On Tue Feb 10 12:43:10 2026 +0000, Wilson Simanjuntak wrote:
the wiki does not contain an LLM policy, so I don't know if this is eligible to be merged at all. I'm not trying to be like the "everything about Gen AI is bad" guy, but Wine is a clean-room reverse engineering project, and LLMs as usual are trained in many ways including web scraping, which depending on the model and how it's trained, the LLM could actually seep in some snippets of illegally obtained leaked Windows code, and that could introduce a legal problem for clean-room reverse engineering projects like Wine . Yes, I was wondering if these changes could help with a clean-room implementation in any way. E.g. knowing the amount of code required for XDG Desktop Portal integration, logs from the patched wine and the information that certain cases require are not a straightforward translation from win32 to the D-Bus interface.
I think the integration would provide a nice UX improvement, but I couldn't implement this myself. Also, I wondered if an LLM could be used to write the specification. But then that would still need to be reviewed by someone who may look at tainted code. Or if letting an LLM like Comma v0.1, trained only on openly licensed text, implement it from scratch would be acceptable: https://arxiv.org/pdf/2506.05209 -- https://gitlab.winehq.org/wine/wine/-/merge_requests/10060#note_129336