c++ - Understanding the origin of a linker duplicate symbol error

Question

I have a c++ program that compiled previously, but after mucking with the Jamfiles, the program no longer compiled and ld emitted a duplicate symbol error. This persisted after successively reverting to the original Jamfiles, running bjam clean, removing the objects by hand, and switching from clang with the gcc front end to gcc 4.2.1 on MacOs 10.6.7.

A simplified description of the program is that there is main.cpp and four files, a.h,cpp and b.h,cpp, which are compiled into a static library which is linked to main.o. Both, main.cpp and b.cpp depend on the file containing the offending symbol, off.h, through two different intermediate files, but neither a.h nor a.cpp depend in any way on off.h.

Before you ask, I made sure that all files were wrapped in multiple definition guards (#ifndef, #define, #endif), and while I did find a file that was missing them, it did not reference off.h. More importantly, b.h does not include anything that references off.h, only the implementation, b.cpp, makes any reference to off.h. This alone had me puzzled.

To add to my confusion, I was able to remove the reference to off.h from b.cpp and, as expected, it recompiled successfully. However, when I added the reference back in, it also compiled successfully, and continued to do so after cleaning out the object files. I am still at a loss for why it was failing to compile, especially considering that the symbols should not have conflicted, I had prevented symbol duplication, and I had gotten rid of any prior/incomplete builds.

Since I was able to successfully compile my program, I doubt I'll be able to reproduce it to test out any suggestions. However, I am curious as to how this can happen, and if I run across this behavior in the future, what, if anything beyond what I've done, might I do to fix it?

Answer 1

This is often the result of defining an object in a header file, rather than merely declaring it. Consider:

h.h :

#ifndef H_H_
#define H_H_
int i;
#endif

a.cpp :

#include "h.h"

b.cpp :

#include "h.h"
int main() {}

This will produce a duplicate symbol i. The solution is to declare the object in the header file: extern int i; and to define it in exactly one of the source-code files: int i;.