For both compilers, once LLVM bitcode is created, a tool called llc can be used to compile the bitcode to a binary.

C family essentially has two outputs: a binary, and a header that describes the binary (Kotlin/Native is the same). The header tells a consuming library information about what symbols are available, the structure of data, and some ABI information.

Actually, I am still missing some bits… You say that cinterop extracts the symbols from Objective-C code and creates Kotlin code that can call that symbol. But actually we start from Kotlin code, not from Objective-C code. We actually want to translate Kotlin code into an Objective-C framework.

Interesting extension of this: the Kotlin compiler currently requires all Kotlin files to be sent to the compiler at once, unlike clang, which can compile an individual file. If you wanted to compile Kotlin one file at a time for some definition of ‘fun’, you could technically compile the least dependent Kotlin file, then use cinterop to convert the generated header to Kotlin serialized IR, which you add as a dependency to another Kotlin file, which you compile, and continue this until the whole project is compiled. Not suggesting this is a good idea, just interesting.