[HN Gopher] Vibe-Coded Ext4 for OpenBSD
___________________________________________________________________
Vibe-Coded Ext4 for OpenBSD
Author : corbet
Score : 53 points
Date : 2026-03-27 18:51 UTC (4 hours ago)
(HTM) web link (lwn.net)
(TXT) w3m dump (lwn.net)
| bitwizeshift wrote:
| Paywalled article on something vibe-coded? That seems like a bold
| strategy.
| dana321 wrote:
| click to continue
| LeFantome wrote:
| Vibe coding and OpenBSD. The perfect combination.
| croes wrote:
| Vibe coding and file systems are even better
| himata4113 wrote:
| _trying to load with linux ext4_ hmm doesn 't load, but it
| works with my version!
|
| Must be a bug in the linux kernel, let me git clone and build
| an out-of-tree module...
| LeFantome wrote:
| Kent Overstreet has already blazed that trail.
| api wrote:
| It's clearly an experiment.
| whalesalad wrote:
| I vibe-configured an Edgerouter 4 as a hot-drop box that would
| establish a secure tunnel and create a fake WAN for some
| servers that had to be temporarily pulled from service but
| remain operational in someones home garage. I overnight shipped
| it to them with two of the ports labeled, they plugged in home
| internet on one port, the rack on the other port, and it secure
| tunneled to a Linode VPS to get a public IP, circumventing all
| the Verizon home internet crap. I used OpenBSD. Claude did most
| of the work.
| nurettin wrote:
| It is amusing to see that the only concern seems to be about a
| confusion around licensing, not the validity or maintainability
| of the code itself.
| g0xA52A2A wrote:
| Is it worth the effort to review until such implications are
| understood?
| nurettin wrote:
| No of course not, bike shedding licenses is where it is at.
| kvuj wrote:
| How is that different than a human writing the code? Whether an
| AI or a human wrote it, I would expect the same bar of
| validity/maintainability.
| nurettin wrote:
| To me, SOTA is just bad at DRY, KISS, succint, well
| architected, top down, easy to test code and has to be
| constantly steered to come close. Even the article suggests
| that. YMMV.
| serf wrote:
| TDD and strong goals help..
|
| ..much like with human development.
| scuff3d wrote:
| Because humans make design decisions, AI just bangs it's head
| against the problem until it gets something that "works".
| tolciho wrote:
| Eh, well, if your guns are trained on the "copyright" portion
| of the ship and you can sink it from there, no need to waste
| ammo or time trying to figure out if code bits are as explosive
| as the copyright bits are. Probably the code is just as
| sinkable, e.g. here's a recent response to some other AI slop:
| I didn't look closely at most of the code but one thing that
| caught my eye, pid is not safe for tempfile name generation,
| another user of the system can easily generate files that
| conflict with this. Functions like mktemp and mkstemp are there
| for a reason. Some of the other "safety" checks make no sense.
| If the LLM code generator is coming up with things which any
| competent unix sysadmin (let alone programmer) can tell are
| obviously wrong, it doesn't bode well for the rest.
|
| https://marc.info/?l=openbsd-ports&m=177460682403496&w=2
|
| The next AI winter can't come soon enough...
| throwatdem12311 wrote:
| Can someone just copyright wash Windows already.
| greyface- wrote:
| ReactOS did this without any need for an LLM.
| ziml77 wrote:
| No they didn't. It would be copyright washing if someone
| contributed to ReactOS who remembered large portions of the
| Windows code and wrote the ReactOS implementations based on
| that.
| wongarsu wrote:
| The Windows 2000 and Windows XP sources are readily available
| and must have made it into the training data. But most software
| has dropped XP support. You really need at least some of the
| Win 8 and Win 10 APIs to claim compatibility with modern
| software, and I doubt claude has seen those from the inside
| FeepingCreature wrote:
| > So as of today, the Copyright system does not have a way for
| the output of a non-human produced set of files to contain the
| grant of permissions which the OpenBSD project needs to perform
| combination and redistribution.
|
| This seems extremely confused. The copyright system does not have
| a way to grant these permissions because the material is not
| covered under copyright! You can distribute it at will, not due
| to any sort of legal grant but simply because you have the
| ability and the law says nothing to stop you.
