duplicate _r_debug symbol - binutils question or glibc question?

[Tom Kacvinsky]

On Thu, Jan 28, 2021 at 4:21 PM Tom Kacvinsky <tkacvins@gmail.com> wrote:

> Hi all,
>
> I am seeing this:
>
> -bash-4.4$ readelf -a /home/Users/tjk/vector/build/vc20/Linux/vc//clicast
> | grep _r_debug
> 000000e34320  0d0c00000005 R_X86_64_COPY     0000000000e34320
> _r_debug@GLIBC_2.2.5 + 0
>   3340: 0000000000e34320    40 OBJECT  GLOBAL DEFAULT   29
> _r_debug@GLIBC_2.2.5 (10)
>  18463: 0000000000e34320    40 OBJECT  GLOBAL DEFAULT   29 _r_debug
>
> This is for our product (which is quite complicated).  If I make a simple
> program, I
> don't see this symbol/variable (unless I run in the debugger):
>
> (gdb) info variables _r_debug
> All variables matching regular expression "_r_debug":
>
> File dl-debug.c:
> struct r_debug _r_debug;
>
> If I run our application under gdb, I see this
>
> (gdb) info variables _r_debug
> All variables matching regular expression "_r_debug":
>
> File dl-debug.c:
> struct r_debug _r_debug;
>
> Non-debugging symbols:
> 0x0000000000e34320  _r_debug
>
> So, my question is why does the symbol show up three times in readelf for
> our program,
> only once for a simple program (in gdb) and twice for our product (again,
> under gdb).
> I am wondering how much of this has to do with one of the _r_debug symbols
> having type
> R_X86_64_COPY?  Not sure if this is a binutils question or a glibc
> question.  Any tips
> as to how to find where the duplicate symbols are coming from would be
> greatly appreciated.
> I have tried with binutils 2.34 and 2.35.1, and am running with glibc 2.17
> on CentOS 7 and
> this also happens on CentOS 8 according to my co-worker.
>
>
I figured it out.  we were linking libgnat.a, which does some stuffwith
_r_debug - which
version of GNAT introduced this, I have yet to determine.  But if I link in
libgnat shared
the problem goes away.

Tom
Tom