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

Comparing libev/ev.pod (file contents):
Revision 1.138 by root, Mon Mar 31 01:14:12 2008 UTC vs.
Revision 1.257 by root, Wed Jul 15 16:08:24 2009 UTC

2 2
3libev - a high performance full-featured event loop written in C 3libev - a high performance full-featured event loop written in C
4 4
5=head1 SYNOPSIS 5=head1 SYNOPSIS
6 6
7 #include <ev.h> 7 #include <ev.h>
8 8
9=head2 EXAMPLE PROGRAM 9=head2 EXAMPLE PROGRAM
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 #include <stdio.h> // for puts
15
14 // every watcher type has its own typedef'd struct 16 // every watcher type has its own typedef'd struct
15 // with the name ev_<type> 17 // with the name ev_TYPE
16 ev_io stdin_watcher; 18 ev_io stdin_watcher;
17 ev_timer timeout_watcher; 19 ev_timer timeout_watcher;
18 20
19 // all watcher callbacks have a similar signature 21 // all watcher callbacks have a similar signature
20 // this callback is called when data is readable on stdin 22 // this callback is called when data is readable on stdin
21 static void 23 static void
22 stdin_cb (EV_P_ struct ev_io *w, int revents) 24 stdin_cb (EV_P_ ev_io *w, int revents)
23 { 25 {
24 puts ("stdin ready"); 26 puts ("stdin ready");
25 // for one-shot events, one must manually stop the watcher 27 // for one-shot events, one must manually stop the watcher
26 // with its corresponding stop function. 28 // with its corresponding stop function.
27 ev_io_stop (EV_A_ w); 29 ev_io_stop (EV_A_ w);
28 30
29 // this causes all nested ev_loop's to stop iterating 31 // this causes all nested ev_loop's to stop iterating
30 ev_unloop (EV_A_ EVUNLOOP_ALL); 32 ev_unloop (EV_A_ EVUNLOOP_ALL);
31 } 33 }
32 34
33 // another callback, this time for a time-out 35 // another callback, this time for a time-out
34 static void 36 static void
35 timeout_cb (EV_P_ struct ev_timer *w, int revents) 37 timeout_cb (EV_P_ ev_timer *w, int revents)
36 { 38 {
37 puts ("timeout"); 39 puts ("timeout");
38 // this causes the innermost ev_loop to stop iterating 40 // this causes the innermost ev_loop to stop iterating
39 ev_unloop (EV_A_ EVUNLOOP_ONE); 41 ev_unloop (EV_A_ EVUNLOOP_ONE);
40 } 42 }
41 43
42 int 44 int
43 main (void) 45 main (void)
44 { 46 {
45 // use the default event loop unless you have special needs 47 // use the default event loop unless you have special needs
46 struct ev_loop *loop = ev_default_loop (0); 48 struct ev_loop *loop = ev_default_loop (0);
47 49
48 // initialise an io watcher, then start it 50 // initialise an io watcher, then start it
49 // this one will watch for stdin to become readable 51 // this one will watch for stdin to become readable
50 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ); 52 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
51 ev_io_start (loop, &stdin_watcher); 53 ev_io_start (loop, &stdin_watcher);
52 54
53 // initialise a timer watcher, then start it 55 // initialise a timer watcher, then start it
54 // simple non-repeating 5.5 second timeout 56 // simple non-repeating 5.5 second timeout
55 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.); 57 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
56 ev_timer_start (loop, &timeout_watcher); 58 ev_timer_start (loop, &timeout_watcher);
57 59
58 // now wait for events to arrive 60 // now wait for events to arrive
59 ev_loop (loop, 0); 61 ev_loop (loop, 0);
60 62
61 // unloop was called, so exit 63 // unloop was called, so exit
62 return 0; 64 return 0;
63 } 65 }
64 66
65=head1 DESCRIPTION 67=head1 ABOUT THIS DOCUMENT
68
69This document documents the libev software package.
66 70
67The newest version of this document is also available as an html-formatted 71The newest version of this document is also available as an html-formatted
68web page you might find easier to navigate when reading it for the first 72web page you might find easier to navigate when reading it for the first
69time: L<http://cvs.schmorp.de/libev/ev.html>. 73time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
74
75While this document tries to be as complete as possible in documenting
76libev, its usage and the rationale behind its design, it is not a tutorial
77on event-based programming, nor will it introduce event-based programming
78with libev.
79
80Familarity with event based programming techniques in general is assumed
81throughout this document.
82
83=head1 ABOUT LIBEV
70 84
71Libev is an event loop: you register interest in certain events (such as a 85Libev is an event loop: you register interest in certain events (such as a
72file descriptor being readable or a timeout occurring), and it will manage 86file descriptor being readable or a timeout occurring), and it will manage
73these event sources and provide your program with events. 87these event sources and provide your program with events.
74 88
103Libev is very configurable. In this manual the default (and most common) 117Libev is very configurable. In this manual the default (and most common)
104configuration will be described, which supports multiple event loops. For 118configuration will be described, which supports multiple event loops. For
105more info about various configuration options please have a look at 119more info about various configuration options please have a look at
106B<EMBED> section in this manual. If libev was configured without support 120B<EMBED> section in this manual. If libev was configured without support
107for multiple event loops, then all functions taking an initial argument of 121for multiple event loops, then all functions taking an initial argument of
108name C<loop> (which is always of type C<struct ev_loop *>) will not have 122name C<loop> (which is always of type C<ev_loop *>) will not have
109this argument. 123this argument.
110 124
111=head2 TIME REPRESENTATION 125=head2 TIME REPRESENTATION
112 126
113Libev represents time as a single floating point number, representing the 127Libev represents time as a single floating point number, representing
114(fractional) number of seconds since the (POSIX) epoch (somewhere near 128the (fractional) number of seconds since the (POSIX) epoch (somewhere
115the beginning of 1970, details are complicated, don't ask). This type is 129near the beginning of 1970, details are complicated, don't ask). This
116called C<ev_tstamp>, which is what you should use too. It usually aliases 130type is called C<ev_tstamp>, which is what you should use too. It usually
117to the C<double> type in C, and when you need to do any calculations on 131aliases to the C<double> type in C. When you need to do any calculations
118it, you should treat it as some floatingpoint value. Unlike the name 132on it, you should treat it as some floating point value. Unlike the name
119component C<stamp> might indicate, it is also used for time differences 133component C<stamp> might indicate, it is also used for time differences
120throughout libev. 134throughout libev.
135
136=head1 ERROR HANDLING
137
138Libev knows three classes of errors: operating system errors, usage errors
139and internal errors (bugs).
140
141When libev catches an operating system error it cannot handle (for example
142a system call indicating a condition libev cannot fix), it calls the callback
143set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
144abort. The default is to print a diagnostic message and to call C<abort
145()>.
146
147When libev detects a usage error such as a negative timer interval, then
148it will print a diagnostic message and abort (via the C<assert> mechanism,
149so C<NDEBUG> will disable this checking): these are programming errors in
150the libev caller and need to be fixed there.
151
152Libev also has a few internal error-checking C<assert>ions, and also has
153extensive consistency checking code. These do not trigger under normal
154circumstances, as they indicate either a bug in libev or worse.
155
121 156
122=head1 GLOBAL FUNCTIONS 157=head1 GLOBAL FUNCTIONS
123 158
124These functions can be called anytime, even before initialising the 159These functions can be called anytime, even before initialising the
125library in any way. 160library in any way.
134 169
135=item ev_sleep (ev_tstamp interval) 170=item ev_sleep (ev_tstamp interval)
136 171
137Sleep for the given interval: The current thread will be blocked until 172Sleep for the given interval: The current thread will be blocked until
138either it is interrupted or the given time interval has passed. Basically 173either it is interrupted or the given time interval has passed. Basically
139this is a subsecond-resolution C<sleep ()>. 174this is a sub-second-resolution C<sleep ()>.
140 175
141=item int ev_version_major () 176=item int ev_version_major ()
142 177
143=item int ev_version_minor () 178=item int ev_version_minor ()
144 179
157not a problem. 192not a problem.
158 193
159Example: Make sure we haven't accidentally been linked against the wrong 194Example: Make sure we haven't accidentally been linked against the wrong
160version. 195version.
161 196
162 assert (("libev version mismatch", 197 assert (("libev version mismatch",
163 ev_version_major () == EV_VERSION_MAJOR 198 ev_version_major () == EV_VERSION_MAJOR
164 && ev_version_minor () >= EV_VERSION_MINOR)); 199 && ev_version_minor () >= EV_VERSION_MINOR));
165 200
166=item unsigned int ev_supported_backends () 201=item unsigned int ev_supported_backends ()
167 202
168Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*> 203Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
169value) compiled into this binary of libev (independent of their 204value) compiled into this binary of libev (independent of their
171a description of the set values. 206a description of the set values.
172 207
173Example: make sure we have the epoll method, because yeah this is cool and 208Example: make sure we have the epoll method, because yeah this is cool and
174a must have and can we have a torrent of it please!!!11 209a must have and can we have a torrent of it please!!!11
175 210
176 assert (("sorry, no epoll, no sex", 211 assert (("sorry, no epoll, no sex",
177 ev_supported_backends () & EVBACKEND_EPOLL)); 212 ev_supported_backends () & EVBACKEND_EPOLL));
178 213
179=item unsigned int ev_recommended_backends () 214=item unsigned int ev_recommended_backends ()
180 215
181Return the set of all backends compiled into this binary of libev and also 216Return the set of all backends compiled into this binary of libev and also
182recommended for this platform. This set is often smaller than the one 217recommended for this platform. This set is often smaller than the one
183returned by C<ev_supported_backends>, as for example kqueue is broken on 218returned by C<ev_supported_backends>, as for example kqueue is broken on
184most BSDs and will not be autodetected unless you explicitly request it 219most BSDs and will not be auto-detected unless you explicitly request it
185(assuming you know what you are doing). This is the set of backends that 220(assuming you know what you are doing). This is the set of backends that
186libev will probe for if you specify no backends explicitly. 221libev will probe for if you specify no backends explicitly.
187 222
188=item unsigned int ev_embeddable_backends () 223=item unsigned int ev_embeddable_backends ()
189 224
193C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for 228C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
194recommended ones. 229recommended ones.
195 230
196See the description of C<ev_embed> watchers for more info. 231See the description of C<ev_embed> watchers for more info.
197 232
198=item ev_set_allocator (void *(*cb)(void *ptr, long size)) 233=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
199 234
200Sets the allocation function to use (the prototype is similar - the 235Sets the allocation function to use (the prototype is similar - the
201semantics is identical - to the realloc C function). It is used to 236semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
202allocate and free memory (no surprises here). If it returns zero when 237used to allocate and free memory (no surprises here). If it returns zero
203memory needs to be allocated, the library might abort or take some 238when memory needs to be allocated (C<size != 0>), the library might abort
204potentially destructive action. The default is your system realloc 239or take some potentially destructive action.
205function. 240
241Since some systems (at least OpenBSD and Darwin) fail to implement
242correct C<realloc> semantics, libev will use a wrapper around the system
243C<realloc> and C<free> functions by default.
206 244
207You could override this function in high-availability programs to, say, 245You could override this function in high-availability programs to, say,
208free some memory if it cannot allocate memory, to use a special allocator, 246free some memory if it cannot allocate memory, to use a special allocator,
209or even to sleep a while and retry until some memory is available. 247or even to sleep a while and retry until some memory is available.
210 248
211Example: Replace the libev allocator with one that waits a bit and then 249Example: Replace the libev allocator with one that waits a bit and then
212retries). 250retries (example requires a standards-compliant C<realloc>).
213 251
214 static void * 252 static void *
215 persistent_realloc (void *ptr, size_t size) 253 persistent_realloc (void *ptr, size_t size)
216 { 254 {
217 for (;;) 255 for (;;)
226 } 264 }
227 265
228 ... 266 ...
229 ev_set_allocator (persistent_realloc); 267 ev_set_allocator (persistent_realloc);
230 268
231=item ev_set_syserr_cb (void (*cb)(const char *msg)); 269=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
232 270
233Set the callback function to call on a retryable syscall error (such 271Set the callback function to call on a retryable system call error (such
234as failed select, poll, epoll_wait). The message is a printable string 272as failed select, poll, epoll_wait). The message is a printable string
235indicating the system call or subsystem causing the problem. If this 273indicating the system call or subsystem causing the problem. If this
236callback is set, then libev will expect it to remedy the sitution, no 274callback is set, then libev will expect it to remedy the situation, no
237matter what, when it returns. That is, libev will generally retry the 275matter what, when it returns. That is, libev will generally retry the
238requested operation, or, if the condition doesn't go away, do bad stuff 276requested operation, or, if the condition doesn't go away, do bad stuff
239(such as abort). 277(such as abort).
240 278
241Example: This is basically the same thing that libev does internally, too. 279Example: This is basically the same thing that libev does internally, too.
252 290
253=back 291=back
254 292
255=head1 FUNCTIONS CONTROLLING THE EVENT LOOP 293=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
256 294
257An event loop is described by a C<struct ev_loop *>. The library knows two 295An event loop is described by a C<struct ev_loop *> (the C<struct>
258types of such loops, the I<default> loop, which supports signals and child 296is I<not> optional in this case, as there is also an C<ev_loop>
259events, and dynamically created loops which do not. 297I<function>).
260 298
261If you use threads, a common model is to run the default event loop 299The library knows two types of such loops, the I<default> loop, which
262in your main thread (or in a separate thread) and for each thread you 300supports signals and child events, and dynamically created loops which do
263create, you also create another event loop. Libev itself does no locking 301not.
264whatsoever, so if you mix calls to the same event loop in different
265threads, make sure you lock (this is usually a bad idea, though, even if
266done correctly, because it's hideous and inefficient).
267 302
268=over 4 303=over 4
269 304
270=item struct ev_loop *ev_default_loop (unsigned int flags) 305=item struct ev_loop *ev_default_loop (unsigned int flags)
271 306
275flags. If that is troubling you, check C<ev_backend ()> afterwards). 310flags. If that is troubling you, check C<ev_backend ()> afterwards).
276 311
277If you don't know what event loop to use, use the one returned from this 312If you don't know what event loop to use, use the one returned from this
278function. 313function.
279 314
315Note that this function is I<not> thread-safe, so if you want to use it
316from multiple threads, you have to lock (note also that this is unlikely,
317as loops cannot be shared easily between threads anyway).
318
280The default loop is the only loop that can handle C<ev_signal> and 319The default loop is the only loop that can handle C<ev_signal> and
281C<ev_child> watchers, and to do this, it always registers a handler 320C<ev_child> watchers, and to do this, it always registers a handler
282for C<SIGCHLD>. If this is a problem for your app you can either 321for C<SIGCHLD>. If this is a problem for your application you can either
283create a dynamic loop with C<ev_loop_new> that doesn't do that, or you 322create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
284can simply overwrite the C<SIGCHLD> signal handler I<after> calling 323can simply overwrite the C<SIGCHLD> signal handler I<after> calling
285C<ev_default_init>. 324C<ev_default_init>.
286 325
287The flags argument can be used to specify special behaviour or specific 326The flags argument can be used to specify special behaviour or specific
296The default flags value. Use this if you have no clue (it's the right 335The default flags value. Use this if you have no clue (it's the right
297thing, believe me). 336thing, believe me).
298 337
299=item C<EVFLAG_NOENV> 338=item C<EVFLAG_NOENV>
300 339
301If this flag bit is ored into the flag value (or the program runs setuid 340If this flag bit is or'ed into the flag value (or the program runs setuid
302or setgid) then libev will I<not> look at the environment variable 341or setgid) then libev will I<not> look at the environment variable
303C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will 342C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
304override the flags completely if it is found in the environment. This is 343override the flags completely if it is found in the environment. This is
305useful to try out specific backends to test their performance, or to work 344useful to try out specific backends to test their performance, or to work
306around bugs. 345around bugs.
313 352
314This works by calling C<getpid ()> on every iteration of the loop, 353This works by calling C<getpid ()> on every iteration of the loop,
315and thus this might slow down your event loop if you do a lot of loop 354and thus this might slow down your event loop if you do a lot of loop
316iterations and little real work, but is usually not noticeable (on my 355iterations and little real work, but is usually not noticeable (on my
317GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence 356GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
318without a syscall and thus I<very> fast, but my GNU/Linux system also has 357without a system call and thus I<very> fast, but my GNU/Linux system also has
319C<pthread_atfork> which is even faster). 358C<pthread_atfork> which is even faster).
320 359
321The big advantage of this flag is that you can forget about fork (and 360The big advantage of this flag is that you can forget about fork (and
322forget about forgetting to tell libev about forking) when you use this 361forget about forgetting to tell libev about forking) when you use this
323flag. 362flag.
324 363
325This flag setting cannot be overriden or specified in the C<LIBEV_FLAGS> 364This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
326environment variable. 365environment variable.
327 366
328=item C<EVBACKEND_SELECT> (value 1, portable select backend) 367=item C<EVBACKEND_SELECT> (value 1, portable select backend)
329 368
330This is your standard select(2) backend. Not I<completely> standard, as 369This is your standard select(2) backend. Not I<completely> standard, as
332but if that fails, expect a fairly low limit on the number of fds when 371but if that fails, expect a fairly low limit on the number of fds when
333using this backend. It doesn't scale too well (O(highest_fd)), but its 372using this backend. It doesn't scale too well (O(highest_fd)), but its
334usually the fastest backend for a low number of (low-numbered :) fds. 373usually the fastest backend for a low number of (low-numbered :) fds.
335 374
336To get good performance out of this backend you need a high amount of 375To get good performance out of this backend you need a high amount of
337parallelity (most of the file descriptors should be busy). If you are 376parallelism (most of the file descriptors should be busy). If you are
338writing a server, you should C<accept ()> in a loop to accept as many 377writing a server, you should C<accept ()> in a loop to accept as many
339connections as possible during one iteration. You might also want to have 378connections as possible during one iteration. You might also want to have
340a look at C<ev_set_io_collect_interval ()> to increase the amount of 379a look at C<ev_set_io_collect_interval ()> to increase the amount of
341readyness notifications you get per iteration. 380readiness notifications you get per iteration.
381
382This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
383C<writefds> set (and to work around Microsoft Windows bugs, also onto the
384C<exceptfds> set on that platform).
342 385
343=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows) 386=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
344 387
345And this is your standard poll(2) backend. It's more complicated 388And this is your standard poll(2) backend. It's more complicated
346than select, but handles sparse fds better and has no artificial 389than select, but handles sparse fds better and has no artificial
347limit on the number of fds you can use (except it will slow down 390limit on the number of fds you can use (except it will slow down
348considerably with a lot of inactive fds). It scales similarly to select, 391considerably with a lot of inactive fds). It scales similarly to select,
349i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for 392i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
350performance tips. 393performance tips.
351 394
395This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
396C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
397
352=item C<EVBACKEND_EPOLL> (value 4, Linux) 398=item C<EVBACKEND_EPOLL> (value 4, Linux)
353 399
354For few fds, this backend is a bit little slower than poll and select, 400For few fds, this backend is a bit little slower than poll and select,
355but it scales phenomenally better. While poll and select usually scale 401but it scales phenomenally better. While poll and select usually scale
356like O(total_fds) where n is the total number of fds (or the highest fd), 402like O(total_fds) where n is the total number of fds (or the highest fd),
357epoll scales either O(1) or O(active_fds). The epoll design has a number 403epoll scales either O(1) or O(active_fds).
358of shortcomings, such as silently dropping events in some hard-to-detect 404
359cases and rewiring a syscall per fd change, no fork support and bad 405The epoll mechanism deserves honorable mention as the most misdesigned
360support for dup. 406of the more advanced event mechanisms: mere annoyances include silently
407dropping file descriptors, requiring a system call per change per file
408descriptor (and unnecessary guessing of parameters), problems with dup and
409so on. The biggest issue is fork races, however - if a program forks then
410I<both> parent and child process have to recreate the epoll set, which can
411take considerable time (one syscall per file descriptor) and is of course
412hard to detect.
413
414Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
415of course I<doesn't>, and epoll just loves to report events for totally
416I<different> file descriptors (even already closed ones, so one cannot
417even remove them from the set) than registered in the set (especially
418on SMP systems). Libev tries to counter these spurious notifications by
419employing an additional generation counter and comparing that against the
420events to filter out spurious ones, recreating the set when required.
361 421
362While stopping, setting and starting an I/O watcher in the same iteration 422While stopping, setting and starting an I/O watcher in the same iteration
363will result in some caching, there is still a syscall per such incident 423will result in some caching, there is still a system call per such
364(because the fd could point to a different file description now), so its 424incident (because the same I<file descriptor> could point to a different
365best to avoid that. Also, C<dup ()>'ed file descriptors might not work 425I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
366very well if you register events for both fds. 426file descriptors might not work very well if you register events for both
367 427file descriptors.
368Please note that epoll sometimes generates spurious notifications, so you
369need to use non-blocking I/O or other means to avoid blocking when no data
370(or space) is available.
371 428
372Best performance from this backend is achieved by not unregistering all 429Best performance from this backend is achieved by not unregistering all
373watchers for a file descriptor until it has been closed, if possible, i.e. 430watchers for a file descriptor until it has been closed, if possible,
374keep at least one watcher active per fd at all times. 431i.e. keep at least one watcher active per fd at all times. Stopping and
432starting a watcher (without re-setting it) also usually doesn't cause
433extra overhead. A fork can both result in spurious notifications as well
434as in libev having to destroy and recreate the epoll object, which can
435take considerable time and thus should be avoided.
375 436
437All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
438faster than epoll for maybe up to a hundred file descriptors, depending on
439the usage. So sad.
440
376While nominally embeddeble in other event loops, this feature is broken in 441While nominally embeddable in other event loops, this feature is broken in
377all kernel versions tested so far. 442all kernel versions tested so far.
443
444This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
445C<EVBACKEND_POLL>.
378 446
379=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones) 447=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
380 448
381Kqueue deserves special mention, as at the time of this writing, it 449Kqueue deserves special mention, as at the time of this writing, it
382was broken on all BSDs except NetBSD (usually it doesn't work reliably 450was broken on all BSDs except NetBSD (usually it doesn't work reliably
383with anything but sockets and pipes, except on Darwin, where of course 451with anything but sockets and pipes, except on Darwin, where of course
384it's completely useless). For this reason it's not being "autodetected" 452it's completely useless). Unlike epoll, however, whose brokenness
453is by design, these kqueue bugs can (and eventually will) be fixed
454without API changes to existing programs. For this reason it's not being
385unless you explicitly specify it explicitly in the flags (i.e. using 455"auto-detected" unless you explicitly specify it in the flags (i.e. using
386C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough) 456C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
387system like NetBSD. 457system like NetBSD.
388 458
389You still can embed kqueue into a normal poll or select backend and use it 459You still can embed kqueue into a normal poll or select backend and use it
390only for sockets (after having made sure that sockets work with kqueue on 460only for sockets (after having made sure that sockets work with kqueue on
391the target platform). See C<ev_embed> watchers for more info. 461the target platform). See C<ev_embed> watchers for more info.
392 462
393It scales in the same way as the epoll backend, but the interface to the 463It scales in the same way as the epoll backend, but the interface to the
394kernel is more efficient (which says nothing about its actual speed, of 464kernel is more efficient (which says nothing about its actual speed, of
395course). While stopping, setting and starting an I/O watcher does never 465course). While stopping, setting and starting an I/O watcher does never
396cause an extra syscall as with C<EVBACKEND_EPOLL>, it still adds up to 466cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
397two event changes per incident, support for C<fork ()> is very bad and it 467two event changes per incident. Support for C<fork ()> is very bad (but
398drops fds silently in similarly hard-to-detect cases. 468sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
469cases
399 470
400This backend usually performs well under most conditions. 471This backend usually performs well under most conditions.
401 472
402While nominally embeddable in other event loops, this doesn't work 473While nominally embeddable in other event loops, this doesn't work
403everywhere, so you might need to test for this. And since it is broken 474everywhere, so you might need to test for this. And since it is broken
404almost everywhere, you should only use it when you have a lot of sockets 475almost everywhere, you should only use it when you have a lot of sockets
405(for which it usually works), by embedding it into another event loop 476(for which it usually works), by embedding it into another event loop
406(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>) and using it only for 477(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
407sockets. 478also broken on OS X)) and, did I mention it, using it only for sockets.
479
480This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
481C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
482C<NOTE_EOF>.
408 483
409=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8) 484=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
410 485
411This is not implemented yet (and might never be, unless you send me an 486This is not implemented yet (and might never be, unless you send me an
412implementation). According to reports, C</dev/poll> only supports sockets 487implementation). According to reports, C</dev/poll> only supports sockets
416=item C<EVBACKEND_PORT> (value 32, Solaris 10) 491=item C<EVBACKEND_PORT> (value 32, Solaris 10)
417 492
418This uses the Solaris 10 event port mechanism. As with everything on Solaris, 493This uses the Solaris 10 event port mechanism. As with everything on Solaris,
419it's really slow, but it still scales very well (O(active_fds)). 494it's really slow, but it still scales very well (O(active_fds)).
420 495
421Please note that solaris event ports can deliver a lot of spurious 496Please note that Solaris event ports can deliver a lot of spurious
422notifications, so you need to use non-blocking I/O or other means to avoid 497notifications, so you need to use non-blocking I/O or other means to avoid
423blocking when no data (or space) is available. 498blocking when no data (or space) is available.
424 499
425While this backend scales well, it requires one system call per active 500While this backend scales well, it requires one system call per active
426file descriptor per loop iteration. For small and medium numbers of file 501file descriptor per loop iteration. For small and medium numbers of file
427descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend 502descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
428might perform better. 503might perform better.
429 504
430On the positive side, ignoring the spurious readyness notifications, this 505On the positive side, with the exception of the spurious readiness
431backend actually performed to specification in all tests and is fully 506notifications, this backend actually performed fully to specification
432embeddable, which is a rare feat among the OS-specific backends. 507in all tests and is fully embeddable, which is a rare feat among the
508OS-specific backends (I vastly prefer correctness over speed hacks).
509
510This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
511C<EVBACKEND_POLL>.
433 512
434=item C<EVBACKEND_ALL> 513=item C<EVBACKEND_ALL>
435 514
436Try all backends (even potentially broken ones that wouldn't be tried 515Try all backends (even potentially broken ones that wouldn't be tried
437with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as 516with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
439 518
440It is definitely not recommended to use this flag. 519It is definitely not recommended to use this flag.
441 520
442=back 521=back
443 522
444If one or more of these are ored into the flags value, then only these 523If one or more of these are or'ed into the flags value, then only these
445backends will be tried (in the reverse order as listed here). If none are 524backends will be tried (in the reverse order as listed here). If none are
446specified, all backends in C<ev_recommended_backends ()> will be tried. 525specified, all backends in C<ev_recommended_backends ()> will be tried.
447 526
448The most typical usage is like this: 527Example: This is the most typical usage.
