Waldemar Kornewald <Waldemar.Kornewald@xxxxxx> wrote: > Axel Dörfler wrote: > > If you use x86-64 there is little point in maintaining backwards > > compatibility (IMO); as with the PPC port, we would just use the > > latest > > compiler. > But there is one little problem with future backwards compatibility: > We currently reserve some memory in our classes (int32 > _reserved[10]). > When we change the class members we can use that reserved memory. > But when we add a pointer we currently take one int32 away (as > pointer > are 32bits in IA32 ;). This would break binary compatibility with > 64bit > systems. > Should we use two memory areas?: > addr_t _reserved_addr[3]; > int32 _reserved_mem[7]; > Or should we do something like this when we allocate a pointer: > int32 _reserved[10 - sizeof(addr_t)]. > The latter will probably lead to people forgetting to use > sizeof(addr_t) > and thus to hard-to-find bugs. True, the current way would possibly create binary compatibility problems. But I am not so worried about that at this point - I don't think we'll have lots of officially supported (especially 64bit) platforms for R1. But you raised a very good point; we definitely have to come up with a solution for that problem then. Two different reserved areas would certainly be a solution, but not that nice, either, ah well. Bye, Axel.