Pavel Vozenilek wrote:
"David Abrahams" <dave@boost-consulting.com> wrote
I can't understand this claim. If the counter works properly, then it will stay valid.
This example was too simplified (and thus not to the point).
More complex one:
----------------------- struct my_class { void* database_handle;
my_class() { database_handle = allocate_handle(); } ~my_class() { destroy_db_handle(database_handle); } my_class& operator=(const my_class& other) { // no copy of database handle, not needed, // not useful here } my_class(const my_class& other) { database_handle = allocate_handle(); // needs unique db connection } };
circular_buffer<my_class> buffer(1);
{ my_class a; buffer.push_back(a); // 2 handles after function returns ... }
{ my_class b; buffer.push_back(b); // desctructor not called, handle lost ... }
I don't see why a handle is lost. { // Only one handle in existence at this point, the one // previously added to buffer by "buffer.push_back(a);". my_class b; // Now two handles exist, one in b, and one in buffer. buffer.push_back(b); // Two handles still exist -- push_back uses assignment, which // doesn't affect handles at all. No new handles have been // created by push_back(b). } // Now only one handle exists, in buffer, since b has // gone out of scope and has been destroyed. What am I missing? Bob