> Since we already have a jmp_buf which is much larger than > necisarry, all > that is needed is to take the first offset into it which is properly > aligned for the given architecture and use that as the system jmp_buf. > when the compiler can guarentee the proper alignment this becomes a > 'nop', otherwise we waste at most alignment - sizeof(intmax_t) - 1 > bytes. not a big deal. best of all, this can be implemented portably > quite easily. > John
This will allow adapting to misc code compatible CPU designs that do have but different alignment constrains. XServer can resolve that at runtime. This might happen when e.g. a more evolved CPU suddenly allows more freedom or some "shrinked" derivate introduces more restrictions. In contrast to this, if dynamic resolvement takes place you have to trim all your binaries for worst case or you do have to provide multiple binaries. Nothing that e.g. Linux distributions would like. Hmmm, if there would be a way for making those table a dynamically callocateable resource rather than a static one... -Alex. _______________________________________________ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
