Egregoros

19w

simon is very kind to the zig project. the rationale is rather nonsensical.

lain

@mischievoustomato@tsundere.love remote

19w

@s8n the problem is in the last line:

> We don’t do this just because it’s the “right” thing to do, but also because it’s the smart thing to do.

what's smart about not merging 4x improvements, because you hope that you get more contributors that way? a very shaky assumption, too.

Replying to @lain@fediffusion.art

tomatobert

19w

@lain @s8n hypothetical currently, but what happens to a project if a pro-ai/ok-with-ai fork ends up being more successful than itself?

Replying to @mischievoustomato@tsundere.love

lain

19w

@mischievoustomato @s8n they'll get all the great contributors, of course

Replying to @lain@fediffusion.art

hat

19w

@lain @s8n a lot of these people are basically artniggers for software. they care less about objective stuff like "does it compile" and "does it run" and "what are the performance / efficiency characteristics" and more about the guy who wrote this or that line of code. they think a public project is like a digital CHAZ where we just need to get along and have fun, man.

lain

19w

@s8n i think that's what they mean by 'smart', though. if not "this is better for the project long-term", what do they mean?

Replying to @lain@fediffusion.art

hat

19w

@lain @s8n @mischievoustomato and those great contributora won't need to forsake the use of a novel, interesting tool because of the maintainers' zeal to quash that tool, meaning the same excellent contributors will be even more excellent in their work on the fork.

hat

19w

@s8n @lain also i have yet to see your foreskin, jew.

hat

19w

@s8n @lain this is not a picture of your foreskin.

Replying to @lain@fediffusion.art

Wanderer atop the sea of clouds or whatever

19w

I think it does make sense. If you want competent people in your maintainer pool and your time is limited, there is no point in allowing LLM commits. They seem to believe that progress on the language is fast enough as is.

Replying to @HatkeshiatorTND@annihilation.social

lain

19w

@HatkeshiatorTND @s8n guys guys this isn't poast

Replying to @WandererUber@poa.st

lain

19w

@WandererUber i understand what they are going for, but i still think they are making lot of assumptions that are unlikely to hold

Replying to @lain@fediffusion.art

Wanderer atop the sea of clouds or whatever

19w

Which are those?
A software does not attract maintainers just because it is fast/well-written, so even if the majority of LLM commits' quality was to their standards, it's not a bad assumption to bet on community instead.

And I think we are also making assumptions such as "they will never have this perf benefit", which is not necessarily true. If one of their maintainers studies the architecture of that patch and writes one that does it, they will get the improvement, it's just that they don't offload direct work onto LLMs right now.

Replying to @WandererUber@poa.st

lain

19w

@WandererUber sorry I’ll reply later

Replying to @WandererUber@poa.st

lain

18w

@WandererUber about the perf benefit, if they have a strict "not even issue reports" policy, i doubt it they'd want someone to take a look at the patch and then just rewrite it to not have the wrong colored bits on it.

> A software does not attract maintainers just because it is fast/well-written

i think it does, especially if those maintainers can use all the use as well. Will be interesting to see how this all plays out over the next years.

Phantasm reposted

lain

19w

> Bun operates its own fork of Zig, and recently achieved a 4x performance improvement on Bun compile after adding "parallel semantic analysis and multiple codegen units to the llvm backend". Here's that code. But @bunjavascript says:
> We do not currently plan to upstream this, as Zig has a strict ban on LLM-authored contributions.

https://simonwillison.net/2026/Apr/30/zig-anti-ai/

Replying to @lain@fediffusion.art

Phantasm

18w

@lain
zig-rust-c-languages-carmack.webp

Phantasm

@hj@shigusegubu.club remote

18w

@s8n @HatkeshiatorTND @lain
>So they do not want any code in the codebase that none of them understands or is willing to maintain.
This is completely the same with human contributions. You don't merge something you don't want to deal with later. Simple as.

Replying to @phnt@fluffytail.org

Fedi Developer (of crushes)

18w

@phnt @lain JS and webdev:

Replying to @hj@shigusegubu.club

Phantasm

18w

@hj @lain
js_devs.jpg
javascript-for-kids.png

Replying to @phnt@fluffytail.org

lain

18w

@phnt @HatkeshiatorTND @s8n i some cases i'd be a lot more willing to merge both LLM and human code because I can get two or three LLM reviewers to look through it instead of just tired old me.

Replying to @lain@fediffusion.art

Q.U.I.N.N.

18w

@lain @s8n not a lot of people out there shiping zig and doing a type 1[1] gives them an audience to grand stand about.

that being said if i see the coding agent shitting directly to the repo i'm giving you the side eye

[1]: https://iceworks.cc/blog/uctd/

Replying to @lain@fediffusion.art

Q.U.I.N.N.

18w

@lain @HatkeshiatorTND @phnt @s8n i don't think they're concerned about the QC aspect. there are projects which have QC policies about ai assisted code and they're pretty transparent about the 100% inspection policy.

Replying to @icedquinn@blob.cat

hat

18w

@icedquinn @lain @phnt @s8n are you implying certain projects may be making decisions based on something other than practicality? such as, perhaps... ideology?
unfathomable.

Replying to @HatkeshiatorTND@annihilation.social

lain

18w

@HatkeshiatorTND @icedquinn @phnt @s8n i mean, if they want to be a by-humans-for-humans community feelgood project, that's great, but i just never had the impression that zig was about that. I guess i was just wrong.

