ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev.pod
(Generate patch)

Comparing libev/ev.pod (file contents):
Revision 1.26 by root, Tue Nov 13 03:11:57 2007 UTC vs.
Revision 1.29 by root, Thu Nov 22 12:28:28 2007 UTC

… …
145C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will 145C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
146override the flags completely if it is found in the environment. This is 146override the flags completely if it is found in the environment. This is
147useful to try out specific backends to test their performance, or to work 147useful to try out specific backends to test their performance, or to work
148around bugs. 148around bugs.
149 149
150=item C<EVMETHOD_SELECT> (portable select backend) 150=item C<EVMETHOD_SELECT> (value 1, portable select backend)
151 151
152This is your standard select(2) backend. Not I<completely> standard, as
153libev tries to roll its own fd_set with no limits on the number of fds,
154but if that fails, expect a fairly low limit on the number of fds when
155using this backend. It doesn't scale too well (O(highest_fd)), but its usually
156the fastest backend for a low number of fds.
157
152=item C<EVMETHOD_POLL> (poll backend, available everywhere except on windows) 158=item C<EVMETHOD_POLL> (value 2, poll backend, available everywhere except on windows)
153 159
160And this is your standard poll(2) backend. It's more complicated than
161select, but handles sparse fds better and has no artificial limit on the
162number of fds you can use (except it will slow down considerably with a
163lot of inactive fds). It scales similarly to select, i.e. O(total_fds).
164
154=item C<EVMETHOD_EPOLL> (linux only) 165=item C<EVMETHOD_EPOLL> (value 4, Linux)
155 166
156=item C<EVMETHOD_KQUEUE> (some bsds only) 167For few fds, this backend is a bit little slower than poll and select,
168but it scales phenomenally better. While poll and select usually scale like
169O(total_fds) where n is the total number of fds (or the highest fd), epoll scales
170either O(1) or O(active_fds).
157 171
172While stopping and starting an I/O watcher in the same iteration will
173result in some caching, there is still a syscall per such incident
174(because the fd could point to a different file description now), so its
175best to avoid that. Also, dup()ed file descriptors might not work very
176well if you register events for both fds.
177
178=item C<EVMETHOD_KQUEUE> (value 8, most BSD clones)
179
180Kqueue deserves special mention, as at the time of this writing, it
181was broken on all BSDs except NetBSD (usually it doesn't work with
182anything but sockets and pipes, except on Darwin, where of course its
183completely useless). For this reason its not being "autodetected" unless
184you explicitly specify the flags (i.e. you don't use EVFLAG_AUTO).
185
186It scales in the same way as the epoll backend, but the interface to the
187kernel is more efficient (which says nothing about its actual speed, of
188course). While starting and stopping an I/O watcher does not cause an
189extra syscall as with epoll, it still adds up to four event changes per
190incident, so its best to avoid that.
191
158=item C<EVMETHOD_DEVPOLL> (solaris 8 only) 192=item C<EVMETHOD_DEVPOLL> (value 16, Solaris 8)
159 193
194This is not implemented yet (and might never be).
195
160=item C<EVMETHOD_PORT> (solaris 10 only) 196=item C<EVMETHOD_PORT> (value 32, Solaris 10)
197
198This uses the Solaris 10 port mechanism. As with everything on Solaris,
199it's really slow, but it still scales very well (O(active_fds)).
200
201=item C<EVMETHOD_ALL>
202
203Try all backends (even potentially broken ones that wouldn't be tried
204with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
205C<EVMETHOD_ALL & ~EVMETHOD_KQUEUE>.
206
207=back
161 208
162If one or more of these are ored into the flags value, then only these 209If one or more of these are ored into the flags value, then only these
163backends will be tried (in the reverse order as given here). If one are 210backends will be tried (in the reverse order as given here). If none are
164specified, any backend will do. 211specified, most compiled-in backend will be tried, usually in reverse
165 212order of their flag values :)
166=back
167 213
168=item struct ev_loop *ev_loop_new (unsigned int flags) 214=item struct ev_loop *ev_loop_new (unsigned int flags)
169 215
170Similar to C<ev_default_loop>, but always creates a new event loop that is 216Similar to C<ev_default_loop>, but always creates a new event loop that is
171always distinct from the default loop. Unlike the default loop, it cannot 217always distinct from the default loop. Unlike the default loop, it cannot
… …
238one iteration of the loop. 284one iteration of the loop.
239 285
240This flags value could be used to implement alternative looping 286This flags value could be used to implement alternative looping
