Login
You're viewing the discuss.systems public feed.
  • jonny (nonvenomous)jonny@neuromatch.social
    Oct 13, 2026, 9:56 PM

    RE: cosocial.ca/@mhoye/11743539240

    "Git has an elegant design, where a few simple concepts support very general needs and a decentralized, forkable, independent work. This doesn't allow us to slam through a million lines a day because our coding products can't possible consider the full design of the code and create a runaway nuclear slop reaction recreating the same thing hundreds of times. So we fixed it by centralizing it, converting simple offline file trees into an always-on server, treating partial information as a desirable trait, and bloating the elegant data structure into arbitrary JSON blobs"

    💬 5🔄 31⭐ 75

Replies

  • 💬 14🔄 3⭐ 33
  • 💬 3🔄 3⭐ 16
  • 💬 4🔄 0⭐ 7
  • 💬 1🔄 0⭐ 6
  • 💬 0🔄 0⭐ 0
  • Sheldonsysop408@sfba.social
    Oct 13, 2026, 10:29 PM

    @Viss

    Broke: Git
    Woke: Github
    Bespoke: Github Hub

    Evidently the time has come for repos that only merge other repos.

    @jonny

    💬 1🔄 0⭐ 7
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 3
  • 💬 1🔄 0⭐ 7
  • Varxvarx@defcon.social
    Oct 13, 2026, 10:24 PM

    @ariadne @jonny

    Who would look at modern distributed change control system like git or mercurial and go "naw bro. We need a centralized code repo" and then reinvent svn.

    Fuck. These people would wouldn't they? No nonono. I won't use SVN again. You cant make me! We replaced it for a reason!

    💬 1🔄 0⭐ 9
  • Ariadne, bunyote friend :therian:ariadne@treehouse.systems
    Oct 13, 2026, 10:27 PM

    @varx @jonny git is not a particularly great fit for giant monorepos, in fact, nothing except giant-monorepo-specific source control is a good fit for giant monorepos, which kinda demonstrates that giant monorepos are probably a bad approach to source control.

    💬 1🔄 0⭐ 8
  • 💬 0🔄 0⭐ 1
  • 🎃 David Z Pumpkins 🎃diazona@techhub.social
    Oct 13, 2026, 10:29 PM

    @varx @ariadne @jonny Eh to be fair I think their logic makes sense. I mean, they've managed to manufacture a crazy development ecosystem, but that's hardly unprecedented, and given that particular set of constraints it's reasonable that Git leaves room for improvement.

    💬 1🔄 0⭐ 1
  • 💬 1🔄 0⭐ 5
  • :: Asta ::aud@spectre.asta.lgbt
    Oct 14, 2026, 12:36 AM

    @jonny @diazona @varx @ariadne I just have to wonder what possible point there is to version control if you're making 30 fucking thousand changes per day.

    Even if you say "okay, the 'agent' can bisect the changes and make a fix", you're already burning infinite compute, why even bother with that?  Why not just spin the wheel till whatever is 'broken' is 'fixed'??  Have they ever reverted a change?  Would they even know if they had??

    💬 1🔄 0⭐ 0
  • :: Asta ::aud@spectre.asta.lgbt
    Oct 14, 2026, 12:37 AM

    @jonny @diazona @varx @ariadne What theoretical 'clean state' are you even going to roll back to??

    just... why bother.  just copy it to a tape every damn day and if it breaks, restore from backup and spin the wheel till it works again

    💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • 💬 1🔄 1⭐ 9
  • 💬 2🔄 0⭐ 0
  • 💬 1🔄 0⭐ 3
  • 💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 1
  • :: Asta ::aud@spectre.asta.lgbt
    Oct 14, 2026, 1:50 AM

    @glyph @jonny @diazona @varx @ariadne how hard could cancer even be, really.  it's not like it's a whole bunch of different things or anything, I mean, they wouldn't lump it all into a single word if that were the case.

    anyway I'm sure the cure is *some* non-linear combination of every reddit post that's ever existed.

    💬 0🔄 0⭐ 0
  • 💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Kaiser C Kitty - Professional GeniusKaiserKitty@brain.worm.pink
    Oct 14, 2026, 9:47 PM
    @ariadne @aud @glyph @jonny @diazona @varx as a bioinformatist I can tell you Claude means nothing. The field has been glutted with machine learning assisted findings for decades before LLMs and the bottle neck has always been verification. There is likely hundreds of thousands of papers listing possible genes to target to deal with any number of cancers no one has time to check. Even if Claude found 1 million more it wouldn’t mean much. In fact at some places I’ve hearing roughly 2 million ML assisted drug findings generates like 2 candidates which even make it to animal testing.

    It’s just software people assuming all or STEM works like programming.
    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • :: Asta ::aud@spectre.asta.lgbt
    Oct 14, 2026, 9:58 PM

    @glyph @KaiserKitty @varx @jonny @ariadne @diazona Yeah, it's just them covering their asses and making big promises that they are woefully unqualified to even *understand* how to keep.  They trot out this "cure cancer" thing everytime someone starts saying "maybe spending a trillion dollars to brute force language isn't a good idea, especially when we can already speak to each other".

    I'm *quite* sick of them dangling some theoretical benefit like that to cover up their shenanigans.  It's not just unscientific bullshit, it's cruel and unethical to do so.  As noted, "cancer" isn't a single thing with a single cause and effect so the whole thing is just... they may as well claim to solve FTL travel, given their total lack of understanding of *anything* relevant.

