On Oct 3, 12:20 am, Paul Pluzhnikov <[EMAIL PROTECTED]> wrote: > jeremy barrett <[EMAIL PROTECTED]> writes: > > > int* A::x = new int(5); > > Note that 'A::x' is initialized to a non-constant value. That means > 'gcc' has to initialize it dynamically. In effect, 'gcc' writes a > new function (called static_initialization_and_destruction()), > which is called via exactly the same mechanism as your initAx(). > > The order of static_init...() and initAx() is not specified, > gcc-3.3.3 for Linux/x86 calls initAx() after static_init...(), > while gcc-4.3.0 calls them in reverse order.
Thanks for this insight. I discovered (anecdotally) the same thing last night, while trying to reproduce this error on a different platform (gcc-4.1.3 on Linux/x86 at work, gcc-4.0.1 on Mac/x86 at home). ((snip)) > > It should be enlightening for you to compile the following separate > snippets into assembly and understand the result: > > int x = 5; // compile-time constant > > int x; > int *px = &x; // link-time constant > > int foo(); > int x = foo(); // runtime intialization required I'll perform this exercise. Thanks. > > Finally, it is entirely pointless to provide initAx() and initBx(). > If you want to initialize A::x and B::x to some values, just do > so directly: Here's the point. A and B are very generic classes that are included in a number of different libraries I've built. So the static initializers set x to something suitably generic for use in standard cases. In order keep data appropriately hidden (and save myself some bookkeeping), I add this constructor function to various of the shared libraries to initialize x to something meaningful (conceptual zero) for the given application/environment. Any (portable) ideas about how I can accomplish my goal? I understand that __attribute__ ((constructor)) is not portable, but is there at least a method that will work with the generic condition: #ifdef __GNUC__ //some brilliant idea #endif ? Thanks! jeremy _______________________________________________ help-gplusplus mailing list help-gplusplus@gnu.org http://lists.gnu.org/mailman/listinfo/help-gplusplus
