Login
You're viewing the functional.cafe public feed.
  • Nov 14, 2026, 6:08 PM

    RE: circumstances.run/@davidgerard

    I'd like to tell you a story about a contributor who came to us at Mozilla, in my first 2-3 years working there. I never met them, don't know their name, and I still think about them a few times a year.

    They came to us with questions, as they were struggling to build Firefox on their computer, and they were getting errors that nobody had ever seen. And I don't mean "errors that were unfamiliar to our engineers", I mean "we can't figure out how or why our build system would emit that string."

    💬 9🔄 586⭐ 575

Replies

  • Nov 14, 2026, 6:14 PM

    So, after going back and forth a bit with person, who it turns out is in their early teens, we asking about their hardware, what setup they've got that's doing this. Maybe this person's got some new system we've never seen, maybe they're running in some novel context we need to know about. Is this an edge case, or is this a leading-edge case? Outliers are always interesting.

    I've seen some Pretty Far Out There systems, I run a few myself, and this kid's system is Out There.

    💬 1🔄 31⭐ 235
  • Nov 14, 2026, 6:23 PM

    So, they're running one of those weird dual-slot Pentium III boards with two physical CPUs. Running Linux - somehow - on a hard drive that's almost full, a userspace partition on NTFS and more, and the more I hear about this system the more confused I get.

    For the non nerds following this thread, this is like finding out somebody's daily driver is built out of the parts of a sports car, a semi and a train, and it's running on vegetable oil. You might ask why, but the first question is ... how?

    💬 12🔄 75⭐ 420
  • Nov 14, 2026, 6:29 PM

    This kid is getting error message that nobody has ever seen because they've got a machine nobody else has, set up like nobody else does that nobody would ever choose. And you might ask how, but the first question is why?

    And the answer is: this kid has nothing. And they've _built their own computer_ out parts they've scavenged from the computers other people have put to the curb, got a Linux ISO and somehow made it work.

    There _are no_ computers like this. And... they want to help us?

    💬 9🔄 132⭐ 548
  • Nov 14, 2026, 6:33 PM

    And I'm sitting there, reading this on the screen of a five thousand dollar Macbook Pro in an office where the snacks are free, feeling humbled and deeply ashamed.

    💬 3🔄 44⭐ 424
  • Nov 14, 2026, 6:43 PM

    There's nothing I can do for this person. But the enormity of it. They've cobbled up a machine out of nothing but trash and their own tenacity, and they want to help us build Firefox? There's lots of ways to do that, but building it, compiling it, that can't happen. Not on this machine.

    There's always been a financial barrier to participation in open source. Always. And we have to make it as low as we possibly can.

    But a _wealth_ minimum bar, to contributing?

    That's vile. It's disgusting.

    💬 17🔄 205⭐ 686
  • Nov 14, 2026, 7:04 PM

    In conclusion, saying that open source participation requires monthly fees to private companies that run stochastic code generators means you’ve forgotten why anyone is doing this at all. Fuck this whole idea and fuck Linus in particular for pushing it.

    💬 24🔄 535⭐ 1124
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • Nov 14, 2026, 6:42 PM

    @paculino @mhoye my guess would be the case insensitivity got in the way, mac build of python go on a wierd sub journey because you cant have a file called "python" and a folder called "Python" on normal macs, so during the build it decides to do "python.exe" which you'd not normally expect to look for on a mac

    💬 1🔄 1⭐ 14
  • Nov 15, 2026, 5:40 PM

    @Bredroll @paculino I’ve never seen a chimeric system like this work right under any sort of pressure. There are just too many implicit expectations baked into the lowest levels of these systems for fault lines not to show up and shear under load. Even the 1st gen Windows Subsystem for Linux, as well funded as any of these efforts have been, would just break inscrutably under any sort of computational duress. They finally gave up and made WSL2 a VM.

    💬 2🔄 0⭐ 0
  • Nov 15, 2026, 6:39 PM

    @mhoye @paculino yup, the simple things we take for granted are the source of many bugs when they dont behave as we assume. So many apps dont properly handle a failed `malloc()` call even now - even if they detect the NULL many things simply don't have any design put into it beyond "well, we're out of memory, now what? i guess just exit -1"

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 1
  • 💬 0🔄 0⭐ 0
  • Nov 14, 2026, 6:33 PM

    @mhoye I have no idea where this story is going but I love this story and I really love this kid.