449 528
450 if (!ev_default_loop (0)) 529 if (!ev_default_loop (0))
451 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?"); 530 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
452 531
453Restrict libev to the select and poll backends, and do not allow 532Example: Restrict libev to the select and poll backends, and do not allow
454environment settings to be taken into account: 533environment settings to be taken into account:
455 534
456 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV); 535 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
457 536
458Use whatever libev has to offer, but make sure that kqueue is used if 537Example: Use whatever libev has to offer, but make sure that kqueue is
459available (warning, breaks stuff, best use only with your own private 538used if available (warning, breaks stuff, best use only with your own
460event loop and only if you know the OS supports your types of fds): 539private event loop and only if you know the OS supports your types of
540fds):
461 541
462 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE); 542 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
463 543
464=item struct ev_loop *ev_loop_new (unsigned int flags) 544=item struct ev_loop *ev_loop_new (unsigned int flags)
465 545
466Similar to C<ev_default_loop>, but always creates a new event loop that is 546Similar to C<ev_default_loop>, but always creates a new event loop that is
467always distinct from the default loop. Unlike the default loop, it cannot 547always distinct from the default loop. Unlike the default loop, it cannot
468handle signal and child watchers, and attempts to do so will be greeted by 548handle signal and child watchers, and attempts to do so will be greeted by
469undefined behaviour (or a failed assertion if assertions are enabled). 549undefined behaviour (or a failed assertion if assertions are enabled).
470 550
551Note that this function I<is> thread-safe, and the recommended way to use
552libev with threads is indeed to create one loop per thread, and using the
553default loop in the "main" or "initial" thread.
554
471Example: Try to create a event loop that uses epoll and nothing else. 555Example: Try to create a event loop that uses epoll and nothing else.
472 556
473 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV); 557 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
474 if (!epoller) 558 if (!epoller)
475 fatal ("no epoll found here, maybe it hides under your chair"); 559 fatal ("no epoll found here, maybe it hides under your chair");
476 560
477=item ev_default_destroy () 561=item ev_default_destroy ()
478 562
479Destroys the default loop again (frees all memory and kernel state 563Destroys the default loop again (frees all memory and kernel state
480etc.). None of the active event watchers will be stopped in the normal 564etc.). None of the active event watchers will be stopped in the normal
481sense, so e.g. C<ev_is_active> might still return true. It is your 565sense, so e.g. C<ev_is_active> might still return true. It is your
482responsibility to either stop all watchers cleanly yoursef I<before> 566responsibility to either stop all watchers cleanly yourself I<before>
483calling this function, or cope with the fact afterwards (which is usually 567calling this function, or cope with the fact afterwards (which is usually
484the easiest thing, you can just ignore the watchers and/or C<free ()> them 568the easiest thing, you can just ignore the watchers and/or C<free ()> them
485for example). 569for example).
486 570
487Note that certain global state, such as signal state, will not be freed by 571Note that certain global state, such as signal state (and installed signal
488this function, and related watchers (such as signal and child watchers) 572handlers), will not be freed by this function, and related watchers (such
489would need to be stopped manually. 573as signal and child watchers) would need to be stopped manually.
490 574
491In general it is not advisable to call this function except in the 575In general it is not advisable to call this function except in the
492rare occasion where you really need to free e.g. the signal handling 576rare occasion where you really need to free e.g. the signal handling
493pipe fds. If you need dynamically allocated loops it is better to use 577pipe fds. If you need dynamically allocated loops it is better to use
494C<ev_loop_new> and C<ev_loop_destroy>). 578C<ev_loop_new> and C<ev_loop_destroy>).
519 603
520=item ev_loop_fork (loop) 604=item ev_loop_fork (loop)
521 605
522Like C<ev_default_fork>, but acts on an event loop created by 606Like C<ev_default_fork>, but acts on an event loop created by
523C<ev_loop_new>. Yes, you have to call this on every allocated event loop 607C<ev_loop_new>. Yes, you have to call this on every allocated event loop
524after fork, and how you do this is entirely your own problem. 608after fork that you want to re-use in the child, and how you do this is
609entirely your own problem.
525 610
526=item int ev_is_default_loop (loop) 611=item int ev_is_default_loop (loop)
527 612
528Returns true when the given loop actually is the default loop, false otherwise. 613Returns true when the given loop is, in fact, the default loop, and false
614otherwise.
529 615
530=item unsigned int ev_loop_count (loop) 616=item unsigned int ev_loop_count (loop)
531 617
532Returns the count of loop iterations for the loop, which is identical to 618Returns the count of loop iterations for the loop, which is identical to
533the number of times libev did poll for new events. It starts at C<0> and 619the number of times libev did poll for new events. It starts at C<0> and
534happily wraps around with enough iterations. 620happily wraps around with enough iterations.
535 621
536This value can sometimes be useful as a generation counter of sorts (it 622This value can sometimes be useful as a generation counter of sorts (it
537"ticks" the number of loop iterations), as it roughly corresponds with 623"ticks" the number of loop iterations), as it roughly corresponds with
538C<ev_prepare> and C<ev_check> calls. 624C<ev_prepare> and C<ev_check> calls.
625
626=item unsigned int ev_loop_depth (loop)
627
628Returns the number of times C<ev_loop> was entered minus the number of
629times C<ev_loop> was exited, in other words, the recursion depth.
630
631Outside C<ev_loop>, this number is zero. In a callback, this number is
632C<1>, unless C<ev_loop> was invoked recursively (or from another thread),
633in which case it is higher.
634
635Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread
636etc.), doesn't count as exit.
539 637
540=item unsigned int ev_backend (loop) 638=item unsigned int ev_backend (loop)
541 639
542Returns one of the C<EVBACKEND_*> flags indicating the event backend in 640Returns one of the C<EVBACKEND_*> flags indicating the event backend in
543use. 641use.
548received events and started processing them. This timestamp does not 646received events and started processing them. This timestamp does not
549change as long as callbacks are being processed, and this is also the base 647change as long as callbacks are being processed, and this is also the base
550time used for relative timers. You can treat it as the timestamp of the 648time used for relative timers. You can treat it as the timestamp of the
551event occurring (or more correctly, libev finding out about it). 649event occurring (or more correctly, libev finding out about it).
552 650
651=item ev_now_update (loop)
652
653Establishes the current time by querying the kernel, updating the time
654returned by C<ev_now ()> in the progress. This is a costly operation and
655is usually done automatically within C<ev_loop ()>.
656
657This function is rarely useful, but when some event callback runs for a
658very long time without entering the event loop, updating libev's idea of
659the current time is a good idea.
660
661See also L<The special problem of time updates> in the C<ev_timer> section.
662
663=item ev_suspend (loop)
664
665=item ev_resume (loop)
666
667These two functions suspend and resume a loop, for use when the loop is
668not used for a while and timeouts should not be processed.
669
670A typical use case would be an interactive program such as a game: When
671the user presses C<^Z> to suspend the game and resumes it an hour later it
672would be best to handle timeouts as if no time had actually passed while
673the program was suspended. This can be achieved by calling C<ev_suspend>
674in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
675C<ev_resume> directly afterwards to resume timer processing.
676
677Effectively, all C<ev_timer> watchers will be delayed by the time spend
678between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
679will be rescheduled (that is, they will lose any events that would have
680occured while suspended).
681
682After calling C<ev_suspend> you B<must not> call I<any> function on the
683given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
684without a previous call to C<ev_suspend>.
685
686Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
687event loop time (see C<ev_now_update>).
688
553=item ev_loop (loop, int flags) 689=item ev_loop (loop, int flags)
554 690
555Finally, this is it, the event handler. This function usually is called 691Finally, this is it, the event handler. This function usually is called
556after you initialised all your watchers and you want to start handling 692after you initialised all your watchers and you want to start handling
557events. 693events.
559If the flags argument is specified as C<0>, it will not return until 695If the flags argument is specified as C<0>, it will not return until
560either no event watchers are active anymore or C<ev_unloop> was called. 696either no event watchers are active anymore or C<ev_unloop> was called.
561 697
562Please note that an explicit C<ev_unloop> is usually better than 698Please note that an explicit C<ev_unloop> is usually better than
563relying on all watchers to be stopped when deciding when a program has 699relying on all watchers to be stopped when deciding when a program has
564finished (especially in interactive programs), but having a program that 700finished (especially in interactive programs), but having a program
565automatically loops as long as it has to and no longer by virtue of 701that automatically loops as long as it has to and no longer by virtue
566relying on its watchers stopping correctly is a thing of beauty. 702of relying on its watchers stopping correctly, that is truly a thing of
703beauty.
567 704
568A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle 705A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
569those events and any outstanding ones, but will not block your process in 706those events and any already outstanding ones, but will not block your
570case there are no events and will return after one iteration of the loop. 707process in case there are no events and will return after one iteration of
708the loop.
571 709
572A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if 710A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
573neccessary) and will handle those and any outstanding ones. It will block 711necessary) and will handle those and any already outstanding ones. It
574your process until at least one new event arrives, and will return after 712will block your process until at least one new event arrives (which could
575one iteration of the loop. This is useful if you are waiting for some 713be an event internal to libev itself, so there is no guarantee that a
576external event in conjunction with something not expressible using other 714user-registered callback will be called), and will return after one
715iteration of the loop.
716
717This is useful if you are waiting for some external event in conjunction
718with something not expressible using other libev watchers (i.e. "roll your
577libev watchers. However, a pair of C<ev_prepare>/C<ev_check> watchers is 719own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
578usually a better approach for this kind of thing. 720usually a better approach for this kind of thing.
579 721
580Here are the gory details of what C<ev_loop> does: 722Here are the gory details of what C<ev_loop> does:
581 723
582 - Before the first iteration, call any pending watchers. 724 - Before the first iteration, call any pending watchers.
583 * If EVFLAG_FORKCHECK was used, check for a fork. 725 * If EVFLAG_FORKCHECK was used, check for a fork.
584 - If a fork was detected, queue and call all fork watchers. 726 - If a fork was detected (by any means), queue and call all fork watchers.
585 - Queue and call all prepare watchers. 727 - Queue and call all prepare watchers.
586 - If we have been forked, recreate the kernel state. 728 - If we have been forked, detach and recreate the kernel state
729 as to not disturb the other process.
587 - Update the kernel state with all outstanding changes. 730 - Update the kernel state with all outstanding changes.
588 - Update the "event loop time". 731 - Update the "event loop time" (ev_now ()).
589 - Calculate for how long to sleep or block, if at all 732 - Calculate for how long to sleep or block, if at all
590 (active idle watchers, EVLOOP_NONBLOCK or not having 733 (active idle watchers, EVLOOP_NONBLOCK or not having
591 any active watchers at all will result in not sleeping). 734 any active watchers at all will result in not sleeping).
592 - Sleep if the I/O and timer collect interval say so. 735 - Sleep if the I/O and timer collect interval say so.
593 - Block the process, waiting for any events. 736 - Block the process, waiting for any events.
594 - Queue all outstanding I/O (fd) events. 737 - Queue all outstanding I/O (fd) events.
595 - Update the "event loop time" and do time jump handling. 738 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
596 - Queue all outstanding timers. 739 - Queue all expired timers.
597 - Queue all outstanding periodics. 740 - Queue all expired periodics.
598 - If no events are pending now, queue all idle watchers. 741 - Unless any events are pending now, queue all idle watchers.
599 - Queue all check watchers. 742 - Queue all check watchers.
600 - Call all queued watchers in reverse order (i.e. check watchers first). 743 - Call all queued watchers in reverse order (i.e. check watchers first).
601 Signals and child watchers are implemented as I/O watchers, and will 744 Signals and child watchers are implemented as I/O watchers, and will
602 be handled here by queueing them when their watcher gets executed. 745 be handled here by queueing them when their watcher gets executed.
603 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK 746 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
608anymore. 751anymore.
609 752
610 ... queue jobs here, make sure they register event watchers as long 753 ... queue jobs here, make sure they register event watchers as long
611 ... as they still have work to do (even an idle watcher will do..) 754 ... as they still have work to do (even an idle watcher will do..)
612 ev_loop (my_loop, 0); 755 ev_loop (my_loop, 0);
613 ... jobs done. yeah! 756 ... jobs done or somebody called unloop. yeah!
614 757
615=item ev_unloop (loop, how) 758=item ev_unloop (loop, how)
616 759
617Can be used to make a call to C<ev_loop> return early (but only after it 760Can be used to make a call to C<ev_loop> return early (but only after it
618has processed all outstanding events). The C<how> argument must be either 761has processed all outstanding events). The C<how> argument must be either
619C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or 762C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
620C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return. 763C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
621 764
622This "unloop state" will be cleared when entering C<ev_loop> again. 765This "unloop state" will be cleared when entering C<ev_loop> again.
623 766
767It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
768
624=item ev_ref (loop) 769=item ev_ref (loop)
625 770
626=item ev_unref (loop) 771=item ev_unref (loop)
627 772
628Ref/unref can be used to add or remove a reference count on the event 773Ref/unref can be used to add or remove a reference count on the event
629loop: Every watcher keeps one reference, and as long as the reference 774loop: Every watcher keeps one reference, and as long as the reference
630count is nonzero, C<ev_loop> will not return on its own. If you have 775count is nonzero, C<ev_loop> will not return on its own.
776
631a watcher you never unregister that should not keep C<ev_loop> from 777If you have a watcher you never unregister that should not keep C<ev_loop>
632returning, ev_unref() after starting, and ev_ref() before stopping it. For 778from returning, call ev_unref() after starting, and ev_ref() before
779stopping it.
780
633example, libev itself uses this for its internal signal pipe: It is not 781As an example, libev itself uses this for its internal signal pipe: It
634visible to the libev user and should not keep C<ev_loop> from exiting if 782is not visible to the libev user and should not keep C<ev_loop> from
635no event watchers registered by it are active. It is also an excellent 783exiting if no event watchers registered by it are active. It is also an
636way to do this for generic recurring timers or from within third-party 784excellent way to do this for generic recurring timers or from within
637libraries. Just remember to I<unref after start> and I<ref before stop> 785third-party libraries. Just remember to I<unref after start> and I<ref
638(but only if the watcher wasn't active before, or was active before, 786before stop> (but only if the watcher wasn't active before, or was active
639respectively). 787before, respectively. Note also that libev might stop watchers itself
788(e.g. non-repeating timers) in which case you have to C<ev_ref>
789in the callback).
640 790
641Example: Create a signal watcher, but keep it from keeping C<ev_loop> 791Example: Create a signal watcher, but keep it from keeping C<ev_loop>
642running when nothing else is active. 792running when nothing else is active.
643 793
644 struct ev_signal exitsig; 794 ev_signal exitsig;
645 ev_signal_init (&exitsig, sig_cb, SIGINT); 795 ev_signal_init (&exitsig, sig_cb, SIGINT);
646 ev_signal_start (loop, &exitsig); 796 ev_signal_start (loop, &exitsig);
647 evf_unref (loop); 797 evf_unref (loop);
648 798
649Example: For some weird reason, unregister the above signal handler again. 799Example: For some weird reason, unregister the above signal handler again.
650 800
651 ev_ref (loop); 801 ev_ref (loop);
652 ev_signal_stop (loop, &exitsig); 802 ev_signal_stop (loop, &exitsig);
653 803
654=item ev_set_io_collect_interval (loop, ev_tstamp interval) 804=item ev_set_io_collect_interval (loop, ev_tstamp interval)
655 805
656=item ev_set_timeout_collect_interval (loop, ev_tstamp interval) 806=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
657 807
658These advanced functions influence the time that libev will spend waiting 808These advanced functions influence the time that libev will spend waiting
659for events. Both are by default C<0>, meaning that libev will try to 809for events. Both time intervals are by default C<0>, meaning that libev
660invoke timer/periodic callbacks and I/O callbacks with minimum latency. 810will try to invoke timer/periodic callbacks and I/O callbacks with minimum
811latency.
661 812
662Setting these to a higher value (the C<interval> I<must> be >= C<0>) 813Setting these to a higher value (the C<interval> I<must> be >= C<0>)
663allows libev to delay invocation of I/O and timer/periodic callbacks to 814allows libev to delay invocation of I/O and timer/periodic callbacks
664increase efficiency of loop iterations. 815to increase efficiency of loop iterations (or to increase power-saving
816opportunities).
665 817
666The background is that sometimes your program runs just fast enough to 818The idea is that sometimes your program runs just fast enough to handle
667handle one (or very few) event(s) per loop iteration. While this makes 819one (or very few) event(s) per loop iteration. While this makes the
668the program responsive, it also wastes a lot of CPU time to poll for new 820program responsive, it also wastes a lot of CPU time to poll for new
669events, especially with backends like C<select ()> which have a high 821events, especially with backends like C<select ()> which have a high
670overhead for the actual polling but can deliver many events at once. 822overhead for the actual polling but can deliver many events at once.
671 823
672By setting a higher I<io collect interval> you allow libev to spend more 824By setting a higher I<io collect interval> you allow libev to spend more
673time collecting I/O events, so you can handle more events per iteration, 825time collecting I/O events, so you can handle more events per iteration,
674at the cost of increasing latency. Timeouts (both C<ev_periodic> and 826at the cost of increasing latency. Timeouts (both C<ev_periodic> and
675C<ev_timer>) will be not affected. Setting this to a non-null value will 827C<ev_timer>) will be not affected. Setting this to a non-null value will
676introduce an additional C<ev_sleep ()> call into most loop iterations. 828introduce an additional C<ev_sleep ()> call into most loop iterations. The
829sleep time ensures that libev will not poll for I/O events more often then
830once per this interval, on average.
677 831
678Likewise, by setting a higher I<timeout collect interval> you allow libev 832Likewise, by setting a higher I<timeout collect interval> you allow libev
679to spend more time collecting timeouts, at the expense of increased 833to spend more time collecting timeouts, at the expense of increased
680latency (the watcher callback will be called later). C<ev_io> watchers 834latency/jitter/inexactness (the watcher callback will be called
681will not be affected. Setting this to a non-null value will not introduce 835later). C<ev_io> watchers will not be affected. Setting this to a non-null
682any overhead in libev. 836value will not introduce any overhead in libev.
683 837
684Many (busy) programs can usually benefit by setting the io collect 838Many (busy) programs can usually benefit by setting the I/O collect
685interval to a value near C<0.1> or so, which is often enough for 839interval to a value near C<0.1> or so, which is often enough for
686interactive servers (of course not for games), likewise for timeouts. It 840interactive servers (of course not for games), likewise for timeouts. It
687usually doesn't make much sense to set it to a lower value than C<0.01>, 841usually doesn't make much sense to set it to a lower value than C<0.01>,
688as this approsaches the timing granularity of most systems. 842as this approaches the timing granularity of most systems. Note that if
843you do transactions with the outside world and you can't increase the
844parallelity, then this setting will limit your transaction rate (if you
845need to poll once per transaction and the I/O collect interval is 0.01,
846then you can't do more than 100 transations per second).
847
848Setting the I<timeout collect interval> can improve the opportunity for
849saving power, as the program will "bundle" timer callback invocations that
850are "near" in time together, by delaying some, thus reducing the number of
851times the process sleeps and wakes up again. Another useful technique to
852reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
853they fire on, say, one-second boundaries only.
854
855Example: we only need 0.1s timeout granularity, and we wish not to poll
856more often than 100 times per second:
857
858 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
859 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
860
861=item ev_invoke_pending (loop)
862
863This call will simply invoke all pending watchers while resetting their
864pending state. Normally, C<ev_loop> does this automatically when required,
865but when overriding the invoke callback this call comes handy.
866
867=item int ev_pending_count (loop)
868
869Returns the number of pending watchers - zero indicates that no watchers
870are pending.
871
872=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
873
874This overrides the invoke pending functionality of the loop: Instead of
875invoking all pending watchers when there are any, C<ev_loop> will call
876this callback instead. This is useful, for example, when you want to
877invoke the actual watchers inside another context (another thread etc.).
878
879If you want to reset the callback, use C<ev_invoke_pending> as new
880callback.
881
882=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
883
884Sometimes you want to share the same loop between multiple threads. This
885can be done relatively simply by putting mutex_lock/unlock calls around
886each call to a libev function.
887
888However, C<ev_loop> can run an indefinite time, so it is not feasible to
889wait for it to return. One way around this is to wake up the loop via
890C<ev_unloop> and C<av_async_send>, another way is to set these I<release>
891and I<acquire> callbacks on the loop.
892
893When set, then C<release> will be called just before the thread is
894suspended waiting for new events, and C<acquire> is called just
895afterwards.
896
897Ideally, C<release> will just call your mutex_unlock function, and
898C<acquire> will just call the mutex_lock function again.
899
900While event loop modifications are allowed between invocations of
901C<release> and C<acquire> (that's their only purpose after all), no
902modifications done will affect the event loop, i.e. adding watchers will
903have no effect on the set of file descriptors being watched, or the time
904waited. USe an C<ev_async> watcher to wake up C<ev_loop> when you want it
905to take note of any changes you made.
906
907In theory, threads executing C<ev_loop> will be async-cancel safe between
908invocations of C<release> and C<acquire>.
909
910See also the locking example in the C<THREADS> section later in this
911document.
912
913=item ev_set_userdata (loop, void *data)
914
915=item ev_userdata (loop)
916
917Set and retrieve a single C<void *> associated with a loop. When
918C<ev_set_userdata> has never been called, then C<ev_userdata> returns
919C<0.>
920
921These two functions can be used to associate arbitrary data with a loop,
922and are intended solely for the C<invoke_pending_cb>, C<release> and
923C<acquire> callbacks described above, but of course can be (ab-)used for
924any other purpose as well.
925
926=item ev_loop_verify (loop)
927
928This function only does something when C<EV_VERIFY> support has been
929compiled in, which is the default for non-minimal builds. It tries to go
930through all internal structures and checks them for validity. If anything
931is found to be inconsistent, it will print an error message to standard
932error and call C<abort ()>.
933
934This can be used to catch bugs inside libev itself: under normal
935circumstances, this function will never abort as of course libev keeps its
936data structures consistent.
689 937
690=back 938=back
691 939
692 940
693=head1 ANATOMY OF A WATCHER 941=head1 ANATOMY OF A WATCHER
942
943In the following description, uppercase C<TYPE> in names stands for the
944watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
945watchers and C<ev_io_start> for I/O watchers.
694 946
695A watcher is a structure that you create and register to record your 947A watcher is a structure that you create and register to record your
696interest in some event. For instance, if you want to wait for STDIN to 948interest in some event. For instance, if you want to wait for STDIN to
697become readable, you would create an C<ev_io> watcher for that: 949become readable, you would create an C<ev_io> watcher for that:
698 950
699 static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents) 951 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
700 { 952 {
701 ev_io_stop (w); 953 ev_io_stop (w);
702 ev_unloop (loop, EVUNLOOP_ALL); 954 ev_unloop (loop, EVUNLOOP_ALL);
703 } 955 }
704 956
705 struct ev_loop *loop = ev_default_loop (0); 957 struct ev_loop *loop = ev_default_loop (0);
958
706 struct ev_io stdin_watcher; 959 ev_io stdin_watcher;
960
707 ev_init (&stdin_watcher, my_cb); 961 ev_init (&stdin_watcher, my_cb);
708 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ); 962 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
709 ev_io_start (loop, &stdin_watcher); 963 ev_io_start (loop, &stdin_watcher);
964
710 ev_loop (loop, 0); 965 ev_loop (loop, 0);
711 966
712As you can see, you are responsible for allocating the memory for your 967As you can see, you are responsible for allocating the memory for your
713watcher structures (and it is usually a bad idea to do this on the stack, 968watcher structures (and it is I<usually> a bad idea to do this on the
714although this can sometimes be quite valid). 969stack).
970
971Each watcher has an associated watcher structure (called C<struct ev_TYPE>
972or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
715 973
716Each watcher structure must be initialised by a call to C<ev_init 974Each watcher structure must be initialised by a call to C<ev_init
717(watcher *, callback)>, which expects a callback to be provided. This 975(watcher *, callback)>, which expects a callback to be provided. This
718callback gets invoked each time the event occurs (or, in the case of io 976callback gets invoked each time the event occurs (or, in the case of I/O
719watchers, each time the event loop detects that the file descriptor given 977watchers, each time the event loop detects that the file descriptor given
720is readable and/or writable). 978is readable and/or writable).
721 979
722Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro 980Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
723with arguments specific to this watcher type. There is also a macro 981macro to configure it, with arguments specific to the watcher type. There
724to combine initialisation and setting in one call: C<< ev_<type>_init 982is also a macro to combine initialisation and setting in one call: C<<
725(watcher *, callback, ...) >>. 983ev_TYPE_init (watcher *, callback, ...) >>.
726 984
727To make the watcher actually watch out for events, you have to start it 985To make the watcher actually watch out for events, you have to start it
728with a watcher-specific start function (C<< ev_<type>_start (loop, watcher 986with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
729*) >>), and you can stop watching for events at any time by calling the 987*) >>), and you can stop watching for events at any time by calling the
730corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>. 988corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
731 989
732As long as your watcher is active (has been started but not stopped) you 990As long as your watcher is active (has been started but not stopped) you
733must not touch the values stored in it. Most specifically you must never 991must not touch the values stored in it. Most specifically you must never
734reinitialise it or call its C<set> macro. 992reinitialise it or call its C<ev_TYPE_set> macro.
735 993
736Each and every callback receives the event loop pointer as first, the 994Each and every callback receives the event loop pointer as first, the
737registered watcher structure as second, and a bitset of received events as 995registered watcher structure as second, and a bitset of received events as
738third argument. 996third argument.
739 997
797 1055
798=item C<EV_ASYNC> 1056=item C<EV_ASYNC>
799 1057
800The given async watcher has been asynchronously notified (see C<ev_async>). 1058The given async watcher has been asynchronously notified (see C<ev_async>).
801 1059
1060=item C<EV_CUSTOM>
1061
1062Not ever sent (or otherwise used) by libev itself, but can be freely used
1063by libev users to signal watchers (e.g. via C<ev_feed_event>).
1064
802=item C<EV_ERROR> 1065=item C<EV_ERROR>
803 1066
804An unspecified error has occured, the watcher has been stopped. This might 1067An unspecified error has occurred, the watcher has been stopped. This might
805happen because the watcher could not be properly started because libev 1068happen because the watcher could not be properly started because libev
806ran out of memory, a file descriptor was found to be closed or any other 1069ran out of memory, a file descriptor was found to be closed or any other
1070problem. Libev considers these application bugs.
1071
807problem. You best act on it by reporting the problem and somehow coping 1072You best act on it by reporting the problem and somehow coping with the
808with the watcher being stopped. 1073watcher being stopped. Note that well-written programs should not receive
1074an error ever, so when your watcher receives it, this usually indicates a
1075bug in your program.
809 1076
810Libev will usually signal a few "dummy" events together with an error, 1077Libev will usually signal a few "dummy" events together with an error, for
811for example it might indicate that a fd is readable or writable, and if 1078example it might indicate that a fd is readable or writable, and if your
812your callbacks is well-written it can just attempt the operation and cope 1079callbacks is well-written it can just attempt the operation and cope with
813with the error from read() or write(). This will not work in multithreaded 1080the error from read() or write(). This will not work in multi-threaded
814programs, though, so beware. 1081programs, though, as the fd could already be closed and reused for another
1082thing, so beware.
815 1083
816=back 1084=back
817 1085
818=head2 GENERIC WATCHER FUNCTIONS 1086=head2 GENERIC WATCHER FUNCTIONS
819
820In the following description, C<TYPE> stands for the watcher type,
821e.g. C<timer> for C<ev_timer> watchers and C<io> for C<ev_io> watchers.
822 1087
823=over 4 1088=over 4
824 1089
825=item C<ev_init> (ev_TYPE *watcher, callback) 1090=item C<ev_init> (ev_TYPE *watcher, callback)
826 1091
832which rolls both calls into one. 1097which rolls both calls into one.
833 1098
834You can reinitialise a watcher at any time as long as it has been stopped 1099You can reinitialise a watcher at any time as long as it has been stopped
835(or never started) and there are no pending events outstanding. 1100(or never started) and there are no pending events outstanding.
836 1101
837The callback is always of type C<void (*)(ev_loop *loop, ev_TYPE *watcher, 1102The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
838int revents)>. 1103int revents)>.
