LLVM is very big, takes ages to compile and compiles code relatively slowly. Also because it was designed to compile C there are occasionally C-isms that make it suboptimal for other languages with different properties. (Although a hobby project is never going to beat its performance even accounting for that.)

That’s what mlir is for

Probably the compilation time is better than it would be with LLVM. On the other hand I doubt that codegen and therefore performance is on par with LLVM.

Definitely a weird thing to advertise.

This is an advertisement, and I usually don't put “no LLVM” in the title or use it as a promotional feature.

I posted this project on HackNews a few times, but it quickly sank to the bottom. Maybe “no LLVM” might pique some people's interest in the project, so I added it. Actually, I posted this link a few days ago and had already given up on it, but unexpectedly, it suddenly appeared on the HackNews homepage.

I now need some attention, which will give the Nature project more capital.

I guess AI and LLMs became an anti-clickbait (clickdetterent?).

So maybe a way to gather people that are tired of Yet Another AI Tool or vibe coding experiment that will be abandoned by Tuesday?

LLVM != LLM

https://en.wikipedia.org/wiki/LLVM

Misread. Acronym fever.

He said LLVM, not LLM.

Maybe it's the same reason as Go, to be able to guarantee that codegen runs really fast.

What does it have to do with using LLVM?

Faster compilation speeds, more customized error handling, and convenient cross-compilation. I believe this is a promising direction for the future development of programming languages. Zig has already made significant progress in this area.