On Sonntag, 6. M?rz 2005 17:37, Udo Richter wrote: > Sascha Volkenandt wrote: > > I assume this is a compiler bug, since the error should be caught at > > compile-time (I can't see any error though, since the called object is a > > derivation and the method is derived, too). Anyway, if I "shift around" those > > methods a bit, it throws a "pure virtual method" error with gcc 2.95, but not > > with gcc 3.x. > > I've seen pure virtual method errors before with cThread's. There is a > way how these pure virtuals can get active: At destructor time, as soon > as the derived class is destructed, the virtuals get reverted to the > pure virtuals of the base class, before the base class destructor takes > over. Reason is that the derived object is gone as soon as the derived > destructor ends, and only the base object needs to destruct. > > In this case (just a guess by a quick look at source code): > ~cStreamdevStreamer() calls Stop(), and Stop() calls Detach(). Since the > derived cStreamdevLiveStreamer is already destructed, its virtual > Detach() is gone too, and the pure virtual cStreamdevStreamer::Detach() > gets called. That's exact that was I've seen when I run vdr from gdb. The line reported was line 103 from streamer.c : Detach (). > > Virtuals in destructors are dangerous. Better make sure the thread is > stopped before the object destructs... -- Stefan Lucke
