On Sunday, October 18th, 2026 at 4:30 AM, Carsten Haitzler <[email protected]> wrote: > On Sun, 18 Oct 2026 05:40:55 -0300 Vinícius dos Santos Oliveira > <[email protected]> said: >
> > Em sáb., 17 de out. de 2026 às 20:15, Carsten Haitzler > > <[email protected]> escreveu: > > > It's a prototype. It could have better code. But then again I cooked it up > > > basically inside of a day with codex (damn LLMs make this so > > > productive/fast). > > > > Do you think LLMs could help to quickly port the still remaining Eo > > interfaces that are blocking the release of EFL 2.0? > > probably The answer is actually it can and I am almost done working on it. The current branch is about +50k lines of code. It include getting the C++, C# and a Go binding working with modern version of their respective toolchain and feel native in that language. I am waiting for the release cut to start spitting it and making it proper PR upstream, but the main issue we have now is that this tool enable us to make a lot of code with very few people available to review the work done. Especially in the case of Eo interface, this is all about API and taste, not really something a CI/tests suite catch (except for the part that prove that a C#, C++ and Go program work each the same). So that is still the hard part and what is taking time. Example that is easier. Adding Android support. You give enough guidance to the LLM plus an Android device and of it goes. Efl run on Android. Still a lot of work, but the definition of done is elementary_test run on Android and provide screenshot with interaction of it. A llm can deal with that and do a lot autonomously especially when we already have a good abstraction for supporting multiple OS. I have a demo working and I am waiting for the release cut to make a clean PR for it. Things that are even easier for us in efl is optimization. We do have tools like expedite for benchmark, you can tell a llm to generate AVX2 optimization and just let it do it. That's in our master and coming when we do this release. Same was applied to terminology and got it faster on low end arm device (I have a pinebook pro that really need things to be optimized). Anyway, tool like claude and codex work. If you can give it a clear goal with a definition of done and what is a working software, then it can help. Cedric _______________________________________________ enlightenment-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/enlightenment-users
