[ViewVC] Diff of: cvs/libev/ev.pod

Comparing libev/ev.pod (file contents):
Revision 1.198 by root, Thu Oct 23 06:30:48 2008 UTC vs.
Revision 1.213 by root, Wed Nov 5 02:48:45 2008 UTC

…		…
10		10
11	// a single header file is required	11	// a single header file is required
12	#include <ev.h>	12	#include <ev.h>
13		13
14	// every watcher type has its own typedef'd struct	14	// every watcher type has its own typedef'd struct
15	// with the name ev_<type>	15	// with the name ev_TYPE
16	ev_io stdin_watcher;	16	ev_io stdin_watcher;
17	ev_timer timeout_watcher;	17	ev_timer timeout_watcher;
18		18
19	// all watcher callbacks have a similar signature	19	// all watcher callbacks have a similar signature
20	// this callback is called when data is readable on stdin	20	// this callback is called when data is readable on stdin
…		…
276		276
277	=back	277	=back
278		278
279	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP	279	=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
280		280
281	An event loop is described by a C<ev_loop *>. The library knows two	281	An event loop is described by a C<struct ev_loop *> (the C<struct>
282	types of such loops, the I<default> loop, which supports signals and child	282	is I<not> optional in this case, as there is also an C<ev_loop>
283	events, and dynamically created loops which do not.	283	I<function>).
		284
		285	The library knows two types of such loops, the I<default> loop, which
		286	supports signals and child events, and dynamically created loops which do
		287	not.
284		288
285	=over 4	289	=over 4
286		290
287	=item struct ev_loop *ev_default_loop (unsigned int flags)	291	=item struct ev_loop *ev_default_loop (unsigned int flags)
288		292
…		…
294	If you don't know what event loop to use, use the one returned from this	298	If you don't know what event loop to use, use the one returned from this
295	function.	299	function.
296		300
297	Note that this function is I<not> thread-safe, so if you want to use it	301	Note that this function is I<not> thread-safe, so if you want to use it
298	from multiple threads, you have to lock (note also that this is unlikely,	302	from multiple threads, you have to lock (note also that this is unlikely,
299	as loops cannot bes hared easily between threads anyway).	303	as loops cannot be shared easily between threads anyway).
300		304
301	The default loop is the only loop that can handle C<ev_signal> and	305	The default loop is the only loop that can handle C<ev_signal> and
302	C<ev_child> watchers, and to do this, it always registers a handler	306	C<ev_child> watchers, and to do this, it always registers a handler
303	for C<SIGCHLD>. If this is a problem for your application you can either	307	for C<SIGCHLD>. If this is a problem for your application you can either
304	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you	308	create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
…		…
380	=item C<EVBACKEND_EPOLL> (value 4, Linux)	384	=item C<EVBACKEND_EPOLL> (value 4, Linux)
381		385
382	For few fds, this backend is a bit little slower than poll and select,	386	For few fds, this backend is a bit little slower than poll and select,
383	but it scales phenomenally better. While poll and select usually scale	387	but it scales phenomenally better. While poll and select usually scale
384	like O(total_fds) where n is the total number of fds (or the highest fd),	388	like O(total_fds) where n is the total number of fds (or the highest fd),
385	epoll scales either O(1) or O(active_fds). The epoll design has a number	389	epoll scales either O(1) or O(active_fds).
386	of shortcomings, such as silently dropping events in some hard-to-detect	390
387	cases and requiring a system call per fd change, no fork support and bad	391	The epoll mechanism deserves honorable mention as the most misdesigned
388	support for dup.	392	of the more advanced event mechanisms: mere annoyances include silently
		393	dropping file descriptors, requiring a system call per change per file
		394	descriptor (and unnecessary guessing of parameters), problems with dup and
		395	so on. The biggest issue is fork races, however - if a program forks then
		396	I<both> parent and child process have to recreate the epoll set, which can
		397	take considerable time (one syscall per file descriptor) and is of course
		398	hard to detect.
		399
		400	Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
		401	of course I<doesn't>, and epoll just loves to report events for totally
		402	I<different> file descriptors (even already closed ones, so one cannot
		403	even remove them from the set) than registered in the set (especially
		404	on SMP systems). Libev tries to counter these spurious notifications by
		405	employing an additional generation counter and comparing that against the
		406	events to filter out spurious ones, recreating the set when required.
389		407
390	While stopping, setting and starting an I/O watcher in the same iteration	408	While stopping, setting and starting an I/O watcher in the same iteration
391	will result in some caching, there is still a system call per such incident	409	will result in some caching, there is still a system call per such
392	(because the fd could point to a different file description now), so its	410	incident (because the same I<file descriptor> could point to a different
393	best to avoid that. Also, C<dup ()>'ed file descriptors might not work	411	I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
394	very well if you register events for both fds.	412	file descriptors might not work very well if you register events for both
395		413	file descriptors.
396	Please note that epoll sometimes generates spurious notifications, so you
397	need to use non-blocking I/O or other means to avoid blocking when no data
398	(or space) is available.
399		414
400	Best performance from this backend is achieved by not unregistering all	415	Best performance from this backend is achieved by not unregistering all
401	watchers for a file descriptor until it has been closed, if possible,	416	watchers for a file descriptor until it has been closed, if possible,
402	i.e. keep at least one watcher active per fd at all times. Stopping and	417	i.e. keep at least one watcher active per fd at all times. Stopping and
403	starting a watcher (without re-setting it) also usually doesn't cause	418	starting a watcher (without re-setting it) also usually doesn't cause
404	extra overhead.	419	extra overhead. A fork can both result in spurious notifications as well
		420	as in libev having to destroy and recreate the epoll object, which can
		421	take considerable time and thus should be avoided.