1104
1105Example: Initialise an C<ev_io> watcher in two steps.
1106
1107 ev_io w;
1108 ev_init (&w, my_cb);
1109 ev_io_set (&w, STDIN_FILENO, EV_READ);
839 1110
840=item C<ev_TYPE_set> (ev_TYPE *, [args]) 1111=item C<ev_TYPE_set> (ev_TYPE *, [args])
841 1112
842This macro initialises the type-specific parts of a watcher. You need to 1113This macro initialises the type-specific parts of a watcher. You need to
843call C<ev_init> at least once before you call this macro, but you can 1114call C<ev_init> at least once before you call this macro, but you can
846difference to the C<ev_init> macro). 1117difference to the C<ev_init> macro).
847 1118
848Although some watcher types do not have type-specific arguments 1119Although some watcher types do not have type-specific arguments
849(e.g. C<ev_prepare>) you still need to call its C<set> macro. 1120(e.g. C<ev_prepare>) you still need to call its C<set> macro.
850 1121
1122See C<ev_init>, above, for an example.
1123
851=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args]) 1124=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
852 1125
853This convinience macro rolls both C<ev_init> and C<ev_TYPE_set> macro 1126This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
854calls into a single call. This is the most convinient method to initialise 1127calls into a single call. This is the most convenient method to initialise
855a watcher. The same limitations apply, of course. 1128a watcher. The same limitations apply, of course.
1129
1130Example: Initialise and set an C<ev_io> watcher in one step.
1131
1132 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
856 1133
857=item C<ev_TYPE_start> (loop *, ev_TYPE *watcher) 1134=item C<ev_TYPE_start> (loop *, ev_TYPE *watcher)
858 1135
859Starts (activates) the given watcher. Only active watchers will receive 1136Starts (activates) the given watcher. Only active watchers will receive
860events. If the watcher is already active nothing will happen. 1137events. If the watcher is already active nothing will happen.
861 1138
1139Example: Start the C<ev_io> watcher that is being abused as example in this
1140whole section.
1141
1142 ev_io_start (EV_DEFAULT_UC, &w);
1143
862=item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher) 1144=item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher)
863 1145
864Stops the given watcher again (if active) and clears the pending 1146Stops the given watcher if active, and clears the pending status (whether
1147the watcher was active or not).
1148
865status. It is possible that stopped watchers are pending (for example, 1149It is possible that stopped watchers are pending - for example,
866non-repeating timers are being stopped when they become pending), but 1150non-repeating timers are being stopped when they become pending - but
867C<ev_TYPE_stop> ensures that the watcher is neither active nor pending. If 1151calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
868you want to free or reuse the memory used by the watcher it is therefore a 1152pending. If you want to free or reuse the memory used by the watcher it is
869good idea to always call its C<ev_TYPE_stop> function. 1153therefore a good idea to always call its C<ev_TYPE_stop> function.
870 1154
871=item bool ev_is_active (ev_TYPE *watcher) 1155=item bool ev_is_active (ev_TYPE *watcher)
872 1156
873Returns a true value iff the watcher is active (i.e. it has been started 1157Returns a true value iff the watcher is active (i.e. it has been started
874and not yet been stopped). As long as a watcher is active you must not modify 1158and not yet been stopped). As long as a watcher is active you must not modify
900integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI> 1184integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
901(default: C<-2>). Pending watchers with higher priority will be invoked 1185(default: C<-2>). Pending watchers with higher priority will be invoked
902before watchers with lower priority, but priority will not keep watchers 1186before watchers with lower priority, but priority will not keep watchers
903from being executed (except for C<ev_idle> watchers). 1187from being executed (except for C<ev_idle> watchers).
904 1188
905This means that priorities are I<only> used for ordering callback
906invocation after new events have been received. This is useful, for
907example, to reduce latency after idling, or more often, to bind two
908watchers on the same event and make sure one is called first.
909
910If you need to suppress invocation when higher priority events are pending 1189If you need to suppress invocation when higher priority events are pending
911you need to look at C<ev_idle> watchers, which provide this functionality. 1190you need to look at C<ev_idle> watchers, which provide this functionality.
912 1191
913You I<must not> change the priority of a watcher as long as it is active or 1192You I<must not> change the priority of a watcher as long as it is active or
914pending. 1193pending.
915 1194
1195Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1196fine, as long as you do not mind that the priority value you query might
1197or might not have been clamped to the valid range.
1198
916The default priority used by watchers when no priority has been set is 1199The default priority used by watchers when no priority has been set is
917always C<0>, which is supposed to not be too high and not be too low :). 1200always C<0>, which is supposed to not be too high and not be too low :).
918 1201
919Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is 1202See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
920fine, as long as you do not mind that the priority value you query might 1203priorities.
921or might not have been adjusted to be within valid range.
922 1204
923=item ev_invoke (loop, ev_TYPE *watcher, int revents) 1205=item ev_invoke (loop, ev_TYPE *watcher, int revents)
924 1206
925Invoke the C<watcher> with the given C<loop> and C<revents>. Neither 1207Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
926C<loop> nor C<revents> need to be valid as long as the watcher callback 1208C<loop> nor C<revents> need to be valid as long as the watcher callback
927can deal with that fact. 1209can deal with that fact, as both are simply passed through to the
1210callback.
928 1211
929=item int ev_clear_pending (loop, ev_TYPE *watcher) 1212=item int ev_clear_pending (loop, ev_TYPE *watcher)
930 1213
931If the watcher is pending, this function returns clears its pending status 1214If the watcher is pending, this function clears its pending status and
932and returns its C<revents> bitset (as if its callback was invoked). If the 1215returns its C<revents> bitset (as if its callback was invoked). If the
933watcher isn't pending it does nothing and returns C<0>. 1216watcher isn't pending it does nothing and returns C<0>.
934 1217
1218Sometimes it can be useful to "poll" a watcher instead of waiting for its
1219callback to be invoked, which can be accomplished with this function.
1220
935=back 1221=back
936 1222
937 1223
938=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER 1224=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
939 1225
940Each watcher has, by default, a member C<void *data> that you can change 1226Each watcher has, by default, a member C<void *data> that you can change
941and read at any time, libev will completely ignore it. This can be used 1227and read at any time: libev will completely ignore it. This can be used
942to associate arbitrary data with your watcher. If you need more data and 1228to associate arbitrary data with your watcher. If you need more data and
943don't want to allocate memory and store a pointer to it in that data 1229don't want to allocate memory and store a pointer to it in that data
944member, you can also "subclass" the watcher type and provide your own 1230member, you can also "subclass" the watcher type and provide your own
945data: 1231data:
946 1232
947 struct my_io 1233 struct my_io
948 { 1234 {
949 struct ev_io io; 1235 ev_io io;
950 int otherfd; 1236 int otherfd;
951 void *somedata; 1237 void *somedata;
952 struct whatever *mostinteresting; 1238 struct whatever *mostinteresting;
953 } 1239 };
1240
1241 ...
1242 struct my_io w;
1243 ev_io_init (&w.io, my_cb, fd, EV_READ);
954 1244
955And since your callback will be called with a pointer to the watcher, you 1245And since your callback will be called with a pointer to the watcher, you
956can cast it back to your own type: 1246can cast it back to your own type:
957 1247
958 static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents) 1248 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
959 { 1249 {
960 struct my_io *w = (struct my_io *)w_; 1250 struct my_io *w = (struct my_io *)w_;
961 ... 1251 ...
962 } 1252 }
963 1253
964More interesting and less C-conformant ways of casting your callback type 1254More interesting and less C-conformant ways of casting your callback type
965instead have been omitted. 1255instead have been omitted.
966 1256
967Another common scenario is having some data structure with multiple 1257Another common scenario is to use some data structure with multiple
968watchers: 1258embedded watchers:
969 1259
970 struct my_biggy 1260 struct my_biggy
971 { 1261 {
972 int some_data; 1262 int some_data;
973 ev_timer t1; 1263 ev_timer t1;
974 ev_timer t2; 1264 ev_timer t2;
975 } 1265 }
976 1266
977In this case getting the pointer to C<my_biggy> is a bit more complicated, 1267In this case getting the pointer to C<my_biggy> is a bit more
978you need to use C<offsetof>: 1268complicated: Either you store the address of your C<my_biggy> struct
1269in the C<data> member of the watcher (for woozies), or you need to use
1270some pointer arithmetic using C<offsetof> inside your watchers (for real
1271programmers):
979 1272
980 #include <stddef.h> 1273 #include <stddef.h>
981 1274
982 static void 1275 static void
983 t1_cb (EV_P_ struct ev_timer *w, int revents) 1276 t1_cb (EV_P_ ev_timer *w, int revents)
984 { 1277 {
985 struct my_biggy big = (struct my_biggy * 1278 struct my_biggy big = (struct my_biggy *)
986 (((char *)w) - offsetof (struct my_biggy, t1)); 1279 (((char *)w) - offsetof (struct my_biggy, t1));
987 } 1280 }
988 1281
989 static void 1282 static void
990 t2_cb (EV_P_ struct ev_timer *w, int revents) 1283 t2_cb (EV_P_ ev_timer *w, int revents)
991 { 1284 {
992 struct my_biggy big = (struct my_biggy * 1285 struct my_biggy big = (struct my_biggy *)
993 (((char *)w) - offsetof (struct my_biggy, t2)); 1286 (((char *)w) - offsetof (struct my_biggy, t2));
994 } 1287 }
1288
1289=head2 WATCHER PRIORITY MODELS
1290
1291Many event loops support I<watcher priorities>, which are usually small
1292integers that influence the ordering of event callback invocation
1293between watchers in some way, all else being equal.
1294
1295In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1296description for the more technical details such as the actual priority
1297range.
1298
1299There are two common ways how these these priorities are being interpreted
1300by event loops:
1301
1302In the more common lock-out model, higher priorities "lock out" invocation
1303of lower priority watchers, which means as long as higher priority
1304watchers receive events, lower priority watchers are not being invoked.
1305
1306The less common only-for-ordering model uses priorities solely to order
1307callback invocation within a single event loop iteration: Higher priority
1308watchers are invoked before lower priority ones, but they all get invoked
1309before polling for new events.
1310
1311Libev uses the second (only-for-ordering) model for all its watchers
1312except for idle watchers (which use the lock-out model).
1313
1314The rationale behind this is that implementing the lock-out model for
1315watchers is not well supported by most kernel interfaces, and most event
1316libraries will just poll for the same events again and again as long as
1317their callbacks have not been executed, which is very inefficient in the
1318common case of one high-priority watcher locking out a mass of lower
1319priority ones.
1320
1321Static (ordering) priorities are most useful when you have two or more
1322watchers handling the same resource: a typical usage example is having an
1323C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1324timeouts. Under load, data might be received while the program handles
1325other jobs, but since timers normally get invoked first, the timeout
1326handler will be executed before checking for data. In that case, giving
1327the timer a lower priority than the I/O watcher ensures that I/O will be
1328handled first even under adverse conditions (which is usually, but not
1329always, what you want).
1330
1331Since idle watchers use the "lock-out" model, meaning that idle watchers
1332will only be executed when no same or higher priority watchers have
1333received events, they can be used to implement the "lock-out" model when
1334required.
1335
1336For example, to emulate how many other event libraries handle priorities,
1337you can associate an C<ev_idle> watcher to each such watcher, and in
1338the normal watcher callback, you just start the idle watcher. The real
1339processing is done in the idle watcher callback. This causes libev to
1340continously poll and process kernel event data for the watcher, but when
1341the lock-out case is known to be rare (which in turn is rare :), this is
1342workable.
1343
1344Usually, however, the lock-out model implemented that way will perform
1345miserably under the type of load it was designed to handle. In that case,
1346it might be preferable to stop the real watcher before starting the
1347idle watcher, so the kernel will not have to process the event in case
1348the actual processing will be delayed for considerable time.
1349
1350Here is an example of an I/O watcher that should run at a strictly lower
1351priority than the default, and which should only process data when no
1352other events are pending:
1353
1354 ev_idle idle; // actual processing watcher
1355 ev_io io; // actual event watcher
1356
1357 static void
1358 io_cb (EV_P_ ev_io *w, int revents)
1359 {
1360 // stop the I/O watcher, we received the event, but
1361 // are not yet ready to handle it.
1362 ev_io_stop (EV_A_ w);
1363
1364 // start the idle watcher to ahndle the actual event.
1365 // it will not be executed as long as other watchers
1366 // with the default priority are receiving events.
1367 ev_idle_start (EV_A_ &idle);
1368 }
1369
1370 static void
1371 idle_cb (EV_P_ ev_idle *w, int revents)
1372 {
1373 // actual processing
1374 read (STDIN_FILENO, ...);
1375
1376 // have to start the I/O watcher again, as
1377 // we have handled the event
1378 ev_io_start (EV_P_ &io);
1379 }
1380
1381 // initialisation
1382 ev_idle_init (&idle, idle_cb);
1383 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1384 ev_io_start (EV_DEFAULT_ &io);
1385
1386In the "real" world, it might also be beneficial to start a timer, so that
1387low-priority connections can not be locked out forever under load. This
1388enables your program to keep a lower latency for important connections
1389during short periods of high load, while not completely locking out less
1390important ones.
995 1391
996 1392
997=head1 WATCHER TYPES 1393=head1 WATCHER TYPES
998 1394
999This section describes each watcher in detail, but will not repeat 1395This section describes each watcher in detail, but will not repeat
1023In general you can register as many read and/or write event watchers per 1419In general you can register as many read and/or write event watchers per
1024fd as you want (as long as you don't confuse yourself). Setting all file 1420fd as you want (as long as you don't confuse yourself). Setting all file
1025descriptors to non-blocking mode is also usually a good idea (but not 1421descriptors to non-blocking mode is also usually a good idea (but not
1026required if you know what you are doing). 1422required if you know what you are doing).
1027 1423
1028If you must do this, then force the use of a known-to-be-good backend 1424If you cannot use non-blocking mode, then force the use of a
1029(at the time of this writing, this includes only C<EVBACKEND_SELECT> and 1425known-to-be-good backend (at the time of this writing, this includes only
1030C<EVBACKEND_POLL>). 1426C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
1427descriptors for which non-blocking operation makes no sense (such as
1428files) - libev doesn't guarentee any specific behaviour in that case.
1031 1429
1032Another thing you have to watch out for is that it is quite easy to 1430Another thing you have to watch out for is that it is quite easy to
1033receive "spurious" readyness notifications, that is your callback might 1431receive "spurious" readiness notifications, that is your callback might
1034be called with C<EV_READ> but a subsequent C<read>(2) will actually block 1432be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1035because there is no data. Not only are some backends known to create a 1433because there is no data. Not only are some backends known to create a
1036lot of those (for example solaris ports), it is very easy to get into 1434lot of those (for example Solaris ports), it is very easy to get into
1037this situation even with a relatively standard program structure. Thus 1435this situation even with a relatively standard program structure. Thus
1038it is best to always use non-blocking I/O: An extra C<read>(2) returning 1436it is best to always use non-blocking I/O: An extra C<read>(2) returning
1039C<EAGAIN> is far preferable to a program hanging until some data arrives. 1437C<EAGAIN> is far preferable to a program hanging until some data arrives.
1040 1438
1041If you cannot run the fd in non-blocking mode (for example you should not 1439If you cannot run the fd in non-blocking mode (for example you should
1042play around with an Xlib connection), then you have to seperately re-test 1440not play around with an Xlib connection), then you have to separately
1043whether a file descriptor is really ready with a known-to-be good interface 1441re-test whether a file descriptor is really ready with a known-to-be good
1044such as poll (fortunately in our Xlib example, Xlib already does this on 1442interface such as poll (fortunately in our Xlib example, Xlib already
1045its own, so its quite safe to use). 1443does this on its own, so its quite safe to use). Some people additionally
1444use C<SIGALRM> and an interval timer, just to be sure you won't block
1445indefinitely.
1446
1447But really, best use non-blocking mode.
1046 1448
1047=head3 The special problem of disappearing file descriptors 1449=head3 The special problem of disappearing file descriptors
1048 1450
1049Some backends (e.g. kqueue, epoll) need to be told about closing a file 1451Some backends (e.g. kqueue, epoll) need to be told about closing a file
1050descriptor (either by calling C<close> explicitly or by any other means, 1452descriptor (either due to calling C<close> explicitly or any other means,
1051such as C<dup>). The reason is that you register interest in some file 1453such as C<dup2>). The reason is that you register interest in some file
1052descriptor, but when it goes away, the operating system will silently drop 1454descriptor, but when it goes away, the operating system will silently drop
1053this interest. If another file descriptor with the same number then is 1455this interest. If another file descriptor with the same number then is
1054registered with libev, there is no efficient way to see that this is, in 1456registered with libev, there is no efficient way to see that this is, in
1055fact, a different file descriptor. 1457fact, a different file descriptor.
1056 1458
1087enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or 1489enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
1088C<EVBACKEND_POLL>. 1490C<EVBACKEND_POLL>.
1089 1491
1090=head3 The special problem of SIGPIPE 1492=head3 The special problem of SIGPIPE
1091 1493
1092While not really specific to libev, it is easy to forget about SIGPIPE: 1494While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1093when reading from a pipe whose other end has been closed, your program 1495when writing to a pipe whose other end has been closed, your program gets
1094gets send a SIGPIPE, which, by default, aborts your program. For most 1496sent a SIGPIPE, which, by default, aborts your program. For most programs
1095programs this is sensible behaviour, for daemons, this is usually 1497this is sensible behaviour, for daemons, this is usually undesirable.
1096undesirable.
1097 1498
1098So when you encounter spurious, unexplained daemon exits, make sure you 1499So when you encounter spurious, unexplained daemon exits, make sure you
1099ignore SIGPIPE (and maybe make sure you log the exit status of your daemon 1500ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1100somewhere, as that would have given you a big clue). 1501somewhere, as that would have given you a big clue).
1101 1502
1107=item ev_io_init (ev_io *, callback, int fd, int events) 1508=item ev_io_init (ev_io *, callback, int fd, int events)
1108 1509
1109=item ev_io_set (ev_io *, int fd, int events) 1510=item ev_io_set (ev_io *, int fd, int events)
1110 1511
1111Configures an C<ev_io> watcher. The C<fd> is the file descriptor to 1512Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
1112rceeive events for and events is either C<EV_READ>, C<EV_WRITE> or 1513receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
1113C<EV_READ | EV_WRITE> to receive the given events. 1514C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
1114 1515
1115=item int fd [read-only] 1516=item int fd [read-only]
1116 1517
1117The file descriptor being watched. 1518The file descriptor being watched.
1118 1519
1126 1527
1127Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well 1528Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1128readable, but only once. Since it is likely line-buffered, you could 1529readable, but only once. Since it is likely line-buffered, you could
1129attempt to read a whole line in the callback. 1530attempt to read a whole line in the callback.
1130 1531
1131 static void 1532 static void
1132 stdin_readable_cb (struct ev_loop *loop, struct ev_io *w, int revents) 1533 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1133 { 1534 {
1134 ev_io_stop (loop, w); 1535 ev_io_stop (loop, w);
1135 .. read from stdin here (or from w->fd) and haqndle any I/O errors 1536 .. read from stdin here (or from w->fd) and handle any I/O errors
1136 } 1537 }
1137 1538
1138 ... 1539 ...
1139 struct ev_loop *loop = ev_default_init (0); 1540 struct ev_loop *loop = ev_default_init (0);
1140 struct ev_io stdin_readable; 1541 ev_io stdin_readable;
1141 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ); 1542 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1142 ev_io_start (loop, &stdin_readable); 1543 ev_io_start (loop, &stdin_readable);
1143 ev_loop (loop, 0); 1544 ev_loop (loop, 0);
1144 1545
1145 1546
1146=head2 C<ev_timer> - relative and optionally repeating timeouts 1547=head2 C<ev_timer> - relative and optionally repeating timeouts
1147 1548
1148Timer watchers are simple relative timers that generate an event after a 1549Timer watchers are simple relative timers that generate an event after a
1149given time, and optionally repeating in regular intervals after that. 1550given time, and optionally repeating in regular intervals after that.
1150 1551
1151The timers are based on real time, that is, if you register an event that 1552The timers are based on real time, that is, if you register an event that
1152times out after an hour and you reset your system clock to last years 1553times out after an hour and you reset your system clock to January last
1153time, it will still time out after (roughly) and hour. "Roughly" because 1554year, it will still time out after (roughly) one hour. "Roughly" because
1154detecting time jumps is hard, and some inaccuracies are unavoidable (the 1555detecting time jumps is hard, and some inaccuracies are unavoidable (the
1155monotonic clock option helps a lot here). 1556monotonic clock option helps a lot here).
1557
1558The callback is guaranteed to be invoked only I<after> its timeout has
1559passed (not I<at>, so on systems with very low-resolution clocks this
1560might introduce a small delay). If multiple timers become ready during the
1561same loop iteration then the ones with earlier time-out values are invoked
1562before ones of the same priority with later time-out values (but this is
1563no longer true when a callback calls C<ev_loop> recursively).
1564
1565=head3 Be smart about timeouts
1566
1567Many real-world problems involve some kind of timeout, usually for error
1568recovery. A typical example is an HTTP request - if the other side hangs,
1569you want to raise some error after a while.
1570
1571What follows are some ways to handle this problem, from obvious and
1572inefficient to smart and efficient.
1573
1574In the following, a 60 second activity timeout is assumed - a timeout that
1575gets reset to 60 seconds each time there is activity (e.g. each time some
1576data or other life sign was received).
1577
1578=over 4
1579
1580=item 1. Use a timer and stop, reinitialise and start it on activity.
1581
1582This is the most obvious, but not the most simple way: In the beginning,
1583start the watcher:
1584
1585 ev_timer_init (timer, callback, 60., 0.);
1586 ev_timer_start (loop, timer);
1587
1588Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1589and start it again:
1590
1591 ev_timer_stop (loop, timer);
1592 ev_timer_set (timer, 60., 0.);
1593 ev_timer_start (loop, timer);
1594
1595This is relatively simple to implement, but means that each time there is
1596some activity, libev will first have to remove the timer from its internal
1597data structure and then add it again. Libev tries to be fast, but it's
1598still not a constant-time operation.
1599
1600=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1601
1602This is the easiest way, and involves using C<ev_timer_again> instead of
1603C<ev_timer_start>.
1604
1605To implement this, configure an C<ev_timer> with a C<repeat> value
1606of C<60> and then call C<ev_timer_again> at start and each time you
1607successfully read or write some data. If you go into an idle state where
1608you do not expect data to travel on the socket, you can C<ev_timer_stop>
1609the timer, and C<ev_timer_again> will automatically restart it if need be.
1610
1611That means you can ignore both the C<ev_timer_start> function and the
1612C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1613member and C<ev_timer_again>.
1614
1615At start:
1616
1617 ev_init (timer, callback);
1618 timer->repeat = 60.;
1619 ev_timer_again (loop, timer);
1620
1621Each time there is some activity:
1622
1623 ev_timer_again (loop, timer);
1624
1625It is even possible to change the time-out on the fly, regardless of
1626whether the watcher is active or not:
1627
1628 timer->repeat = 30.;
1629 ev_timer_again (loop, timer);
1630
1631This is slightly more efficient then stopping/starting the timer each time
1632you want to modify its timeout value, as libev does not have to completely
1633remove and re-insert the timer from/into its internal data structure.
1634
1635It is, however, even simpler than the "obvious" way to do it.
1636
1637=item 3. Let the timer time out, but then re-arm it as required.
1638
1639This method is more tricky, but usually most efficient: Most timeouts are
1640relatively long compared to the intervals between other activity - in
1641our example, within 60 seconds, there are usually many I/O events with
1642associated activity resets.
1643
1644In this case, it would be more efficient to leave the C<ev_timer> alone,
1645but remember the time of last activity, and check for a real timeout only
1646within the callback:
1647
1648 ev_tstamp last_activity; // time of last activity
1649
1650 static void
1651 callback (EV_P_ ev_timer *w, int revents)
1652 {
1653 ev_tstamp now = ev_now (EV_A);
1654 ev_tstamp timeout = last_activity + 60.;
1655
1656 // if last_activity + 60. is older than now, we did time out
1657 if (timeout < now)
1658 {
1659 // timeout occured, take action
1660 }
1661 else
1662 {
1663 // callback was invoked, but there was some activity, re-arm
1664 // the watcher to fire in last_activity + 60, which is
1665 // guaranteed to be in the future, so "again" is positive:
1666 w->repeat = timeout - now;
1667 ev_timer_again (EV_A_ w);
1668 }
1669 }
1670
1671To summarise the callback: first calculate the real timeout (defined
1672as "60 seconds after the last activity"), then check if that time has
1673been reached, which means something I<did>, in fact, time out. Otherwise
1674the callback was invoked too early (C<timeout> is in the future), so
1675re-schedule the timer to fire at that future time, to see if maybe we have
1676a timeout then.
1677
1678Note how C<ev_timer_again> is used, taking advantage of the
1679C<ev_timer_again> optimisation when the timer is already running.
1680
1681This scheme causes more callback invocations (about one every 60 seconds
1682minus half the average time between activity), but virtually no calls to
1683libev to change the timeout.
1684
1685To start the timer, simply initialise the watcher and set C<last_activity>
1686to the current time (meaning we just have some activity :), then call the
1687callback, which will "do the right thing" and start the timer:
1688
1689 ev_init (timer, callback);
1690 last_activity = ev_now (loop);
1691 callback (loop, timer, EV_TIMEOUT);
1692
1693And when there is some activity, simply store the current time in
1694C<last_activity>, no libev calls at all:
1695
1696 last_actiivty = ev_now (loop);
1697
1698This technique is slightly more complex, but in most cases where the
1699time-out is unlikely to be triggered, much more efficient.
1700
1701Changing the timeout is trivial as well (if it isn't hard-coded in the
1702callback :) - just change the timeout and invoke the callback, which will
1703fix things for you.
1704
1705=item 4. Wee, just use a double-linked list for your timeouts.
1706
1707If there is not one request, but many thousands (millions...), all
1708employing some kind of timeout with the same timeout value, then one can
1709do even better:
1710
1711When starting the timeout, calculate the timeout value and put the timeout
1712at the I<end> of the list.
1713
1714Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1715the list is expected to fire (for example, using the technique #3).
1716
1717When there is some activity, remove the timer from the list, recalculate
1718the timeout, append it to the end of the list again, and make sure to
1719update the C<ev_timer> if it was taken from the beginning of the list.
1720
1721This way, one can manage an unlimited number of timeouts in O(1) time for
1722starting, stopping and updating the timers, at the expense of a major
1723complication, and having to use a constant timeout. The constant timeout
1724ensures that the list stays sorted.
1725
1726=back
1727
1728So which method the best?
1729
1730Method #2 is a simple no-brain-required solution that is adequate in most
1731situations. Method #3 requires a bit more thinking, but handles many cases
1732better, and isn't very complicated either. In most case, choosing either
1733one is fine, with #3 being better in typical situations.
1734
1735Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1736rather complicated, but extremely efficient, something that really pays
1737off after the first million or so of active timers, i.e. it's usually
1738overkill :)
1739
1740=head3 The special problem of time updates
1741
1742Establishing the current time is a costly operation (it usually takes at
1743least two system calls): EV therefore updates its idea of the current
1744time only before and after C<ev_loop> collects new events, which causes a
1745growing difference between C<ev_now ()> and C<ev_time ()> when handling
1746lots of events in one iteration.
