https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88684
Bug ID: 88684
Summary: Please make SANITIZER_NON_UNIQUE_TYPEINFO a runtime
flag (or always true)
Product: gcc
Version: unknown
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: sanitizer
Assignee: unassigned at gcc dot gnu.org
Reporter: rafael at espindo dot la
CC: dodji at gcc dot gnu.org, dvyukov at gcc dot gnu.org,
jakub at gcc dot gnu.org, kcc at gcc dot gnu.org, marxin at
gcc dot gnu.org
Target Milestone: ---
Even on ABIs that normally unique typeinfo names, it is easy to end up in
situations where that fails.
Consider a shared library implemented with
lib.hh:
struct foo {
virtual ~foo(){}
};
struct bar : public foo {
virtual void zed();
};
lib.cc:
#include "lib.hh"
void bar::zed() {}
and being used by the program (could be another library):
test.cc:
#include "lib.hh"
int main(int argc, char** argv) { bar t; }
if the program is compiled with -fvisibility=hidden, it will have a hidden
_ZTI3foo which isDerivedFromAtOffset will think doesn't match the _ZTI3foo in
the library.
The above test is a reduction of
#include <boost/test/unit_test.hpp>
int main(int argc, char **argv) {
return 0;
}
compiled with -fvisibility=hidden, which complains that
/usr/include/boost/test/unit_test_log.hpp:112:23: runtime error: member call on
address 0x000006583060 which does not point to an object of type
'test_observer'