241constructs, but the C<prepare> and C<check> watchers provide a better and 287constructs, but the C<prepare> and C<check> watchers provide a better and
242more generic mechanism. 288more generic mechanism.
289
290Here are the gory details of what ev_loop does:
291
292 1. If there are no active watchers (reference count is zero), return.
293 2. Queue and immediately call all prepare watchers.
294 3. If we have been forked, recreate the kernel state.
295 4. Update the kernel state with all outstanding changes.
296 5. Update the "event loop time".
297 6. Calculate for how long to block.
298 7. Block the process, waiting for events.
299 8. Update the "event loop time" and do time jump handling.
300 9. Queue all outstanding timers.
301 10. Queue all outstanding periodics.
302 11. If no events are pending now, queue all idle watchers.
303 12. Queue all check watchers.
304 13. Call all queued watchers in reverse order (i.e. check watchers first).
305 14. If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
306 was used, return, otherwise continue with step #1.
243 307
244=item ev_unloop (loop, how) 308=item ev_unloop (loop, how)
245 309
246Can be used to make a call to C<ev_loop> return early (but only after it 310Can be used to make a call to C<ev_loop> return early (but only after it
247has processed all outstanding events). The C<how> argument must be either 311has processed all outstanding events). The C<how> argument must be either
… …
454given time, and optionally repeating in regular intervals after that. 518given time, and optionally repeating in regular intervals after that.
455 519
456The timers are based on real time, that is, if you register an event that 520The timers are based on real time, that is, if you register an event that
457times out after an hour and you reset your system clock to last years 521times out after an hour and you reset your system clock to last years
458time, it will still time out after (roughly) and hour. "Roughly" because 522time, it will still time out after (roughly) and hour. "Roughly" because
459detecting time jumps is hard, and soem inaccuracies are unavoidable (the 523detecting time jumps is hard, and some inaccuracies are unavoidable (the
460monotonic clock option helps a lot here). 524monotonic clock option helps a lot here).
461 525
462The relative timeouts are calculated relative to the C<ev_now ()> 526The relative timeouts are calculated relative to the C<ev_now ()>
463time. This is usually the right thing as this timestamp refers to the time 527time. This is usually the right thing as this timestamp refers to the time
464of the event triggering whatever timeout you are modifying/starting. If 528of the event triggering whatever timeout you are modifying/starting. If
465you suspect event processing to be delayed and you *need* to base the timeout 529you suspect event processing to be delayed and you I<need> to base the timeout
466on the current time, use something like this to adjust for this: 530on the current time, use something like this to adjust for this:
467 531
468 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.); 532 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
533
534The callback is guarenteed to be invoked only when its timeout has passed,
535but if multiple timers become ready during the same loop iteration then
536order of execution is undefined.
469 537
470=over 4 538=over 4
471 539
472=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat) 540=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
473 541
… …
520again). 588again).
521 589
522They can also be used to implement vastly more complex timers, such as 590They can also be used to implement vastly more complex timers, such as
523triggering an event on eahc midnight, local time. 591triggering an event on eahc midnight, local time.
524 592
593As with timers, the callback is guarenteed to be invoked only when the
594time (C<at>) has been passed, but if multiple periodic timers become ready
595during the same loop iteration then order of execution is undefined.
596
525=over 4 597=over 4
526 598
527=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb) 599=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb)
528 600
529=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb) 601=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb)
530 602
531Lots of arguments, lets sort it out... There are basically three modes of 603Lots of arguments, lets sort it out... There are basically three modes of
532operation, and we will explain them from simplest to complex: 604operation, and we will explain them from simplest to complex:
533
534 605
535=over 4 606=over 4
536 607
537=item * absolute timer (interval = reschedule_cb = 0) 608=item * absolute timer (interval = reschedule_cb = 0)
538 609

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines