alumina VS llvm-cbe

You can go surprisingly far with C, though LLVM is probably a better long-term option for a serious compiler, because it's a tool made for the job (unless you target exotic and/or embedded platforms that don't have LLVM support - but that's fairly unlikely).

C is very easy to get started with if you don't already know LLVM. The downside is that once your compiler is reasonably complete, you may spend quite a bit of time working around quirks of C (e.g. int promotion is very annoying when you already have full type information, so your compiler either has to understand C semantics fairly well or defensively cast every subexpression).

I have a C backend in my compiler (https://github.com/alumina-lang/alumina) and it works really well, though the generated C is really ugly and assembly-like. With #line directives, you can also get source-level debugging (gdb/lldb) that just works out of the box.

There are a few goodies that LLVM gives you that you don't get with C, like coverage (https://clang.llvm.org/docs/SourceBasedCodeCoverage.html). It works when done through clang, but easily be made to track the original sources.

Totally agreed about FFI. I wanted to make it easy to interop with C code and write expressive bindings.

Check for example the language bindings to LLVM's C API (fairly low level) and Tree-Sitter which is used internally (a bit higher level bindings)

https://github.com/tibordp/alumina/tree/master/libraries/llv...

https://github.com/tibordp/alumina/blob/master/libraries/tre...

However, I am not super happy with the current structure of the compiler and for the self-hosted one I'd like to fix those mistakes. For example, mono.rs is a giant module that does everything from mixin expansion, lowering from AST to IR, type checking and monomorphization. There are some bugs in Alumina that are quite hard to fix without a big refactoring (e.g. if a nested function binds generic parameters of the parent) and I'd like to get it right the next time around.

So how does the LLVM C backend work then?

https://github.com/JuliaHubOSS/llvm-cbe

One alternative worth mentioning, though, would be the LLVM C Backend maintained by the Julia community.

If you drop "easily" and "human" (/s) from your requirements list, then the C backend for LLVM might work. Then you can choose any programming language you want that has LLVM 10-compatible frontend.

If you really must have something that compiles in C (e.g. for a platform where you only have a C compiler) there's an LLVM backend that outputs C code: https://github.com/JuliaComputingOSS/llvm-cbe

You can convert llvm bitcode to C and then use C compiler, there is such project https://github.com/JuliaComputingOSS/llvm-cbe .

I'm really surprised that the LLVM C backends have continually been resurrected then abandoned over the years. It's a good solution to this sort of thing and would enable a lot of cool stuff like Rust to weird embedded platforms. The most recent one is the Julia backend: https://github.com/JuliaComputingOSS/llvm-cbe

Check this project out: https://github.com/JuliaComputingOSS/llvm-cbe.

Well IMO it can definitely be rewritten in Julia, and to an easier degree than python since Julia allows hooking into the compiler pipeline at many areas of the stack. It's lispy an built from the ground up for codegen, with libraries like (https://github.com/JuliaSymbolics/Metatheory.jl) that provide high level pattern matching with e-graphs. The question is whether it's worth your time to learn Julia to do so.

You could also do it at the LLVM level: https://github.com/JuliaComputingOSS/llvm-cbe

For interesting takes on that, you can see https://github.com/JuliaLinearAlgebra/Octavian.jl which relies on loopvectorization.jl to do transforms on Julia AST beyond what LLVM does. Because of that, Octavian.jl beats openblas on many linalg benchmarks

You can try your luck with the "resurrected" C backend: https://github.com/JuliaComputingOSS/llvm-cbe

I don't understand why I see so many requests for LLVM-based languages to change around their backend or IR, that seems to be a huge amount of work for comparatively little benefit. The correct thing to do there is to just add support for those to LLVM.