abi

structing C++ functions. What's _actually_ wrong with that, in terms of robustness? The strings are generated not based on arbitrary user input, but based on Clang QualTypes and Decls. i.e. you need valid Clang values to actually get there anyway. Given that the ABI situation is an absolute mess, and that jank is already using Clang's JIT C++ compiler, I think this is a very viable solution. However, in terms of robustness, I go back to Clang's error handling, lack of grace, and poor tooling for use cases like this. Based on my experience, _that_ is what will cause robustness issues. Please don't take my response as unreceptive or defensive. I really do appreciate the discussion and if I'm saying something wrong, or if you want to explain further, please do. For alternatives, you linked to https://github.com/Mr-Anyone/abi which is 3 months old and has 0 stars (and so I assume 0 users and 0 years of battle testing). You also linked to https://discourse.llvm.org/t/llvm-introduce-an-abi-lowering-... which I agree would be great, _if/when it becomes available_. So, out of all of the options, I'll ask clearly and sincerely: is there really a _better_ option which exists today? CppInterOp is an implementation detail of jank. If we can replace C++ string generation with more IR generation and a portable ABI mechanism, _and_ if Clang can provide the sufficient libraries to make it so that I don't need to rely on C++ strings to be certain that my template specializations get the correct instantiation, I am definitely open to replacing CppInterOp. From all I've seen, we're not there yet.

y all this with maximum love in my heart for a language that would have first class cpp interop - i would immediately become jank's biggest proponent/user if its cpp interop were robust. EDIT: for people wanting/needing receipts, you can skim through https://github.com/compiler-research/CppInterOp/blob/main/li...