| jagged-chisel wrote:
| Eh ... the argument will likely be things created by Thing at
| the behest of Author is owned by the Author. It'll take a few
| cases going through the courts, or an Act of Congress to
| solidify this stuff.
| HappySweeney wrote:
| Haven't there already been a few cases, each of which found
| that mechanically-produced works are not copywritable?
| senko wrote:
| no
| wongarsu wrote:
| Just like we settled on photographers havin copyright on the
| works created by their camera. The same arguments seem to
| apply
|
| The US Copyright Office has published a piece that argues
| otherwise, but a) unless they pass regulation their opinion
| doesn't really matter, and b) there is way too much money
| resting on the assumption code can be copyrighted despite AI
| involvement.
| fragmede wrote:
| It's not settled. The monkey selfie copyright dispute ruled
| that a monkey that pressed the button to take a selfie,
| does not and cannot open the copyright to that photo, and
| neither does the photographer who's camera it was. How that
| extends to AI generated code is for the courts to decide,
| but there are some parallels to that case.
|
| https://en.wikipedia.org/wiki/Monkey_selfie_copyright_dispu
| t...
| charcircuit wrote:
| This filesystem driver was made by a human using AI, not
| a monkey.
| wongarsu wrote:
| But with the monkey there are two levels of separation
| from the artist: the human makes the creative decision to
| hand the camera to a monkey, who presses the trigger, and
| the camera makes the picture. Compared to the single
| layer of separation of a photographer choosing framing
| and camera parameters, pressing the trigger and the
| camera taking the picture. Or the zero levels of
| separation when the artist paints the picture.
|
| A programmer writing code would be like the painter, and
| the programmer writing a prompt for Claude looks a lot
| like the photographer. The prompt is the creative work
| that makes it copyrightable, just like the artistic
| choices of the photographer make the photo copyrightable
|
| You could argue that the prompt is more like a technical
| description than a creative work. But then the same
| should probably be true of the code itself, and
| consequently copyright should not apply to code at all
|
| The copyright office's argument is that the AI is more
| like a freelancer than like a machine like a camera.
| Which you might equate to the monkey, who's also a bit
| freelancer like. But I have my doubts that holds up in
| court. Monkeys are a lot more sentient than AIs
| KallDrexx wrote:
| The copyright office is pretty clear on this if you read:
| https://www.copyright.gov/ai/Copyright-and-Artificial-
| Intell....
|
| There is case law surrounding the fact that just because
| you commission a work to another entity doesn't give you
| co-authorship, the entity doing the work and making
| creative decisions is the entity that gets copyright.
|
| In order for you to have co-authorship of the
| commissioned work you have to be involved and pretty much
| giving instruction level detail to the real author. The
| opinion shows many cases that its not the case with how
| LLM prompts work.
|
| The monkey selfie case is relevant also because since it
| also solidifies that non-persons cannot claim copyright,
| that means the LLM cannot claim copyright, and therefore
| it does not have copyright that can be passed onto the
| LLM operator.
| michaelmrose wrote:
| The law is whatever it needs to be to satisfy monied
| interests with the degree of acceptable of adaptation
| being a function of the unity of those interests and the
| political ascendancy of those in favor.
|
| Overwhelmingly this is in favor of treating ai as a tool
| like Photoshop.
|
| Even those against AI disagree on different matters and
| will overwhelmingly want a cut not a different
| interpretation.
| themafia wrote:
| Just because you can distribute something doesn't mean you
| aren't violating someone _else 's_ copyright. You cannot assume
| that just because a language model popped out some code for you
| that it is clear of any other claims.
|
| This is just lazy copyright whitewashing.
| plorg wrote:
| This all relies, as the article points out, on everyone looking
| directly at code that both looks like and works like the only
| extant codebase for EXT4 and nonetheless concluding that in
| fact the computer conjured it from the aether. If I wrote a
| program that zipped up the Linux kernel source, unzipped it,
| and grepped -v for comments it would not then be magically
| transformed into unattributable public domain software.
