There is a bug (http://sources.redhat.com/bugzilla/show_bug.cgi?id=1158) in how glibc 2.3.x orders functions registered with cxa_atexit across DSOs. This has been fixed upstream, but only on the CVS trunk, which I assume will (eventually) become glibc 2.4: http://sources.redhat.com/cgi-bin/cvsweb.cgi/libc/stdlib/cxa_atexit.c.diff?r1=1.5&r2=1.6&cvsroot=glibc Could this patch be considered for inclusion in the 2.3.x patchset? (I'm using it locally without any problems, and it fixes a number of segfault-on-program-termination issues in a local application that makes heavy use of C++ function-scope statics in DSOs loaded with dlopen).
thanks, ive added it to our 2.3.5 and 2.3.6 patchsets and it'll be in the next versions we put out