Am Donnerstag, dem 19.03.2026 um 09:48 +0000 schrieb Steffen Möller via Emc-developers: > * LLMs: Code hardening.
of course it is good practice to write correct code that doesn't suffer from obscure race conditions in corner cases or buffer overflows in debug messages when running multi-gb g-code files. you shouldn't do general computing stuff on your linuxcnc controller, or directly connect it to the internet. part of linuxcnc runs setuid root, and can load arbitrary code into the realtime domain that is executed as root into the kernel, so in principle, anything that has access to that setuid binary can get "root" trivially. maybe LLM assistance can help further the "non-setuid" branch but I doubt it. > * Help with the installation of LinuxCNC. I already had the pleasure to experience this first hand, a local guy is using some LLM to help generate HAL files, and while usually the result is not completely wrong, it didn't yield working configurations either because of subtle details, like including third-party drivers for some VFD that doesn't exist / doesn't exit any more, mixing up signals etc. digging through such a mess where it is not clear where some hal configuration is coming from and what the intention was to do it like that is no fun. avoid. > * Support for writing error reports when installation fails? I don't think it is useful there either because the typical user is supposed to run stable version on a supported OS and per definition there are no "installation fails" there. LLM won't help if buildbot doesn't build or artifacts from github ci are faulty. getting LLM generated failure reports that contain even slight inaccuracies or "slop" will become a nuisance really quickly, I wouldn't want to dissect such reports. -- Robert Schöftner <[email protected]> _______________________________________________ Emc-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/emc-developers
