Redeclaration with incompatible signature may use unexpected signature

[TheDan64]

Newlib does a little bit more complicated version of this:

#include <limits.h>

double __exp10(unsigned x);

int main(void) {
  __exp10(1U);

  return 0;
}

Where double __exp10(unsigned x) is defined in another local file.

When translating this code you get a warning from C:

/home/dkolsoi/repos/c2rust/test.c:3:8: warning: incompatible redeclaration of library function
      '__exp10' [-Wincompatible-library-redeclaration]
double __exp10(unsigned x);
       ^
/home/dkolsoi/repos/c2rust/test.c:3:8: note: '__exp10' is a builtin with type 'double (double)'

But the translated code:

extern "C" {
    #[no_mangle]
    fn __exp10(_: libc::c_double) -> libc::c_double;
}
unsafe fn main_0() -> libc::c_int { __exp10(1u32); return 0i32; }

Will fail to compile because it uses the builtin signature which takes a double and not an int.

Interestingly (but maybe not totally surprising), passing the -ffreestanding flag to clang via the transpiler will bypass this issue and use the correct signature:

extern "C" {
    #[no_mangle]
    fn __exp10(x: libc::c_uint) -> libc::c_double;
}
#[export_name = "main"]
pub unsafe extern "C" fn main_0() -> libc::c_int {
    __exp10(1u32);
    return 0i32;
}

(also interesting, but not problematic, note that it's using x not _ in the param sig)

This can be seen in newlib when configuring with --disable-newlib-fno-builtin