All classes that use G4Allocator<Type> should check in operator new that the size of the allocation is equal to sizeof(Type). This will prevent difficult-to-debug segmentation faults in constructors of classes derived from classes that use G4Allocator. This subtle point should also be commented in the source of all such classes. I suggest something like this: inline void* G4Trajectory::operator new(size_t nbytes) { // G4Allocator<G4Trajectory> only allocates objects of that size // derived classes _must_ provide their own operators new and delete assert(nbytes == sizeof(G4Trajectory)); void* aTrajectory = (void*)aTrajectoryAllocator.MallocSingle(); return aTrajectory; } G4Trajectory caused me great pain until I figured this out.
This can "work for you", and yet still be a ticking time bomb for all users. To test it, you must add at least (malloc quantization [=8?]) bytes of new data to the class, and must actually use the new data in an allocated object, which is then deleted. This must happen many times to really test it. If, for example, you only added 4 bytes to the object, but the original class happened to have >= 4 bytes unused in the quantization, it would just happen to work for you, and yet cause problems for others. I _STRONGLY_ urge that all uses of G4Allocator be modified as I suggested (to test the size of the object being allocated), and the documentation updated. This is a ticking time bomb that can cause unexplained segmentation faults ANYWHERE in the program, because the heap gets corrupted. I _KNOW_ this caused me considerable pain, and I'll bet many others.
