 18:51:16 Ah. Gotcha'. If you're not having any problems, just keep running with that. 18:51:38 thanks ek ! 18:52:22 And, obviously, if you DO have any problems, please report them. :) 18:52:45 sure, I will 20:00:51 Any ideas how to keep several bootable kernels (and their modules) on same machine/rootfs ? I came to this simple but bogus solution: I create /boot-14.4 and /boot-15.1, I put corresponding kernel and modules there, than I symlink /boot to one of these before reboot. Is there any better solution ? Also can I change where loader finds kernel and modules during boot time ? 20:01:38 checkpoint: Using root on ZFS? 20:02:19 If so, use the Boot Environment Controller (bectl) 20:02:57 That works with UFS as well doesn't it? 20:03:26 Or at least, _someone_ got it to work. 20:03:49 boru: I'm not sure. Obviously, snapshotting the BE won't be an option on UFS. But, maybe it uses something else if it detects UFS? 20:03:54 I haven't used UFS in years. 20:04:19 Been a few years for me, as well. 20:04:25 Ah, here we go: https://vermaden.wordpress.com/2021/04/02/ufs-boot-environments/ 20:04:31 From 2021, though 20:04:58 Not nearly as convenient as zfs, naturally. 20:05:24 Definitely not. 20:06:12 UFS on Solaris 20:06:18 Certainly not surprising that it was Vermaden, though. Haha 20:06:19 ek: bectl is for zfs only 20:06:53 checkpoint: See boru's link above. It works with UFS as well (albeit, differently) 20:06:53 ek: I'm on UFS and prefer to leave it that way on laptops 20:07:14 Yeah, but in that page, I recommend reading the last part first. 20:07:17 all this reminds me that I need to get porting bectl to linux 20:07:19 I'm not sure if it will suffice for your needs on UFS. 20:07:35 Yeah, I won't vouch for it, because I haven't tried it. 20:07:35 LXGHTNXNG: That would definitely be nice! 20:07:41 checkpoint: er, loader searches all of the directories in /boot for kernels 20:07:54 checkpoint: the normal convention is to just install them into /boot/kernel.foo 20:08:59 kevenas: yes, but I would like to make it search some other directory, like /boot-15.1 in my case. Is that possible ? 20:09:50 checkpoint: why, exactly? I'm not sure why you wouldn't just install a 15.1 kernel into /boot/kernel.15 or some such 20:10:12 it wouldn't be hard to fix this, but i'm curious of the use case 20:12:03 kevans: because besides kernel there are modules and tons of other kernel related files, all of them a spread under single /boot/ 20:13:08 checkpoint: most things are relatively self-contained, presumably you're mostly concerned about installing kmods from ports? 20:13:53 kevans: it would be great to tell loader and kernel where to search. Say, if I want to boot 15.1-RELEASe I would setenv boot_dir=/boot.15 in boot prompts. As simple as that. 20:15:29 kevans: not only kmods, there tons of firmware files which depend on kernel (wonder why?). 20:18:23 as I said I can switch now between kernels with something like "rm /boot && ln -s /boot.15 /boot && reboot", but I would like to be able to to that during boot time, in boot prompt or even in Boot Menu. :) 20:19:29 yeah, firmware and untied modules make it a little more complicated 20:19:39 hmm 20:20:35 from reading boru's link: "The idea behind UFS Boot Environments lays in several additional root (/) partitions that will be used as alternate boot environments." -- kind of same Idea, but uses mount(8) instead of "ln -s" 20:21:19 I mean, you if you have /boot/bananas and /boot/badgers, you can `set kernel=badgers` 20:21:42 boru: right, but firmware and untied modules live in generic paths that are baked into module_path 20:22:56 Firmware should be mostly kernel agnostic, no? KOs on the other hand, yeah, depends where they are, I guess. 20:23:08 new-style firmware is, old-style firmware wasn't 20:23:21 Could always just make multiple /boot partitions and then efibootmgr it or something. Lots of ways to skin this particular cat, imho 20:23:23 old-style we wrapped the firmware into a .ko and it was ugly 20:23:27 boru: ok, if you go `set kernel=badgers`, where kernel finds its modules and firmware ? 20:23:56 See above, kevans is probably right. 20:24:44 boru: firmware should NOT be kernel agnostic, firmware depends only on hardware. That's how it works in Linux, btw. 20:25:01 Do you know what agnostic means? 20:26:22 boru: please explain ? 20:27:18 If you get a dictionary, go to the A section, find "agnostic" and read the definition of what the word means. 20:28:56 boru: thank you, bro. ;) 20:29:11 You're welcome. 20:30:30 anyways, the idea of having multiple /boot partitions seems ok, but it's hard to implement it on already deployed system. 20:39:30 i'm still munching on some notion of 'kernel profiles' or something to swap out the bits you need 20:42:02 kevans: I think something similar can be achieved by manipulating kern.module_path sysctl and/or module_path kenv during boot 20:49:54 checkpoint: yes, this is what it would do 20:51:13 * checkpoint wonders if that can be done using just lua scripting 20:52:21 the boot menu we have (1. Normal boot, etc), how is it made ? in lua ? 20:57:00 yes, pua 20:57:04 lua 20:57:39 you can add a local.lua for personalization likw this 20:58:16 happy to advise if you go that route 20:58:48 * kevans wrote a lot of these parts 21:01:40 Haha. "pua" is never going to leave my mind now. 21:18:21 kevans: did you write that bootloader lua stuff ? 21:31:53 checkpoint: yes, quite a bit of it 21:40:45 kevens: Cool. Thank you very much for doing that! 21:44:31 kevans: If you find time to implement "Kernel Profiles" feature that would be just great, it would help people to test/debug differnels kernels on real hardware without much effort. :) 21:46:55 * checkpoint is currently toying with 13.5, 14.4 and 15.1 kernels on same laptop using base rootfs using from 13.5. 21:54:00 amusing 22:25:44 checkpoint: yeah, unfortunately it might go on the ol' backlog for a while. need to remember if there are any other useful things besides per-profile module_path that should be included 22:26:24 maybe per-profile module sets 22:26:53 foo.iwx_load=YES -> foo profile loads iwx 22:27:22 iwx_load=YES -> all profiles load iwx 23:05:17 but why 23:15:29 because it actually is useful functionality 23:15:59 particularly for their case where the default values for some things aren't really globally applicable 23:17:40 (it dawned on me after the iwx_load messages that it probably just makes sense to infer the variable sets from the kernel path; /boot/kernel.foo -> foo.module_path, /boot/foo -> foo.module_path 23:18:08 A little more complicated if you try to name a kernel like `kernel.15`, but maybe let's just not do that