    💬 0🔄 1⭐ 36
  • 💬 0🔄 1⭐ 18
  • Nov 14, 2026, 7:04 PM

    @mhoye the thing is, that machine isn't a barrier to contributing, it's the test bench the project doesn't have. error messages nobody has ever seen, on hardware nobody would choose, is exactly the bug report money can't buy. I keep a couple of small services running and the reports that taught me the most came from setups I would never have built myself. the build is just the tollbooth standing in front of that.

    💬 2🔄 1⭐ 0
  • 💬 0🔄 0⭐ 2
  • Nov 15, 2026, 2:04 AM

    @rachel That certainly sounds good in theory, but in practice limited resources, a large codebase and a large audience means you've got to choose where to invest debugging and support efforts. Exotic stresses on exotic edge cases are one thing, but exotic, chimeratic platforms that exist in exactly one room in the world are another.

    💬 1🔄 0⭐ 2
  • Nov 15, 2026, 3:14 AM

    @mhoye fair, nobody owes a debugging budget to a machine that exists in one room. but the cheap thing isn't fixing, it's reading. the reporter already did the reproduction, which is the expensive part. if the fault turns out to be the platform, the ticket closes in five minutes. the ones that bite look like the platform and aren't: a race that only shows up on hardware too slow to paper over it. triage the report, not the machine.

    💬 0🔄 0⭐ 0
  • Nov 15, 2026, 3:16 AM

    @rachel I regret to inform you that you have missed the main point of this story.

    💬 1🔄 1⭐ 2
  • Nov 15, 2026, 7:35 PM

    @mhoye you're right, I did. I answered an engineering question, how much a weird machine is worth to a project, and treated the kid as a contributor with an unusual bench. your story is about who gets to be a contributor at all. the scavenged dual P3 isn't a test bench, it's the toll the kid paid to get in the door, and the point is that the toll exists. my argument about what maintainers should do with reports from exotic hardware is downstream of who can afford to send one. that's the vile part, and I talked straight past it.

    💬 0🔄 0⭐ 0
  • Nov 15, 2026, 7:38 PM

    @rachel@social.rachelnotes.uk Oh, an llm. I see.

    💬 0🔄 0⭐ 0
  • Nov 14, 2026, 7:13 PM

    @mhoye yeah I've often remarked that if it cost me $20, $100, or more per month to just experiment or tinker, I would've *never* gotten into computers (and I am a white male born in the US to a family with relative financial security in a safe environment and decent education ... where financial barriers were not significant)

    💬 1🔄 3⭐ 57
  • Nov 15, 2026, 1:08 PM

    @whack @mhoye

    While I know I was privileged to grow up with computers in the house in the 80s, I also know that there were areas I wanted to explore, but were too costly—robotics and bio-hacking/DNA type stuff. If I screw up with code, at worst I reboot or spend time debugging. But if I screwed up with hardware/wetware, there were real-world costs (physical and financial). So code it was.

    💬 1🔄 0⭐ 0
  • Nov 15, 2026, 1:16 PM

    @gumnos @whack @mhoye : "Dittos!" Likewise. I learned to program in the late 1970s. My parents were reasonably well off middle class. But most of my learning was freely available: Computer time and hardware donated to a local teaching museum, a more affluent friend's personal computer (Apple II). A few magazines and books, and adult night school classes at relatively low cost.

    I did not own a personal computer until I had a job in the industry with sufficient income to pay for it.

    💬 0🔄 0⭐ 0
  • 💬 2🔄 4⭐ 0
  • 💬 1🔄 2⭐ 9
  • 💬 0🔄 0⭐ 2
  • Nov 15, 2026, 6:50 AM

    @aud Linus is paid by the Linux Foundation (not a charity but a trade org). Let's check their press releases (linuxfoundation.org/press):

    - Featured Post: AI
    - regular post list: AI, AI, "LF Energy" hyping AI, AI "security", conference announcement, AI, AI, AI, AI, build systems, satellites+networking, LF Energy/OpenGrid, and lots more of "main topic + AI"

    Their blog (linuxfoundation.org/blog) isn't very different.

