Re: c++/7708: g++ -O inserts definition of std::string::replace in library module

[Bradford Castalia <Castalia at azstarnet dot com>]

I've just installed the gcc32 package from the FreePSG ports site:

idaeim:castalia>> g++ --version
g++ (GCC) 3.2.1

After running the test cases, as well as the full recompile of my project code,
I can confirm that the -O switch no longer generates library symbol definitions
for the std::string::replace method.

Using a Test.h file:

#include        <string>

class Test
{
public:
Test (const std::string& message = "");
void message (const std::string& message);

private:
std::string Message;
};

And a Test.cpp file:

#include        "Test.h"

Test::Test (const std::string& message) : Message (message)
{
if (! Message.empty () && Message[Message.size () - 1] == '\n')
        Message.erase (Message.size () - 1);
}

void Test::message (const std::string& new_message)
{
Message.replace (std::string::size_type(0), Message.length (), new_message);
if (! Message.empty () && Message[Message.size () - 1] == '\n')
        Message.erase (Message.size () - 1);
}

Building an object module compiled with g++ 3.1.1 20020617 (prerelease) [FreeBSD]
(g++31 -O -c -o opt-Test.o Test.cpp) and then examing the object symbols reveals,
in addition to the weak symbols, a text symbol for "replace":

idaeim:castalia>> nm -g -o -C opt-Test.o | grep replace
opt-Test.o:00000000 W std::string& std::string::_M_replace<char const*>(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, char const*, char const*, std::input_iterator_tag)
opt-Test.o:00000000 W std::string& std::string::_M_replace_safe<__gnu_cxx::__normal_iterator<char*, std::string> >(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>)
opt-Test.o:00000000 W std::string& std::string::_M_replace_safe<char const*>(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, char const*, char const*)
opt-Test.o:000001f0 T std::string::replace(unsigned, unsigned, char const*, unsigned)

Building an object module compiled with g++ 3.2.1 and then examing the symbols
reveals that the text symbol is now an undefined reference to "replace" as expected:

idaeim:castalia>> nm -g -o -C new-opt-Test.o | grep replace
new-opt-Test.o:00000000 W std::string& std::string::_M_replace<char const*>(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, char const*, char const*, std::input_iterator_tag)
new-opt-Test.o:00000000 W std::string& std::string::_M_replace_safe<__gnu_cxx::__normal_iterator<char*, std::string> >(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>)
new-opt-Test.o:00000000 W std::string& std::string::_M_replace_safe<char const*>(__gnu_cxx::__normal_iterator<char*, std::string>, __gnu_cxx::__normal_iterator<char*, std::string>, char const*, char const*)
new-opt-Test.o:         U std::string::replace(unsigned, unsigned, char const*, unsigned)

The problem of problem of multiple definitions at link time has now been fixed.

Thnx!

-- 

Bradford Castalia
Systems Analyst


bangerth@dealii.org wrote:
> 
> Synopsis: g++ -O inserts definition of std::string::replace in library module
> 
> State-Changed-From-To: open->feedback
> State-Changed-By: bangerth
> State-Changed-When: Thu Dec 19 18:19:52 2002
> State-Changed-Why:
>     I tried your program with gcc3.2, but on x86 linux. I can
>     confirm that the mentioned symbols are present in the
>     object file. However, they are weak:
>     00000000 W std::string& std::string::_M_replace<....
> 
>     Weak symbols means: they can be in multiple object files,
>     and one of these copies is picked at link time.
> 
>     Could you possibly check whether the problem still exists
>     on your platform with gcc3.2.1?
> 
>     Thanks
>       Wolfgang
> 
> http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=7708

Wolfgang Bangerth wrote:
> 
> On Sun, 22 Dec 2002, Bradford Castalia wrote:
> 
> > When I get gcc3.2.1 (or greater) installed on the FreeBSD workstation
> > (hasn't happened yet, sorry) I will definately run the test files
> > again. You will probably get a faster response from one of the FreeBSD
> > developers (http://www.FreeBSD.org).
> 
> Please let us know about your results. Maybe someone of the FeeBSD
> maintainers could also try to look into this?
> 
> Regards
>   Wolfgang
> 
> -------------------------------------------------------------------------
> Wolfgang Bangerth              email:           bangerth@ticam.utexas.edu
>                                www: http://www.ticam.utexas.edu/~bangerth