Re: Multiple definition of gnulib symbols

[Simon Josefsson]

Bruno Haible <address@hidden> writes:

> Simon Josefsson wrote:
>> But yes, gc-libgcrypt and gc-gnulib are intended to be two mutually  
>> exclusive modules, since they provide the same symbols
>
> ... except for gc_hash_hmac_setkey.

Would implementing that function in gc-gnulib.c solve the problem?

>> I had no idea that defining multiple symbols in a single object  
>> file could lead to trouble.  Sigh.
>
> That alone is not a problem. But if you have two object files with
> _different_ but _overlapping_ set of symbols, you are in trouble.
> Say,  first.o  defines 'aaa', and  second.o  defines 'aaa' and 'bbb'.
> Then, when you link these:
>
>      cc   (an object or library needing aaa) \
>           (a library containing first.o) \
>           (an object or library needing bbb) \
>           (a library containing second.o)
>
> you will get an error, because the linker will
>   - see that it needs aaa, thus include first.o in the link,
>   - see that it needs bbb, thus include second.o in the link,
>   - and therefore give an error.
> The linker does *not* remove first.o from the link when it decides to include
> second.o.

Aha, that makes more sense to me.  It seems this is a gc-design problem
then.  Perhaps there should be some #define's happening, mapping the
function names to gnulib_gc_* or into libgcrypt_gc_*, and have the files
implement those symbols.

I'm wondering if that will be sufficient.  When is the namespace change
hack that you mentioned Ralf identified necessary?

Thanks,
Simon