		422
		423	All this means that, in practise, C<EVBACKEND_SELECT> is as fast or faster
		424	then epoll for maybe up to a hundred file descriptors. So sad.
405		425
406	While nominally embeddable in other event loops, this feature is broken in	426	While nominally embeddable in other event loops, this feature is broken in
407	all kernel versions tested so far.	427	all kernel versions tested so far.
408		428
409	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as	429	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
410	C<EVBACKEND_POLL>.	430	C<EVBACKEND_POLL>.
411		431
412	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)	432	=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
413		433
414	Kqueue deserves special mention, as at the time of this writing, it was	434	Kqueue deserves special mention, as at the time of this writing, it
415	broken on all BSDs except NetBSD (usually it doesn't work reliably with	435	was broken on all BSDs except NetBSD (usually it doesn't work reliably
416	anything but sockets and pipes, except on Darwin, where of course it's	436	with anything but sockets and pipes, except on Darwin, where of course
417	completely useless). For this reason it's not being "auto-detected" unless	437	it's completely useless). Unlike epoll, however, whose brokenness
418	you explicitly specify it in the flags (i.e. using C<EVBACKEND_KQUEUE>) or	438	is by design, these kqueue bugs can (and eventually will) be fixed
419	libev was compiled on a known-to-be-good (-enough) system like NetBSD.	439	without API changes to existing programs. For this reason it's not being
		440	"auto-detected" unless you explicitly specify it in the flags (i.e. using
		441	C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
		442	system like NetBSD.
420		443
421	You still can embed kqueue into a normal poll or select backend and use it	444	You still can embed kqueue into a normal poll or select backend and use it
422	only for sockets (after having made sure that sockets work with kqueue on	445	only for sockets (after having made sure that sockets work with kqueue on
423	the target platform). See C<ev_embed> watchers for more info.	446	the target platform). See C<ev_embed> watchers for more info.
424		447
425	It scales in the same way as the epoll backend, but the interface to the	448	It scales in the same way as the epoll backend, but the interface to the
426	kernel is more efficient (which says nothing about its actual speed, of	449	kernel is more efficient (which says nothing about its actual speed, of
427	course). While stopping, setting and starting an I/O watcher does never	450	course). While stopping, setting and starting an I/O watcher does never
428	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to	451	cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
429	two event changes per incident. Support for C<fork ()> is very bad and it	452	two event changes per incident. Support for C<fork ()> is very bad (but
430	drops fds silently in similarly hard-to-detect cases.	453	sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
		454	cases
431		455
432	This backend usually performs well under most conditions.	456	This backend usually performs well under most conditions.
433		457
434	While nominally embeddable in other event loops, this doesn't work	458	While nominally embeddable in other event loops, this doesn't work
435	everywhere, so you might need to test for this. And since it is broken	459	everywhere, so you might need to test for this. And since it is broken
…		…
464	might perform better.	488	might perform better.
465		489
466	On the positive side, with the exception of the spurious readiness	490	On the positive side, with the exception of the spurious readiness
467	notifications, this backend actually performed fully to specification	491	notifications, this backend actually performed fully to specification
468	in all tests and is fully embeddable, which is a rare feat among the	492	in all tests and is fully embeddable, which is a rare feat among the
469	OS-specific backends.	493	OS-specific backends (I vastly prefer correctness over speed hacks).
470		494
471	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as	495	This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
472	C<EVBACKEND_POLL>.	496	C<EVBACKEND_POLL>.
473		497
474	=item C<EVBACKEND_ALL>	498	=item C<EVBACKEND_ALL>
…		…
527	responsibility to either stop all watchers cleanly yourself I<before>	551	responsibility to either stop all watchers cleanly yourself I<before>
528	calling this function, or cope with the fact afterwards (which is usually	552	calling this function, or cope with the fact afterwards (which is usually
529	the easiest thing, you can just ignore the watchers and/or C<free ()> them	553	the easiest thing, you can just ignore the watchers and/or C<free ()> them
530	for example).	554	for example).
531		555
532	Note that certain global state, such as signal state, will not be freed by	556	Note that certain global state, such as signal state (and installed signal
533	this function, and related watchers (such as signal and child watchers)	557	handlers), will not be freed by this function, and related watchers (such
534	would need to be stopped manually.	558	as signal and child watchers) would need to be stopped manually.
535		559
536	In general it is not advisable to call this function except in the	560	In general it is not advisable to call this function except in the
537	rare occasion where you really need to free e.g. the signal handling	561	rare occasion where you really need to free e.g. the signal handling
538	pipe fds. If you need dynamically allocated loops it is better to use	562	pipe fds. If you need dynamically allocated loops it is better to use
539	C<ev_loop_new> and C<ev_loop_destroy>).	563	C<ev_loop_new> and C<ev_loop_destroy>).
…		…
631	the loop.	655	the loop.