| FeepingCreature wrote:
| Under the premise advanced in the quote, copyright is not
| being violated because there is none. Thus, the quote makes
| no sense as stated. It may be that, additionally, copyright
| is in fact being violated (I don't believe it myself), but if
| so that's a separate argument.
| g0xA52A2A wrote:
| Wow that thread just kept going. Whilst the LWN article covered
| most of the "highlights" I think this reply from Theo is pretty
| suscient on the topic at large [1].
|
| [1] https://marc.info/?l=openbsd-tech&m=177425035627562&w=2
| bt1a wrote:
| > Lacking Copyright (or similarily a Public Domain declaration
| by a human), we don't receive sufficient rights grants which
| would permit us to include it into the aggregate body of source
| code, without that aggregate body becoming less free than it is
| now.
|
| Thats awesome lmao
| raggi wrote:
| that's not a statement from a lawyer, and it's confused.
| there is one true thing in there which is that at least under
| US considerations the LLM output may not be copyrightable due
| to insufficient human involvement, but the rest of the
| implications are poorly extrapolated.
|
| there are lots of portions of code today, prior to AI
| authorship, that are already not copyrightable due to the way
| they are produced. the existence of such code does not
| decimate the copyright of an overall collective work.
| LeFantome wrote:
| The article is largely about the copyright concerns of LLM
| generated code that was almost certainly trained on the GPL
| original.
|
| Also, it is essentially an ext2 filesystem as it does not support
| journaling.
| charcircuit wrote:
| >incorporate knowledge carrying an illiberal license.
|
| Copyright prevents copying. It doesn't prevent using knowledge.
| bigfishrunning wrote:
| Good luck proving an LLM has "Knowledge", and isn't just a
| statistical model that tries to form outputs as a copy of it's
| training data...
| CodeWriter23 wrote:
| Well this is ironic, GPL advocate(s) declaring a clean
| implementation based on specifications infringing due to
| someone/something reading specs provided under license. Didn't
| Oracle lose that argument in court as pertains to Android
| implementation of Java libraries?
| corbet wrote:
| I'm not sure what you're reading; there is a distinct lack of
| GPL advocates in that conversation.
| longislandguido wrote:
| ~20 years ago, the Linux camp accused OpenBSD of importing GPL'd
| code (a wireless driver IIRC) and cried foul. The code was
| removed.
|
| Fast forward to 2026, Theo says no to vibe-coded slop, prove to
| me your magic oracle LLM didn't ingest gobs of GPL code before
| spitting out an answer.
|
| People are big mad of course, but you want me to believe Theo is
| the bad guy here for playing it conservatively?
| ksherlock wrote:
| The history is a bit backwards but the point is good. OpenBSD
| atheros wireless code was imported into linux, the BSD
| attributions were removed, and it was re-declared as GPL. That
| was later changed back.
| longislandguido wrote:
| https://marc.info/?l=linux-wireless&m=117579116031296&w=2
| hypeatei wrote:
| > This obsession with copyrights between different free software
| ecosystems - who put the lawyers in charge?
|
| This comment on the article is spot on. I don't vibe code or care
| about AI really, but it's so exhausting to see people playing
| lawyer in threads about LLM-generated code. No one knows, a ton
| of people are using LLMs, the companies behind these models
| torrented content themselves, and why would you spend your time
| defending copyright / use it as a tool to spread FUD? Copyright
| is a made up concept that exists to kill competition and protect
| those who suck at executing on ideas.
| ethin wrote:
| > Lacking Copyright (or similarily a Public Domain declaration by
| a human), we don't receive sufficient rights grants which would
| permit us to include it into the aggregate body of source code,
| without that aggregate body becoming less free than it is now.
|
| Can someone explain this to me? I was under the impression that
| if a work of authorship was not copyrightable because it was AI
| generated and not authored by a human, it was in the public
| domain and therefore you could do whatever you wanted with it.
| Normal copyright restrictions would not apply here.
| Joel_Mckay wrote:
| Data theft of service or piracy from the web and "AI" users
| content are used in the model training sets, and when codified
| the statistical saliency is significant if popular content is
| present.
