I believe I have found a small bug in Event-0.81 that causes the event loop to
use too much CPU time. This is with perl 5.6.0 on Solaris 2.6.
The symptoms I see are: when a timer comes due for action, Event spends one
second doing a tight loop calling poll() with zero timeout before it executes
the callback of the of the timer.
The problem appears to be in pe_timeables_check() in c/timeable.c. Here's the
context diff:
*** timeable.c Mon Apr 16 08:12:15 2001
--- timeable.c-ORIG Tue Jun 13 08:24:24 2000
***************
*** 19,25 ****
pe_timeable *tm = (pe_timeable*) Timeables.ring.next;
double now = NVtime();
/* warn("timeables at %.2f\n", now); db_show_timeables();/**/
! while (tm->ring.self && tm->at <= now) {
pe_watcher *ev = (pe_watcher*) tm->ring.self;
pe_timeable *next = (pe_timeable*) tm->ring.next;
D_TIMEABLE({
--- 19,25 ----
pe_timeable *tm = (pe_timeable*) Timeables.ring.next;
double now = NVtime();
/* warn("timeables at %.2f\n", now); db_show_timeables();/**/
! while (tm->ring.self && tm->at < now) {
pe_watcher *ev = (pe_watcher*) tm->ring.self;
pe_timeable *next = (pe_timeable*) tm->ring.next;
D_TIMEABLE({
With this change, Event executes the timer's callback as soon as it is due,
without doing a bunch of poll() calls. But I am not certain that this doesn't
break something else in Event; this C in perl is pretty new to me!
--
Pete Lombard
Seismological Laboratory
215 McCone Hall
University of California
Berkeley, CA 94720-4760
[EMAIL PROTECTED]
(510) 642-0073