632		656
633	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if	657	A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
634	necessary) and will handle those and any already outstanding ones. It	658	necessary) and will handle those and any already outstanding ones. It
635	will block your process until at least one new event arrives (which could	659	will block your process until at least one new event arrives (which could
636	be an event internal to libev itself, so there is no guarentee that a	660	be an event internal to libev itself, so there is no guarantee that a
637	user-registered callback will be called), and will return after one	661	user-registered callback will be called), and will return after one
638	iteration of the loop.	662	iteration of the loop.
639		663
640	This is useful if you are waiting for some external event in conjunction	664	This is useful if you are waiting for some external event in conjunction
641	with something not expressible using other libev watchers (i.e. "roll your	665	with something not expressible using other libev watchers (i.e. "roll your
…		…
768	they fire on, say, one-second boundaries only.	792	they fire on, say, one-second boundaries only.
769		793
770	=item ev_loop_verify (loop)	794	=item ev_loop_verify (loop)
771		795
772	This function only does something when C<EV_VERIFY> support has been	796	This function only does something when C<EV_VERIFY> support has been
773	compiled in. which is the default for non-minimal builds. It tries to go	797	compiled in, which is the default for non-minimal builds. It tries to go
774	through all internal structures and checks them for validity. If anything	798	through all internal structures and checks them for validity. If anything
775	is found to be inconsistent, it will print an error message to standard	799	is found to be inconsistent, it will print an error message to standard
776	error and call C<abort ()>.	800	error and call C<abort ()>.
777		801
778	This can be used to catch bugs inside libev itself: under normal	802	This can be used to catch bugs inside libev itself: under normal
…		…
781		805
782	=back	806	=back
783		807
784		808
785	=head1 ANATOMY OF A WATCHER	809	=head1 ANATOMY OF A WATCHER
		810
		811	In the following description, uppercase C<TYPE> in names stands for the
		812	watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
		813	watchers and C<ev_io_start> for I/O watchers.
786		814
787	A watcher is a structure that you create and register to record your	815	A watcher is a structure that you create and register to record your
788	interest in some event. For instance, if you want to wait for STDIN to	816	interest in some event. For instance, if you want to wait for STDIN to
789	become readable, you would create an C<ev_io> watcher for that:	817	become readable, you would create an C<ev_io> watcher for that:
790		818
…		…
793	ev_io_stop (w);	821	ev_io_stop (w);
794	ev_unloop (loop, EVUNLOOP_ALL);	822	ev_unloop (loop, EVUNLOOP_ALL);
795	}	823	}
796		824
797	struct ev_loop *loop = ev_default_loop (0);	825	struct ev_loop *loop = ev_default_loop (0);
		826
798	ev_io stdin_watcher;	827	ev_io stdin_watcher;
		828
799	ev_init (&stdin_watcher, my_cb);	829	ev_init (&stdin_watcher, my_cb);
800	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);	830	ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
801	ev_io_start (loop, &stdin_watcher);	831	ev_io_start (loop, &stdin_watcher);
		832
802	ev_loop (loop, 0);	833	ev_loop (loop, 0);
803		834
804	As you can see, you are responsible for allocating the memory for your	835	As you can see, you are responsible for allocating the memory for your
805	watcher structures (and it is usually a bad idea to do this on the stack,	836	watcher structures (and it is I<usually> a bad idea to do this on the
806	although this can sometimes be quite valid).	837	stack).
		838
		839	Each watcher has an associated watcher structure (called C<struct ev_TYPE>
		840	or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
807		841
808	Each watcher structure must be initialised by a call to C<ev_init	842	Each watcher structure must be initialised by a call to C<ev_init
809	(watcher *, callback)>, which expects a callback to be provided. This	843	(watcher *, callback)>, which expects a callback to be provided. This
810	callback gets invoked each time the event occurs (or, in the case of I/O	844	callback gets invoked each time the event occurs (or, in the case of I/O
811	watchers, each time the event loop detects that the file descriptor given	845	watchers, each time the event loop detects that the file descriptor given
812	is readable and/or writable).	846	is readable and/or writable).
813		847
814	Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro	848	Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
815	with arguments specific to this watcher type. There is also a macro	849	macro to configure it, with arguments specific to the watcher type. There
816	to combine initialisation and setting in one call: C<< ev_<type>_init	850	is also a macro to combine initialisation and setting in one call: C<<
817	(watcher *, callback, ...) >>.	851	ev_TYPE_init (watcher *, callback, ...) >>.
818		852
819	To make the watcher actually watch out for events, you have to start it	853	To make the watcher actually watch out for events, you have to start it
820	with a watcher-specific start function (C<< ev_<type>_start (loop, watcher	854	with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
821	*) >>), and you can stop watching for events at any time by calling the	855	*) >>), and you can stop watching for events at any time by calling the
822	corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>.	856	corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
823		857
824	As long as your watcher is active (has been started but not stopped) you	858	As long as your watcher is active (has been started but not stopped) you
825	must not touch the values stored in it. Most specifically you must never	859	must not touch the values stored in it. Most specifically you must never
826	reinitialise it or call its C<set> macro.	860	reinitialise it or call its C<ev_TYPE_set> macro.
827		861
828	Each and every callback receives the event loop pointer as first, the	862	Each and every callback receives the event loop pointer as first, the
829	registered watcher structure as second, and a bitset of received events as	863	registered watcher structure as second, and a bitset of received events as
830	third argument.	864	third argument.
