That's a compiler bug. Can you extract a test-case from your source?
Glad to. Find attached the GNUStep makefile and source files.
The failure doesn't reproduce for me on Darwin with top-of-tree clang. (I did mock up a quick Divide_Result.h, though, so the file would build.)
Which version of clang are you using? If a previous release, it's possible this has been fixed since then.
If it does reproduce for you with top-of-tree clang, then there's something different on Windows that's exposing the errant behavior. In which case, we can try reproducing it via a bitcode file. To get that, run the command line that builds the calc.o file and add "-emit-llvm -o calc.bc" to it. That'll output a self-contained intermediate file we can use to try to reproduce the behavior elsewhere.