On Thursday, November 01, 2012 22:21:11 Dan wrote:
> struct S {
> int[] a; // array is privately owned by this instance
> this(this) {
> a = a.dup;
> }
> ~this() {
> delete a;
> }
> }
>
> Is the delete call, then per TDPL not necessary? Is it harmful or
> harmless?It's not necessary at all. delete is _never_ necessary, and it's not safe. delete frees GC-allocated memory. If you just leave it alone, and there are really no other references to it, then the GC will eventually free it if it needs more memory. Deleting it makes it so that it's freed now rather than freed who-knows-when later, but if any other references to that data still exist when it's deleted, then they'll end up pointing to garbage behavior, giving you nasty bugs. Because of all of this, delete is going to be deprecated if it hasn't been already. core.memory will still provide functions for freeing GC memory if you really want to and are willing to go the extra mile to make sure that your code is safe, but there will no longer be a language primitive for doing so. clear (which was recently renamed to destroy) specifically destroys an object but does _not_ free its memory. So, you won't end up with bugs due to other references to that data operating on garbage. In the case of classes, because destroy zeroes out the vtbl, calling virtual functions on the destroyed object will cause a segfault. In the case of primitives such as int, I believe that they're set to their init property. And in the case of arrays, I believe that it's no different from setting them to null, so nothing else is actually affected by calling destroy on them. Regardless, with destroy, you're not going to run into nasty memory issues due to memory being freed when other references to it still existed, because it doesn't actually free any memory. It just destroys what's there so that fewer references to it exist and so that any non-GC-allocated resources which the destroyed object had get released. - Jonathan M Davis