1156 1747
1157The relative timeouts are calculated relative to the C<ev_now ()> 1748The relative timeouts are calculated relative to the C<ev_now ()>
1158time. This is usually the right thing as this timestamp refers to the time 1749time. This is usually the right thing as this timestamp refers to the time
1159of the event triggering whatever timeout you are modifying/starting. If 1750of the event triggering whatever timeout you are modifying/starting. If
1160you suspect event processing to be delayed and you I<need> to base the timeout 1751you suspect event processing to be delayed and you I<need> to base the
1161on the current time, use something like this to adjust for this: 1752timeout on the current time, use something like this to adjust for this:
1162 1753
1163 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.); 1754 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
1164 1755
1165The callback is guarenteed to be invoked only when its timeout has passed, 1756If the event loop is suspended for a long time, you can also force an
1166but if multiple timers become ready during the same loop iteration then 1757update of the time returned by C<ev_now ()> by calling C<ev_now_update
1167order of execution is undefined. 1758()>.
1759
1760=head3 The special problems of suspended animation
1761
1762When you leave the server world it is quite customary to hit machines that
1763can suspend/hibernate - what happens to the clocks during such a suspend?
1764
1765Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
1766all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
1767to run until the system is suspended, but they will not advance while the
1768system is suspended. That means, on resume, it will be as if the program
1769was frozen for a few seconds, but the suspend time will not be counted
1770towards C<ev_timer> when a monotonic clock source is used. The real time
1771clock advanced as expected, but if it is used as sole clocksource, then a
1772long suspend would be detected as a time jump by libev, and timers would
1773be adjusted accordingly.
1774
1775I would not be surprised to see different behaviour in different between
1776operating systems, OS versions or even different hardware.
1777
1778The other form of suspend (job control, or sending a SIGSTOP) will see a
1779time jump in the monotonic clocks and the realtime clock. If the program
1780is suspended for a very long time, and monotonic clock sources are in use,
1781then you can expect C<ev_timer>s to expire as the full suspension time
1782will be counted towards the timers. When no monotonic clock source is in
1783use, then libev will again assume a timejump and adjust accordingly.
1784
1785It might be beneficial for this latter case to call C<ev_suspend>
1786and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
1787deterministic behaviour in this case (you can do nothing against
1788C<SIGSTOP>).
1168 1789
1169=head3 Watcher-Specific Functions and Data Members 1790=head3 Watcher-Specific Functions and Data Members
1170 1791
1171=over 4 1792=over 4
1172 1793
1173=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat) 1794=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
1174 1795
1175=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat) 1796=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
1176 1797
1177Configure the timer to trigger after C<after> seconds. If C<repeat> is 1798Configure the timer to trigger after C<after> seconds. If C<repeat>
1178C<0.>, then it will automatically be stopped. If it is positive, then the 1799is C<0.>, then it will automatically be stopped once the timeout is
1179timer will automatically be configured to trigger again C<repeat> seconds 1800reached. If it is positive, then the timer will automatically be
1180later, again, and again, until stopped manually. 1801configured to trigger again C<repeat> seconds later, again, and again,
1802until stopped manually.
1181 1803
1182The timer itself will do a best-effort at avoiding drift, that is, if you 1804The timer itself will do a best-effort at avoiding drift, that is, if
1183configure a timer to trigger every 10 seconds, then it will trigger at 1805you configure a timer to trigger every 10 seconds, then it will normally
1184exactly 10 second intervals. If, however, your program cannot keep up with 1806trigger at exactly 10 second intervals. If, however, your program cannot
1185the timer (because it takes longer than those 10 seconds to do stuff) the 1807keep up with the timer (because it takes longer than those 10 seconds to
1186timer will not fire more than once per event loop iteration. 1808do stuff) the timer will not fire more than once per event loop iteration.
1187 1809
1188=item ev_timer_again (loop, ev_timer *) 1810=item ev_timer_again (loop, ev_timer *)
1189 1811
1190This will act as if the timer timed out and restart it again if it is 1812This will act as if the timer timed out and restart it again if it is
1191repeating. The exact semantics are: 1813repeating. The exact semantics are:
1192 1814
1193If the timer is pending, its pending status is cleared. 1815If the timer is pending, its pending status is cleared.
1194 1816
1195If the timer is started but nonrepeating, stop it (as if it timed out). 1817If the timer is started but non-repeating, stop it (as if it timed out).
1196 1818
1197If the timer is repeating, either start it if necessary (with the 1819If the timer is repeating, either start it if necessary (with the
1198C<repeat> value), or reset the running timer to the C<repeat> value. 1820C<repeat> value), or reset the running timer to the C<repeat> value.
1199 1821
1200This sounds a bit complicated, but here is a useful and typical 1822This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
1201example: Imagine you have a tcp connection and you want a so-called idle 1823usage example.
1202timeout, that is, you want to be called when there have been, say, 60
1203seconds of inactivity on the socket. The easiest way to do this is to
1204configure an C<ev_timer> with a C<repeat> value of C<60> and then call
1205C<ev_timer_again> each time you successfully read or write some data. If
1206you go into an idle state where you do not expect data to travel on the
1207socket, you can C<ev_timer_stop> the timer, and C<ev_timer_again> will
1208automatically restart it if need be.
1209
1210That means you can ignore the C<after> value and C<ev_timer_start>
1211altogether and only ever use the C<repeat> value and C<ev_timer_again>:
1212
1213 ev_timer_init (timer, callback, 0., 5.);
1214 ev_timer_again (loop, timer);
1215 ...
1216 timer->again = 17.;
1217 ev_timer_again (loop, timer);
1218 ...
1219 timer->again = 10.;
1220 ev_timer_again (loop, timer);
1221
1222This is more slightly efficient then stopping/starting the timer each time
1223you want to modify its timeout value.
1224 1824
1225=item ev_tstamp repeat [read-write] 1825=item ev_tstamp repeat [read-write]
1226 1826
1227The current C<repeat> value. Will be used each time the watcher times out 1827The current C<repeat> value. Will be used each time the watcher times out
1228or C<ev_timer_again> is called and determines the next timeout (if any), 1828or C<ev_timer_again> is called, and determines the next timeout (if any),
1229which is also when any modifications are taken into account. 1829which is also when any modifications are taken into account.
1230 1830
1231=back 1831=back
1232 1832
1233=head3 Examples 1833=head3 Examples
1234 1834
1235Example: Create a timer that fires after 60 seconds. 1835Example: Create a timer that fires after 60 seconds.
1236 1836
1237 static void 1837 static void
1238 one_minute_cb (struct ev_loop *loop, struct ev_timer *w, int revents) 1838 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
1239 { 1839 {
1240 .. one minute over, w is actually stopped right here 1840 .. one minute over, w is actually stopped right here
1241 } 1841 }
1242 1842
1243 struct ev_timer mytimer; 1843 ev_timer mytimer;
1244 ev_timer_init (&mytimer, one_minute_cb, 60., 0.); 1844 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
1245 ev_timer_start (loop, &mytimer); 1845 ev_timer_start (loop, &mytimer);
1246 1846
1247Example: Create a timeout timer that times out after 10 seconds of 1847Example: Create a timeout timer that times out after 10 seconds of
1248inactivity. 1848inactivity.
1249 1849
1250 static void 1850 static void
1251 timeout_cb (struct ev_loop *loop, struct ev_timer *w, int revents) 1851 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
1252 { 1852 {
1253 .. ten seconds without any activity 1853 .. ten seconds without any activity
1254 } 1854 }
1255 1855
1256 struct ev_timer mytimer; 1856 ev_timer mytimer;
1257 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */ 1857 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
1258 ev_timer_again (&mytimer); /* start timer */ 1858 ev_timer_again (&mytimer); /* start timer */
1259 ev_loop (loop, 0); 1859 ev_loop (loop, 0);
1260 1860
1261 // and in some piece of code that gets executed on any "activity": 1861 // and in some piece of code that gets executed on any "activity":
1262 // reset the timeout to start ticking again at 10 seconds 1862 // reset the timeout to start ticking again at 10 seconds
1263 ev_timer_again (&mytimer); 1863 ev_timer_again (&mytimer);
1264 1864
1265 1865
1266=head2 C<ev_periodic> - to cron or not to cron? 1866=head2 C<ev_periodic> - to cron or not to cron?
1267 1867
1268Periodic watchers are also timers of a kind, but they are very versatile 1868Periodic watchers are also timers of a kind, but they are very versatile
1269(and unfortunately a bit complex). 1869(and unfortunately a bit complex).
1270 1870
1271Unlike C<ev_timer>'s, they are not based on real time (or relative time) 1871Unlike C<ev_timer>, periodic watchers are not based on real time (or
1272but on wallclock time (absolute time). You can tell a periodic watcher 1872relative time, the physical time that passes) but on wall clock time
1273to trigger "at" some specific point in time. For example, if you tell a 1873(absolute time, the thing you can read on your calender or clock). The
1274periodic watcher to trigger in 10 seconds (by specifiying e.g. C<ev_now () 1874difference is that wall clock time can run faster or slower than real
1275+ 10.>) and then reset your system clock to the last year, then it will 1875time, and time jumps are not uncommon (e.g. when you adjust your
1276take a year to trigger the event (unlike an C<ev_timer>, which would trigger 1876wrist-watch).
1277roughly 10 seconds later).
1278 1877
1279They can also be used to implement vastly more complex timers, such as 1878You can tell a periodic watcher to trigger after some specific point
1280triggering an event on each midnight, local time or other, complicated, 1879in time: for example, if you tell a periodic watcher to trigger "in 10
1281rules. 1880seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
1881not a delay) and then reset your system clock to January of the previous
1882year, then it will take a year or more to trigger the event (unlike an
1883C<ev_timer>, which would still trigger roughly 10 seconds after starting
1884it, as it uses a relative timeout).
1282 1885
1886C<ev_periodic> watchers can also be used to implement vastly more complex
1887timers, such as triggering an event on each "midnight, local time", or
1888other complicated rules. This cannot be done with C<ev_timer> watchers, as
1889those cannot react to time jumps.
1890
1283As with timers, the callback is guarenteed to be invoked only when the 1891As with timers, the callback is guaranteed to be invoked only when the
1284time (C<at>) has been passed, but if multiple periodic timers become ready 1892point in time where it is supposed to trigger has passed. If multiple
1285during the same loop iteration then order of execution is undefined. 1893timers become ready during the same loop iteration then the ones with
1894earlier time-out values are invoked before ones with later time-out values
1895(but this is no longer true when a callback calls C<ev_loop> recursively).
1286 1896
1287=head3 Watcher-Specific Functions and Data Members 1897=head3 Watcher-Specific Functions and Data Members
1288 1898
1289=over 4 1899=over 4
1290 1900
1291=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb) 1901=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
1292 1902
1293=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb) 1903=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
1294 1904
1295Lots of arguments, lets sort it out... There are basically three modes of 1905Lots of arguments, let's sort it out... There are basically three modes of
1296operation, and we will explain them from simplest to complex: 1906operation, and we will explain them from simplest to most complex:
1297 1907
1298=over 4 1908=over 4
1299 1909
1300=item * absolute timer (at = time, interval = reschedule_cb = 0) 1910=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
1301 1911
1302In this configuration the watcher triggers an event at the wallclock time 1912In this configuration the watcher triggers an event after the wall clock
1303C<at> and doesn't repeat. It will not adjust when a time jump occurs, 1913time C<offset> has passed. It will not repeat and will not adjust when a
1304that is, if it is to be run at January 1st 2011 then it will run when the 1914time jump occurs, that is, if it is to be run at January 1st 2011 then it
1305system time reaches or surpasses this time. 1915will be stopped and invoked when the system clock reaches or surpasses
1916this point in time.
1306 1917
1307=item * repeating interval timer (at = offset, interval > 0, reschedule_cb = 0) 1918=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
1308 1919
1309In this mode the watcher will always be scheduled to time out at the next 1920In this mode the watcher will always be scheduled to time out at the next
1310C<at + N * interval> time (for some integer N, which can also be negative) 1921C<offset + N * interval> time (for some integer N, which can also be
1311and then repeat, regardless of any time jumps. 1922negative) and then repeat, regardless of any time jumps. The C<offset>
1923argument is merely an offset into the C<interval> periods.
1312 1924
1313This can be used to create timers that do not drift with respect to system 1925This can be used to create timers that do not drift with respect to the
1314time: 1926system clock, for example, here is an C<ev_periodic> that triggers each
1927hour, on the hour (with respect to UTC):
1315 1928
1316 ev_periodic_set (&periodic, 0., 3600., 0); 1929 ev_periodic_set (&periodic, 0., 3600., 0);
1317 1930
1318This doesn't mean there will always be 3600 seconds in between triggers, 1931This doesn't mean there will always be 3600 seconds in between triggers,
1319but only that the the callback will be called when the system time shows a 1932but only that the callback will be called when the system time shows a
1320full hour (UTC), or more correctly, when the system time is evenly divisible 1933full hour (UTC), or more correctly, when the system time is evenly divisible
1321by 3600. 1934by 3600.
1322 1935
1323Another way to think about it (for the mathematically inclined) is that 1936Another way to think about it (for the mathematically inclined) is that
1324C<ev_periodic> will try to run the callback in this mode at the next possible 1937C<ev_periodic> will try to run the callback in this mode at the next possible
1325time where C<time = at (mod interval)>, regardless of any time jumps. 1938time where C<time = offset (mod interval)>, regardless of any time jumps.
1326 1939
1327For numerical stability it is preferable that the C<at> value is near 1940For numerical stability it is preferable that the C<offset> value is near
1328C<ev_now ()> (the current time), but there is no range requirement for 1941C<ev_now ()> (the current time), but there is no range requirement for
1329this value. 1942this value, and in fact is often specified as zero.
1330 1943
1944Note also that there is an upper limit to how often a timer can fire (CPU
1945speed for example), so if C<interval> is very small then timing stability
1946will of course deteriorate. Libev itself tries to be exact to be about one
1947millisecond (if the OS supports it and the machine is fast enough).
1948
1331=item * manual reschedule mode (at and interval ignored, reschedule_cb = callback) 1949=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
1332 1950
1333In this mode the values for C<interval> and C<at> are both being 1951In this mode the values for C<interval> and C<offset> are both being
1334ignored. Instead, each time the periodic watcher gets scheduled, the 1952ignored. Instead, each time the periodic watcher gets scheduled, the
1335reschedule callback will be called with the watcher as first, and the 1953reschedule callback will be called with the watcher as first, and the
1336current time as second argument. 1954current time as second argument.
1337 1955
1338NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, 1956NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
1339ever, or make any event loop modifications>. If you need to stop it, 1957or make ANY other event loop modifications whatsoever, unless explicitly
1340return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by 1958allowed by documentation here>.
1341starting an C<ev_prepare> watcher, which is legal).
1342 1959
1960If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
1961it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
1962only event loop modification you are allowed to do).
1963
1343Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w, 1964The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
1344ev_tstamp now)>, e.g.: 1965*w, ev_tstamp now)>, e.g.:
1345 1966
1967 static ev_tstamp
1346 static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now) 1968 my_rescheduler (ev_periodic *w, ev_tstamp now)
1347 { 1969 {
1348 return now + 60.; 1970 return now + 60.;
1349 } 1971 }
1350 1972
1351It must return the next time to trigger, based on the passed time value 1973It must return the next time to trigger, based on the passed time value
1352(that is, the lowest time value larger than to the second argument). It 1974(that is, the lowest time value larger than to the second argument). It
1353will usually be called just before the callback will be triggered, but 1975will usually be called just before the callback will be triggered, but
1354might be called at other times, too. 1976might be called at other times, too.
1355 1977
1356NOTE: I<< This callback must always return a time that is later than the 1978NOTE: I<< This callback must always return a time that is higher than or
1357passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger. 1979equal to the passed C<now> value >>.
1358 1980
1359This can be used to create very complex timers, such as a timer that 1981This can be used to create very complex timers, such as a timer that
1360triggers on each midnight, local time. To do this, you would calculate the 1982triggers on "next midnight, local time". To do this, you would calculate the
1361next midnight after C<now> and return the timestamp value for this. How 1983next midnight after C<now> and return the timestamp value for this. How
1362you do this is, again, up to you (but it is not trivial, which is the main 1984you do this is, again, up to you (but it is not trivial, which is the main
1363reason I omitted it as an example). 1985reason I omitted it as an example).
1364 1986
1365=back 1987=back
1369Simply stops and restarts the periodic watcher again. This is only useful 1991Simply stops and restarts the periodic watcher again. This is only useful
1370when you changed some parameters or the reschedule callback would return 1992when you changed some parameters or the reschedule callback would return
1371a different time than the last time it was called (e.g. in a crond like 1993a different time than the last time it was called (e.g. in a crond like
1372program when the crontabs have changed). 1994program when the crontabs have changed).
1373 1995
1996=item ev_tstamp ev_periodic_at (ev_periodic *)
1997
1998When active, returns the absolute time that the watcher is supposed
1999to trigger next. This is not the same as the C<offset> argument to
2000C<ev_periodic_set>, but indeed works even in interval and manual
2001rescheduling modes.
2002
1374=item ev_tstamp offset [read-write] 2003=item ev_tstamp offset [read-write]
1375 2004
1376When repeating, this contains the offset value, otherwise this is the 2005When repeating, this contains the offset value, otherwise this is the
1377absolute point in time (the C<at> value passed to C<ev_periodic_set>). 2006absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2007although libev might modify this value for better numerical stability).
1378 2008
1379Can be modified any time, but changes only take effect when the periodic 2009Can be modified any time, but changes only take effect when the periodic
1380timer fires or C<ev_periodic_again> is being called. 2010timer fires or C<ev_periodic_again> is being called.
1381 2011
1382=item ev_tstamp interval [read-write] 2012=item ev_tstamp interval [read-write]
1383 2013
1384The current interval value. Can be modified any time, but changes only 2014The current interval value. Can be modified any time, but changes only
1385take effect when the periodic timer fires or C<ev_periodic_again> is being 2015take effect when the periodic timer fires or C<ev_periodic_again> is being
1386called. 2016called.
1387 2017
1388=item ev_tstamp (*reschedule_cb)(struct ev_periodic *w, ev_tstamp now) [read-write] 2018=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
1389 2019
1390The current reschedule callback, or C<0>, if this functionality is 2020The current reschedule callback, or C<0>, if this functionality is
1391switched off. Can be changed any time, but changes only take effect when 2021switched off. Can be changed any time, but changes only take effect when
1392the periodic timer fires or C<ev_periodic_again> is being called. 2022the periodic timer fires or C<ev_periodic_again> is being called.
1393 2023
1394=item ev_tstamp at [read-only]
1395
1396When active, contains the absolute time that the watcher is supposed to
1397trigger next.
1398
1399=back 2024=back
1400 2025
1401=head3 Examples 2026=head3 Examples
1402 2027
1403Example: Call a callback every hour, or, more precisely, whenever the 2028Example: Call a callback every hour, or, more precisely, whenever the
1404system clock is divisible by 3600. The callback invocation times have 2029system time is divisible by 3600. The callback invocation times have
1405potentially a lot of jittering, but good long-term stability. 2030potentially a lot of jitter, but good long-term stability.
1406 2031
1407 static void 2032 static void
1408 clock_cb (struct ev_loop *loop, struct ev_io *w, int revents) 2033 clock_cb (struct ev_loop *loop, ev_io *w, int revents)
1409 { 2034 {
1410 ... its now a full hour (UTC, or TAI or whatever your clock follows) 2035 ... its now a full hour (UTC, or TAI or whatever your clock follows)
1411 } 2036 }
1412 2037
1413 struct ev_periodic hourly_tick; 2038 ev_periodic hourly_tick;
1414 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0); 2039 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
1415 ev_periodic_start (loop, &hourly_tick); 2040 ev_periodic_start (loop, &hourly_tick);
1416 2041
1417Example: The same as above, but use a reschedule callback to do it: 2042Example: The same as above, but use a reschedule callback to do it:
1418 2043
1419 #include <math.h> 2044 #include <math.h>
1420 2045
1421 static ev_tstamp 2046 static ev_tstamp
1422 my_scheduler_cb (struct ev_periodic *w, ev_tstamp now) 2047 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
1423 { 2048 {
1424 return fmod (now, 3600.) + 3600.; 2049 return now + (3600. - fmod (now, 3600.));
1425 } 2050 }
1426 2051
1427 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb); 2052 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
1428 2053
1429Example: Call a callback every hour, starting now: 2054Example: Call a callback every hour, starting now:
1430 2055
1431 struct ev_periodic hourly_tick; 2056 ev_periodic hourly_tick;
1432 ev_periodic_init (&hourly_tick, clock_cb, 2057 ev_periodic_init (&hourly_tick, clock_cb,
1433 fmod (ev_now (loop), 3600.), 3600., 0); 2058 fmod (ev_now (loop), 3600.), 3600., 0);
1434 ev_periodic_start (loop, &hourly_tick); 2059 ev_periodic_start (loop, &hourly_tick);
1435 2060
1436 2061
1437=head2 C<ev_signal> - signal me when a signal gets signalled! 2062=head2 C<ev_signal> - signal me when a signal gets signalled!
1438 2063
1439Signal watchers will trigger an event when the process receives a specific 2064Signal watchers will trigger an event when the process receives a specific
1440signal one or more times. Even though signals are very asynchronous, libev 2065signal one or more times. Even though signals are very asynchronous, libev
1441will try it's best to deliver signals synchronously, i.e. as part of the 2066will try it's best to deliver signals synchronously, i.e. as part of the
1442normal event processing, like any other event. 2067normal event processing, like any other event.
1443 2068
2069If you want signals asynchronously, just use C<sigaction> as you would
2070do without libev and forget about sharing the signal. You can even use
2071C<ev_async> from a signal handler to synchronously wake up an event loop.
2072
1444You can configure as many watchers as you like per signal. Only when the 2073You can configure as many watchers as you like per signal. Only when the
1445first watcher gets started will libev actually register a signal watcher 2074first watcher gets started will libev actually register a signal handler
1446with the kernel (thus it coexists with your own signal handlers as long 2075with the kernel (thus it coexists with your own signal handlers as long as
1447as you don't register any with libev). Similarly, when the last signal 2076you don't register any with libev for the same signal). Similarly, when
1448watcher for a signal is stopped libev will reset the signal handler to 2077the last signal watcher for a signal is stopped, libev will reset the
1449SIG_DFL (regardless of what it was set to before). 2078signal handler to SIG_DFL (regardless of what it was set to before).
1450 2079
1451If possible and supported, libev will install its handlers with 2080If possible and supported, libev will install its handlers with
1452C<SA_RESTART> behaviour enabled, so syscalls should not be unduly 2081C<SA_RESTART> behaviour enabled, so system calls should not be unduly
1453interrupted. If you have a problem with syscalls getting interrupted by 2082interrupted. If you have a problem with system calls getting interrupted by
1454signals you can block all signals in an C<ev_check> watcher and unblock 2083signals you can block all signals in an C<ev_check> watcher and unblock
1455them in an C<ev_prepare> watcher. 2084them in an C<ev_prepare> watcher.
1456 2085
1457=head3 Watcher-Specific Functions and Data Members 2086=head3 Watcher-Specific Functions and Data Members
1458 2087
1471 2100
1472=back 2101=back
1473 2102
1474=head3 Examples 2103=head3 Examples
1475 2104
1476Example: Try to exit cleanly on SIGINT and SIGTERM. 2105Example: Try to exit cleanly on SIGINT.
1477 2106
1478 static void 2107 static void
1479 sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents) 2108 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
1480 { 2109 {
1481 ev_unloop (loop, EVUNLOOP_ALL); 2110 ev_unloop (loop, EVUNLOOP_ALL);
1482 } 2111 }
1483 2112
1484 struct ev_signal signal_watcher; 2113 ev_signal signal_watcher;
1485 ev_signal_init (&signal_watcher, sigint_cb, SIGINT); 2114 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
1486 ev_signal_start (loop, &sigint_cb); 2115 ev_signal_start (loop, &signal_watcher);
1487 2116
1488 2117
1489=head2 C<ev_child> - watch out for process status changes 2118=head2 C<ev_child> - watch out for process status changes
1490 2119
1491Child watchers trigger when your process receives a SIGCHLD in response to 2120Child watchers trigger when your process receives a SIGCHLD in response to
1492some child status changes (most typically when a child of yours dies). It 2121some child status changes (most typically when a child of yours dies or
1493is permissible to install a child watcher I<after> the child has been 2122exits). It is permissible to install a child watcher I<after> the child
1494forked (which implies it might have already exited), as long as the event 2123has been forked (which implies it might have already exited), as long
1495loop isn't entered (or is continued from a watcher). 2124as the event loop isn't entered (or is continued from a watcher), i.e.,
2125forking and then immediately registering a watcher for the child is fine,
2126but forking and registering a watcher a few event loop iterations later or
2127in the next callback invocation is not.
1496 2128
1497Only the default event loop is capable of handling signals, and therefore 2129Only the default event loop is capable of handling signals, and therefore
1498you can only rgeister child watchers in the default event loop. 2130you can only register child watchers in the default event loop.
2131
2132Due to some design glitches inside libev, child watchers will always be
2133handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2134libev)
1499 2135
1500=head3 Process Interaction 2136=head3 Process Interaction
1501 2137
1502Libev grabs C<SIGCHLD> as soon as the default event loop is 2138Libev grabs C<SIGCHLD> as soon as the default event loop is
1503initialised. This is necessary to guarantee proper behaviour even if 2139initialised. This is necessary to guarantee proper behaviour even if
1504the first child watcher is started after the child exits. The occurance 2140the first child watcher is started after the child exits. The occurrence
1505of C<SIGCHLD> is recorded asynchronously, but child reaping is done 2141of C<SIGCHLD> is recorded asynchronously, but child reaping is done
1506synchronously as part of the event loop processing. Libev always reaps all 2142synchronously as part of the event loop processing. Libev always reaps all
1507children, even ones not watched. 2143children, even ones not watched.
1508 2144
1509=head3 Overriding the Built-In Processing 2145=head3 Overriding the Built-In Processing
1513handler, you can override it easily by installing your own handler for 2149handler, you can override it easily by installing your own handler for
1514C<SIGCHLD> after initialising the default loop, and making sure the 2150C<SIGCHLD> after initialising the default loop, and making sure the
1515default loop never gets destroyed. You are encouraged, however, to use an 2151default loop never gets destroyed. You are encouraged, however, to use an
1516event-based approach to child reaping and thus use libev's support for 2152event-based approach to child reaping and thus use libev's support for
1517that, so other libev users can use C<ev_child> watchers freely. 2153that, so other libev users can use C<ev_child> watchers freely.
2154
2155=head3 Stopping the Child Watcher
2156
2157Currently, the child watcher never gets stopped, even when the
2158child terminates, so normally one needs to stop the watcher in the
2159callback. Future versions of libev might stop the watcher automatically
2160when a child exit is detected.
1518 2161
1519=head3 Watcher-Specific Functions and Data Members 2162=head3 Watcher-Specific Functions and Data Members
1520 2163
1521=over 4 2164=over 4
1522 2165
1551=head3 Examples 2194=head3 Examples
1552 2195
1553Example: C<fork()> a new process and install a child handler to wait for 2196Example: C<fork()> a new process and install a child handler to wait for
1554its completion. 2197its completion.
