I don't think it is a bug against LLVM: there are basically two way of doing this: either you extend LLVM with more intrinsic or concept that carry your high-le

I don't think it is a bug against LLVM: there are basically two way of doing this: either you extend LLVM with more intrinsic or concept that carry your high-level semantics to the LLVM level (this is what people from Azul are doing [0]) or build an other intermediate representation like Swift and Rust. For the swift case, you can see the talk at last year LLVM Dev Meeting [1]. There are advantages and drawback to each approaches. [0] : "LLVM for a managed language: what we've learned" [1] : "Swift's High-Level IR: A Case Study of Complementing LLVM IR with Language-Specific Optimization" Both have slides and video here: http://llvm.org/devmtg/2015-10/ http://llvm.org/devmtg/2015-10/