Open Source GZDoom Community Splinters After Creator Inserts AI-Generated Code

An anonymous reader quotes a report from Ars Technica: If you've even idly checked in on the robust world of Doom fan development in recent years, you've probably encountered one of the hundreds of gameplay mods, WAD files, or entire commercial games based on GZDoom. The open source Doom port -- which can trace its lineage back to the original launch of ZDoom back in 1998 -- adds modern graphics rendering, quality-of-life additions, and incredibly deep modding features to the original Doom source code that John Carmack released in 1997. Now, though, the community behind GZDoom is publicly fracturing, with a large contingent of developers uniting behind a new fork called UZDoom. The move is in apparent protest of the leadership of GZDoom creator and maintainer Cristoph Oelckers (aka Graf Zahl), who recently admitted to inserting untested AI-generated code into the GZDoom codebase.

"Due to some disagreements -- some recent; some tolerated for close to 2 decades -- with how collaboration should work, we've decided that the best course of action was to fork the project," developer Nash Muhandes wrote on the DoomWorld forums Wednesday. "I don't want to see the GZDoom legacy die, as do most all of us, hence why I think the best thing to do is to continue development through a fork, while introducing a different development model that highly favors transparent collaboration between multiple people." [...] Zahl defended the use of AI-generated snippets for "boilerplate code" that isn't key to underlying game features. "I surely have my reservations about using AI for project specific code," he wrote, "but this here is just superficial checks of system configuration settings that can be found on various websites -- just with 10x the effort required."

But others in the community were adamant that there's no place for AI tools in the workflow of an open source project like this. "If using code slop generated from ChatGPT or any other GenAI/AI chatbots is the future of this project, I'm sorry to say but I'm out," GitHub user Cacodemon345 wrote, summarizing the feelings of many other developers. In a GitHub bug report posted Tuesday, user the-phinet laid out the disagreements over AI-generated code alongside other alleged issues with Zahl's top-down approach to pushing out GZDoom updates.

Re:

[taustin]

Like it or hate it, approve of disapprove, if it (ever) works, it will be used.

Standing in the way of progress is a sure fire way to be run over by it.

Re:

[thecombatwombat]

Follow the links and you'll see that wasn't the issue. It was that it wasn't clear the code was GPL compatible, untested, and then the maintainer tried to remove all history of it by force pushing to git.

It's not like being mad some coder used VS Code instead of vim, "if someone provides free work for you, they decide what tools they use" just really isn't the story.

Sooo many AI code issues are really just the same old issues of cut and paste code but we discuss them like they're new because AI. Those thing

Re:hmm

[nightflameauto]

Calling AI code "slop" is disingenuous. Yes, if the code is shit, it is slop. But modern AI can generate clean functioning code in large segments. If it runs, it runs.

If it runs, it runs? You must be in management.

I have yet to see an AI code a "large segment" without flaw. Small snippets, sometimes, but even there it's a 50/50 shot. The larger the ask, the easier it is to get back complete garbage that may run, but won't actually do what was in the spec / prompt.

Re:hmm

[neilo_1701D]

I have yet to see an AI code a "large segment" without flaw. Small snippets, sometimes, but even there it's a 50/50 shot. The larger the ask, the easier it is to get back complete garbage that may run, but won't actually do what was in the spec / prompt.

I do Microsoft Dynamics 365 ERP systems. As a test, I've been asking Copilot to generate code for me. I am yet to see a piece of code that runs correctly - and I'm talking about something that the AI system should know, since it has feasted on Microsoft documentation. Even the smallest piece of code - generating an event handler for a table insert event - is wrong. It might be 90% correct, but that 10% kills it.

Re: hmm

[liqu1d]

This is the bit that concerns me. We see it making clear mistakes in areas we have knowledge. If we canâ(TM)t trust it then how on earth can we trust it in areas we are not so knowledgable. Too many people are quite happy to and itâ(TM)s scary.

Re: hmm

[liqu1d]

Perhaps the slashdot site developers can use AI to support Unicode because they dont seem to do it on their ownâ¦

Re:

[Strangelover]

...these "site developers". Are they in the room with us now? O.o

Re: hmm

[liqu1d]

If they are they should be ashamed

Re:

[Pascoea]

I do Microsoft Dynamics 365 ERP systems.

Nice! I'm mostly in the 2012 world, but we've been (slowly) migrating to 365. I'm not a dev, but have dabbled.

I wonder if LLM's ability (or lack thereof) to generate proper code for AX is just how close X++ is to other languages, and how relatively niche it is. If I'm playing around with PHP, JS, or futzing in the Arduino world, I've had pretty good luck getting ChatGPT or Copilot spit out small snippets of code for a very specific task.

Re:

[fafalone]

My favorite is when AI tells me thing x is impossible in language y when code to do x has been on a major site for years if not over a decade. When you try to push through it generates actual gibberish, like not a single line remotely related to the task.

Modern AI?

[RitchCraft]

"modern AI" - what the hell is modern AI?

Re:

[thegarbz]

If it runs, it runs.

I take it you were the coder behind the Apple calculator app? https://developers.slashdot.or... [slashdot.org]

I see a similar divide at work

[ffkom]

The divide between rapidly brain-atrophied LLM abusers and old men yelling at clouds (because they want high-quality, hand-optimized software development back) seems all too familiar to me, and I don't see a resolution anytime soon. One might argue it should be possible to use LLMs just as yet another tool to help with some things, without being abused for other things. But it seems between vibe-coders and experienced programmers there is no significant middle ground to speak of.

Re:

[Local ID10T]

IDEs are the middle ground. A properly configured IDE does a ton of automation already. An IDE that integrates a purpose built LLM/AI is just "a smarter IDE".

The problem isnnt AI

[MerlynEmrys67]

recently admitted to inserting untested AI-generated code into the GZDoom codebase.

Wait we are putting untested code into the codebase? I dont care where untested code is coming from - it is never Ok. If AI code passes tests, then I don't care. If Linus Torvalds develops code himself and puts untested code into Linux... FORK immediately

Sounds like "work around me" leadership

[Gravis Zero]

introducing a different development model that highly favors transparent collaboration between multiple people.

Given that this is a change, I've seen this kind of bullshit "leadership". It basically amounts to, "I'm doing what I'm doing and fuck you if you get in the way of that." This means their own shitty view of how to do things (like insisting on git merge instead of rebasing) and are dismissive of rational arguments to change in favor of, "well this is how I do it." Did you have a large patch that fixes stuff that's ready to be merged in? Well, the leader had a bit of inspiration and merged in something he's b

Is this about code or politics?

[Waccoon]

I dunno. There may be more going on here than reported. I find it a bit odd that a bunch of people immediately decide to fork the project rather than take the time to cool off and work out their differences.

I've done my share of working on FOSS projects and there's often even more infighting and politics than in commercial products.

Kinda sad, as I've been a huge fan of ZDoom and GZDoom for well over a decade.