Fedor Indutny

“Virtual File System” PR for Node.js comes with a disclaimer:

"Disclaimer: I've used a significant amount of Claude Code tokens to create this PR. I've reviewed all changes myself.”

LoC added: +18945

Thankfully it hasn't been merged yet. The changes are all over the place and this makes me uncomfortable about running Node.js long term.

https://github.com/nodejs/node/pull/61478

Virtual File System for Node.js by mcollina · Pull Request #61478 · nodejs/node

A first-class virtual file system module (node:vfs) with a provider-based architecture that integrates with Node.js's fs module and module loader. Key Features Provider Architecture - Extensi...

GitHub
Mar 17 at 3:44pmWeb
Show threadFedor IndutnyMar 17
My perhaps overly irritated response to the PR: https://github.com/nodejs/node/pull/61478#issuecomment-4075998105
Show threadFedor IndutnyMar 18
And... this ended with a petition that I want to put before TSC vote in two weeks https://github.com/indutny/no-slop-in-nodejs-core
GitHub - indutny/no-slop-in-nodejs-core: A petition to disallow acceptance of LLM assisted Pull Requests in Node.js core

A petition to disallow acceptance of LLM assisted Pull Requests in Node.js core - indutny/no-slop-in-nodejs-core

GitHub
Show threadFedor IndutnyMar 20

Update: I'm so proud of y'all!

I know AI companies and many employers aren’t making it easy to step out and say "no" to it, but you went in and signed the petition. I respect that a lot!

We still have time to get more names on the list and make stronger case for TSC committee, but seeing all these names under it is already very refreshing in the days where every news cycle either talks about AI changing the world or about job cuts by major companies (in addition to wars, obviously).

Show threadFedor IndutnyMar 20

Some people asked me if I could offer a way to sign the petition without having a github account, so I created a mirror (?) on Change.org:

https://www.change.org/p/no-ai-code-in-node-js-core

Github is still a preferred way to sign this, but if for some reason this doesn't work for you - at least now you have an alternative way to make your voice heard!

Sign the Petition

No AI code in Node​.​js Core

Change.org
Show threadClaudiusMar 18
@indutny is anyone ok to sign? I have never contributed directly to NodeJS. I'm just a user.
Show threadFedor IndutnyMar 18
@claudius absolutely! As long as you care about it - go for it!
Show threadBryan EnglishMar 18

@indutny I'm all for this. And will sign momentarily.

That said, what teeth does this have? Will you be able to present this in the meeting where the vote is held?

Show threadFedor IndutnyMar 18
@bengl that's my plan! Join the meeting and present this to TSC
Show threadyassernascMar 17
@indutny for sure there is the whole license thing, but what triggers me the most is that a complete understanding of the code produced is becoming a secondary thing.
Show threadFedor IndutnyMar 17

@yasser yes, absolutely!

I’m trying to find a common ground on which we could base the decision not to accept such change, and which is agreeable to engineers leaning on LLMs heavily. DCO seemed like a good starting point since in my reading it clearly is incompatible with the LLM assisted contributions.

Show threadAndrew KelleyMar 18

@indutny I left my comment on the HN article just now:

> I used Node.js extensively for about a decade, both professionally and as a hobby. I don't buy the problem statement.

> These arguments don't even make sense, they look LLM generated. I can't even formulate a disagreement against this nonsense.

Show threadFedor IndutnyMar 18

@andrewrk thank you!

There’s now going to be a TSC vote on the matter in about two weeks. I'll hope to present these points succinctly before the committee!

Show threadAndrew KelleyMar 18
@indutny good luck. is signal built using node.js?
Show threadSuper OwlMar 18
@andrewrk @indutny Just what Signal needs: a virtual filesystem. It's OK everyone, the coder audited it themself, so you know it'll be secure. [thick sarcasm]
Show threadFedor IndutnyMar 18

@andrewrk tbh, I feel exactly the same about the proposed change itself. It is hard to buy whatever is being sold...

Thankfully the outrage that I felt happened way before I could even take a look at the code. Asking people to review 19k LoC LLM-generated PR is beyond what I'd consider basic politeness. The disproportionality of the time spent implementing this and the time needed to carefully review and catch slop mistakes is egregious.

Show threadAndrew KelleyMar 18
@indutny there has been so much thought and care poured into node.js over the years... to dilute it feels like spray painting some shitty graffiti on top of a beautiful mural
Show threadFedor IndutnyMar 18
@andrewrk I think I'm gonna rephrase and use this if you don't mind!
Show threadAndrew KelleyMar 18
@indutny go for it, I don't even need attribution
Show threadKyle DavisMar 19

@indutny thanks for challenging this. I agree that the DCO seems largely incompatible with LLM usage it actually read it and think.

Most people haven’t read it or thought about it, just sign off your commit that Claude made and move on.

The LF (sadly) seems to be rather unbothered by it.

Show threadClaudiusMar 17
@indutny ooooof. Not a fan 😞
Show threadevinMar 18
@indutny I was already uncomfortable about running Node.js...
Show threadJan Lehnardt Mar 18
@indutny jesus fuck how can anyone think this is a PR that should be opened?
Show threadRadomír ŽemličkaMar 18
@indutny Yikes! I hope this won't be merged.
Show threadOwl EyesMar 18
@indutny all the grumbling I do about Node.js isn't for nothing. It's like a dolt-magnet.
Show threadLuke HarbyMar 18
@indutny Experiment with LLM's on your own little crappy project, or contribute to Node, but don't mix the two.
Show threadLuke HarbyMar 18
@indutny Also the sheer size of the PR! Bitesize updates please.
Show threadcthos 🐱Mar 18
@indutny what are the odds that, given the first several reviews are “can we use path.join” on a bunch of lines that are using string concatination for constructing paths, they have introduced several directory traversal attacks in this thing?
Show threadNiels AbildgaardMar 19

@cthos @indutny I was thinking about the same thing, also vis-a-vis other replies here about how large and one-shot the PR is.

Claude has gotten really good at producing code that *looks* right (that is, hides errors well), and trying to find them in 19k lines is just...

Show threadCap YbarraMar 18
@indutny jfc 19k lines. are there any real programmers working on this project if this is even considered. what fuckin ninny is ok merging 19k lines
Show threadFedor IndutnyMar 18
@cap_ybarra I know, right! Let's sign the petition to stop this https://github.com/indutny/no-slop-in-nodejs-core
Show threadCap YbarraMar 18
@indutny considering it. my name won't mean anything, i am no one important or illustrious, i retired my github after 19 years because i was tired of these slopmongers stealing from me and giving nothing, and it might puncture my pseudoanonymity
Show threadCap YbarraMar 18
@indutny my name would have (Just Some Fuckin Guy, You Don't Know Him) after it
Show threadErwinMar 18
@indutny
Translation: I spend a big pile of my own cash on this (BTW, thank you), so you should let me merge.
Show threadheapwolfMar 18
@indutny i saw this PR and was surprised and disappointed to see who made it.
Show threadFedor IndutnyMar 18
@heapwolf please sign the petition! https://github.com/indutny/no-slop-in-nodejs-core
GitHub - indutny/no-slop-in-nodejs-core: A petition to disallow acceptance of LLM assisted Pull Requests in Node.js core

A petition to disallow acceptance of LLM assisted Pull Requests in Node.js core - indutny/no-slop-in-nodejs-core

GitHub