- memset(&from,0,sizeof(from));
- len=sizeof(from);
- /* Note: under VMS with SOCKETSHR the fourth parameter is currently
- * of type (int *) whereas under other systems it is (void *) if
- * you don't have a cast it will choke the compiler: if you do
- * have a cast then you can either go for (int *) or (void *).
+ struct {
+ /*
+ * As for following union. Trouble is that there are platforms
+ * that have socklen_t and there are platforms that don't, on
+ * some platforms socklen_t is int and on some size_t. So what
+ * one can do? One can cook #ifdef spaghetti, which is nothing
+ * but masochistic. Or one can do union between int and size_t.
+ * One naturally does it primarily for 64-bit platforms where
+ * sizeof(int) != sizeof(size_t). But would it work? Note that
+ * if size_t member is initialized to 0, then later int member
+ * assignment naturally does the job on little-endian platforms
+ * regardless accept's expectations! What about big-endians?
+ * If accept expects int*, then it works, and if size_t*, then
+ * length value would appear as unreasonably large. But this
+ * won't prevent it from filling in the address structure. The
+ * trouble of course would be if accept returns more data than
+ * actual buffer can accomodate and overwrite stack... That's
+ * where early OPENSSL_assert comes into picture. Besides, the
+ * only 64-bit big-endian platform found so far that expects
+ * size_t* is HP-UX, where stack grows towards higher address.
+ * <appro>