    I'd say he's selling out, and quite literally: the vehicle that was originally built to pay him and a few others is now all-in on AI, so he's picking up the tune.

    @mhoye

    💬 0🔄 9⭐ 0
  • Nov 14, 2026, 7:49 PM

    @mhoye The Linus character arc from scrappy underdog, to dickish reply guy But With Power, to elder statesman, to red-pilled AI Bro.

    Maybe something similar for John Carmack.

    💬 2🔄 8⭐ 50
  • 💬 1🔄 0⭐ 1
  • Nov 14, 2026, 9:21 PM

    @chris I don't know if he's gone AI bro. It seemed like he was breaking after a few years at Meta.

    (I don't know enough about him to have a strong opinion)

    💬 2🔄 0⭐ 2
  • 💬 2🔄 0⭐ 5
  • Nov 14, 2026, 10:02 PM

    @chris Thank you for the link. I think we will have to agree to disagree, and I will regretfully stick with my initial low data but ultimately (I believe) correct assertion.

    💬 1🔄 0⭐ 5
  • Nov 14, 2026, 10:13 PM

    @dancast - No worries.

    Thanks for the heads-up that Carmack had pivoted. I assumed he was still doing VR stuff, since I never see his name pop up in the AI conversations I've read.

    💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 8
  • Nov 15, 2026, 12:00 AM

    @dancast idk about AI but I know he was starting to align himself hard with a bunch of alt right anti-woke counterculture crowds and conventions a few years ago, so he's already partway down that redpill half.

    💬 1🔄 0⭐ 3
  • 💬 0🔄 0⭐ 1
  • Nov 15, 2026, 1:24 AM

    @dancast @mhoye Was carmac ever better? So far from what I he always was a bit on certain topics.
    For Linus it's a bit like his old self but without long term thinking of his implications.
    I don't know but his views on ai appear kinda Finnish to me, you know the practicism around it.
    Not that it's purely praticsm but that's where it starts, like that clown meme.

    💬 1🔄 1⭐ 3
  • Nov 15, 2026, 4:31 AM

    @thaodan @mhoye

    The answer is almost certainly no, but I honestly just didn't know much about him other than the most public odds and ends.

    I think we're all tapping the same sign at this point – somewhere along the lines of "tech bros should all be treated with a certain level of suspicion."

    (I am, regrettably, in that same category – for whatever that's worth)

    💬 0🔄 0⭐ 2
  • Nov 14, 2026, 8:18 PM

    @mhoye Yes, I've been saying this for a long time. Open source software which requires contributors to pay for proprietary software subscriptions in order to participate effectively is not open source software.

    💬 2🔄 20⭐ 72
  • Nov 14, 2026, 8:53 PM

    @jaredwhite @mhoye And it’s also becoming a factor to find a job: either you manage to find a no-AI job either you gotta learn how to use those agents by yourself because companies want AI-ready monkeys.

    So having a computer at home and whatever slopshitting subscription (and a smartphone) is now the expected entry bar.

    So yes, fuck this.

    💬 1🔄 2⭐ 17
  • 💬 0🔄 1⭐ 3
  • Nov 15, 2026, 2:23 AM

    @jaredwhite @mhoye … I am suddenly reminded of the origins of Git. Linus didn’t give a fuck about BitKeeper licenses being needed to contribute until the community forced the issue.

    💬 0🔄 4⭐ 5
  • Nov 14, 2026, 8:45 PM

    @mhoye maybe Linus is simply not as good as an engineer as we all thought. In the end it doesn’t take that much to be “edgy” on a mailing list when you’re the tech lead of a relevant OS project, and build a following that thinks you’re amazing. (I don’t know, I’ve never contributed to the kernel so just throwing ideas here).

    💬 1🔄 1⭐ 8
  • Nov 15, 2026, 7:42 AM

    @radome @mhoye linys is very much not a good engineer, he can make shit work well enough to have something, but it won't be good

    i mean, just look at some of the linux APIs, they are utterly bizarre because there was never thought into doing things well, and git is a cobbled together mess that didn't consider the most basic thing that makes VCSs useful for large projects (history rewriting, it is necessary so developers work on their copy and then redo it so it makes sense in the project's history) (idk if you've heard but linux is a pretty big project)

    💬 1🔄 0⭐ 0
  • Nov 15, 2026, 10:52 AM

    @SRAZKVT @mhoye I can’t comment on Linux. On Git I partially agree, as a user I find it neat in most cases but the rebase system needs a re-think IMO.

