------- Comment #2 from peter_foelsche at agilent dot com 2007-11-16 16:42 ------- the attached piece of source code does exhibit the compiler problem with the new compiler. When compiling using -O3 the code works fine. When compiling using -g the bug happens. Some temporary objects are being created and one constructor call for these temporary objects gets the wrong address. To exhibit this I put calls to printf() into the constructor and destructor showing the this pointer. Look at the end of the code to find the main function with one call to printf() and several temporary objects of my old string class. (Btw I'm not using this class anymore since we have std::string.)
Here comes the output of the test showing the problem. You can see that the destructors are called in opposite order to the destructors. CString::CString(): this==0x7fbfff8d30 CString::CString(): this==0x7fbfff8c80 CString::CString(): this==0x7fbfff8dc0 CString::CString(): this==0x7fbfff8d60 CString::CString(): this==0x7fbfff8d00 testÉtest CString::~CString(): this==0x7fbfff8d00 CString::~CString(): this==0x7fbfff8d60 CString::~CString(): this==0x7fbfff8dc0 CString::~CString(): this==0x7fbfff8d90 free(): invalid pointer 0x7fbfff8c88! CString::~CString(): this==0x7fbfff8d30 If you dig deep into this code you will find that one object is of type derived from CString. This is where the compiler calculates the wrong address. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34121
