Crashes on closing applications

[Raphael Kubo da Costa]

On Sunday 11 July 2010 11:25:39 Rolf Eike Beer wrote:
> Thomas Lübking wrote:
> > Am Sunday 11 July 2010 schrieb Rolf Eike Beer:
> > > I just wonder why this is libstdc++-v3, openSuSE 11.2 uses gcc 4.4 by
> > > default?
> > 
> > I don't think the libstdc++ version ever changed since gcc3 (and indeed,
> > it's -v3 on gcc 4.5 as well)
> > 
> > The whole thing looks quite like a memory ("0x6" ...) corruption, but
> > testing
> > 
> > gcc -O[n] std_string_seg.cpp -lstdc++ -o std_string_seg | n = 0-3
> > 
> > on the -plain std::string, no KDE- attachment doesn't fail at all
> > (i however don't know how -optimized- my libstdc++ was compiled)
> > 
> > So this is either in the particular OpenSuSE libstdc++ or an overflow in
> > some KDE lib.
> > 
> > I also attached a binary, compiled and linked on arch, 32bit x86, gcc4.5
> > prerelease, -O2, lisbstdc++.so.6.0.14 - maybe test it with gdb
> 
> It's not that trivial, otherwise I think it would have been long solved.
> For example if you start dolphin and immediately close it afterwards the
> crash does not happen. If you do some work before closing it it will
> crash.
> 
> I also suspect not the string object itself be the problem, but the memory
> within that (i.e. the string data). We see those crashes with all string
> objects being on the stack as I suspect the report from that google search
> does too.
> 
> And once again: "delete 0" is fine but must return immediately. In the
> backtrace it doesn't but tries to dereference something. Although I find
> that offset 0x6 suspicious, I would have expected a multiple of 4 for any
> offsets holding a pointer.
> 
> Eike

Does Valgrind say anything useful in this case?