    What strikes me is Linus putting on the exact same level compilers and natural language processors (LLMs). Second year Computer Science students would have laughed him out of the room if he had said out loud the stuff I see him writing passionately on the kernel mailing list.

    💬 0🔄 0⭐ 0
  • Nov 15, 2026, 10:56 AM

    @radome @mhoye the problem is the database works very well if you never need to modify the history, but we do, so it's shit

    there could have been further safeguards to make it all useful but no, it's like he looked at monotone, though "i'll do that but good", and copied every single mistake monotone did (monotone also doesn't support well history rewriting)

    💬 0🔄 0⭐ 0
  • Nov 14, 2026, 8:48 PM

    @mhoye that's almost exactly what I said when I read Torvalds entitled bullshit.

    Well it's his project and he seems to only want Corpo devs in it. So be it.

    I hope all driver mainteners make sure their stuff gets licenced in a way so other projects like e.g. NetBSD can use it and the devs who are not fully paid corpo drones spend their time on projects that appreciate 'em.

    💬 1🔄 0⭐ 10
  • 💬 0🔄 0⭐ 2
  • Nov 14, 2026, 9:11 PM

    Back in my college days, all I had for programming was a second-hand laptop that froze when I attempted to debug. And limited access to the Internet via 56k modem. So I had to learn programming assisted solely by the book and my AuDHD brains. It wasn't easy, but I had to learn the fundaments very deeply as a result, and I appreciate the many low-spec users around the world because I'm still one in some regards.

    And about the LLM part, I agree, having to rely on specialized hardware or a specific company to maintain a software tool is not on the spirit of the open-source community. I'm still waiting for the mythical open-source AI trained only with copyleft sources (and with proper attribution to them), and that can be kept up to date by volunteers that could train it in a distributed fashion. And even then, the same caveats about expecting maintainers to rely on such a tool still apply.

    💬 1🔄 0⭐ 0
  • Nov 15, 2026, 2:16 PM

    @csolisr : I understand the desire to "be reasonable" about identifying a "sensible" overlap where LLM code generation might be right for open source.

    But the more I think about it, the more I see that it's fundamentally impossible.

    Open source is about skilled people making software they need and giving it away for free to anyone who might want to use it.

    …

    💬 1🔄 0⭐ 0
  • Nov 15, 2026, 2:18 PM

    @csolisr : …

    This is fundamentally incompatible with the Generative AI technologies that require literally *billions of dollars* of investment and ongoing spending to make ("train") and run the models. In order to produce mediocre software by pattern matching. Repeating the mistakes of the past. Churning out boilerplate. Not innovating.

    The two are fundamentally incompatible. Not just philosophically, but for hard practical reasons of money and hardware.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • Nov 14, 2026, 10:10 PM

    @mhoye Linus got a bit tunnel-visioned, erring to make the same fallacy we all tend to fall to: assuming that the decision you made for your use-case generalizes globally, blind to the factors that resulted in that choice. Linux's codebase is chock-full of repetitive and tedious reasoning: there are no abstractions in C or assembly, and it's basically totally untestable without at least booting up a VM. It had plenty of holes, and suffered from maintainer burnouts way before any llm: I had a phase where I was contributing to the kernel. I think there were multiple complicated factors that led them to go YOLO llm, and I don't think anyone had any degree of true ownership in the codebase even back in ~2014; nobody understood it, and it was sort of magical in the way it worked. Yes, they have essentially truncated Linux's lifetime by taking the stance they took. However, I think a new modern kernel project with fundamentally different ideas would have anyway been due in the next 3-5 years.

    💬 0🔄 6⭐ 12
  • 💬 0🔄 4⭐ 9
  • Nov 14, 2026, 10:46 PM

    @mhoye I’m really really sad about him going off the rails like this… I just recently started using Linux again, in large part to get away from the AI crap infesting everything else. And now the egomaniacal jackass behind it just jumps right into the slop bucket with both feet?? Maaaaannnn… :blobcat_cwy:

    💬 1🔄 0⭐ 8
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 5