|
| For example, when an LLM does a vector search, there is a high
| probability of pirated content bleed-though and isomorphic
| plagiarism in the high dimensional vector space results. Thus,
| often when you coincidentally type in "name a cartoon mouse",
| there is a higher probability Disney "Micky Mouse" will pop out
| in the output rather than "Mighty Mouse". Note Trademarks never
| expire if the fees are paid, and Disney can still technically
| sue anyone that messes with their mouse.
|
| Much like em dashes "--", telling the current set of models to
| stop using them inappropriately often fails. Also, activation
| capping is used to improve the models behavioral vector, and
| have nothing to do with the Anthropic CEO developing political
| ethics.
|
| LLM are useful for context search, but can't function properly
| without constantly stealing from actual humans. Thus, will
| often violate copyright, trademark, and patents. In a
| commercial context it is legally irrelevant how the output has
| misappropriated IP, and one can bet your wallet the lawyers
| won't care either. No, IP is not public domain for a long time
| (17 to 78 years) regardless of peoples delusions, even if some
| kid in a place like India (no software patents) thinks it is..
|
| This channel offers several simplified explanations of the work
| being done with models, and Anthropic posts detailed research
| papers on its website.
|
| https://www.youtube.com/watch?v=YDdKiQNw80c
|
| https://www.youtube.com/watch?v=Xx4Tpsk_fnM
|
| https://www.youtube.com/watch?v=JAcwtV_bFp4
|
| Many YC bots are poisoning discourse -- so this thread will
| likely get negative karma. Some LLM users seem to develop
| emotional or delusional relationships with the algorithms. The
| internet is already >52% generated nonsense and growing. =3
| cachius wrote:
| I'd like to see it AFL fuzzed and compared to the original. Took
| 2 hours to first bug ten years ago in 2016.
|
| Discussion then https://news.ycombinator.com/item?id=11469535
|
| Mirror of the slides
| https://events.static.linuxfound.org/sites/events/files/slid...
| joshstrange wrote:
| > Who is the copyright holder in this case? It clearly draws
| heavily from an existing work, and it's clear the human offering
| the patch didn't do it. It's not the AI, because only persons can
| own copyright. Is it the set of people whose work was represented
| in the training corpus? Was the it the set of people who wrote
| ext4 and whose work was in the training corpus? The company who
| own the AI who wrote the code? Someone else?
|
| I don't love this take. Specifically:
|
| > it's clear the human offering the patch didn't do it
|
| I find it hard to believe that there wasn't a good bit of "blood,
| sweat, and tears" invested by a human directing the LLM to make
| this happen. Yes, LLMs can spit out full projects in 1 prompt but
| that's not what happened here. From his blog the work on this
| spanned 5 months at least. And while he probably wasn't working
| on it exclusively during that time, I find it hard to believe it
| was him sending "continue" periodically to an LLM.
|
| Anyone who has built something large or complicated with LLM
| assistance knows that it takes more than just asking the LLM to
| accomplish your end goal, saying "it's clear the human offering
| the patch didn't do it" is insulting.
|
| I've done a number of things with the help of LLMs, in all but
| the most contrived of cases it required knowledge, input from me,
| and careful guidance to accomplish. Multiple plans, multiple
| rollbacks, the knowledge of when we needed to step back and when
| to push forward. The LLM didn't bring that to the table. It
| brought the ability to crank out code to test a theory, to
| implement a plan only after we had gone 10+ rounds, or to
| function as grep++ or google++.
|
| LLMs are tools, they aren't a magic "Make me ext4 for
| OpenBSD"-button (or at least they sure as hell aren't that today,
| or 5 months ago when this was started).
| kgeist wrote:
| Binaries are copyrightable in both the US and the EU, and they
| are not technically produced by a human either, they're produced
| by a computer program. I honestly don't understand why this isn't
| extended to AI-generated code. Isn't it the same thing? One could
| argue that compilers merely transform source code into binaries
| "as is," while AI models have some "knowledge" baked in that they
| extract and paste as code. But there are compilers that also
| generate binaries by selecting ready-to-use binary patches
| authored by compiler developers and combining them into a
| program. One could also argue that, in the case of compilers, at
| least the input source code is authored by a human. But why can't
| we treat prompts as "source code in natural language" too? Where
| is the line between authorship and non-authorship, and how is the
| line defined? "Your prompt was too basic to constitute
| authorship" doesn't sound like an objectibe criterion.
