I found an epic thread that describes the problem in detail. It does seem quite dated, but they do discuss gcc 3.x which I think still applies. Not sure what the resolution and related GCC changes are tho... http://gcc.gnu.org/ml/gcc/2002-05/msg00869.html There's also a post on Python c++ mail list here (https://svn.boost.org/trac/boost/ticket/3210#comment:3) which indicates that all in all, compiling exception body into a DSO will likely work, but then again it might be very flimsy. Unfortunately I don't think using GLOBAL is an option in my case; some of the Python extensions (namely SciPy) are throwing fits, so I'll have to figure something out, likely resort to statically linking everything in. Thanks for your help, Dimitri. On Mon, Aug 30, 2010 at 9:44 PM, Ian Lance Taylor <iant@xxxxxxxxxx> wrote: > Dimitri Tcaciuc <dtcaciuc@xxxxxxxxx> writes: > >> Right... So then I imagine even the 'fix' where I'm compiling >> exception destructor into a separate .so file is still subject to >> breakage and std::exception catching works because every C++ program >> actually links against libstdc++ and not just dlopens it, yes? > > Yes, I think that is correct. > > Ian >
