As new timestamps typically get higher:-)
it's better to traverse the existing list from
the tail.
This is not completely optimal, but it should be better
than before.
A second optimization could be done for zero timestamps,
we would just remember the last_zero_timer,
but that would change the internal ABI.
Normally thatshould not be a poblem, but the Samba's
source3/lib/events.c abuses tevent_internal.h
from the current source tree, even if an external tevent.h
is used. The other problem is that it makes use of
tevent_common_add_timer() without using
tevent_common_loop_timer_delay().
Signed-off-by: Stefan Metzmacher <metze@samba.org>
const char *handler_name,
const char *location)
{
const char *handler_name,
const char *location)
{
- struct tevent_timer *te, *last_te, *cur_te;
+ struct tevent_timer *te, *prev_te, *cur_te;
te = talloc(mem_ctx?mem_ctx:ev, struct tevent_timer);
if (te == NULL) return NULL;
te = talloc(mem_ctx?mem_ctx:ev, struct tevent_timer);
if (te == NULL) return NULL;
te->additional_data = NULL;
/* keep the list ordered */
te->additional_data = NULL;
/* keep the list ordered */
- last_te = NULL;
- for (cur_te = ev->timer_events; cur_te; cur_te = cur_te->next) {
- /* if the new event comes before the current one break */
- if (tevent_timeval_compare(&te->next_event, &cur_te->next_event) < 0) {
- break;
+ prev_te = NULL;
+ /*
+ * we traverse the list from the tail
+ * because it's much more likely that
+ * timers are added at the end of the list
+ */
+ for (cur_te = DLIST_TAIL(ev->timer_events);
+ cur_te != NULL;
+ cur_te = DLIST_PREV(cur_te))
+ {
+ int ret;
+
+ /*
+ * if the new event comes before the current
+ * we continue searching
+ */
+ ret = tevent_timeval_compare(&te->next_event,
+ &cur_te->next_event);
+ if (ret < 0) {
+ continue;
- DLIST_ADD_AFTER(ev->timer_events, te, last_te);
+ DLIST_ADD_AFTER(ev->timer_events, te, prev_te);
talloc_set_destructor(te, tevent_common_timed_destructor);
talloc_set_destructor(te, tevent_common_timed_destructor);