1555 2198
1556 ev_child cw; 2199 ev_child cw;
1557 2200
1558 static void 2201 static void
1559 child_cb (EV_P_ struct ev_child *w, int revents) 2202 child_cb (EV_P_ ev_child *w, int revents)
1560 { 2203 {
1561 ev_child_stop (EV_A_ w); 2204 ev_child_stop (EV_A_ w);
1562 printf ("process %d exited with status %x\n", w->rpid, w->rstatus); 2205 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
1563 } 2206 }
1564 2207
1565 pid_t pid = fork (); 2208 pid_t pid = fork ();
1566 2209
1567 if (pid < 0) 2210 if (pid < 0)
1568 // error 2211 // error
1569 else if (pid == 0) 2212 else if (pid == 0)
1570 { 2213 {
1571 // the forked child executes here 2214 // the forked child executes here
1572 exit (1); 2215 exit (1);
1573 } 2216 }
1574 else 2217 else
1575 { 2218 {
1576 ev_child_init (&cw, child_cb, pid, 0); 2219 ev_child_init (&cw, child_cb, pid, 0);
1577 ev_child_start (EV_DEFAULT_ &cw); 2220 ev_child_start (EV_DEFAULT_ &cw);
1578 } 2221 }
1579 2222
1580 2223
1581=head2 C<ev_stat> - did the file attributes just change? 2224=head2 C<ev_stat> - did the file attributes just change?
1582 2225
1583This watches a filesystem path for attribute changes. That is, it calls 2226This watches a file system path for attribute changes. That is, it calls
1584C<stat> regularly (or when the OS says it changed) and sees if it changed 2227C<stat> on that path in regular intervals (or when the OS says it changed)
1585compared to the last time, invoking the callback if it did. 2228and sees if it changed compared to the last time, invoking the callback if
2229it did.
1586 2230
1587The path does not need to exist: changing from "path exists" to "path does 2231The path does not need to exist: changing from "path exists" to "path does
1588not exist" is a status change like any other. The condition "path does 2232not exist" is a status change like any other. The condition "path does not
1589not exist" is signified by the C<st_nlink> field being zero (which is 2233exist" (or more correctly "path cannot be stat'ed") is signified by the
1590otherwise always forced to be at least one) and all the other fields of 2234C<st_nlink> field being zero (which is otherwise always forced to be at
1591the stat buffer having unspecified contents. 2235least one) and all the other fields of the stat buffer having unspecified
2236contents.
1592 2237
1593The path I<should> be absolute and I<must not> end in a slash. If it is 2238The path I<must not> end in a slash or contain special components such as
2239C<.> or C<..>. The path I<should> be absolute: If it is relative and
1594relative and your working directory changes, the behaviour is undefined. 2240your working directory changes, then the behaviour is undefined.
1595 2241
1596Since there is no standard to do this, the portable implementation simply 2242Since there is no portable change notification interface available, the
1597calls C<stat (2)> regularly on the path to see if it changed somehow. You 2243portable implementation simply calls C<stat(2)> regularly on the path
1598can specify a recommended polling interval for this case. If you specify 2244to see if it changed somehow. You can specify a recommended polling
1599a polling interval of C<0> (highly recommended!) then a I<suitable, 2245interval for this case. If you specify a polling interval of C<0> (highly
1600unspecified default> value will be used (which you can expect to be around 2246recommended!) then a I<suitable, unspecified default> value will be used
1601five seconds, although this might change dynamically). Libev will also 2247(which you can expect to be around five seconds, although this might
1602impose a minimum interval which is currently around C<0.1>, but thats 2248change dynamically). Libev will also impose a minimum interval which is
1603usually overkill. 2249currently around C<0.1>, but that's usually overkill.
1604 2250
1605This watcher type is not meant for massive numbers of stat watchers, 2251This watcher type is not meant for massive numbers of stat watchers,
1606as even with OS-supported change notifications, this can be 2252as even with OS-supported change notifications, this can be
1607resource-intensive. 2253resource-intensive.
1608 2254
1609At the time of this writing, only the Linux inotify interface is 2255At the time of this writing, the only OS-specific interface implemented
1610implemented (implementing kqueue support is left as an exercise for the 2256is the Linux inotify interface (implementing kqueue support is left as an
1611reader). Inotify will be used to give hints only and should not change the 2257exercise for the reader. Note, however, that the author sees no way of
1612semantics of C<ev_stat> watchers, which means that libev sometimes needs 2258implementing C<ev_stat> semantics with kqueue, except as a hint).
1613to fall back to regular polling again even with inotify, but changes are
1614usually detected immediately, and if the file exists there will be no
1615polling.
1616 2259
1617=head3 ABI Issues (Largefile Support) 2260=head3 ABI Issues (Largefile Support)
1618 2261
1619Libev by default (unless the user overrides this) uses the default 2262Libev by default (unless the user overrides this) uses the default
1620compilation environment, which means that on systems with optionally 2263compilation environment, which means that on systems with large file
1621disabled large file support, you get the 32 bit version of the stat 2264support disabled by default, you get the 32 bit version of the stat
1622structure. When using the library from programs that change the ABI to 2265structure. When using the library from programs that change the ABI to
1623use 64 bit file offsets the programs will fail. In that case you have to 2266use 64 bit file offsets the programs will fail. In that case you have to
1624compile libev with the same flags to get binary compatibility. This is 2267compile libev with the same flags to get binary compatibility. This is
1625obviously the case with any flags that change the ABI, but the problem is 2268obviously the case with any flags that change the ABI, but the problem is
1626most noticably with ev_stat and largefile support. 2269most noticeably displayed with ev_stat and large file support.
1627 2270
1628=head3 Inotify 2271The solution for this is to lobby your distribution maker to make large
2272file interfaces available by default (as e.g. FreeBSD does) and not
2273optional. Libev cannot simply switch on large file support because it has
2274to exchange stat structures with application programs compiled using the
2275default compilation environment.
1629 2276
2277=head3 Inotify and Kqueue
2278
1630When C<inotify (7)> support has been compiled into libev (generally only 2279When C<inotify (7)> support has been compiled into libev and present at
1631available on Linux) and present at runtime, it will be used to speed up 2280runtime, it will be used to speed up change detection where possible. The
1632change detection where possible. The inotify descriptor will be created lazily 2281inotify descriptor will be created lazily when the first C<ev_stat>
1633when the first C<ev_stat> watcher is being started. 2282watcher is being started.
1634 2283
1635Inotify presense does not change the semantics of C<ev_stat> watchers 2284Inotify presence does not change the semantics of C<ev_stat> watchers
1636except that changes might be detected earlier, and in some cases, to avoid 2285except that changes might be detected earlier, and in some cases, to avoid
1637making regular C<stat> calls. Even in the presense of inotify support 2286making regular C<stat> calls. Even in the presence of inotify support
1638there are many cases where libev has to resort to regular C<stat> polling. 2287there are many cases where libev has to resort to regular C<stat> polling,
2288but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2289many bugs), the path exists (i.e. stat succeeds), and the path resides on
2290a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2291xfs are fully working) libev usually gets away without polling.
1639 2292
1640(There is no support for kqueue, as apparently it cannot be used to 2293There is no support for kqueue, as apparently it cannot be used to
1641implement this functionality, due to the requirement of having a file 2294implement this functionality, due to the requirement of having a file
1642descriptor open on the object at all times). 2295descriptor open on the object at all times, and detecting renames, unlinks
2296etc. is difficult.
2297
2298=head3 C<stat ()> is a synchronous operation
2299
2300Libev doesn't normally do any kind of I/O itself, and so is not blocking
2301the process. The exception are C<ev_stat> watchers - those call C<stat
2302()>, which is a synchronous operation.
2303
2304For local paths, this usually doesn't matter: unless the system is very
2305busy or the intervals between stat's are large, a stat call will be fast,
2306as the path data is usually in memory already (except when starting the
2307watcher).
2308
2309For networked file systems, calling C<stat ()> can block an indefinite
2310time due to network issues, and even under good conditions, a stat call
2311often takes multiple milliseconds.
2312
2313Therefore, it is best to avoid using C<ev_stat> watchers on networked
2314paths, although this is fully supported by libev.
1643 2315
1644=head3 The special problem of stat time resolution 2316=head3 The special problem of stat time resolution
1645 2317
1646The C<stat ()> syscall only supports full-second resolution portably, and 2318The C<stat ()> system call only supports full-second resolution portably,
1647even on systems where the resolution is higher, many filesystems still 2319and even on systems where the resolution is higher, most file systems
1648only support whole seconds. 2320still only support whole seconds.
1649 2321
1650That means that, if the time is the only thing that changes, you might 2322That means that, if the time is the only thing that changes, you can
1651miss updates: on the first update, C<ev_stat> detects a change and calls 2323easily miss updates: on the first update, C<ev_stat> detects a change and
1652your callback, which does something. When there is another update within 2324calls your callback, which does something. When there is another update
1653the same second, C<ev_stat> will be unable to detect it. 2325within the same second, C<ev_stat> will be unable to detect unless the
2326stat data does change in other ways (e.g. file size).
1654 2327
1655The solution to this is to delay acting on a change for a second (or till 2328The solution to this is to delay acting on a change for slightly more
1656the next second boundary), using a roughly one-second delay C<ev_timer> 2329than a second (or till slightly after the next full second boundary), using
1657(C<ev_timer_set (w, 0., 1.01); ev_timer_again (loop, w)>). The C<.01> 2330a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
1658is added to work around small timing inconsistencies of some operating 2331ev_timer_again (loop, w)>).
1659systems. 2332
2333The C<.02> offset is added to work around small timing inconsistencies
2334of some operating systems (where the second counter of the current time
2335might be be delayed. One such system is the Linux kernel, where a call to
2336C<gettimeofday> might return a timestamp with a full second later than
2337a subsequent C<time> call - if the equivalent of C<time ()> is used to
2338update file times then there will be a small window where the kernel uses
2339the previous second to update file times but libev might already execute
2340the timer callback).
1660 2341
1661=head3 Watcher-Specific Functions and Data Members 2342=head3 Watcher-Specific Functions and Data Members
1662 2343
1663=over 4 2344=over 4
1664 2345
1670C<path>. The C<interval> is a hint on how quickly a change is expected to 2351C<path>. The C<interval> is a hint on how quickly a change is expected to
1671be detected and should normally be specified as C<0> to let libev choose 2352be detected and should normally be specified as C<0> to let libev choose
1672a suitable value. The memory pointed to by C<path> must point to the same 2353a suitable value. The memory pointed to by C<path> must point to the same
1673path for as long as the watcher is active. 2354path for as long as the watcher is active.
1674 2355
1675The callback will be receive C<EV_STAT> when a change was detected, 2356The callback will receive an C<EV_STAT> event when a change was detected,
1676relative to the attributes at the time the watcher was started (or the 2357relative to the attributes at the time the watcher was started (or the
1677last change was detected). 2358last change was detected).
1678 2359
1679=item ev_stat_stat (loop, ev_stat *) 2360=item ev_stat_stat (loop, ev_stat *)
1680 2361
1681Updates the stat buffer immediately with new values. If you change the 2362Updates the stat buffer immediately with new values. If you change the
1682watched path in your callback, you could call this fucntion to avoid 2363watched path in your callback, you could call this function to avoid
1683detecting this change (while introducing a race condition). Can also be 2364detecting this change (while introducing a race condition if you are not
1684useful simply to find out the new values. 2365the only one changing the path). Can also be useful simply to find out the
2366new values.
1685 2367
1686=item ev_statdata attr [read-only] 2368=item ev_statdata attr [read-only]
1687 2369
1688The most-recently detected attributes of the file. Although the type is of 2370The most-recently detected attributes of the file. Although the type is
1689C<ev_statdata>, this is usually the (or one of the) C<struct stat> types 2371C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
1690suitable for your system. If the C<st_nlink> member is C<0>, then there 2372suitable for your system, but you can only rely on the POSIX-standardised
2373members to be present. If the C<st_nlink> member is C<0>, then there was
1691was some error while C<stat>ing the file. 2374some error while C<stat>ing the file.
1692 2375
1693=item ev_statdata prev [read-only] 2376=item ev_statdata prev [read-only]
1694 2377
1695The previous attributes of the file. The callback gets invoked whenever 2378The previous attributes of the file. The callback gets invoked whenever
1696C<prev> != C<attr>. 2379C<prev> != C<attr>, or, more precisely, one or more of these members
2380differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2381C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
1697 2382
1698=item ev_tstamp interval [read-only] 2383=item ev_tstamp interval [read-only]
1699 2384
1700The specified interval. 2385The specified interval.
1701 2386
1702=item const char *path [read-only] 2387=item const char *path [read-only]
1703 2388
1704The filesystem path that is being watched. 2389The file system path that is being watched.
1705 2390
1706=back 2391=back
1707 2392
1708=head3 Examples 2393=head3 Examples
1709 2394
1710Example: Watch C</etc/passwd> for attribute changes. 2395Example: Watch C</etc/passwd> for attribute changes.
1711 2396
1712 static void 2397 static void
1713 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents) 2398 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
1714 { 2399 {
1715 /* /etc/passwd changed in some way */ 2400 /* /etc/passwd changed in some way */
1716 if (w->attr.st_nlink) 2401 if (w->attr.st_nlink)
1717 { 2402 {
1718 printf ("passwd current size %ld\n", (long)w->attr.st_size); 2403 printf ("passwd current size %ld\n", (long)w->attr.st_size);
1719 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime); 2404 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
1720 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime); 2405 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
1721 } 2406 }
1722 else 2407 else
1723 /* you shalt not abuse printf for puts */ 2408 /* you shalt not abuse printf for puts */
1724 puts ("wow, /etc/passwd is not there, expect problems. " 2409 puts ("wow, /etc/passwd is not there, expect problems. "
1725 "if this is windows, they already arrived\n"); 2410 "if this is windows, they already arrived\n");
1726 } 2411 }
1727 2412
1728 ... 2413 ...
1729 ev_stat passwd; 2414 ev_stat passwd;
1730 2415
1731 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.); 2416 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
1732 ev_stat_start (loop, &passwd); 2417 ev_stat_start (loop, &passwd);
1733 2418
1734Example: Like above, but additionally use a one-second delay so we do not 2419Example: Like above, but additionally use a one-second delay so we do not
1735miss updates (however, frequent updates will delay processing, too, so 2420miss updates (however, frequent updates will delay processing, too, so
1736one might do the work both on C<ev_stat> callback invocation I<and> on 2421one might do the work both on C<ev_stat> callback invocation I<and> on
1737C<ev_timer> callback invocation). 2422C<ev_timer> callback invocation).
1738 2423
1739 static ev_stat passwd; 2424 static ev_stat passwd;
1740 static ev_timer timer; 2425 static ev_timer timer;
1741 2426
1742 static void 2427 static void
1743 timer_cb (EV_P_ ev_timer *w, int revents) 2428 timer_cb (EV_P_ ev_timer *w, int revents)
1744 { 2429 {
1745 ev_timer_stop (EV_A_ w); 2430 ev_timer_stop (EV_A_ w);
1746 2431
1747 /* now it's one second after the most recent passwd change */ 2432 /* now it's one second after the most recent passwd change */
1748 } 2433 }
1749 2434
1750 static void 2435 static void
1751 stat_cb (EV_P_ ev_stat *w, int revents) 2436 stat_cb (EV_P_ ev_stat *w, int revents)
1752 { 2437 {
1753 /* reset the one-second timer */ 2438 /* reset the one-second timer */
1754 ev_timer_again (EV_A_ &timer); 2439 ev_timer_again (EV_A_ &timer);
1755 } 2440 }
1756 2441
1757 ... 2442 ...
1758 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.); 2443 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
1759 ev_stat_start (loop, &passwd); 2444 ev_stat_start (loop, &passwd);
1760 ev_timer_init (&timer, timer_cb, 0., 1.01); 2445 ev_timer_init (&timer, timer_cb, 0., 1.02);
1761 2446
1762 2447
1763=head2 C<ev_idle> - when you've got nothing better to do... 2448=head2 C<ev_idle> - when you've got nothing better to do...
1764 2449
1765Idle watchers trigger events when no other events of the same or higher 2450Idle watchers trigger events when no other events of the same or higher
1766priority are pending (prepare, check and other idle watchers do not 2451priority are pending (prepare, check and other idle watchers do not count
1767count). 2452as receiving "events").
1768 2453
1769That is, as long as your process is busy handling sockets or timeouts 2454That is, as long as your process is busy handling sockets or timeouts
1770(or even signals, imagine) of the same or higher priority it will not be 2455(or even signals, imagine) of the same or higher priority it will not be
1771triggered. But when your process is idle (or only lower-priority watchers 2456triggered. But when your process is idle (or only lower-priority watchers
1772are pending), the idle watchers are being called once per event loop 2457are pending), the idle watchers are being called once per event loop
1783 2468
1784=head3 Watcher-Specific Functions and Data Members 2469=head3 Watcher-Specific Functions and Data Members
1785 2470
1786=over 4 2471=over 4
1787 2472
1788=item ev_idle_init (ev_signal *, callback) 2473=item ev_idle_init (ev_idle *, callback)
1789 2474
1790Initialises and configures the idle watcher - it has no parameters of any 2475Initialises and configures the idle watcher - it has no parameters of any
1791kind. There is a C<ev_idle_set> macro, but using it is utterly pointless, 2476kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
1792believe me. 2477believe me.
1793 2478
1796=head3 Examples 2481=head3 Examples
1797 2482
1798Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the 2483Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
1799callback, free it. Also, use no error checking, as usual. 2484callback, free it. Also, use no error checking, as usual.
1800 2485
1801 static void 2486 static void
1802 idle_cb (struct ev_loop *loop, struct ev_idle *w, int revents) 2487 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
1803 { 2488 {
1804 free (w); 2489 free (w);
1805 // now do something you wanted to do when the program has 2490 // now do something you wanted to do when the program has
1806 // no longer anything immediate to do. 2491 // no longer anything immediate to do.
1807 } 2492 }
1808 2493
1809 struct ev_idle *idle_watcher = malloc (sizeof (struct ev_idle)); 2494 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
1810 ev_idle_init (idle_watcher, idle_cb); 2495 ev_idle_init (idle_watcher, idle_cb);
1811 ev_idle_start (loop, idle_cb); 2496 ev_idle_start (loop, idle_watcher);
1812 2497
1813 2498
1814=head2 C<ev_prepare> and C<ev_check> - customise your event loop! 2499=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
1815 2500
1816Prepare and check watchers are usually (but not always) used in tandem: 2501Prepare and check watchers are usually (but not always) used in pairs:
1817prepare watchers get invoked before the process blocks and check watchers 2502prepare watchers get invoked before the process blocks and check watchers
1818afterwards. 2503afterwards.
1819 2504
1820You I<must not> call C<ev_loop> or similar functions that enter 2505You I<must not> call C<ev_loop> or similar functions that enter
1821the current event loop from either C<ev_prepare> or C<ev_check> 2506the current event loop from either C<ev_prepare> or C<ev_check>
1824those watchers, i.e. the sequence will always be C<ev_prepare>, blocking, 2509those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
1825C<ev_check> so if you have one watcher of each kind they will always be 2510C<ev_check> so if you have one watcher of each kind they will always be
1826called in pairs bracketing the blocking call. 2511called in pairs bracketing the blocking call.
1827 2512
1828Their main purpose is to integrate other event mechanisms into libev and 2513Their main purpose is to integrate other event mechanisms into libev and
1829their use is somewhat advanced. This could be used, for example, to track 2514their use is somewhat advanced. They could be used, for example, to track
1830variable changes, implement your own watchers, integrate net-snmp or a 2515variable changes, implement your own watchers, integrate net-snmp or a
1831coroutine library and lots more. They are also occasionally useful if 2516coroutine library and lots more. They are also occasionally useful if
1832you cache some data and want to flush it before blocking (for example, 2517you cache some data and want to flush it before blocking (for example,
1833in X programs you might want to do an C<XFlush ()> in an C<ev_prepare> 2518in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
1834watcher). 2519watcher).
1835 2520
1836This is done by examining in each prepare call which file descriptors need 2521This is done by examining in each prepare call which file descriptors
1837to be watched by the other library, registering C<ev_io> watchers for 2522need to be watched by the other library, registering C<ev_io> watchers
1838them and starting an C<ev_timer> watcher for any timeouts (many libraries 2523for them and starting an C<ev_timer> watcher for any timeouts (many
1839provide just this functionality). Then, in the check watcher you check for 2524libraries provide exactly this functionality). Then, in the check watcher,
1840any events that occured (by checking the pending status of all watchers 2525you check for any events that occurred (by checking the pending status
1841and stopping them) and call back into the library. The I/O and timer 2526of all watchers and stopping them) and call back into the library. The
1842callbacks will never actually be called (but must be valid nevertheless, 2527I/O and timer callbacks will never actually be called (but must be valid
1843because you never know, you know?). 2528nevertheless, because you never know, you know?).
1844 2529
1845As another example, the Perl Coro module uses these hooks to integrate 2530As another example, the Perl Coro module uses these hooks to integrate
1846coroutines into libev programs, by yielding to other active coroutines 2531coroutines into libev programs, by yielding to other active coroutines
1847during each prepare and only letting the process block if no coroutines 2532during each prepare and only letting the process block if no coroutines
1848are ready to run (it's actually more complicated: it only runs coroutines 2533are ready to run (it's actually more complicated: it only runs coroutines
1851loop from blocking if lower-priority coroutines are active, thus mapping 2536loop from blocking if lower-priority coroutines are active, thus mapping
1852low-priority coroutines to idle/background tasks). 2537low-priority coroutines to idle/background tasks).
1853 2538
1854It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>) 2539It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
1855priority, to ensure that they are being run before any other watchers 2540priority, to ensure that they are being run before any other watchers
2541after the poll (this doesn't matter for C<ev_prepare> watchers).
2542
1856after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers, 2543Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
1857too) should not activate ("feed") events into libev. While libev fully 2544activate ("feed") events into libev. While libev fully supports this, they
1858supports this, they will be called before other C<ev_check> watchers 2545might get executed before other C<ev_check> watchers did their job. As
1859did their job. As C<ev_check> watchers are often used to embed other 2546C<ev_check> watchers are often used to embed other (non-libev) event
1860(non-libev) event loops those other event loops might be in an unusable 2547loops those other event loops might be in an unusable state until their
1861state until their C<ev_check> watcher ran (always remind yourself to 2548C<ev_check> watcher ran (always remind yourself to coexist peacefully with
1862coexist peacefully with others). 2549others).
1863 2550
1864=head3 Watcher-Specific Functions and Data Members 2551=head3 Watcher-Specific Functions and Data Members
1865 2552
1866=over 4 2553=over 4
1867 2554
1869 2556
1870=item ev_check_init (ev_check *, callback) 2557=item ev_check_init (ev_check *, callback)
1871 2558
1872Initialises and configures the prepare or check watcher - they have no 2559Initialises and configures the prepare or check watcher - they have no
1873parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set> 2560parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
1874macros, but using them is utterly, utterly and completely pointless. 2561macros, but using them is utterly, utterly, utterly and completely
2562pointless.
1875 2563
1876=back 2564=back
1877 2565
1878=head3 Examples 2566=head3 Examples
1879 2567
1880There are a number of principal ways to embed other event loops or modules 2568There are a number of principal ways to embed other event loops or modules
1881into libev. Here are some ideas on how to include libadns into libev 2569into libev. Here are some ideas on how to include libadns into libev
1882(there is a Perl module named C<EV::ADNS> that does this, which you could 2570(there is a Perl module named C<EV::ADNS> that does this, which you could
1883use for an actually working example. Another Perl module named C<EV::Glib> 2571use as a working example. Another Perl module named C<EV::Glib> embeds a
1884embeds a Glib main context into libev, and finally, C<Glib::EV> embeds EV 2572Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
1885into the Glib event loop). 2573Glib event loop).
1886 2574
1887Method 1: Add IO watchers and a timeout watcher in a prepare handler, 2575Method 1: Add IO watchers and a timeout watcher in a prepare handler,
1888and in a check watcher, destroy them and call into libadns. What follows 2576and in a check watcher, destroy them and call into libadns. What follows
1889is pseudo-code only of course. This requires you to either use a low 2577is pseudo-code only of course. This requires you to either use a low
1890priority for the check watcher or use C<ev_clear_pending> explicitly, as 2578priority for the check watcher or use C<ev_clear_pending> explicitly, as
1891the callbacks for the IO/timeout watchers might not have been called yet. 2579the callbacks for the IO/timeout watchers might not have been called yet.
1892 2580
1893 static ev_io iow [nfd]; 2581 static ev_io iow [nfd];
1894 static ev_timer tw; 2582 static ev_timer tw;
1895 2583
1896 static void 2584 static void
1897 io_cb (ev_loop *loop, ev_io *w, int revents) 2585 io_cb (struct ev_loop *loop, ev_io *w, int revents)
1898 { 2586 {
1899 } 2587 }
1900 2588
1901 // create io watchers for each fd and a timer before blocking 2589 // create io watchers for each fd and a timer before blocking
1902 static void 2590 static void
1903 adns_prepare_cb (ev_loop *loop, ev_prepare *w, int revents) 2591 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
1904 { 2592 {
1905 int timeout = 3600000; 2593 int timeout = 3600000;
1906 struct pollfd fds [nfd]; 2594 struct pollfd fds [nfd];
1907 // actual code will need to loop here and realloc etc. 2595 // actual code will need to loop here and realloc etc.
1908 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ())); 2596 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
1909 2597
1910 /* the callback is illegal, but won't be called as we stop during check */ 2598 /* the callback is illegal, but won't be called as we stop during check */
1911 ev_timer_init (&tw, 0, timeout * 1e-3); 2599 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
1912 ev_timer_start (loop, &tw); 2600 ev_timer_start (loop, &tw);
1913 2601
1914 // create one ev_io per pollfd 2602 // create one ev_io per pollfd
1915 for (int i = 0; i < nfd; ++i) 2603 for (int i = 0; i < nfd; ++i)
1916 { 2604 {
1917 ev_io_init (iow + i, io_cb, fds [i].fd, 2605 ev_io_init (iow + i, io_cb, fds [i].fd,
1918 ((fds [i].events & POLLIN ? EV_READ : 0) 2606 ((fds [i].events & POLLIN ? EV_READ : 0)
1919 | (fds [i].events & POLLOUT ? EV_WRITE : 0))); 2607 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
1920 2608
1921 fds [i].revents = 0; 2609 fds [i].revents = 0;
1922 ev_io_start (loop, iow + i); 2610 ev_io_start (loop, iow + i);
1923 } 2611 }
1924 } 2612 }
1925 2613
1926 // stop all watchers after blocking 2614 // stop all watchers after blocking
1927 static void 2615 static void
1928 adns_check_cb (ev_loop *loop, ev_check *w, int revents) 2616 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
1929 { 2617 {
1930 ev_timer_stop (loop, &tw); 2618 ev_timer_stop (loop, &tw);
1931 2619
1932 for (int i = 0; i < nfd; ++i) 2620 for (int i = 0; i < nfd; ++i)
1933 { 2621 {
1934 // set the relevant poll flags 2622 // set the relevant poll flags
1935 // could also call adns_processreadable etc. here 2623 // could also call adns_processreadable etc. here
1936 struct pollfd *fd = fds + i; 2624 struct pollfd *fd = fds + i;
1937 int revents = ev_clear_pending (iow + i); 2625 int revents = ev_clear_pending (iow + i);
1938 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN; 2626 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
1939 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT; 2627 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
1940 2628
1941 // now stop the watcher 2629 // now stop the watcher
1942 ev_io_stop (loop, iow + i); 2630 ev_io_stop (loop, iow + i);
1943 } 2631 }
1944 2632
1945 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop)); 2633 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
1946 } 2634 }
1947 2635
1948Method 2: This would be just like method 1, but you run C<adns_afterpoll> 2636Method 2: This would be just like method 1, but you run C<adns_afterpoll>
1949in the prepare watcher and would dispose of the check watcher. 2637in the prepare watcher and would dispose of the check watcher.