|
| Maybe for lawyers, AI is some kind of magical thing on its own.
| But having successfully created a working inference engine for
| Qwen3, and seeing how the core loop is just ~50 lines of very
| simple matrix multiplication code, I can't see LLMs as anything
| more than pretty simple interpreters that process "neural network
| bytecode," which can output code from pre-existing templates just
| like some compilers. And I'm not sure how this is different from
| transpilers or autogenerated code (like server generators based
| on an OpenAPI schema)
|
| Sure, if an LLM was trained on GPL code, it's possible it may
| output GPL-licensed code verbatim, but that's a different matter
| from the question of whether AI-generated code is copyrightable
| in principle.
|
| Interestingly, I found an opinion here [0] that binaries
| technically shouldn't be copyrightable, and currently they are
| because: the copyright office listened to
| software publishers, and they wanted binaries protected by
| copyright so they could sell them that way
|
| [0]
| https://freesoftwaremagazine.com/articles/what_if_copyright_...
| wahern wrote:
| That linked opinion overstates the case. In the real-world, two
| different programs performing any non-trivial but functionally
| identical task will look substantially dissimilar in their
| source code, and that dissimilarity will carry over to the
| compiled binary, meaning what was expressive (if anything) is
| largely preserved. To the extent two different programs do end
| up with identical code, then that aspect was likely primarily
| functional and non-copyrightable, or at least the expressive
| character didn't carry over to the binary. Ordering and naming
| of APIs in source code can be expressive, and that indeed is
| often lost (literally or at least the expressive character)
| during the compilation process, but there are other expressive
| aspects to software programing that will be preserved and
| protected in the binary form.
|
| IMO, your intuition regarding AI is right--it's not a magic
| copyright laundering machine, and AFAIU courts have very
| quickly agreed that infringement is occurring. But in copyright
| law establishing infringement (or the possibility of
| infringement) is the easy, straight-forward part. Copyright
| infringement _liability_ is a much more complex question.
| Transformative uses in particular are a Fair Use, and Fair Use
| is technically treated as an affirmative defense to
| infringement.[1] If something is Fair Use, infringement is
| effectively presumed. But Fair Uses are typically very fact-
| intensive questions, and unlike the case with search engines I
| 'm not sure we'll get to the point where there's a well-defined
| fence protecting "AI".
|
| [1] There's a scholarly pedantic debate about whether Fair Use
| is properly a "defense", rather than "exception" to
| infringement, but it walks and talks like a defense in the
| sense that the defendant has the burden of proving Fair Use
| after the plaintiff has established infringement. There's a
| similarly pedantic (though slightly more substantive) debate in
| criminal law regarding affirmative defenses. But the very term
| "affirmative defense" was coined to recognize and avoid these
| pedantic debates.
| ptidhomme wrote:
| I liked this reply in the thread :
|
| _There 's another issue surrounding developer skill atrophy or
| stunting that I find \ particularly concerning on an existential
| level.
|
| If we allow people to use LLMs to write code for a given
| project/platform, experience \ in that platform will potentially
| atrophy or under develop as contributors \ increasingly rely on
| out sourcing their applicable skills and decisions to "AI".
|
| Even if you believe out sourcing the minutia of coding is a net
| positive, the \ "enshitification" principal in general should
| give you pause; as soon as the net \ developer skill for a
| project has degraded to a point of reliance, even somewhat, I \
| think we can be confident those AI tools will NOT get less
| expensive.
|
| I'd rather be independently less productive, than dependent on
| some MegaCorp(TM)'s \ good will to rent us back access to our
| brains at a fair price.
|
| - achaean_
|
| https://marc.info/?l=openbsd-tech&m=177430829313972&w=2
| hulitu wrote:
| > Vibe-Coded Ext4 for OpenBSD
|
| Who wants to test it ? Preferably on real hardware. /s
___________________________________________________________________
(page generated 2026-03-27 23:01 UTC)