831		865
…		…
912		946
913	=back	947	=back
914		948
915	=head2 GENERIC WATCHER FUNCTIONS	949	=head2 GENERIC WATCHER FUNCTIONS
916		950
917	In the following description, C<TYPE> stands for the watcher type,
918	e.g. C<timer> for C<ev_timer> watchers and C<io> for C<ev_io> watchers.
919
920	=over 4	951	=over 4
921		952
922	=item C<ev_init> (ev_TYPE *watcher, callback)	953	=item C<ev_init> (ev_TYPE *watcher, callback)
923		954
924	This macro initialises the generic portion of a watcher. The contents	955	This macro initialises the generic portion of a watcher. The contents
…		…
1032	The default priority used by watchers when no priority has been set is	1063	The default priority used by watchers when no priority has been set is
1033	always C<0>, which is supposed to not be too high and not be too low :).	1064	always C<0>, which is supposed to not be too high and not be too low :).
1034		1065
1035	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is	1066	Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1036	fine, as long as you do not mind that the priority value you query might	1067	fine, as long as you do not mind that the priority value you query might
1037	or might not have been adjusted to be within valid range.	1068	or might not have been clamped to the valid range.
1038		1069
1039	=item ev_invoke (loop, ev_TYPE *watcher, int revents)	1070	=item ev_invoke (loop, ev_TYPE *watcher, int revents)
1040		1071
1041	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither	1072	Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1042	C<loop> nor C<revents> need to be valid as long as the watcher callback	1073	C<loop> nor C<revents> need to be valid as long as the watcher callback
…		…
1288	passed, but if multiple timers become ready during the same loop iteration	1319	passed, but if multiple timers become ready during the same loop iteration
1289	then order of execution is undefined.	1320	then order of execution is undefined.
1290		1321
1291	=head3 Be smart about timeouts	1322	=head3 Be smart about timeouts
1292		1323
1293	Many real-world problems invole some kind of time-out, usually for error	1324	Many real-world problems involve some kind of timeout, usually for error
1294	recovery. A typical example is an HTTP request - if the other side hangs,	1325	recovery. A typical example is an HTTP request - if the other side hangs,
1295	you want to raise some error after a while.	1326	you want to raise some error after a while.
1296		1327
1297	Here are some ways on how to handle this problem, from simple and	1328	What follows are some ways to handle this problem, from obvious and
1298	inefficient to very efficient.	1329	inefficient to smart and efficient.
1299		1330
1300	In the following examples a 60 second activity timeout is assumed - a	1331	In the following, a 60 second activity timeout is assumed - a timeout that
1301	timeout that gets reset to 60 seconds each time some data ("a lifesign")	1332	gets reset to 60 seconds each time there is activity (e.g. each time some
1302	was received.	1333	data or other life sign was received).
1303		1334
1304	=over 4	1335	=over 4
1305		1336
1306	=item 1. Use a timer and stop, reinitialise, start it on activity.	1337	=item 1. Use a timer and stop, reinitialise and start it on activity.
1307		1338
1308	This is the most obvious, but not the most simple way: In the beginning,	1339	This is the most obvious, but not the most simple way: In the beginning,
1309	start the watcher:	1340	start the watcher:
1310		1341
1311	ev_timer_init (timer, callback, 60., 0.);	1342	ev_timer_init (timer, callback, 60., 0.);
1312	ev_timer_start (loop, timer);	1343	ev_timer_start (loop, timer);
1313		1344
1314	Then, each time there is some activity, C<ev_timer_stop> the timer,	1345	Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1315	initialise it again, and start it:	1346	and start it again:
1316		1347
1317	ev_timer_stop (loop, timer);	1348	ev_timer_stop (loop, timer);
1318	ev_timer_set (timer, 60., 0.);	1349	ev_timer_set (timer, 60., 0.);
1319	ev_timer_start (loop, timer);	1350	ev_timer_start (loop, timer);
1320		1351
1321	This is relatively simple to implement, but means that each time there	1352	This is relatively simple to implement, but means that each time there is
1322	is some activity, libev will first have to remove the timer from it's	1353	some activity, libev will first have to remove the timer from its internal
1323	internal data strcuture and then add it again.	1354	data structure and then add it again. Libev tries to be fast, but it's
		1355	still not a constant-time operation.
1324		1356
1325	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.	1357	=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1326		1358
1327	This is the easiest way, and involves using C<ev_timer_again> instead of	1359	This is the easiest way, and involves using C<ev_timer_again> instead of
1328	C<ev_timer_start>.	1360	C<ev_timer_start>.
1329		1361
1330	For this, configure an C<ev_timer> with a C<repeat> value of C<60> and	1362	To implement this, configure an C<ev_timer> with a C<repeat> value
1331	then call C<ev_timer_again> at start and each time you successfully read	1363	of C<60> and then call C<ev_timer_again> at start and each time you
1332	or write some data. If you go into an idle state where you do not expect	1364	successfully read or write some data. If you go into an idle state where
1333	data to travel on the socket, you can C<ev_timer_stop> the timer, and	1365	you do not expect data to travel on the socket, you can C<ev_timer_stop>
1334	C<ev_timer_again> will automatically restart it if need be.	1366	the timer, and C<ev_timer_again> will automatically restart it if need be.
1335		1367
1336	That means you can ignore the C<after> value and C<ev_timer_start>	1368	That means you can ignore both the C<ev_timer_start> function and the
1337	altogether and only ever use the C<repeat> value and C<ev_timer_again>.	1369	C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
		1370	member and C<ev_timer_again>.