1950 2638
1951Method 3: If the module to be embedded supports explicit event 2639Method 3: If the module to be embedded supports explicit event
1952notification (adns does), you can also make use of the actual watcher 2640notification (libadns does), you can also make use of the actual watcher
1953callbacks, and only destroy/create the watchers in the prepare watcher. 2641callbacks, and only destroy/create the watchers in the prepare watcher.
1954 2642
1955 static void 2643 static void
1956 timer_cb (EV_P_ ev_timer *w, int revents) 2644 timer_cb (EV_P_ ev_timer *w, int revents)
1957 { 2645 {
1958 adns_state ads = (adns_state)w->data; 2646 adns_state ads = (adns_state)w->data;
1959 update_now (EV_A); 2647 update_now (EV_A);
1960 2648
1961 adns_processtimeouts (ads, &tv_now); 2649 adns_processtimeouts (ads, &tv_now);
1962 } 2650 }
1963 2651
1964 static void 2652 static void
1965 io_cb (EV_P_ ev_io *w, int revents) 2653 io_cb (EV_P_ ev_io *w, int revents)
1966 { 2654 {
1967 adns_state ads = (adns_state)w->data; 2655 adns_state ads = (adns_state)w->data;
1968 update_now (EV_A); 2656 update_now (EV_A);
1969 2657
1970 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now); 2658 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
1971 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now); 2659 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
1972 } 2660 }
1973 2661
1974 // do not ever call adns_afterpoll 2662 // do not ever call adns_afterpoll
1975 2663
1976Method 4: Do not use a prepare or check watcher because the module you 2664Method 4: Do not use a prepare or check watcher because the module you
1977want to embed is too inflexible to support it. Instead, youc na override 2665want to embed is not flexible enough to support it. Instead, you can
1978their poll function. The drawback with this solution is that the main 2666override their poll function. The drawback with this solution is that the
1979loop is now no longer controllable by EV. The C<Glib::EV> module does 2667main loop is now no longer controllable by EV. The C<Glib::EV> module uses
1980this. 2668this approach, effectively embedding EV as a client into the horrible
2669libglib event loop.
1981 2670
1982 static gint 2671 static gint
1983 event_poll_func (GPollFD *fds, guint nfds, gint timeout) 2672 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
1984 { 2673 {
1985 int got_events = 0; 2674 int got_events = 0;
1986 2675
1987 for (n = 0; n < nfds; ++n) 2676 for (n = 0; n < nfds; ++n)
1988 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events 2677 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
1989 2678
1990 if (timeout >= 0) 2679 if (timeout >= 0)
1991 // create/start timer 2680 // create/start timer
1992 2681
1993 // poll 2682 // poll
1994 ev_loop (EV_A_ 0); 2683 ev_loop (EV_A_ 0);
1995 2684
1996 // stop timer again 2685 // stop timer again
1997 if (timeout >= 0) 2686 if (timeout >= 0)
1998 ev_timer_stop (EV_A_ &to); 2687 ev_timer_stop (EV_A_ &to);
1999 2688
2000 // stop io watchers again - their callbacks should have set 2689 // stop io watchers again - their callbacks should have set
2001 for (n = 0; n < nfds; ++n) 2690 for (n = 0; n < nfds; ++n)
2002 ev_io_stop (EV_A_ iow [n]); 2691 ev_io_stop (EV_A_ iow [n]);
2003 2692
2004 return got_events; 2693 return got_events;
2005 } 2694 }
2006 2695
2007 2696
2008=head2 C<ev_embed> - when one backend isn't enough... 2697=head2 C<ev_embed> - when one backend isn't enough...
2009 2698
2010This is a rather advanced watcher type that lets you embed one event loop 2699This is a rather advanced watcher type that lets you embed one event loop
2016prioritise I/O. 2705prioritise I/O.
2017 2706
2018As an example for a bug workaround, the kqueue backend might only support 2707As an example for a bug workaround, the kqueue backend might only support
2019sockets on some platform, so it is unusable as generic backend, but you 2708sockets on some platform, so it is unusable as generic backend, but you
2020still want to make use of it because you have many sockets and it scales 2709still want to make use of it because you have many sockets and it scales
2021so nicely. In this case, you would create a kqueue-based loop and embed it 2710so nicely. In this case, you would create a kqueue-based loop and embed
2022into your default loop (which might use e.g. poll). Overall operation will 2711it into your default loop (which might use e.g. poll). Overall operation
2023be a bit slower because first libev has to poll and then call kevent, but 2712will be a bit slower because first libev has to call C<poll> and then
2024at least you can use both at what they are best. 2713C<kevent>, but at least you can use both mechanisms for what they are
2714best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2025 2715
2026As for prioritising I/O: rarely you have the case where some fds have 2716As for prioritising I/O: under rare circumstances you have the case where
2027to be watched and handled very quickly (with low latency), and even 2717some fds have to be watched and handled very quickly (with low latency),
2028priorities and idle watchers might have too much overhead. In this case 2718and even priorities and idle watchers might have too much overhead. In
2029you would put all the high priority stuff in one loop and all the rest in 2719this case you would put all the high priority stuff in one loop and all
2030a second one, and embed the second one in the first. 2720the rest in a second one, and embed the second one in the first.
2031 2721
2032As long as the watcher is active, the callback will be invoked every time 2722As long as the watcher is active, the callback will be invoked every
2033there might be events pending in the embedded loop. The callback must then 2723time there might be events pending in the embedded loop. The callback
2034call C<ev_embed_sweep (mainloop, watcher)> to make a single sweep and invoke 2724must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2035their callbacks (you could also start an idle watcher to give the embedded 2725sweep and invoke their callbacks (the callback doesn't need to invoke the
2036loop strictly lower priority for example). You can also set the callback 2726C<ev_embed_sweep> function directly, it could also start an idle watcher
2037to C<0>, in which case the embed watcher will automatically execute the 2727to give the embedded loop strictly lower priority for example).
2038embedded loop sweep.
2039 2728
2040As long as the watcher is started it will automatically handle events. The 2729You can also set the callback to C<0>, in which case the embed watcher
2041callback will be invoked whenever some events have been handled. You can 2730will automatically execute the embedded loop sweep whenever necessary.
2042set the callback to C<0> to avoid having to specify one if you are not
2043interested in that.
2044 2731
2045Also, there have not currently been made special provisions for forking: 2732Fork detection will be handled transparently while the C<ev_embed> watcher
2046when you fork, you not only have to call C<ev_loop_fork> on both loops, 2733is active, i.e., the embedded loop will automatically be forked when the
2047but you will also have to stop and restart any C<ev_embed> watchers 2734embedding loop forks. In other cases, the user is responsible for calling
2048yourself. 2735C<ev_loop_fork> on the embedded loop.
2049 2736
2050Unfortunately, not all backends are embeddable, only the ones returned by 2737Unfortunately, not all backends are embeddable: only the ones returned by
2051C<ev_embeddable_backends> are, which, unfortunately, does not include any 2738C<ev_embeddable_backends> are, which, unfortunately, does not include any
2052portable one. 2739portable one.
2053 2740
2054So when you want to use this feature you will always have to be prepared 2741So when you want to use this feature you will always have to be prepared
2055that you cannot get an embeddable loop. The recommended way to get around 2742that you cannot get an embeddable loop. The recommended way to get around
2056this is to have a separate variables for your embeddable loop, try to 2743this is to have a separate variables for your embeddable loop, try to
2057create it, and if that fails, use the normal loop for everything. 2744create it, and if that fails, use the normal loop for everything.
2058 2745
2746=head3 C<ev_embed> and fork
2747
2748While the C<ev_embed> watcher is running, forks in the embedding loop will
2749automatically be applied to the embedded loop as well, so no special
2750fork handling is required in that case. When the watcher is not running,
2751however, it is still the task of the libev user to call C<ev_loop_fork ()>
2752as applicable.
2753
2059=head3 Watcher-Specific Functions and Data Members 2754=head3 Watcher-Specific Functions and Data Members
2060 2755
2061=over 4 2756=over 4
2062 2757
2063=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop) 2758=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2066 2761
2067Configures the watcher to embed the given loop, which must be 2762Configures the watcher to embed the given loop, which must be
2068embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be 2763embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
2069invoked automatically, otherwise it is the responsibility of the callback 2764invoked automatically, otherwise it is the responsibility of the callback
2070to invoke it (it will continue to be called until the sweep has been done, 2765to invoke it (it will continue to be called until the sweep has been done,
2071if you do not want thta, you need to temporarily stop the embed watcher). 2766if you do not want that, you need to temporarily stop the embed watcher).
2072 2767
2073=item ev_embed_sweep (loop, ev_embed *) 2768=item ev_embed_sweep (loop, ev_embed *)
2074 2769
2075Make a single, non-blocking sweep over the embedded loop. This works 2770Make a single, non-blocking sweep over the embedded loop. This works
2076similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most 2771similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
2077apropriate way for embedded loops. 2772appropriate way for embedded loops.
2078 2773
2079=item struct ev_loop *other [read-only] 2774=item struct ev_loop *other [read-only]
2080 2775
2081The embedded event loop. 2776The embedded event loop.
2082 2777
2084 2779
2085=head3 Examples 2780=head3 Examples
2086 2781
2087Example: Try to get an embeddable event loop and embed it into the default 2782Example: Try to get an embeddable event loop and embed it into the default
2088event loop. If that is not possible, use the default loop. The default 2783event loop. If that is not possible, use the default loop. The default
2089loop is stored in C<loop_hi>, while the mebeddable loop is stored in 2784loop is stored in C<loop_hi>, while the embeddable loop is stored in
2090C<loop_lo> (which is C<loop_hi> in the acse no embeddable loop can be 2785C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
2091used). 2786used).
2092 2787
2093 struct ev_loop *loop_hi = ev_default_init (0); 2788 struct ev_loop *loop_hi = ev_default_init (0);
2094 struct ev_loop *loop_lo = 0; 2789 struct ev_loop *loop_lo = 0;
2095 struct ev_embed embed; 2790 ev_embed embed;
2096 2791
2097 // see if there is a chance of getting one that works 2792 // see if there is a chance of getting one that works
2098 // (remember that a flags value of 0 means autodetection) 2793 // (remember that a flags value of 0 means autodetection)
2099 loop_lo = ev_embeddable_backends () & ev_recommended_backends () 2794 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
2100 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ()) 2795 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
2101 : 0; 2796 : 0;
2102 2797
2103 // if we got one, then embed it, otherwise default to loop_hi 2798 // if we got one, then embed it, otherwise default to loop_hi
2104 if (loop_lo) 2799 if (loop_lo)
2105 { 2800 {
2106 ev_embed_init (&embed, 0, loop_lo); 2801 ev_embed_init (&embed, 0, loop_lo);
2107 ev_embed_start (loop_hi, &embed); 2802 ev_embed_start (loop_hi, &embed);
2108 } 2803 }
2109 else 2804 else
2110 loop_lo = loop_hi; 2805 loop_lo = loop_hi;
2111 2806
2112Example: Check if kqueue is available but not recommended and create 2807Example: Check if kqueue is available but not recommended and create
2113a kqueue backend for use with sockets (which usually work with any 2808a kqueue backend for use with sockets (which usually work with any
2114kqueue implementation). Store the kqueue/socket-only event loop in 2809kqueue implementation). Store the kqueue/socket-only event loop in
2115C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too). 2810C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
2116 2811
2117 struct ev_loop *loop = ev_default_init (0); 2812 struct ev_loop *loop = ev_default_init (0);
2118 struct ev_loop *loop_socket = 0; 2813 struct ev_loop *loop_socket = 0;
2119 struct ev_embed embed; 2814 ev_embed embed;
2120 2815
2121 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE) 2816 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2122 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE)) 2817 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2123 { 2818 {
2124 ev_embed_init (&embed, 0, loop_socket); 2819 ev_embed_init (&embed, 0, loop_socket);
2125 ev_embed_start (loop, &embed); 2820 ev_embed_start (loop, &embed);
2126 } 2821 }
2127 2822
2128 if (!loop_socket) 2823 if (!loop_socket)
2129 loop_socket = loop; 2824 loop_socket = loop;
2130 2825
2131 // now use loop_socket for all sockets, and loop for everything else 2826 // now use loop_socket for all sockets, and loop for everything else
2132 2827
2133 2828
2134=head2 C<ev_fork> - the audacity to resume the event loop after a fork 2829=head2 C<ev_fork> - the audacity to resume the event loop after a fork
2135 2830
2136Fork watchers are called when a C<fork ()> was detected (usually because 2831Fork watchers are called when a C<fork ()> was detected (usually because
2139event loop blocks next and before C<ev_check> watchers are being called, 2834event loop blocks next and before C<ev_check> watchers are being called,
2140and only in the child after the fork. If whoever good citizen calling 2835and only in the child after the fork. If whoever good citizen calling
2141C<ev_default_fork> cheats and calls it in the wrong process, the fork 2836C<ev_default_fork> cheats and calls it in the wrong process, the fork
2142handlers will be invoked, too, of course. 2837handlers will be invoked, too, of course.
2143 2838
2839=head3 The special problem of life after fork - how is it possible?
2840
2841Most uses of C<fork()> consist of forking, then some simple calls to ste
2842up/change the process environment, followed by a call to C<exec()>. This
2843sequence should be handled by libev without any problems.
2844
2845This changes when the application actually wants to do event handling
2846in the child, or both parent in child, in effect "continuing" after the
2847fork.
2848
2849The default mode of operation (for libev, with application help to detect
2850forks) is to duplicate all the state in the child, as would be expected
2851when I<either> the parent I<or> the child process continues.
2852
2853When both processes want to continue using libev, then this is usually the
2854wrong result. In that case, usually one process (typically the parent) is
2855supposed to continue with all watchers in place as before, while the other
2856process typically wants to start fresh, i.e. without any active watchers.
2857
2858The cleanest and most efficient way to achieve that with libev is to
2859simply create a new event loop, which of course will be "empty", and
2860use that for new watchers. This has the advantage of not touching more
2861memory than necessary, and thus avoiding the copy-on-write, and the
2862disadvantage of having to use multiple event loops (which do not support
2863signal watchers).
2864
2865When this is not possible, or you want to use the default loop for
2866other reasons, then in the process that wants to start "fresh", call
2867C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
2868the default loop will "orphan" (not stop) all registered watchers, so you
2869have to be careful not to execute code that modifies those watchers. Note
2870also that in that case, you have to re-register any signal watchers.
2871
2144=head3 Watcher-Specific Functions and Data Members 2872=head3 Watcher-Specific Functions and Data Members
2145 2873
2146=over 4 2874=over 4
2147 2875
2148=item ev_fork_init (ev_signal *, callback) 2876=item ev_fork_init (ev_signal *, callback)
2180is that the author does not know of a simple (or any) algorithm for a 2908is that the author does not know of a simple (or any) algorithm for a
2181multiple-writer-single-reader queue that works in all cases and doesn't 2909multiple-writer-single-reader queue that works in all cases and doesn't
2182need elaborate support such as pthreads. 2910need elaborate support such as pthreads.
2183 2911
2184That means that if you want to queue data, you have to provide your own 2912That means that if you want to queue data, you have to provide your own
2185queue. But at least I can tell you would implement locking around your 2913queue. But at least I can tell you how to implement locking around your
2186queue: 2914queue:
2187 2915
2188=over 4 2916=over 4
2189 2917
2190=item queueing from a signal handler context 2918=item queueing from a signal handler context
2191 2919
2192To implement race-free queueing, you simply add to the queue in the signal 2920To implement race-free queueing, you simply add to the queue in the signal
2193handler but you block the signal handler in the watcher callback. Here is an example that does that for 2921handler but you block the signal handler in the watcher callback. Here is
2194some fictitiuous SIGUSR1 handler: 2922an example that does that for some fictitious SIGUSR1 handler:
2195 2923
2196 static ev_async mysig; 2924 static ev_async mysig;
2197 2925
2198 static void 2926 static void
2199 sigusr1_handler (void) 2927 sigusr1_handler (void)
2265=over 4 2993=over 4
2266 2994
2267=item ev_async_init (ev_async *, callback) 2995=item ev_async_init (ev_async *, callback)
2268 2996
2269Initialises and configures the async watcher - it has no parameters of any 2997Initialises and configures the async watcher - it has no parameters of any
2270kind. There is a C<ev_asynd_set> macro, but using it is utterly pointless, 2998kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
2271believe me. 2999trust me.
2272 3000
2273=item ev_async_send (loop, ev_async *) 3001=item ev_async_send (loop, ev_async *)
2274 3002
2275Sends/signals/activates the given C<ev_async> watcher, that is, feeds 3003Sends/signals/activates the given C<ev_async> watcher, that is, feeds
2276an C<EV_ASYNC> event on the watcher into the event loop. Unlike 3004an C<EV_ASYNC> event on the watcher into the event loop. Unlike
2277C<ev_feed_event>, this call is safe to do in other threads, signal or 3005C<ev_feed_event>, this call is safe to do from other threads, signal or
2278similar contexts (see the dicusssion of C<EV_ATOMIC_T> in the embedding 3006similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
2279section below on what exactly this means). 3007section below on what exactly this means).
2280 3008
3009Note that, as with other watchers in libev, multiple events might get
3010compressed into a single callback invocation (another way to look at this
3011is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
3012reset when the event loop detects that).
3013
2281This call incurs the overhead of a syscall only once per loop iteration, 3014This call incurs the overhead of a system call only once per event loop
2282so while the overhead might be noticable, it doesn't apply to repeated 3015iteration, so while the overhead might be noticeable, it doesn't apply to
2283calls to C<ev_async_send>. 3016repeated calls to C<ev_async_send> for the same event loop.
3017
3018=item bool = ev_async_pending (ev_async *)
3019
3020Returns a non-zero value when C<ev_async_send> has been called on the
3021watcher but the event has not yet been processed (or even noted) by the
3022event loop.
3023
3024C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3025the loop iterates next and checks for the watcher to have become active,
3026it will reset the flag again. C<ev_async_pending> can be used to very
3027quickly check whether invoking the loop might be a good idea.
3028
3029Not that this does I<not> check whether the watcher itself is pending,
3030only whether it has been requested to make this watcher pending: there
3031is a time window between the event loop checking and resetting the async
3032notification, and the callback being invoked.
2284 3033
2285=back 3034=back
2286 3035
2287 3036
2288=head1 OTHER FUNCTIONS 3037=head1 OTHER FUNCTIONS
2292=over 4 3041=over 4
2293 3042
2294=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback) 3043=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
2295 3044
2296This function combines a simple timer and an I/O watcher, calls your 3045This function combines a simple timer and an I/O watcher, calls your
2297callback on whichever event happens first and automatically stop both 3046callback on whichever event happens first and automatically stops both
2298watchers. This is useful if you want to wait for a single event on an fd 3047watchers. This is useful if you want to wait for a single event on an fd
2299or timeout without having to allocate/configure/start/stop/free one or 3048or timeout without having to allocate/configure/start/stop/free one or
2300more watchers yourself. 3049more watchers yourself.
2301 3050
2302If C<fd> is less than 0, then no I/O watcher will be started and events 3051If C<fd> is less than 0, then no I/O watcher will be started and the
2303is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and 3052C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
2304C<events> set will be craeted and started. 3053the given C<fd> and C<events> set will be created and started.
2305 3054
2306If C<timeout> is less than 0, then no timeout watcher will be 3055If C<timeout> is less than 0, then no timeout watcher will be
2307started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and 3056started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
2308repeat = 0) will be started. While C<0> is a valid timeout, it is of 3057repeat = 0) will be started. C<0> is a valid timeout.
2309dubious value.
2310 3058
2311The callback has the type C<void (*cb)(int revents, void *arg)> and gets 3059The callback has the type C<void (*cb)(int revents, void *arg)> and gets
2312passed an C<revents> set like normal event callbacks (a combination of 3060passed an C<revents> set like normal event callbacks (a combination of
2313C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg> 3061C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
2314value passed to C<ev_once>: 3062value passed to C<ev_once>. Note that it is possible to receive I<both>
3063a timeout and an io event at the same time - you probably should give io
3064events precedence.
2315 3065
3066Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3067
2316 static void stdin_ready (int revents, void *arg) 3068 static void stdin_ready (int revents, void *arg)
2317 { 3069 {
2318 if (revents & EV_TIMEOUT)
2319 /* doh, nothing entered */;
2320 else if (revents & EV_READ) 3070 if (revents & EV_READ)
2321 /* stdin might have data for us, joy! */; 3071 /* stdin might have data for us, joy! */;
3072 else if (revents & EV_TIMEOUT)
3073 /* doh, nothing entered */;
2322 } 3074 }
2323 3075
2324 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0); 3076 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
2325 3077
2326=item ev_feed_event (ev_loop *, watcher *, int revents) 3078=item ev_feed_event (struct ev_loop *, watcher *, int revents)
2327 3079
2328Feeds the given event set into the event loop, as if the specified event 3080Feeds the given event set into the event loop, as if the specified event
2329had happened for the specified watcher (which must be a pointer to an 3081had happened for the specified watcher (which must be a pointer to an
2330initialised but not necessarily started event watcher). 3082initialised but not necessarily started event watcher).
2331 3083
2332=item ev_feed_fd_event (ev_loop *, int fd, int revents) 3084=item ev_feed_fd_event (struct ev_loop *, int fd, int revents)
2333 3085
2334Feed an event on the given fd, as if a file descriptor backend detected 3086Feed an event on the given fd, as if a file descriptor backend detected
2335the given events it. 3087the given events it.
2336 3088
2337=item ev_feed_signal_event (ev_loop *loop, int signum) 3089=item ev_feed_signal_event (struct ev_loop *loop, int signum)
2338 3090
2339Feed an event as if the given signal occured (C<loop> must be the default 3091Feed an event as if the given signal occurred (C<loop> must be the default
2340loop!). 3092loop!).
2341 3093
2342=back 3094=back
2343 3095
2344 3096
2360 3112
2361=item * Priorities are not currently supported. Initialising priorities 3113=item * Priorities are not currently supported. Initialising priorities
2362will fail and all watchers will have the same priority, even though there 3114will fail and all watchers will have the same priority, even though there
2363is an ev_pri field. 3115is an ev_pri field.
2364 3116
3117=item * In libevent, the last base created gets the signals, in libev, the
3118first base created (== the default loop) gets the signals.
3119
2365=item * Other members are not supported. 3120=item * Other members are not supported.
2366 3121
2367=item * The libev emulation is I<not> ABI compatible to libevent, you need 3122=item * The libev emulation is I<not> ABI compatible to libevent, you need
2368to use the libev header file and library. 3123to use the libev header file and library.
2369 3124
2370=back 3125=back
2371 3126
2372=head1 C++ SUPPORT 3127=head1 C++ SUPPORT
2373 3128
2374Libev comes with some simplistic wrapper classes for C++ that mainly allow 3129Libev comes with some simplistic wrapper classes for C++ that mainly allow
2375you to use some convinience methods to start/stop watchers and also change 3130you to use some convenience methods to start/stop watchers and also change
2376the callback model to a model using method callbacks on objects. 3131the callback model to a model using method callbacks on objects.
2377 3132
2378To use it, 3133To use it,
2379 3134
2380 #include <ev++.h> 3135 #include <ev++.h>
2381 3136
2382This automatically includes F<ev.h> and puts all of its definitions (many 3137This automatically includes F<ev.h> and puts all of its definitions (many
2383of them macros) into the global namespace. All C++ specific things are 3138of them macros) into the global namespace. All C++ specific things are
2384put into the C<ev> namespace. It should support all the same embedding 3139put into the C<ev> namespace. It should support all the same embedding
2385options as F<ev.h>, most notably C<EV_MULTIPLICITY>. 3140options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
2452your compiler is good :), then the method will be fully inlined into the 3207your compiler is good :), then the method will be fully inlined into the
2453thunking function, making it as fast as a direct C callback. 3208thunking function, making it as fast as a direct C callback.
2454 3209
2455Example: simple class declaration and watcher initialisation 3210Example: simple class declaration and watcher initialisation
2456 3211
2457 struct myclass 3212 struct myclass
2458 { 3213 {
2459 void io_cb (ev::io &w, int revents) { } 3214 void io_cb (ev::io &w, int revents) { }
2460 } 3215 }
2461 3216
2462 myclass obj; 3217 myclass obj;
2463 ev::io iow; 3218 ev::io iow;
2464 iow.set <myclass, &myclass::io_cb> (&obj); 3219 iow.set <myclass, &myclass::io_cb> (&obj);
3220
3221=item w->set (object *)
3222
3223This is an B<experimental> feature that might go away in a future version.
3224
3225This is a variation of a method callback - leaving out the method to call
3226will default the method to C<operator ()>, which makes it possible to use
3227functor objects without having to manually specify the C<operator ()> all
3228the time. Incidentally, you can then also leave out the template argument
3229list.
3230
3231The C<operator ()> method prototype must be C<void operator ()(watcher &w,
3232int revents)>.
3233
3234See the method-C<set> above for more details.
3235
3236Example: use a functor object as callback.
3237
3238 struct myfunctor
3239 {
3240 void operator() (ev::io &w, int revents)
3241 {
3242 ...
3243 }
3244 }
3245
3246 myfunctor f;
3247
3248 ev::io w;
3249 w.set (&f);
2465 3250
2466=item w->set<function> (void *data = 0) 3251=item w->set<function> (void *data = 0)
2467 3252
2468Also sets a callback, but uses a static method or plain function as 3253Also sets a callback, but uses a static method or plain function as
2469callback. The optional C<data> argument will be stored in the watcher's 3254callback. The optional C<data> argument will be stored in the watcher's
2471 3256
2472The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>. 3257The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
2473 3258
2474See the method-C<set> above for more details. 3259See the method-C<set> above for more details.
2475 3260
2476Example: 3261Example: Use a plain function as callback.
2477 3262
2478 static void io_cb (ev::io &w, int revents) { } 3263 static void io_cb (ev::io &w, int revents) { }
2479 iow.set <io_cb> (); 3264 iow.set <io_cb> ();
2480 3265
2481=item w->set (struct ev_loop *) 3266=item w->set (struct ev_loop *)
2482 3267
2483Associates a different C<struct ev_loop> with this watcher. You can only 3268Associates a different C<struct ev_loop> with this watcher. You can only
2484do this when the watcher is inactive (and not pending either). 3269do this when the watcher is inactive (and not pending either).
2485 3270
2486=item w->set ([args]) 3271=item w->set ([arguments])
2487 3272
2488Basically the same as C<ev_TYPE_set>, with the same args. Must be 3273Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
2489called at least once. Unlike the C counterpart, an active watcher gets 3274called at least once. Unlike the C counterpart, an active watcher gets
2490automatically stopped and restarted when reconfiguring it with this 3275automatically stopped and restarted when reconfiguring it with this
2491method. 3276method.
2492 3277
2493=item w->start () 3278=item w->start ()
2517=back 3302=back
2518 3303
2519Example: Define a class with an IO and idle watcher, start one of them in 3304Example: Define a class with an IO and idle watcher, start one of them in
2520the constructor. 3305the constructor.