    (my own background is in protein folding, so while less relevant than bioinformatics I have enough knowledge to where their little "we'll cure cancer" has annoyed me to the point where I feel people should absolutely call them out on it, especially as they trotted that back out a few weeks.  "We'll cure cancer" "really cause it looks like you're dicking around with re-inventing a lopsided wheel and you don't even *understand* what 'cancer' is")

    (having said that I appreciate the input!  But felt maybe I should give a little background to my own off-the-cuff annoyed joke since I actively do NOT want people to think either that A. that it's at *all* plausible to regurgitate reddit posts into a 'cancer cure', or that B. I believe they can)

    Buuut all these freaking puff piece interviews and press releases... no one feels the need to ask, "didn't you say you were gonna 'cure cancer'?  Why are you spending all your compute on solved problems", clearly.

    💬 0🔄 0⭐ 0
  • ¡a!niain@kolektiva.social
    Oct 13, 2026, 10:21 PM

    @jonny well, a commit is an actual record of the changes, whereas a scn changeset is whatever the LLM they inevitably use to run diff on the code hallucinated as its output

    💬 0🔄 0⭐ 4
  • @sarasara@hachyderm.io
    Oct 13, 2026, 10:21 PM

    @jonny and exactly zero humans were harmed (or involved) in the writing of this paragraph

    💬 0🔄 0⭐ 3
  • 💬 0🔄 1⭐ 7
  • 💬 0🔄 0⭐ 1
  • 💬 0🔄 0⭐ 3
  • 💬 0🔄 0⭐ 0
  • Mark Dominusmjd@mathstodon.xyz
    Oct 13, 2026, 10:29 PM

    @jonny

    Commits are not changesets. A commit is a snapshot of the source tree. It is not a record of the changes that produced that tree. For that you have to compare the trees in two commits, which is what `git diff` does.

    Git is different in this way from many other VCSes (SVN for example), which do explicitly track changes. In Git's early days the choice to track trees rather than changes was quite controversial.

    💬 1🔄 0⭐ 7
  • jonny (nonvenomous)jonny@neuromatch.social
    Oct 13, 2026, 10:56 PM

    @mjd i know, but that's because the reference to the whole file is a frontend/format concern and representing it as a delta or a full file is an implementation detail, which seems like a split that makes sense to me? I only care about the delta when i am cloning, storing, and diffing. two of those i should never have to think about, and the other is good to have as a computed/derived outcome of comparing two full things rather than encoded directly because then i can choose a different diffing algorithm while keeping a single unitary hash tree datastructure (and not, e.g., have the hash of a changeset depend on the diffing algorithm)

    💬 1🔄 0⭐ 0
  • bnuuybnuuy@mathstodon.xyz
    Oct 13, 2026, 11:20 PM

    @jonny @mjd with git, linus in his infinite wisdom decided on storing the full file to keep git simple, and to massively speed up checkouts. anthropic realized this approach also has drawbacks, actually, and talks to us as if changesets are this radical new invention they came up with, and not the norm outside of basically every other version control system other than git.

    idk, the fact that they go from "now version control is becoming the bottleneck from code churn and gigabytes(!!) of code" to "we will write (vibe?) our own version control to fix this" rather than ask if they aren't fundamentally going in the wrong direction in some way, is deeply unimpressive.

    💬 0🔄 0⭐ 7
  • tekheddtekhedd@byteheaven.net
    Oct 13, 2026, 10:44 PM

    @jonny I guess pedantically (my happy place) a "commit" is whatever causes your scm to produce one changeset, which is whatever yoiur SCM uses to represent a delta from a previous state.

    Everything else is implementation details and nothing you should know about at all about unless somebody really screwed up and you don't have backups. The fact that they internally use "commits" or "changesets" to represent internal implementation is (should be) irrelevant but usually isn't because developers.

    💬 1🔄 0⭐ 0
  • tekheddtekhedd@byteheaven.net
    Oct 14, 2026, 2:40 AM

    @jonny y'know, all this bullshit is why I still prefer Mercurial as my goto. If you need something it doesn't have, it doesn't need enhancement, instead maybe your development process needs review.

    💬 0🔄 0⭐ 0
  • JPfroztbyte@mastodon.social
    Oct 13, 2026, 11:10 PM

    @jonny I see they’re going to rediscover what darcs and fossil have known for years

    should be entertaining

    💬 0🔄 0⭐ 3
  • Dr. Bruce Simpsonbms@bsd.cafe
    Oct 13, 2026, 11:38 PM

    @jonny Wait, they named it the same as the Swiss-originated, excessively-centralized control-structured Internet Protocol backbone replacement?

    That is truly epic

    💬 0🔄 0⭐ 2
  • TunafishTigertunafishtiger@mastodon.online
    Oct 14, 2026, 3:32 AM

    @jonny it may be ontologically exactly the same but it is different in one key respect: one is bullshit and the other is new bullshit

    💬 0🔄 0⭐ 2
  • 💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 2
  • 💬 1🔄 0⭐ 2
  • FoolishOwlfoolishowl@social.coop
    Oct 15, 2026, 5:26 PM

    @jni @jonny @mhoye I'm wondering if someday we'll see encouraging frequent patching to be as elementary a blunder as measuring productivity in lines of code.

    💬 0🔄 0⭐ 0
  • Ariel Chinnyellyc.bsky.social@bsky.brid.gy
    Oct 14, 2026, 5:00 AM

    In fairness, Google moved off git many years ago because it genuinely couldn’t keep up with what they were doing.
    I’m sure this project will be much more stupid though.

    💬 0🔄 0⭐ 0
  • Alien software, human hardwaremavu@mastodon.social
    Oct 14, 2026, 8:39 AM

    @jonny "Plus, Anthropic is unique in that our ability to spend tokens for internal use far exceeds what any other business would realistically want to spend."

    Ah.. Not sure i would want that sentence in an official public post while preparing an IPO.

    💬 1🔄 0⭐ 1
  • 💬 0🔄 0⭐ 1