------- Comment #3 from rguenther at suse dot de 2009-01-16 18:27 ------- Subject: Re: field-insensitive PTA causes libstdc++ miscompiles
On Fri, 16 Jan 2009, pinskia at gcc dot gnu dot org wrote: > ------- Comment #2 from pinskia at gcc dot gnu dot org 2009-01-16 18:22 > ------- > return reinterpret_cast<_Tp*>(reinterpret_cast<_UIntPtrType>(this) > + _M_diff); > > > (const struct _Fwd_list_node_base *) this p+ (long unsigned int) ((const > struct _Relative_pointer_impl *) this)->_M_diff > > > That seems fine. In fact If we are failing because of aliasing issues, then > we > have a bigger issue really because the code could have been written as: > return reinterpret_cast<_Tp*>(reinterpret_cast<char*>(this) + > _M_diff); > > And still be defined so I don't think this is really PR 36227. No, char * is not allowed to change objects either. What the above code does is use the pointer -> intptr -> pointer rule by storing a pointer as a relative offset to the store, like so: void *p; intptr_t iptr = (intptr_t)&iptr - (intptr_t)p; and reconstructing p via (void *)((intptr_t)&iptr + iptr); which is perfectly valid, but we break that because we fold the latter to (void *)(&iptr + iptr) which is of course invalid. Richard. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38835