Replying to @lain@fediffusion.art

hat

@hj@shigusegubu.club remote

18w

@lain @icedquinn @phnt @s8n project culture is a very real thing and everybody has convictions. there's always a margin where you'll refuse +0.5% productivity on principle, ditto +5000% productivity.
this is why we have forks! forking a project can help it continue when the culture is no longer conducive to the project's spirit. if zig happens to be in the hands of people who care deeply about not accepting ai contributions, i don't want or need the project reft from their hold or anything. i can just operate a fork.
which is what bun is doing, which is why bun/zig is going to go places zig/zig isn't.

Replying to @phnt@fluffytail.org

Fedi Developer (of crushes)

18w

@phnt @lain i wish it wasn't stigmatized...

Replying to @HatkeshiatorTND@annihilation.social

Q.U.I.N.N.

@toiletpaper@shitposter.world remote

18w

@HatkeshiatorTND @lain @phnt @s8n
> DORA 2025 found that AI adoption led to +154% PR size, +91% review time, and +9% more bugs shipped. Reviewers are drowning in noise. (Ataraxy Labs, https://github.com/ataraxy-labs/inspect)

Replying to @icedquinn@blob.cat

⚡⛥ Maður Margvalds ⛦⚡

18w

@icedquinn @HatkeshiatorTND @lain @phnt @s8n

That makes sense. I can understand not wanting to deal with a bunch of noise, but having a hard limit on no LLM generated code seems like a losing battle, even in the short term. The real criteria should be whether the code is sound, not by whom/what it was created. Also if the amount of AI slop is becoming an issue, that's a perfect use-case for leveraging LLMs as an initial review mechanism. Fight fire with fire.

Replying to @toiletpaper@shitposter.world

Q.U.I.N.N.

18w

@toiletpaper @HatkeshiatorTND @lain @phnt @s8n there are people that are vibe maintaining but also mostly the quality control side is just shuttering PRs unless they know you and the whole thing is going to reputation networks

Replying to @icedquinn@blob.cat

Phantasm

18w

@icedquinn @HatkeshiatorTND @lain @s8n
>> DORA 2025 found that AI adoption led to +154% PR size, +91% review time, and +9% more bugs shipped. Reviewers are drowning in noise.
I tend to agree with this.

Replying to @hj@shigusegubu.club

Phantasm

@hj@shigusegubu.club remote

18w

@hj @lain It's deserved. The language sucks and browsers suck even more.

Replying to @phnt@fluffytail.org

Fedi Developer (of crushes)

18w

@phnt @lain i don't think it sucks, people just have skill issue and stdlib for it sucks, but stdlib suckage isn't exclusive to the language

Replying to @phnt@fluffytail.org

lain

18w

@phnt @HatkeshiatorTND @icedquinn @s8n i do think drive-by LLM contributions of significant features can be hard for projects. In the future there probably has to be a lot more coordination between maintainers and new contributors. But this will all fizzle out. Many huge projects by now just ignore all PRs that don't from either a core member or are obviously valuable, which is fine. On the other hand, many people run their own forks of stuff with tons of added features that will never find their way back, because that's also easy now. It's more of a move to personal software instead of community software.

Replying to @lain@fediffusion.art

lain

@protos@tsundere.love remote

18w

@phnt @HatkeshiatorTND @icedquinn @s8n i mean, look at something like this

πρωτος 卐🇲🇹☪︎🇸🇦☭🇷🇺

18w

@s8n @lain and users don't value contributors, users value code

Replying to @lain@fediffusion.art

Wanderer atop the sea of clouds or whatever

18w

I read somewhere over the last few days that they built something really fragile and also Zig upstream was already working on a more general solution

>i think it does
(xkcd about guy in Nebraska maintaining critical piece of infra goes here)

Replying to @WandererUber@poa.st

lain

18w

@WandererUber those projects are often neither fast nor well written and have unfriendly maintainers

Phantasm

18w

@lil5 @hj @lain Both plain C++ Qt UIs or C++ Qt glued with QML (which is diluted JS) are much more pleasant to me than trying to decipher whatever the $js_framework does and how it works (Vue is probably the least stupid from what I've dealt with, Svelte and especially React). Gtk on the other hand is irredeemably bad compared to even webapps, it cannot be saved.

Not sorry for the shitty meme :)
frontend-development-webapps-qt-react-svelte-vue-gui.png

Replying to @lain@fediffusion.art

Wanderer atop the sea of clouds or whatever

18w

Sorry I meant to say Dubai.

I found it btw
https://ziggit.dev/t/bun-s-zig-fork-got-4x-faster-comp...

Replying to @WandererUber@poa.st

lain

18w

@WandererUber interesting.

Phantasm

18w

@lil5 @lain @hj I don't have any recommendations. My style is to dive head first into something and figure it out as I go, usually adding a new feature into something already existing or fixing a bug. Tutorials don't do it for me.

Phantasm

18w

@lil5 @hj @lain Only the "official" documentation and asking LLM models for help/advice/finding solutions when documentation with simple search engine queries isn't enough. So for Qt, this: https://doc.qt.io/qt-6/reference-overview.html

Most of the time, I don't do gui things, so plain and simple shell scripting and C/C++ maybe with some Python.

Q.U.I.N.N.