1338		1371
1339	At start:	1372	At start:
1340		1373
1341	ev_timer_init (timer, callback, 0., 60.);	1374	ev_timer_init (timer, callback);
		1375	timer->repeat = 60.;
1342	ev_timer_again (loop, timer);	1376	ev_timer_again (loop, timer);
1343		1377
1344	Each time you receive some data:	1378	Each time there is some activity:
1345		1379
1346	ev_timer_again (loop, timer);	1380	ev_timer_again (loop, timer);
1347		1381
1348	It is even possible to change the time-out on the fly:	1382	It is even possible to change the time-out on the fly, regardless of
		1383	whether the watcher is active or not:
1349		1384
1350	timer->repeat = 30.;	1385	timer->repeat = 30.;
1351	ev_timer_again (loop, timer);	1386	ev_timer_again (loop, timer);
1352		1387
1353	This is slightly more efficient then stopping/starting the timer each time	1388	This is slightly more efficient then stopping/starting the timer each time
1354	you want to modify its timeout value, as libev does not have to completely	1389	you want to modify its timeout value, as libev does not have to completely
1355	remove and re-insert the timer from/into it's internal data structure.	1390	remove and re-insert the timer from/into its internal data structure.
		1391
		1392	It is, however, even simpler than the "obvious" way to do it.
1356		1393
1357	=item 3. Let the timer time out, but then re-arm it as required.	1394	=item 3. Let the timer time out, but then re-arm it as required.
1358		1395
1359	This method is more tricky, but usually most efficient: Most timeouts are	1396	This method is more tricky, but usually most efficient: Most timeouts are
1360	relatively long compared to the loop iteration time - in our example,	1397	relatively long compared to the intervals between other activity - in
1361	within 60 seconds, there are usually many I/O events with associated	1398	our example, within 60 seconds, there are usually many I/O events with
1362	activity resets.	1399	associated activity resets.
1363		1400
1364	In this case, it would be more efficient to leave the C<ev_timer> alone,	1401	In this case, it would be more efficient to leave the C<ev_timer> alone,
1365	but remember the time of last activity, and check for a real timeout only	1402	but remember the time of last activity, and check for a real timeout only
1366	within the callback:	1403	within the callback:
1367		1404
1368	ev_tstamp last_activity; // time of last activity	1405	ev_tstamp last_activity; // time of last activity
1369		1406
1370	static void	1407	static void
1371	callback (EV_P_ ev_timer *w, int revents)	1408	callback (EV_P_ ev_timer *w, int revents)
1372	{	1409	{
1373	ev_tstamp now = ev_now (EV_A);	1410	ev_tstamp now = ev_now (EV_A);
1374	ev_tstamp timeout = last_activity + 60.;	1411	ev_tstamp timeout = last_activity + 60.;
1375		1412
1376	// if last_activity is older than now - timeout, we did time out	1413	// if last_activity + 60. is older than now, we did time out
1377	if (timeout < now)	1414	if (timeout < now)
1378	{	1415	{
1379	// timeout occured, take action	1416	// timeout occured, take action
1380	}	1417	}
1381	else	1418	else
1382	{	1419	{
1383	// callback was invoked, but there was some activity, re-arm	1420	// callback was invoked, but there was some activity, re-arm
1384	// to fire in last_activity + 60.	1421	// the watcher to fire in last_activity + 60, which is
		1422	// guaranteed to be in the future, so "again" is positive:
1385	w->again = timeout - now;	1423	w->again = timeout - now;
1386	ev_timer_again (EV_A_ w);	1424	ev_timer_again (EV_A_ w);
1387	}	1425	}
1388	}	1426	}
1389		1427
1390	To summarise the callback: first calculate the real time-out (defined as	1428	To summarise the callback: first calculate the real timeout (defined
1391	"60 seconds after the last activity"), then check if that time has been	1429	as "60 seconds after the last activity"), then check if that time has
1392	reached, which means there was a real timeout. Otherwise the callback was	1430	been reached, which means something I<did>, in fact, time out. Otherwise
1393	invoked too early (timeout is in the future), so re-schedule the timer to	1431	the callback was invoked too early (C<timeout> is in the future), so
1394	fire at that future time.	1432	re-schedule the timer to fire at that future time, to see if maybe we have
		1433	a timeout then.
1395		1434
1396	Note how C<ev_timer_again> is used, taking advantage of the	1435	Note how C<ev_timer_again> is used, taking advantage of the
1397	C<ev_timer_again> optimisation when the timer is already running.	1436	C<ev_timer_again> optimisation when the timer is already running.
1398		1437
1399	This scheme causes more callback invocations (about one every 60 seconds),	1438	This scheme causes more callback invocations (about one every 60 seconds
1400	but virtually no calls to libev to change the timeout.	1439	minus half the average time between activity), but virtually no calls to
		1440	libev to change the timeout.
1401		1441
1402	To start the timer, simply intiialise the watcher and C<last_activity>,	1442	To start the timer, simply initialise the watcher and set C<last_activity>
1403	then call the callback:	1443	to the current time (meaning we just have some activity :), then call the
		1444	callback, which will "do the right thing" and start the timer:
1404		1445
1405	ev_timer_init (timer, callback);	1446	ev_timer_init (timer, callback);
1406	last_activity = ev_now (loop);	1447	last_activity = ev_now (loop);
1407	callback (loop, timer, EV_TIMEOUT);	1448	callback (loop, timer, EV_TIMEOUT);
1408		1449
1409	And when there is some activity, simply remember the time in	1450	And when there is some activity, simply store the current time in
1410	C<last_activity>:	1451	C<last_activity>, no libev calls at all:
1411		1452
1412	last_actiivty = ev_now (loop);	1453	last_actiivty = ev_now (loop);
1413		1454
1414	This technique is slightly more complex, but in most cases where the	1455	This technique is slightly more complex, but in most cases where the
1415	time-out is unlikely to be triggered, much more efficient.	1456	time-out is unlikely to be triggered, much more efficient.
