Re: libstdc++ baseline updates for 3.4/HEAD

[Matthias Klose <doko at cs dot tu-berlin dot de>]

Benjamin Kosnik writes:
> I'm not quite sure what is up with this. I've configured exactly as you
> have, and have my same symbols, exactly.

> What I'd like you to do is start with 3.4.0 sources, no patches, default
> configure. Try to reproduce my results.

started with gcc-3.4.0 build, configured with --enable-languages=c++
--enable-__cxa_atexit --enable-clocale=gnu i486-linux

- seems with our version of glibc (patched 2.3.2), --enable-clocale=gnu
  is explicitely needed, without it the configury fails back to generic.

- the --enable-__cxa_atexit is/should be default?

the comparison <yours> <my> is down to

-FUNC:_ZNSdC1EPSt15basic_streambufIcSt11char_traitsIcEE@@GLIBCXX_3.4
-FUNC:_ZNSdC1Ev@@GLIBCXX_3.4
-FUNC:_ZNSdC2EPSt15basic_streambufIcSt11char_traitsIcEE@@GLIBCXX_3.4
-FUNC:_ZNSdC2Ev@@GLIBCXX_3.4
-FUNC:_ZNSdD0Ev@@GLIBCXX_3.4
-FUNC:_ZNSdD1Ev@@GLIBCXX_3.4
-FUNC:_ZNSdD2Ev@@GLIBCXX_3.4

are these glibc related as well?

> >+FUNC:_ZNSdC2Ev@@GLIBCXX_3.4
> 
> Let's start here. This should be weak in io-inst.o. 

$ objdump -t src/.libs/io-inst.o | grep _ZNSdC2Ev
00000000 l    d  .gnu.linkonce.t._ZNSdC2Ev      00000000 
00000000  w    F .gnu.linkonce.t._ZNSdC2Ev      00000045 _ZNSdC2Ev

> Then, once that's done, try to
> figure out what is different between 3.4.0 release and your sources. I
> don't expect divergence in libstdc++ sources, so maybe this is
> compiler-related? Or, if there are libstdc++ differences, what are they?

do you have a baseline for another architecture to check against?

	Matthias