[Back to Index]
[04:55] <-- TAS_2012v left irc: Remote host closed the connection
[07:03] --> i3 joined #scummvm.
[07:05] --> ldevulder joined #scummvm.
[07:06] <-- i3 left irc: Read error: Connection reset by peer
[07:10] <grogbot> [discord] <gu3> https://x.com/antonio__capone/status/2122114645678784571?a=eEGO20uyixyBB09gLho&s=46
[08:05] <-- ldevulder left irc: Quit: Leaving
[08:05] --> ldevulder joined #scummvm.
[08:39] --> borosky joined #scummvm.
[09:38] <-- Unseen2 left irc: Read error: Connection reset by peer
[09:38] --> Unseen2 joined #scummvm.
[09:53] <-- Unseen2 left irc: Ping timeout: 241 seconds
[11:01] --> Unseen2 joined #scummvm.
[11:23] <-- Unseen2 left irc: Read error: Connection reset by peer
[11:29] --> Unseen2 joined #scummvm.
[13:31] --> ccawley2011 joined #scummvm.
[13:31] #scummvm: mode change '+o ccawley2011' by ChanServ!ChanServ@services.libera.chat
[18:36] <-- borosky left irc: Ping timeout: 249 seconds
[19:29] --> TAS_2012v joined #scummvm.
[20:08] <-- ldevulder left irc: Ping timeout: 259 seconds
[20:24] <grogbot> [discord] <vromantp> Hello. Over the past few weeks, I've been working on adding support for The X-Files Game (https://en.wikipedia.org/wiki/The_X-Files_Game) to a fork of ScummVM. I believe that the engine is in a playable state at the moment, and the game can be completed. I was going to ask what the procedure is for onboarding a new engine, but one of the rules that are mentioned when joining here is that LLM code is particularly frowned-upon.
[20:24] <grogbot> [discord] <vromantp> I want to be completely open and say that I've been guiding Claude during this process, setting up a testing harness which captures all input from the original game to record them as 'reference runs', which could then be compared against the engine simulation.
[20:24] <grogbot> [discord] <vromantp> Is the statement regarding LLM-output still true? And if so, does this basically mean that there is no chance (a version of) this engine could get merged?
[20:56] --> borosky joined #scummvm.
[21:10] <grogbot> [discord] <dreammaster> The trouble with LLM generated code is that of maintenance. Once an engine is accepted into ScummVM, any bugs that come up in the future obviously need to be fixed. Say there's a memory leak. You could throw the bug report into Claude, and it might suggest a place where it could be freed. But would that really be the correct place? Or would it cause indeterminate issues with other code using it after free, that Claude didn't pick u
[21:10] <grogbot> [discord] <dreammaster> Using LLM to disassemble and figure a game is fine; even I've been doing it a lot lately. But in the end, you as a developer have to commit to actually understanding it, and be willing to join the team to maintain it for a reasonable period of time.
[21:12] <grogbot> [discord] <dreammaster> Plus, even the best AI agent isn't guaranteed to produce clean code. So even before acceptance, an engine would need proper cleanups, formatting, and work done on it by you. That's the step where a developer could learn the contents of the engine as they're validating what the AI produced.
[23:24] <-- borosky left irc: Ping timeout: 266 seconds