1416		1457
		1458	Changing the timeout is trivial as well (if it isn't hard-coded in the
		1459	callback :) - just change the timeout and invoke the callback, which will
		1460	fix things for you.
		1461
		1462	=item 4. Wee, just use a double-linked list for your timeouts.
		1463
		1464	If there is not one request, but many thousands (millions...), all
		1465	employing some kind of timeout with the same timeout value, then one can
		1466	do even better:
		1467
		1468	When starting the timeout, calculate the timeout value and put the timeout
		1469	at the I<end> of the list.
		1470
		1471	Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
		1472	the list is expected to fire (for example, using the technique #3).
		1473
		1474	When there is some activity, remove the timer from the list, recalculate
		1475	the timeout, append it to the end of the list again, and make sure to
		1476	update the C<ev_timer> if it was taken from the beginning of the list.
		1477
		1478	This way, one can manage an unlimited number of timeouts in O(1) time for
		1479	starting, stopping and updating the timers, at the expense of a major
		1480	complication, and having to use a constant timeout. The constant timeout
		1481	ensures that the list stays sorted.
		1482
1417	=back	1483	=back
		1484
		1485	So which method the best?
		1486
		1487	Method #2 is a simple no-brain-required solution that is adequate in most
		1488	situations. Method #3 requires a bit more thinking, but handles many cases
		1489	better, and isn't very complicated either. In most case, choosing either
		1490	one is fine, with #3 being better in typical situations.
		1491
		1492	Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
		1493	rather complicated, but extremely efficient, something that really pays
		1494	off after the first million or so of active timers, i.e. it's usually
		1495	overkill :)
1418		1496
1419	=head3 The special problem of time updates	1497	=head3 The special problem of time updates
1420		1498
1421	Establishing the current time is a costly operation (it usually takes at	1499	Establishing the current time is a costly operation (it usually takes at
1422	least two system calls): EV therefore updates its idea of the current	1500	least two system calls): EV therefore updates its idea of the current
…		…
1852		1930
1853		1931
1854	=head2 C<ev_stat> - did the file attributes just change?	1932	=head2 C<ev_stat> - did the file attributes just change?
1855		1933
1856	This watches a file system path for attribute changes. That is, it calls	1934	This watches a file system path for attribute changes. That is, it calls
1857	C<stat> regularly (or when the OS says it changed) and sees if it changed	1935	C<stat> on that path in regular intervals (or when the OS says it changed)
1858	compared to the last time, invoking the callback if it did.	1936	and sees if it changed compared to the last time, invoking the callback if
		1937	it did.
1859		1938
1860	The path does not need to exist: changing from "path exists" to "path does	1939	The path does not need to exist: changing from "path exists" to "path does
1861	not exist" is a status change like any other. The condition "path does	1940	not exist" is a status change like any other. The condition "path does not
1862	not exist" is signified by the C<st_nlink> field being zero (which is	1941	exist" (or more correctly "path cannot be stat'ed") is signified by the
1863	otherwise always forced to be at least one) and all the other fields of	1942	C<st_nlink> field being zero (which is otherwise always forced to be at
1864	the stat buffer having unspecified contents.	1943	least one) and all the other fields of the stat buffer having unspecified
		1944	contents.
1865		1945
1866	The path I<should> be absolute and I<must not> end in a slash. If it is	1946	The path I<must not> end in a slash or contain special components such as
		1947	C<.> or C<..>. The path I<should> be absolute: If it is relative and
1867	relative and your working directory changes, the behaviour is undefined.	1948	your working directory changes, then the behaviour is undefined.
1868		1949
1869	Since there is no standard kernel interface to do this, the portable	1950	Since there is no portable change notification interface available, the
1870	implementation simply calls C<stat (2)> regularly on the path to see if	1951	portable implementation simply calls C<stat(2)> regularly on the path
1871	it changed somehow. You can specify a recommended polling interval for	1952	to see if it changed somehow. You can specify a recommended polling
1872	this case. If you specify a polling interval of C<0> (highly recommended!)	1953	interval for this case. If you specify a polling interval of C<0> (highly
1873	then a I<suitable, unspecified default> value will be used (which	1954	recommended!) then a I<suitable, unspecified default> value will be used
1874	you can expect to be around five seconds, although this might change	1955	(which you can expect to be around five seconds, although this might
1875	dynamically). Libev will also impose a minimum interval which is currently	1956	change dynamically). Libev will also impose a minimum interval which is
1876	around C<0.1>, but thats usually overkill.	1957	currently around C<0.1>, but that's usually overkill.
1877		1958
1878	This watcher type is not meant for massive numbers of stat watchers,	1959	This watcher type is not meant for massive numbers of stat watchers,
1879	as even with OS-supported change notifications, this can be	1960	as even with OS-supported change notifications, this can be
1880	resource-intensive.	1961	resource-intensive.