2521 3306
2522 class myclass 3307 class myclass
2523 { 3308 {
2524 ev::io io; void io_cb (ev::io &w, int revents); 3309 ev::io io ; void io_cb (ev::io &w, int revents);
2525 ev:idle idle void idle_cb (ev::idle &w, int revents); 3310 ev::idle idle; void idle_cb (ev::idle &w, int revents);
2526 3311
2527 myclass (int fd) 3312 myclass (int fd)
2528 { 3313 {
2529 io .set <myclass, &myclass::io_cb > (this); 3314 io .set <myclass, &myclass::io_cb > (this);
2530 idle.set <myclass, &myclass::idle_cb> (this); 3315 idle.set <myclass, &myclass::idle_cb> (this);
2531 3316
2532 io.start (fd, ev::READ); 3317 io.start (fd, ev::READ);
2533 } 3318 }
2534 }; 3319 };
2535 3320
2536 3321
2537=head1 OTHER LANGUAGE BINDINGS 3322=head1 OTHER LANGUAGE BINDINGS
2538 3323
2539Libev does not offer other language bindings itself, but bindings for a 3324Libev does not offer other language bindings itself, but bindings for a
2540numbe rof languages exist in the form of third-party packages. If you know 3325number of languages exist in the form of third-party packages. If you know
2541any interesting language binding in addition to the ones listed here, drop 3326any interesting language binding in addition to the ones listed here, drop
2542me a note. 3327me a note.
2543 3328
2544=over 4 3329=over 4
2545 3330
2546=item Perl 3331=item Perl
2547 3332
2548The EV module implements the full libev API and is actually used to test 3333The EV module implements the full libev API and is actually used to test
2549libev. EV is developed together with libev. Apart from the EV core module, 3334libev. EV is developed together with libev. Apart from the EV core module,
2550there are additional modules that implement libev-compatible interfaces 3335there are additional modules that implement libev-compatible interfaces
2551to C<libadns> (C<EV::ADNS>), C<Net::SNMP> (C<Net::SNMP::EV>) and the 3336to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
2552C<libglib> event core (C<Glib::EV> and C<EV::Glib>). 3337C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3338and C<EV::Glib>).
2553 3339
2554It can be found and installed via CPAN, its homepage is found at 3340It can be found and installed via CPAN, its homepage is at
2555L<http://software.schmorp.de/pkg/EV>. 3341L<http://software.schmorp.de/pkg/EV>.
2556 3342
3343=item Python
3344
3345Python bindings can be found at L<http://code.google.com/p/pyev/>. It
3346seems to be quite complete and well-documented.
3347
2557=item Ruby 3348=item Ruby
2558 3349
2559Tony Arcieri has written a ruby extension that offers access to a subset 3350Tony Arcieri has written a ruby extension that offers access to a subset
2560of the libev API and adds filehandle abstractions, asynchronous DNS and 3351of the libev API and adds file handle abstractions, asynchronous DNS and
2561more on top of it. It can be found via gem servers. Its homepage is at 3352more on top of it. It can be found via gem servers. Its homepage is at
2562L<http://rev.rubyforge.org/>. 3353L<http://rev.rubyforge.org/>.
2563 3354
3355Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3356makes rev work even on mingw.
3357
3358=item Haskell
3359
3360A haskell binding to libev is available at
3361L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
3362
2564=item D 3363=item D
2565 3364
2566Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to 3365Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
2567be found at L<http://git.llucax.com.ar/?p=software/ev.d.git;a=summary>. 3366be found at L<http://proj.llucax.com.ar/wiki/evd>.
3367
3368=item Ocaml
3369
3370Erkki Seppala has written Ocaml bindings for libev, to be found at
3371L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
2568 3372
2569=back 3373=back
2570 3374
2571 3375
2572=head1 MACRO MAGIC 3376=head1 MACRO MAGIC
2573 3377
2574Libev can be compiled with a variety of options, the most fundamantal 3378Libev can be compiled with a variety of options, the most fundamental
2575of which is C<EV_MULTIPLICITY>. This option determines whether (most) 3379of which is C<EV_MULTIPLICITY>. This option determines whether (most)
2576functions and callbacks have an initial C<struct ev_loop *> argument. 3380functions and callbacks have an initial C<struct ev_loop *> argument.
2577 3381
2578To make it easier to write programs that cope with either variant, the 3382To make it easier to write programs that cope with either variant, the
2579following macros are defined: 3383following macros are defined:
2584 3388
2585This provides the loop I<argument> for functions, if one is required ("ev 3389This provides the loop I<argument> for functions, if one is required ("ev
2586loop argument"). The C<EV_A> form is used when this is the sole argument, 3390loop argument"). The C<EV_A> form is used when this is the sole argument,
2587C<EV_A_> is used when other arguments are following. Example: 3391C<EV_A_> is used when other arguments are following. Example:
2588 3392
2589 ev_unref (EV_A); 3393 ev_unref (EV_A);
2590 ev_timer_add (EV_A_ watcher); 3394 ev_timer_add (EV_A_ watcher);
2591 ev_loop (EV_A_ 0); 3395 ev_loop (EV_A_ 0);
2592 3396
2593It assumes the variable C<loop> of type C<struct ev_loop *> is in scope, 3397It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
2594which is often provided by the following macro. 3398which is often provided by the following macro.
2595 3399
2596=item C<EV_P>, C<EV_P_> 3400=item C<EV_P>, C<EV_P_>
2597 3401
2598This provides the loop I<parameter> for functions, if one is required ("ev 3402This provides the loop I<parameter> for functions, if one is required ("ev
2599loop parameter"). The C<EV_P> form is used when this is the sole parameter, 3403loop parameter"). The C<EV_P> form is used when this is the sole parameter,
2600C<EV_P_> is used when other parameters are following. Example: 3404C<EV_P_> is used when other parameters are following. Example:
2601 3405
2602 // this is how ev_unref is being declared 3406 // this is how ev_unref is being declared
2603 static void ev_unref (EV_P); 3407 static void ev_unref (EV_P);
2604 3408
2605 // this is how you can declare your typical callback 3409 // this is how you can declare your typical callback
2606 static void cb (EV_P_ ev_timer *w, int revents) 3410 static void cb (EV_P_ ev_timer *w, int revents)
2607 3411
2608It declares a parameter C<loop> of type C<struct ev_loop *>, quite 3412It declares a parameter C<loop> of type C<struct ev_loop *>, quite
2609suitable for use with C<EV_A>. 3413suitable for use with C<EV_A>.
2610 3414
2611=item C<EV_DEFAULT>, C<EV_DEFAULT_> 3415=item C<EV_DEFAULT>, C<EV_DEFAULT_>
2612 3416
2613Similar to the other two macros, this gives you the value of the default 3417Similar to the other two macros, this gives you the value of the default
2614loop, if multiple loops are supported ("ev loop default"). 3418loop, if multiple loops are supported ("ev loop default").
3419
3420=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
3421
3422Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
3423default loop has been initialised (C<UC> == unchecked). Their behaviour
3424is undefined when the default loop has not been initialised by a previous
3425execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
3426
3427It is often prudent to use C<EV_DEFAULT> when initialising the first
3428watcher in a function but use C<EV_DEFAULT_UC> afterwards.
2615 3429
2616=back 3430=back
2617 3431
2618Example: Declare and initialise a check watcher, utilising the above 3432Example: Declare and initialise a check watcher, utilising the above
2619macros so it will work regardless of whether multiple loops are supported 3433macros so it will work regardless of whether multiple loops are supported
2620or not. 3434or not.
2621 3435
2622 static void 3436 static void
2623 check_cb (EV_P_ ev_timer *w, int revents) 3437 check_cb (EV_P_ ev_timer *w, int revents)
2624 { 3438 {
2625 ev_check_stop (EV_A_ w); 3439 ev_check_stop (EV_A_ w);
2626 } 3440 }
2627 3441
2628 ev_check check; 3442 ev_check check;
2629 ev_check_init (&check, check_cb); 3443 ev_check_init (&check, check_cb);
2630 ev_check_start (EV_DEFAULT_ &check); 3444 ev_check_start (EV_DEFAULT_ &check);
2631 ev_loop (EV_DEFAULT_ 0); 3445 ev_loop (EV_DEFAULT_ 0);
2632 3446
2633=head1 EMBEDDING 3447=head1 EMBEDDING
2634 3448
2635Libev can (and often is) directly embedded into host 3449Libev can (and often is) directly embedded into host
2636applications. Examples of applications that embed it include the Deliantra 3450applications. Examples of applications that embed it include the Deliantra
2643libev somewhere in your source tree). 3457libev somewhere in your source tree).
2644 3458
2645=head2 FILESETS 3459=head2 FILESETS
2646 3460
2647Depending on what features you need you need to include one or more sets of files 3461Depending on what features you need you need to include one or more sets of files
2648in your app. 3462in your application.
2649 3463
2650=head3 CORE EVENT LOOP 3464=head3 CORE EVENT LOOP
2651 3465
2652To include only the libev core (all the C<ev_*> functions), with manual 3466To include only the libev core (all the C<ev_*> functions), with manual
2653configuration (no autoconf): 3467configuration (no autoconf):
2654 3468
2655 #define EV_STANDALONE 1 3469 #define EV_STANDALONE 1
2656 #include "ev.c" 3470 #include "ev.c"
2657 3471
2658This will automatically include F<ev.h>, too, and should be done in a 3472This will automatically include F<ev.h>, too, and should be done in a
2659single C source file only to provide the function implementations. To use 3473single C source file only to provide the function implementations. To use
2660it, do the same for F<ev.h> in all files wishing to use this API (best 3474it, do the same for F<ev.h> in all files wishing to use this API (best
2661done by writing a wrapper around F<ev.h> that you can include instead and 3475done by writing a wrapper around F<ev.h> that you can include instead and
2662where you can put other configuration options): 3476where you can put other configuration options):
2663 3477
2664 #define EV_STANDALONE 1 3478 #define EV_STANDALONE 1
2665 #include "ev.h" 3479 #include "ev.h"
2666 3480
2667Both header files and implementation files can be compiled with a C++ 3481Both header files and implementation files can be compiled with a C++
2668compiler (at least, thats a stated goal, and breakage will be treated 3482compiler (at least, that's a stated goal, and breakage will be treated
2669as a bug). 3483as a bug).
2670 3484
2671You need the following files in your source tree, or in a directory 3485You need the following files in your source tree, or in a directory
2672in your include path (e.g. in libev/ when using -Ilibev): 3486in your include path (e.g. in libev/ when using -Ilibev):
2673 3487
2674 ev.h 3488 ev.h
2675 ev.c 3489 ev.c
2676 ev_vars.h 3490 ev_vars.h
2677 ev_wrap.h 3491 ev_wrap.h
2678 3492
2679 ev_win32.c required on win32 platforms only 3493 ev_win32.c required on win32 platforms only
2680 3494
2681 ev_select.c only when select backend is enabled (which is enabled by default) 3495 ev_select.c only when select backend is enabled (which is enabled by default)
2682 ev_poll.c only when poll backend is enabled (disabled by default) 3496 ev_poll.c only when poll backend is enabled (disabled by default)
2683 ev_epoll.c only when the epoll backend is enabled (disabled by default) 3497 ev_epoll.c only when the epoll backend is enabled (disabled by default)
2684 ev_kqueue.c only when the kqueue backend is enabled (disabled by default) 3498 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
2685 ev_port.c only when the solaris port backend is enabled (disabled by default) 3499 ev_port.c only when the solaris port backend is enabled (disabled by default)
2686 3500
2687F<ev.c> includes the backend files directly when enabled, so you only need 3501F<ev.c> includes the backend files directly when enabled, so you only need
2688to compile this single file. 3502to compile this single file.
2689 3503
2690=head3 LIBEVENT COMPATIBILITY API 3504=head3 LIBEVENT COMPATIBILITY API
2691 3505
2692To include the libevent compatibility API, also include: 3506To include the libevent compatibility API, also include:
2693 3507
2694 #include "event.c" 3508 #include "event.c"
2695 3509
2696in the file including F<ev.c>, and: 3510in the file including F<ev.c>, and:
2697 3511
2698 #include "event.h" 3512 #include "event.h"
2699 3513
2700in the files that want to use the libevent API. This also includes F<ev.h>. 3514in the files that want to use the libevent API. This also includes F<ev.h>.
2701 3515
2702You need the following additional files for this: 3516You need the following additional files for this:
2703 3517
2704 event.h 3518 event.h
2705 event.c 3519 event.c
2706 3520
2707=head3 AUTOCONF SUPPORT 3521=head3 AUTOCONF SUPPORT
2708 3522
2709Instead of using C<EV_STANDALONE=1> and providing your config in 3523Instead of using C<EV_STANDALONE=1> and providing your configuration in
2710whatever way you want, you can also C<m4_include([libev.m4])> in your 3524whatever way you want, you can also C<m4_include([libev.m4])> in your
2711F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then 3525F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
2712include F<config.h> and configure itself accordingly. 3526include F<config.h> and configure itself accordingly.
2713 3527
2714For this of course you need the m4 file: 3528For this of course you need the m4 file:
2715 3529
2716 libev.m4 3530 libev.m4
2717 3531
2718=head2 PREPROCESSOR SYMBOLS/MACROS 3532=head2 PREPROCESSOR SYMBOLS/MACROS
2719 3533
2720Libev can be configured via a variety of preprocessor symbols you have to define 3534Libev can be configured via a variety of preprocessor symbols you have to
2721before including any of its files. The default is not to build for multiplicity 3535define before including any of its files. The default in the absence of
2722and only include the select backend. 3536autoconf is documented for every option.
2723 3537
2724=over 4 3538=over 4
2725 3539
2726=item EV_STANDALONE 3540=item EV_STANDALONE
2727 3541
2729keeps libev from including F<config.h>, and it also defines dummy 3543keeps libev from including F<config.h>, and it also defines dummy
2730implementations for some libevent functions (such as logging, which is not 3544implementations for some libevent functions (such as logging, which is not
2731supported). It will also not define any of the structs usually found in 3545supported). It will also not define any of the structs usually found in
2732F<event.h> that are not directly supported by the libev core alone. 3546F<event.h> that are not directly supported by the libev core alone.
2733 3547
3548In stanbdalone mode, libev will still try to automatically deduce the
3549configuration, but has to be more conservative.
3550
2734=item EV_USE_MONOTONIC 3551=item EV_USE_MONOTONIC
2735 3552
2736If defined to be C<1>, libev will try to detect the availability of the 3553If defined to be C<1>, libev will try to detect the availability of the
2737monotonic clock option at both compiletime and runtime. Otherwise no use 3554monotonic clock option at both compile time and runtime. Otherwise no
2738of the monotonic clock option will be attempted. If you enable this, you 3555use of the monotonic clock option will be attempted. If you enable this,
2739usually have to link against librt or something similar. Enabling it when 3556you usually have to link against librt or something similar. Enabling it
2740the functionality isn't available is safe, though, although you have 3557when the functionality isn't available is safe, though, although you have
2741to make sure you link against any libraries where the C<clock_gettime> 3558to make sure you link against any libraries where the C<clock_gettime>
2742function is hiding in (often F<-lrt>). 3559function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
2743 3560
2744=item EV_USE_REALTIME 3561=item EV_USE_REALTIME
2745 3562
2746If defined to be C<1>, libev will try to detect the availability of the 3563If defined to be C<1>, libev will try to detect the availability of the
2747realtime clock option at compiletime (and assume its availability at 3564real-time clock option at compile time (and assume its availability
2748runtime if successful). Otherwise no use of the realtime clock option will 3565at runtime if successful). Otherwise no use of the real-time clock
2749be attempted. This effectively replaces C<gettimeofday> by C<clock_get 3566option will be attempted. This effectively replaces C<gettimeofday>
2750(CLOCK_REALTIME, ...)> and will not normally affect correctness. See the 3567by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
2751note about libraries in the description of C<EV_USE_MONOTONIC>, though. 3568correctness. See the note about libraries in the description of
3569C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
3570C<EV_USE_CLOCK_SYSCALL>.
3571
3572=item EV_USE_CLOCK_SYSCALL
3573
3574If defined to be C<1>, libev will try to use a direct syscall instead
3575of calling the system-provided C<clock_gettime> function. This option
3576exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
3577unconditionally pulls in C<libpthread>, slowing down single-threaded
3578programs needlessly. Using a direct syscall is slightly slower (in
3579theory), because no optimised vdso implementation can be used, but avoids
3580the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
3581higher, as it simplifies linking (no need for C<-lrt>).
2752 3582
2753=item EV_USE_NANOSLEEP 3583=item EV_USE_NANOSLEEP
2754 3584
2755If defined to be C<1>, libev will assume that C<nanosleep ()> is available 3585If defined to be C<1>, libev will assume that C<nanosleep ()> is available
2756and will use it for delays. Otherwise it will use C<select ()>. 3586and will use it for delays. Otherwise it will use C<select ()>.
2757 3587
3588=item EV_USE_EVENTFD
3589
3590If defined to be C<1>, then libev will assume that C<eventfd ()> is
3591available and will probe for kernel support at runtime. This will improve
3592C<ev_signal> and C<ev_async> performance and reduce resource consumption.
3593If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
35942.7 or newer, otherwise disabled.
3595
2758=item EV_USE_SELECT 3596=item EV_USE_SELECT
2759 3597
2760If undefined or defined to be C<1>, libev will compile in support for the 3598If undefined or defined to be C<1>, libev will compile in support for the
2761C<select>(2) backend. No attempt at autodetection will be done: if no 3599C<select>(2) backend. No attempt at auto-detection will be done: if no
2762other method takes over, select will be it. Otherwise the select backend 3600other method takes over, select will be it. Otherwise the select backend
2763will not be compiled in. 3601will not be compiled in.
2764 3602
2765=item EV_SELECT_USE_FD_SET 3603=item EV_SELECT_USE_FD_SET
2766 3604
2767If defined to C<1>, then the select backend will use the system C<fd_set> 3605If defined to C<1>, then the select backend will use the system C<fd_set>
2768structure. This is useful if libev doesn't compile due to a missing 3606structure. This is useful if libev doesn't compile due to a missing
2769C<NFDBITS> or C<fd_mask> definition or it misguesses the bitset layout on 3607C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
2770exotic systems. This usually limits the range of file descriptors to some 3608on exotic systems. This usually limits the range of file descriptors to
2771low limit such as 1024 or might have other limitations (winsocket only 3609some low limit such as 1024 or might have other limitations (winsocket
2772allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation, might 3610only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
2773influence the size of the C<fd_set> used. 3611configures the maximum size of the C<fd_set>.
2774 3612
2775=item EV_SELECT_IS_WINSOCKET 3613=item EV_SELECT_IS_WINSOCKET
2776 3614
2777When defined to C<1>, the select backend will assume that 3615When defined to C<1>, the select backend will assume that
2778select/socket/connect etc. don't understand file descriptors but 3616select/socket/connect etc. don't understand file descriptors but
2798 3636
2799=item EV_USE_EPOLL 3637=item EV_USE_EPOLL
2800 3638
2801If defined to be C<1>, libev will compile in support for the Linux 3639If defined to be C<1>, libev will compile in support for the Linux
2802C<epoll>(7) backend. Its availability will be detected at runtime, 3640C<epoll>(7) backend. Its availability will be detected at runtime,
2803otherwise another method will be used as fallback. This is the 3641otherwise another method will be used as fallback. This is the preferred
2804preferred backend for GNU/Linux systems. 3642backend for GNU/Linux systems. If undefined, it will be enabled if the
3643headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
2805 3644
2806=item EV_USE_KQUEUE 3645=item EV_USE_KQUEUE
2807 3646
2808If defined to be C<1>, libev will compile in support for the BSD style 3647If defined to be C<1>, libev will compile in support for the BSD style
2809C<kqueue>(2) backend. Its actual availability will be detected at runtime, 3648C<kqueue>(2) backend. Its actual availability will be detected at runtime,
2822otherwise another method will be used as fallback. This is the preferred 3661otherwise another method will be used as fallback. This is the preferred
2823backend for Solaris 10 systems. 3662backend for Solaris 10 systems.
2824 3663
2825=item EV_USE_DEVPOLL 3664=item EV_USE_DEVPOLL
2826 3665
2827reserved for future expansion, works like the USE symbols above. 3666Reserved for future expansion, works like the USE symbols above.
2828 3667
2829=item EV_USE_INOTIFY 3668=item EV_USE_INOTIFY
2830 3669
2831If defined to be C<1>, libev will compile in support for the Linux inotify 3670If defined to be C<1>, libev will compile in support for the Linux inotify
2832interface to speed up C<ev_stat> watchers. Its actual availability will 3671interface to speed up C<ev_stat> watchers. Its actual availability will
2833be detected at runtime. 3672be detected at runtime. If undefined, it will be enabled if the headers
3673indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
2834 3674
2835=item EV_ATOMIC_T 3675=item EV_ATOMIC_T
2836 3676
2837Libev requires an integer type (suitable for storing C<0> or C<1>) whose 3677Libev requires an integer type (suitable for storing C<0> or C<1>) whose
2838access is atomic with respect to other threads or signal contexts. No such 3678access is atomic with respect to other threads or signal contexts. No such
2839type is easily found in the C language, so you can provide your own type 3679type is easily found in the C language, so you can provide your own type
2840that you know is safe for your purposes. It is used both for signal handler "locking" 3680that you know is safe for your purposes. It is used both for signal handler "locking"
2841as well as for signal and thread safety in C<ev_async> watchers. 3681as well as for signal and thread safety in C<ev_async> watchers.
2842 3682
2843In the absense of this define, libev will use C<sig_atomic_t volatile> 3683In the absence of this define, libev will use C<sig_atomic_t volatile>
2844(from F<signal.h>), which is usually good enough on most platforms. 3684(from F<signal.h>), which is usually good enough on most platforms.
2845 3685
2846=item EV_H 3686=item EV_H
2847 3687
2848The name of the F<ev.h> header file used to include it. The default if 3688The name of the F<ev.h> header file used to include it. The default if
2887When doing priority-based operations, libev usually has to linearly search 3727When doing priority-based operations, libev usually has to linearly search
2888all the priorities, so having many of them (hundreds) uses a lot of space 3728all the priorities, so having many of them (hundreds) uses a lot of space
2889and time, so using the defaults of five priorities (-2 .. +2) is usually 3729and time, so using the defaults of five priorities (-2 .. +2) is usually
2890fine. 3730fine.
2891 3731
2892If your embedding app does not need any priorities, defining these both to 3732If your embedding application does not need any priorities, defining these
2893C<0> will save some memory and cpu. 3733both to C<0> will save some memory and CPU.
2894 3734
2895=item EV_PERIODIC_ENABLE 3735=item EV_PERIODIC_ENABLE
2896 3736
2897If undefined or defined to be C<1>, then periodic timers are supported. If 3737If undefined or defined to be C<1>, then periodic timers are supported. If
2898defined to be C<0>, then they are not. Disabling them saves a few kB of 3738defined to be C<0>, then they are not. Disabling them saves a few kB of
2905code. 3745code.
2906 3746
2907=item EV_EMBED_ENABLE 3747=item EV_EMBED_ENABLE
2908 3748
2909If undefined or defined to be C<1>, then embed watchers are supported. If 3749If undefined or defined to be C<1>, then embed watchers are supported. If
2910defined to be C<0>, then they are not. 3750defined to be C<0>, then they are not. Embed watchers rely on most other
3751watcher types, which therefore must not be disabled.
2911 3752
2912=item EV_STAT_ENABLE 3753=item EV_STAT_ENABLE
2913 3754
2914If undefined or defined to be C<1>, then stat watchers are supported. If 3755If undefined or defined to be C<1>, then stat watchers are supported. If
2915defined to be C<0>, then they are not. 3756defined to be C<0>, then they are not.
2925defined to be C<0>, then they are not. 3766defined to be C<0>, then they are not.
2926 3767
2927=item EV_MINIMAL 3768=item EV_MINIMAL
2928 3769
2929If you need to shave off some kilobytes of code at the expense of some 3770If you need to shave off some kilobytes of code at the expense of some
2930speed, define this symbol to C<1>. Currently only used for gcc to override 3771speed (but with the full API), define this symbol to C<1>. Currently this
2931some inlining decisions, saves roughly 30% codesize of amd64. 3772is used to override some inlining decisions, saves roughly 30% code size
3773on amd64. It also selects a much smaller 2-heap for timer management over
3774the default 4-heap.
3775
3776You can save even more by disabling watcher types you do not need
3777and setting C<EV_MAXPRI> == C<EV_MINPRI>. Also, disabling C<assert>
3778(C<-DNDEBUG>) will usually reduce code size a lot.
3779
3780Defining C<EV_MINIMAL> to C<2> will additionally reduce the core API to
3781provide a bare-bones event library. See C<ev.h> for details on what parts
3782of the API are still available, and do not complain if this subset changes
3783over time.
2932 3784
2933=item EV_PID_HASHSIZE 3785=item EV_PID_HASHSIZE
2934 3786
2935C<ev_child> watchers use a small hash table to distribute workload by 3787C<ev_child> watchers use a small hash table to distribute workload by
2936pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more 3788pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more
2943inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>), 3795inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
2944usually more than enough. If you need to manage thousands of C<ev_stat> 3796usually more than enough. If you need to manage thousands of C<ev_stat>
2945watchers you might want to increase this value (I<must> be a power of 3797watchers you might want to increase this value (I<must> be a power of
2946two). 3798two).
2947 3799
3800=item EV_USE_4HEAP
3801
3802Heaps are not very cache-efficient. To improve the cache-efficiency of the
3803timer and periodics heaps, libev uses a 4-heap when this symbol is defined
3804to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
3805faster performance with many (thousands) of watchers.
3806
3807The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3808(disabled).
3809
3810=item EV_HEAP_CACHE_AT
3811
3812Heaps are not very cache-efficient. To improve the cache-efficiency of the
3813timer and periodics heaps, libev can cache the timestamp (I<at>) within
3814the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
3815which uses 8-12 bytes more per watcher and a few hundred bytes more code,
3816but avoids random read accesses on heap changes. This improves performance
3817noticeably with many (hundreds) of watchers.
3818
3819The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3820(disabled).
3821
3822=item EV_VERIFY
3823
3824Controls how much internal verification (see C<ev_loop_verify ()>) will
3825be done: If set to C<0>, no internal verification code will be compiled
3826in. If set to C<1>, then verification code will be compiled in, but not
3827called. If set to C<2>, then the internal verification code will be
3828called once per loop, which can slow down libev. If set to C<3>, then the
3829verification code will be called very frequently, which will slow down
3830libev considerably.
3831
3832The default is C<1>, unless C<EV_MINIMAL> is set, in which case it will be
3833C<0>.
3834
2948=item EV_COMMON 3835=item EV_COMMON
2949 3836
2950By default, all watchers have a C<void *data> member. By redefining 3837By default, all watchers have a C<void *data> member. By redefining
2951this macro to a something else you can include more and other types of 3838this macro to a something else you can include more and other types of
2952members. You have to define it each time you include one of the files, 3839members. You have to define it each time you include one of the files,
2953though, and it must be identical each time. 3840though, and it must be identical each time.
2954 3841
2955For example, the perl EV module uses something like this: 3842For example, the perl EV module uses something like this:
2956 3843
2957 #define EV_COMMON \ 3844 #define EV_COMMON \
2958 SV *self; /* contains this struct */ \ 3845 SV *self; /* contains this struct */ \
2959 SV *cb_sv, *fh /* note no trailing ";" */ 3846 SV *cb_sv, *fh /* note no trailing ";" */
2960 3847
2961=item EV_CB_DECLARE (type) 3848=item EV_CB_DECLARE (type)
2962 3849
2963=item EV_CB_INVOKE (watcher, revents) 3850=item EV_CB_INVOKE (watcher, revents)
2964 3851
2969definition and a statement, respectively. See the F<ev.h> header file for 3856definition and a statement, respectively. See the F<ev.h> header file for
2970their default definitions. One possible use for overriding these is to 3857their default definitions. One possible use for overriding these is to
2971avoid the C<struct ev_loop *> as first argument in all cases, or to use 3858avoid the C<struct ev_loop *> as first argument in all cases, or to use
2972method calls instead of plain function calls in C++. 3859method calls instead of plain function calls in C++.