1881		1962
1882	At the time of this writing, the only OS-specific interface implemented	1963	At the time of this writing, the only OS-specific interface implemented
1883	is the Linux inotify interface (implementing kqueue support is left as	1964	is the Linux inotify interface (implementing kqueue support is left as an
1884	an exercise for the reader. Note, however, that the author sees no way	1965	exercise for the reader. Note, however, that the author sees no way of
1885	of implementing C<ev_stat> semantics with kqueue).	1966	implementing C<ev_stat> semantics with kqueue, except as a hint).
1886		1967
1887	=head3 ABI Issues (Largefile Support)	1968	=head3 ABI Issues (Largefile Support)
1888		1969
1889	Libev by default (unless the user overrides this) uses the default	1970	Libev by default (unless the user overrides this) uses the default
1890	compilation environment, which means that on systems with large file	1971	compilation environment, which means that on systems with large file
1891	support disabled by default, you get the 32 bit version of the stat	1972	support disabled by default, you get the 32 bit version of the stat
1892	structure. When using the library from programs that change the ABI to	1973	structure. When using the library from programs that change the ABI to
1893	use 64 bit file offsets the programs will fail. In that case you have to	1974	use 64 bit file offsets the programs will fail. In that case you have to
1894	compile libev with the same flags to get binary compatibility. This is	1975	compile libev with the same flags to get binary compatibility. This is
1895	obviously the case with any flags that change the ABI, but the problem is	1976	obviously the case with any flags that change the ABI, but the problem is
1896	most noticeably disabled with ev_stat and large file support.	1977	most noticeably displayed with ev_stat and large file support.
1897		1978
1898	The solution for this is to lobby your distribution maker to make large	1979	The solution for this is to lobby your distribution maker to make large
1899	file interfaces available by default (as e.g. FreeBSD does) and not	1980	file interfaces available by default (as e.g. FreeBSD does) and not
1900	optional. Libev cannot simply switch on large file support because it has	1981	optional. Libev cannot simply switch on large file support because it has
1901	to exchange stat structures with application programs compiled using the	1982	to exchange stat structures with application programs compiled using the
1902	default compilation environment.	1983	default compilation environment.
1903		1984
1904	=head3 Inotify and Kqueue	1985	=head3 Inotify and Kqueue
1905		1986
1906	When C<inotify (7)> support has been compiled into libev (generally	1987	When C<inotify (7)> support has been compiled into libev and present at
1907	only available with Linux 2.6.25 or above due to bugs in earlier	1988	runtime, it will be used to speed up change detection where possible. The
1908	implementations) and present at runtime, it will be used to speed up	1989	inotify descriptor will be created lazily when the first C<ev_stat>
1909	change detection where possible. The inotify descriptor will be created	1990	watcher is being started.
1910	lazily when the first C<ev_stat> watcher is being started.
1911		1991
1912	Inotify presence does not change the semantics of C<ev_stat> watchers	1992	Inotify presence does not change the semantics of C<ev_stat> watchers
1913	except that changes might be detected earlier, and in some cases, to avoid	1993	except that changes might be detected earlier, and in some cases, to avoid
1914	making regular C<stat> calls. Even in the presence of inotify support	1994	making regular C<stat> calls. Even in the presence of inotify support
1915	there are many cases where libev has to resort to regular C<stat> polling,	1995	there are many cases where libev has to resort to regular C<stat> polling,
1916	but as long as the path exists, libev usually gets away without polling.	1996	but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
		1997	many bugs), the path exists (i.e. stat succeeds), and the path resides on
		1998	a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
		1999	xfs are fully working) libev usually gets away without polling.
1917		2000
1918	There is no support for kqueue, as apparently it cannot be used to	2001	There is no support for kqueue, as apparently it cannot be used to
1919	implement this functionality, due to the requirement of having a file	2002	implement this functionality, due to the requirement of having a file
1920	descriptor open on the object at all times, and detecting renames, unlinks	2003	descriptor open on the object at all times, and detecting renames, unlinks
1921	etc. is difficult.	2004	etc. is difficult.
1922		2005
		2006	=head3 C<stat ()> is a synchronous operation
		2007
		2008	Libev doesn't normally do any kind of I/O itself, and so is not blocking
		2009	the process. The exception are C<ev_stat> watchers - those call C<stat
		2010	()>, which is a synchronous operation.
		2011
		2012	For local paths, this usually doesn't matter: unless the system is very
		2013	busy or the intervals between stat's are large, a stat call will be fast,
		2014	as the path data is suually in memory already (except when starting the
		2015	watcher).
		2016
		2017	For networked file systems, calling C<stat ()> can block an indefinite
		2018	time due to network issues, and even under good conditions, a stat call
		2019	often takes multiple milliseconds.
		2020
		2021	Therefore, it is best to avoid using C<ev_stat> watchers on networked
		2022	paths, although this is fully supported by libev.
		2023
1923	=head3 The special problem of stat time resolution	2024	=head3 The special problem of stat time resolution
1924		2025
1925	The C<stat ()> system call only supports full-second resolution portably, and	2026	The C<stat ()> system call only supports full-second resolution portably,
1926	even on systems where the resolution is higher, most file systems still	2027	and even on systems where the resolution is higher, most file systems
1927	only support whole seconds.	2028	still only support whole seconds.
1928		2029
1929	That means that, if the time is the only thing that changes, you can	2030	That means that, if the time is the only thing that changes, you can
1930	easily miss updates: on the first update, C<ev_stat> detects a change and	2031	easily miss updates: on the first update, C<ev_stat> detects a change and
1931	calls your callback, which does something. When there is another update	2032	calls your callback, which does something. When there is another update
1932	within the same second, C<ev_stat> will be unable to detect unless the	2033	within the same second, C<ev_stat> will be unable to detect unless the
…		…
2571	=over 4	2672	=over 4
2572		2673
2573	=item ev_async_init (ev_async *, callback)	2674	=item ev_async_init (ev_async *, callback)
2574		2675
2575	Initialises and configures the async watcher - it has no parameters of any	2676	Initialises and configures the async watcher - it has no parameters of any
2576	kind. There is a C<ev_asynd_set> macro, but using it is utterly pointless,	2677	kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
2577	trust me.	2678	trust me.
2578		2679
2579	=item ev_async_send (loop, ev_async *)	2680	=item ev_async_send (loop, ev_async *)
2580		2681
2581	Sends/signals/activates the given C<ev_async> watcher, that is, feeds	2682	Sends/signals/activates the given C<ev_async> watcher, that is, feeds
…		…
2900	=item D	3001	=item D
2901		3002
2902	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to	3003	Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
2903	be found at L<http://proj.llucax.com.ar/wiki/evd>.	3004	be found at L<http://proj.llucax.com.ar/wiki/evd>.
2904		3005
		3006	=item Ocaml
		3007
		3008	Erkki Seppala has written Ocaml bindings for libev, to be found at
		3009	L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
		3010
2905	=back	3011	=back
2906		3012
2907		3013
2908	=head1 MACRO MAGIC	3014	=head1 MACRO MAGIC
2909		3015
…		…
3009		3115
3010	#define EV_STANDALONE 1	3116	#define EV_STANDALONE 1
3011	#include "ev.h"	3117	#include "ev.h"
3012		3118
3013	Both header files and implementation files can be compiled with a C++	3119	Both header files and implementation files can be compiled with a C++
3014	compiler (at least, thats a stated goal, and breakage will be treated	3120	compiler (at least, that's a stated goal, and breakage will be treated
3015	as a bug).	3121	as a bug).
3016		3122
3017	You need the following files in your source tree, or in a directory	3123	You need the following files in your source tree, or in a directory
3018	in your include path (e.g. in libev/ when using -Ilibev):	3124	in your include path (e.g. in libev/ when using -Ilibev):
3019		3125
…		…
3491	loop, as long as you don't confuse yourself). The only exception is that	3597	loop, as long as you don't confuse yourself). The only exception is that
3492	you must not do this from C<ev_periodic> reschedule callbacks.	3598	you must not do this from C<ev_periodic> reschedule callbacks.
3493		3599
3494	Care has been taken to ensure that libev does not keep local state inside	3600	Care has been taken to ensure that libev does not keep local state inside
3495	C<ev_loop>, and other calls do not usually allow for coroutine switches as	3601	C<ev_loop>, and other calls do not usually allow for coroutine switches as
3496	they do not clal any callbacks.	3602	they do not call any callbacks.
3497		3603
3498	=head2 COMPILER WARNINGS	3604	=head2 COMPILER WARNINGS
3499		3605
3500	Depending on your compiler and compiler settings, you might get no or a	3606	Depending on your compiler and compiler settings, you might get no or a
3501	lot of warnings when compiling libev code. Some people are apparently	3607	lot of warnings when compiling libev code. Some people are apparently
…		…
3535	==2274== definitely lost: 0 bytes in 0 blocks.	3641	==2274== definitely lost: 0 bytes in 0 blocks.
3536	==2274== possibly lost: 0 bytes in 0 blocks.	3642	==2274== possibly lost: 0 bytes in 0 blocks.
3537	==2274== still reachable: 256 bytes in 1 blocks.	3643	==2274== still reachable: 256 bytes in 1 blocks.
3538		3644
3539	Then there is no memory leak, just as memory accounted to global variables	3645	Then there is no memory leak, just as memory accounted to global variables
3540	is not a memleak - the memory is still being refernced, and didn't leak.	3646	is not a memleak - the memory is still being referenced, and didn't leak.
3541		3647
3542	Similarly, under some circumstances, valgrind might report kernel bugs	3648	Similarly, under some circumstances, valgrind might report kernel bugs
3543	as if it were a bug in libev (e.g. in realloc or in the poll backend,	3649	as if it were a bug in libev (e.g. in realloc or in the poll backend,
3544	although an acceptable workaround has been found here), or it might be	3650	although an acceptable workaround has been found here), or it might be
3545	confused.	3651	confused.
…		…
3783	=back	3889	=back
3784		3890
3785		3891
3786	=head1 AUTHOR	3892	=head1 AUTHOR
3787		3893
3788	Marc Lehmann <libev@schmorp.de>.	3894	Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
3789		3895

Diff Legend

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

Comparing libev/ev.pod (file contents): Revision 1.198 by root, Thu Oct 23 06:30:48 2008 UTC vs. Revision 1.213 by root, Wed Nov 5 02:48:45 2008 UTC

Diff Legend

Comparing libev/ev.pod (file contents):
Revision 1.198 by root, Thu Oct 23 06:30:48 2008 UTC vs.
Revision 1.213 by root, Wed Nov 5 02:48:45 2008 UTC