2973 3860
3861=back
3862
2974=head2 EXPORTED API SYMBOLS 3863=head2 EXPORTED API SYMBOLS
2975 3864
2976If you need to re-export the API (e.g. via a dll) and you need a list of 3865If you need to re-export the API (e.g. via a DLL) and you need a list of
2977exported symbols, you can use the provided F<Symbol.*> files which list 3866exported symbols, you can use the provided F<Symbol.*> files which list
2978all public symbols, one per line: 3867all public symbols, one per line:
2979 3868
2980 Symbols.ev for libev proper 3869 Symbols.ev for libev proper
2981 Symbols.event for the libevent emulation 3870 Symbols.event for the libevent emulation
2982 3871
2983This can also be used to rename all public symbols to avoid clashes with 3872This can also be used to rename all public symbols to avoid clashes with
2984multiple versions of libev linked together (which is obviously bad in 3873multiple versions of libev linked together (which is obviously bad in
2985itself, but sometimes it is inconvinient to avoid this). 3874itself, but sometimes it is inconvenient to avoid this).
2986 3875
2987A sed command like this will create wrapper C<#define>'s that you need to 3876A sed command like this will create wrapper C<#define>'s that you need to
2988include before including F<ev.h>: 3877include before including F<ev.h>:
2989 3878
2990 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h 3879 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
3007file. 3896file.
3008 3897
3009The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file 3898The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
3010that everybody includes and which overrides some configure choices: 3899that everybody includes and which overrides some configure choices:
3011 3900
3012 #define EV_MINIMAL 1 3901 #define EV_MINIMAL 1
3013 #define EV_USE_POLL 0 3902 #define EV_USE_POLL 0
3014 #define EV_MULTIPLICITY 0 3903 #define EV_MULTIPLICITY 0
3015 #define EV_PERIODIC_ENABLE 0 3904 #define EV_PERIODIC_ENABLE 0
3016 #define EV_STAT_ENABLE 0 3905 #define EV_STAT_ENABLE 0
3017 #define EV_FORK_ENABLE 0 3906 #define EV_FORK_ENABLE 0
3018 #define EV_CONFIG_H <config.h> 3907 #define EV_CONFIG_H <config.h>
3019 #define EV_MINPRI 0 3908 #define EV_MINPRI 0
3020 #define EV_MAXPRI 0 3909 #define EV_MAXPRI 0
3021 3910
3022 #include "ev++.h" 3911 #include "ev++.h"
3023 3912
3024And a F<ev_cpp.C> implementation file that contains libev proper and is compiled: 3913And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
3025 3914
3026 #include "ev_cpp.h" 3915 #include "ev_cpp.h"
3027 #include "ev.c" 3916 #include "ev.c"
3028 3917
3918=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
3029 3919
3030=head1 COMPLEXITIES 3920=head2 THREADS AND COROUTINES
3031 3921
3032In this section the complexities of (many of) the algorithms used inside 3922=head3 THREADS
3033libev will be explained. For complexity discussions about backends see the
3034documentation for C<ev_default_init>.
3035 3923
3036All of the following are about amortised time: If an array needs to be 3924All libev functions are reentrant and thread-safe unless explicitly
3037extended, libev needs to realloc and move the whole array, but this 3925documented otherwise, but libev implements no locking itself. This means
3038happens asymptotically never with higher number of elements, so O(1) might 3926that you can use as many loops as you want in parallel, as long as there
3039mean it might do a lengthy realloc operation in rare cases, but on average 3927are no concurrent calls into any libev function with the same loop
3040it is much faster and asymptotically approaches constant time. 3928parameter (C<ev_default_*> calls have an implicit default loop parameter,
3929of course): libev guarantees that different event loops share no data
3930structures that need any locking.
3931
3932Or to put it differently: calls with different loop parameters can be done
3933concurrently from multiple threads, calls with the same loop parameter
3934must be done serially (but can be done from different threads, as long as
3935only one thread ever is inside a call at any point in time, e.g. by using
3936a mutex per loop).
3937
3938Specifically to support threads (and signal handlers), libev implements
3939so-called C<ev_async> watchers, which allow some limited form of
3940concurrency on the same event loop, namely waking it up "from the
3941outside".
3942
3943If you want to know which design (one loop, locking, or multiple loops
3944without or something else still) is best for your problem, then I cannot
3945help you, but here is some generic advice:
3041 3946
3042=over 4 3947=over 4
3043 3948
3044=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers) 3949=item * most applications have a main thread: use the default libev loop
3950in that thread, or create a separate thread running only the default loop.
3045 3951
3046This means that, when you have a watcher that triggers in one hour and 3952This helps integrating other libraries or software modules that use libev
3047there are 100 watchers that would trigger before that then inserting will 3953themselves and don't care/know about threading.
3048have to skip roughly seven (C<ld 100>) of these watchers.
3049 3954
3050=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers) 3955=item * one loop per thread is usually a good model.
3051 3956
3052That means that changing a timer costs less than removing/adding them 3957Doing this is almost never wrong, sometimes a better-performance model
3053as only the relative motion in the event queue has to be paid for. 3958exists, but it is always a good start.
3054 3959
3055=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1) 3960=item * other models exist, such as the leader/follower pattern, where one
3961loop is handed through multiple threads in a kind of round-robin fashion.
3056 3962
3057These just add the watcher into an array or at the head of a list. 3963Choosing a model is hard - look around, learn, know that usually you can do
3964better than you currently do :-)
3058 3965
3059=item Stopping check/prepare/idle/fork/async watchers: O(1) 3966=item * often you need to talk to some other thread which blocks in the
3967event loop.
3060 3968
3061=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE)) 3969C<ev_async> watchers can be used to wake them up from other threads safely
3970(or from signal contexts...).
3062 3971
3063These watchers are stored in lists then need to be walked to find the 3972An example use would be to communicate signals or other events that only
3064correct watcher to remove. The lists are usually short (you don't usually 3973work in the default loop by registering the signal watcher with the
3065have many watchers waiting for the same fd or signal). 3974default loop and triggering an C<ev_async> watcher from the default loop
3066 3975watcher callback into the event loop interested in the signal.
3067=item Finding the next timer in each loop iteration: O(1)
3068
3069By virtue of using a binary heap, the next timer is always found at the
3070beginning of the storage array.
3071
3072=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
3073
3074A change means an I/O watcher gets started or stopped, which requires
3075libev to recalculate its status (and possibly tell the kernel, depending
3076on backend and wether C<ev_io_set> was used).
3077
3078=item Activating one watcher (putting it into the pending state): O(1)
3079
3080=item Priority handling: O(number_of_priorities)
3081
3082Priorities are implemented by allocating some space for each
3083priority. When doing priority-based operations, libev usually has to
3084linearly search all the priorities, but starting/stopping and activating
3085watchers becomes O(1) w.r.t. priority handling.
3086
3087=item Sending an ev_async: O(1)
3088
3089=item Processing ev_async_send: O(number_of_async_watchers)
3090
3091=item Processing signals: O(max_signal_number)
3092
3093Sending involves a syscall I<iff> there were no other C<ev_async_send>
3094calls in the current loop iteration. Checking for async and signal events
3095involves iterating over all running async watchers or all signal numbers.
3096 3976
3097=back 3977=back
3098 3978
3979=head4 THREAD LOCKING EXAMPLE
3099 3980
3100=head1 Win32 platform limitations and workarounds 3981Here is a fictitious example of how to run an event loop in a different
3982thread than where callbacks are being invoked and watchers are
3983created/added/removed.
3984
3985For a real-world example, see the C<EV::Loop::Async> perl module,
3986which uses exactly this technique (which is suited for many high-level
3987languages).
3988
3989The example uses a pthread mutex to protect the loop data, a condition
3990variable to wait for callback invocations, an async watcher to notify the
3991event loop thread and an unspecified mechanism to wake up the main thread.
3992
3993First, you need to associate some data with the event loop:
3994
3995 typedef struct {
3996 mutex_t lock; /* global loop lock */
3997 ev_async async_w;
3998 thread_t tid;
3999 cond_t invoke_cv;
4000 } userdata;
4001
4002 void prepare_loop (EV_P)
4003 {
4004 // for simplicity, we use a static userdata struct.
4005 static userdata u;
4006
4007 ev_async_init (&u->async_w, async_cb);
4008 ev_async_start (EV_A_ &u->async_w);
4009
4010 pthread_mutex_init (&u->lock, 0);
4011 pthread_cond_init (&u->invoke_cv, 0);
4012
4013 // now associate this with the loop
4014 ev_set_userdata (EV_A_ u);
4015 ev_set_invoke_pending_cb (EV_A_ l_invoke);
4016 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
4017
4018 // then create the thread running ev_loop
4019 pthread_create (&u->tid, 0, l_run, EV_A);
4020 }
4021
4022The callback for the C<ev_async> watcher does nothing: the watcher is used
4023solely to wake up the event loop so it takes notice of any new watchers
4024that might have been added:
4025
4026 static void
4027 async_cb (EV_P_ ev_async *w, int revents)
4028 {
4029 // just used for the side effects
4030 }
4031
4032The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
4033protecting the loop data, respectively.
4034
4035 static void
4036 l_release (EV_P)
4037 {
4038 userdata *u = ev_userdata (EV_A);
4039 pthread_mutex_unlock (&u->lock);
4040 }
4041
4042 static void
4043 l_acquire (EV_P)
4044 {
4045 userdata *u = ev_userdata (EV_A);
4046 pthread_mutex_lock (&u->lock);
4047 }
4048
4049The event loop thread first acquires the mutex, and then jumps straight
4050into C<ev_loop>:
4051
4052 void *
4053 l_run (void *thr_arg)
4054 {
4055 struct ev_loop *loop = (struct ev_loop *)thr_arg;
4056
4057 l_acquire (EV_A);
4058 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
4059 ev_loop (EV_A_ 0);
4060 l_release (EV_A);
4061
4062 return 0;
4063 }
4064
4065Instead of invoking all pending watchers, the C<l_invoke> callback will
4066signal the main thread via some unspecified mechanism (signals? pipe
4067writes? C<Async::Interrupt>?) and then waits until all pending watchers
4068have been called (in a while loop because a) spurious wakeups are possible
4069and b) skipping inter-thread-communication when there are no pending
4070watchers is very beneficial):
4071
4072 static void
4073 l_invoke (EV_P)
4074 {
4075 userdata *u = ev_userdata (EV_A);
4076
4077 while (ev_pending_count (EV_A))
4078 {
4079 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
4080 pthread_cond_wait (&u->invoke_cv, &u->lock);
4081 }
4082 }
4083
4084Now, whenever the main thread gets told to invoke pending watchers, it
4085will grab the lock, call C<ev_invoke_pending> and then signal the loop
4086thread to continue:
4087
4088 static void
4089 real_invoke_pending (EV_P)
4090 {
4091 userdata *u = ev_userdata (EV_A);
4092
4093 pthread_mutex_lock (&u->lock);
4094 ev_invoke_pending (EV_A);
4095 pthread_cond_signal (&u->invoke_cv);
4096 pthread_mutex_unlock (&u->lock);
4097 }
4098
4099Whenever you want to start/stop a watcher or do other modifications to an
4100event loop, you will now have to lock:
4101
4102 ev_timer timeout_watcher;
4103 userdata *u = ev_userdata (EV_A);
4104
4105 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
4106
4107 pthread_mutex_lock (&u->lock);
4108 ev_timer_start (EV_A_ &timeout_watcher);
4109 ev_async_send (EV_A_ &u->async_w);
4110 pthread_mutex_unlock (&u->lock);
4111
4112Note that sending the C<ev_async> watcher is required because otherwise
4113an event loop currently blocking in the kernel will have no knowledge
4114about the newly added timer. By waking up the loop it will pick up any new
4115watchers in the next event loop iteration.
4116
4117=head3 COROUTINES
4118
4119Libev is very accommodating to coroutines ("cooperative threads"):
4120libev fully supports nesting calls to its functions from different
4121coroutines (e.g. you can call C<ev_loop> on the same loop from two
4122different coroutines, and switch freely between both coroutines running
4123the loop, as long as you don't confuse yourself). The only exception is
4124that you must not do this from C<ev_periodic> reschedule callbacks.
4125
4126Care has been taken to ensure that libev does not keep local state inside
4127C<ev_loop>, and other calls do not usually allow for coroutine switches as
4128they do not call any callbacks.
4129
4130=head2 COMPILER WARNINGS
4131
4132Depending on your compiler and compiler settings, you might get no or a
4133lot of warnings when compiling libev code. Some people are apparently
4134scared by this.
4135
4136However, these are unavoidable for many reasons. For one, each compiler
4137has different warnings, and each user has different tastes regarding
4138warning options. "Warn-free" code therefore cannot be a goal except when
4139targeting a specific compiler and compiler-version.
4140
4141Another reason is that some compiler warnings require elaborate
4142workarounds, or other changes to the code that make it less clear and less
4143maintainable.
4144
4145And of course, some compiler warnings are just plain stupid, or simply
4146wrong (because they don't actually warn about the condition their message
4147seems to warn about). For example, certain older gcc versions had some
4148warnings that resulted an extreme number of false positives. These have
4149been fixed, but some people still insist on making code warn-free with
4150such buggy versions.
4151
4152While libev is written to generate as few warnings as possible,
4153"warn-free" code is not a goal, and it is recommended not to build libev
4154with any compiler warnings enabled unless you are prepared to cope with
4155them (e.g. by ignoring them). Remember that warnings are just that:
4156warnings, not errors, or proof of bugs.
4157
4158
4159=head2 VALGRIND
4160
4161Valgrind has a special section here because it is a popular tool that is
4162highly useful. Unfortunately, valgrind reports are very hard to interpret.
4163
4164If you think you found a bug (memory leak, uninitialised data access etc.)
4165in libev, then check twice: If valgrind reports something like:
4166
4167 ==2274== definitely lost: 0 bytes in 0 blocks.
4168 ==2274== possibly lost: 0 bytes in 0 blocks.
4169 ==2274== still reachable: 256 bytes in 1 blocks.
4170
4171Then there is no memory leak, just as memory accounted to global variables
4172is not a memleak - the memory is still being referenced, and didn't leak.
4173
4174Similarly, under some circumstances, valgrind might report kernel bugs
4175as if it were a bug in libev (e.g. in realloc or in the poll backend,
4176although an acceptable workaround has been found here), or it might be
4177confused.
4178
4179Keep in mind that valgrind is a very good tool, but only a tool. Don't
4180make it into some kind of religion.
4181
4182If you are unsure about something, feel free to contact the mailing list
4183with the full valgrind report and an explanation on why you think this
4184is a bug in libev (best check the archives, too :). However, don't be
4185annoyed when you get a brisk "this is no bug" answer and take the chance
4186of learning how to interpret valgrind properly.
4187
4188If you need, for some reason, empty reports from valgrind for your project
4189I suggest using suppression lists.
4190
4191
4192=head1 PORTABILITY NOTES
4193
4194=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
3101 4195
3102Win32 doesn't support any of the standards (e.g. POSIX) that libev 4196Win32 doesn't support any of the standards (e.g. POSIX) that libev
3103requires, and its I/O model is fundamentally incompatible with the POSIX 4197requires, and its I/O model is fundamentally incompatible with the POSIX
3104model. Libev still offers limited functionality on this platform in 4198model. Libev still offers limited functionality on this platform in
3105the form of the C<EVBACKEND_SELECT> backend, and only supports socket 4199the form of the C<EVBACKEND_SELECT> backend, and only supports socket
3106descriptors. This only applies when using Win32 natively, not when using 4200descriptors. This only applies when using Win32 natively, not when using
3107e.g. cygwin. 4201e.g. cygwin.
3108 4202
4203Lifting these limitations would basically require the full
4204re-implementation of the I/O system. If you are into these kinds of
4205things, then note that glib does exactly that for you in a very portable
4206way (note also that glib is the slowest event library known to man).
4207
3109There is no supported compilation method available on windows except 4208There is no supported compilation method available on windows except
3110embedding it into other applications. 4209embedding it into other applications.
3111 4210
4211Sensible signal handling is officially unsupported by Microsoft - libev
4212tries its best, but under most conditions, signals will simply not work.
4213
4214Not a libev limitation but worth mentioning: windows apparently doesn't
4215accept large writes: instead of resulting in a partial write, windows will
4216either accept everything or return C<ENOBUFS> if the buffer is too large,
4217so make sure you only write small amounts into your sockets (less than a
4218megabyte seems safe, but this apparently depends on the amount of memory
4219available).
4220
3112Due to the many, low, and arbitrary limits on the win32 platform and the 4221Due to the many, low, and arbitrary limits on the win32 platform and
3113abysmal performance of winsockets, using a large number of sockets is not 4222the abysmal performance of winsockets, using a large number of sockets
3114recommended (and not reasonable). If your program needs to use more than 4223is not recommended (and not reasonable). If your program needs to use
3115a hundred or so sockets, then likely it needs to use a totally different 4224more than a hundred or so sockets, then likely it needs to use a totally
3116implementation for windows, as libev offers the POSIX model, which cannot 4225different implementation for windows, as libev offers the POSIX readiness
3117be implemented efficiently on windows (microsoft monopoly games). 4226notification model, which cannot be implemented efficiently on windows
4227(due to Microsoft monopoly games).
4228
4229A typical way to use libev under windows is to embed it (see the embedding
4230section for details) and use the following F<evwrap.h> header file instead
4231of F<ev.h>:
4232
4233 #define EV_STANDALONE /* keeps ev from requiring config.h */
4234 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
4235
4236 #include "ev.h"
4237
4238And compile the following F<evwrap.c> file into your project (make sure
4239you do I<not> compile the F<ev.c> or any other embedded source files!):
4240
4241 #include "evwrap.h"
4242 #include "ev.c"
3118 4243
3119=over 4 4244=over 4
3120 4245
3121=item The winsocket select function 4246=item The winsocket select function
3122 4247
3123The winsocket C<select> function doesn't follow POSIX in that it requires 4248The winsocket C<select> function doesn't follow POSIX in that it
3124socket I<handles> and not socket I<file descriptors>. This makes select 4249requires socket I<handles> and not socket I<file descriptors> (it is
3125very inefficient, and also requires a mapping from file descriptors 4250also extremely buggy). This makes select very inefficient, and also
3126to socket handles. See the discussion of the C<EV_SELECT_USE_FD_SET>, 4251requires a mapping from file descriptors to socket handles (the Microsoft
3127C<EV_SELECT_IS_WINSOCKET> and C<EV_FD_TO_WIN32_HANDLE> preprocessor 4252C runtime provides the function C<_open_osfhandle> for this). See the
3128symbols for more info. 4253discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
4254C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
3129 4255
3130The configuration for a "naked" win32 using the microsoft runtime 4256The configuration for a "naked" win32 using the Microsoft runtime
3131libraries and raw winsocket select is: 4257libraries and raw winsocket select is:
3132 4258
3133 #define EV_USE_SELECT 1 4259 #define EV_USE_SELECT 1
3134 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */ 4260 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
3135 4261
3136Note that winsockets handling of fd sets is O(n), so you can easily get a 4262Note that winsockets handling of fd sets is O(n), so you can easily get a
3137complexity in the O(n²) range when using win32. 4263complexity in the O(n²) range when using win32.
3138 4264
3139=item Limited number of file descriptors 4265=item Limited number of file descriptors
3140 4266
3141Windows has numerous arbitrary (and low) limits on things. Early versions 4267Windows has numerous arbitrary (and low) limits on things.
3142of winsocket's select only supported waiting for a max. of C<64> handles 4268
4269Early versions of winsocket's select only supported waiting for a maximum
3143(probably owning to the fact that all windows kernels can only wait for 4270of C<64> handles (probably owning to the fact that all windows kernels
3144C<64> things at the same time internally; microsoft recommends spawning a 4271can only wait for C<64> things at the same time internally; Microsoft
3145chain of threads and wait for 63 handles and the previous thread in each). 4272recommends spawning a chain of threads and wait for 63 handles and the
4273previous thread in each. Sounds great!).
3146 4274
3147Newer versions support more handles, but you need to define C<FD_SETSIZE> 4275Newer versions support more handles, but you need to define C<FD_SETSIZE>
3148to some high number (e.g. C<2048>) before compiling the winsocket select 4276to some high number (e.g. C<2048>) before compiling the winsocket select
3149call (which might be in libev or elsewhere, for example, perl does its own 4277call (which might be in libev or elsewhere, for example, perl and many
3150select emulation on windows). 4278other interpreters do their own select emulation on windows).
3151 4279
3152Another limit is the number of file descriptors in the microsoft runtime 4280Another limit is the number of file descriptors in the Microsoft runtime
3153libraries, which by default is C<64> (there must be a hidden I<64> fetish 4281libraries, which by default is C<64> (there must be a hidden I<64>
3154or something like this inside microsoft). You can increase this by calling 4282fetish or something like this inside Microsoft). You can increase this
3155C<_setmaxstdio>, which can increase this limit to C<2048> (another 4283by calling C<_setmaxstdio>, which can increase this limit to C<2048>
3156arbitrary limit), but is broken in many versions of the microsoft runtime 4284(another arbitrary limit), but is broken in many versions of the Microsoft
3157libraries.
3158
3159This might get you to about C<512> or C<2048> sockets (depending on 4285runtime libraries. This might get you to about C<512> or C<2048> sockets
3160windows version and/or the phase of the moon). To get more, you need to 4286(depending on windows version and/or the phase of the moon). To get more,
3161wrap all I/O functions and provide your own fd management, but the cost of 4287you need to wrap all I/O functions and provide your own fd management, but
3162calling select (O(n²)) will likely make this unworkable. 4288the cost of calling select (O(n²)) will likely make this unworkable.
3163 4289
3164=back 4290=back
3165 4291
4292=head2 PORTABILITY REQUIREMENTS
4293
4294In addition to a working ISO-C implementation and of course the
4295backend-specific APIs, libev relies on a few additional extensions:
4296
4297=over 4
4298
4299=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
4300calling conventions regardless of C<ev_watcher_type *>.
4301
4302Libev assumes not only that all watcher pointers have the same internal
4303structure (guaranteed by POSIX but not by ISO C for example), but it also
4304assumes that the same (machine) code can be used to call any watcher
4305callback: The watcher callbacks have different type signatures, but libev
4306calls them using an C<ev_watcher *> internally.
4307
4308=item C<sig_atomic_t volatile> must be thread-atomic as well
4309
4310The type C<sig_atomic_t volatile> (or whatever is defined as
4311C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
4312threads. This is not part of the specification for C<sig_atomic_t>, but is
4313believed to be sufficiently portable.
4314
4315=item C<sigprocmask> must work in a threaded environment
4316
4317Libev uses C<sigprocmask> to temporarily block signals. This is not
4318allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
4319pthread implementations will either allow C<sigprocmask> in the "main
4320thread" or will block signals process-wide, both behaviours would
4321be compatible with libev. Interaction between C<sigprocmask> and
4322C<pthread_sigmask> could complicate things, however.
4323
4324The most portable way to handle signals is to block signals in all threads
4325except the initial one, and run the default loop in the initial thread as
4326well.
4327
4328=item C<long> must be large enough for common memory allocation sizes
4329
4330To improve portability and simplify its API, libev uses C<long> internally
4331instead of C<size_t> when allocating its data structures. On non-POSIX
4332systems (Microsoft...) this might be unexpectedly low, but is still at
4333least 31 bits everywhere, which is enough for hundreds of millions of
4334watchers.
4335
4336=item C<double> must hold a time value in seconds with enough accuracy
4337
4338The type C<double> is used to represent timestamps. It is required to
4339have at least 51 bits of mantissa (and 9 bits of exponent), which is good
4340enough for at least into the year 4000. This requirement is fulfilled by
4341implementations implementing IEEE 754, which is basically all existing
4342ones. With IEEE 754 doubles, you get microsecond accuracy until at least
43432200.
4344
4345=back
4346
4347If you know of other additional requirements drop me a note.
4348
4349
4350=head1 ALGORITHMIC COMPLEXITIES
4351
4352In this section the complexities of (many of) the algorithms used inside
4353libev will be documented. For complexity discussions about backends see
4354the documentation for C<ev_default_init>.
4355
4356All of the following are about amortised time: If an array needs to be
4357extended, libev needs to realloc and move the whole array, but this
4358happens asymptotically rarer with higher number of elements, so O(1) might
4359mean that libev does a lengthy realloc operation in rare cases, but on
4360average it is much faster and asymptotically approaches constant time.
4361
4362=over 4
4363
4364=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
4365
4366This means that, when you have a watcher that triggers in one hour and
4367there are 100 watchers that would trigger before that, then inserting will
4368have to skip roughly seven (C<ld 100>) of these watchers.
4369
4370=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
4371
4372That means that changing a timer costs less than removing/adding them,
4373as only the relative motion in the event queue has to be paid for.
4374
4375=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
4376
4377These just add the watcher into an array or at the head of a list.
4378
4379=item Stopping check/prepare/idle/fork/async watchers: O(1)
4380
4381=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
4382
4383These watchers are stored in lists, so they need to be walked to find the
4384correct watcher to remove. The lists are usually short (you don't usually
4385have many watchers waiting for the same fd or signal: one is typical, two
4386is rare).
4387
4388=item Finding the next timer in each loop iteration: O(1)
4389
4390By virtue of using a binary or 4-heap, the next timer is always found at a
4391fixed position in the storage array.
4392
4393=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
4394
4395A change means an I/O watcher gets started or stopped, which requires
4396libev to recalculate its status (and possibly tell the kernel, depending
4397on backend and whether C<ev_io_set> was used).
4398
4399=item Activating one watcher (putting it into the pending state): O(1)
4400
4401=item Priority handling: O(number_of_priorities)
4402
4403Priorities are implemented by allocating some space for each
4404priority. When doing priority-based operations, libev usually has to
4405linearly search all the priorities, but starting/stopping and activating
4406watchers becomes O(1) with respect to priority handling.
4407
4408=item Sending an ev_async: O(1)
4409
4410=item Processing ev_async_send: O(number_of_async_watchers)
4411
4412=item Processing signals: O(max_signal_number)
4413
4414Sending involves a system call I<iff> there were no other C<ev_async_send>
4415calls in the current loop iteration. Checking for async and signal events
4416involves iterating over all running async watchers or all signal numbers.
4417
4418=back
4419
4420
4421=head1 GLOSSARY
4422
4423=over 4
4424
4425=item active
4426
4427A watcher is active as long as it has been started (has been attached to
4428an event loop) but not yet stopped (disassociated from the event loop).
4429
4430=item application
4431
4432In this document, an application is whatever is using libev.
4433
4434=item callback
4435
4436The address of a function that is called when some event has been
4437detected. Callbacks are being passed the event loop, the watcher that
4438received the event, and the actual event bitset.
4439
4440=item callback invocation
4441
4442The act of calling the callback associated with a watcher.
4443
4444=item event
4445
4446A change of state of some external event, such as data now being available
4447for reading on a file descriptor, time having passed or simply not having
4448any other events happening anymore.
4449
4450In libev, events are represented as single bits (such as C<EV_READ> or
4451C<EV_TIMEOUT>).
4452
4453=item event library
4454
4455A software package implementing an event model and loop.
4456
4457=item event loop
4458
4459An entity that handles and processes external events and converts them
4460into callback invocations.
4461
4462=item event model
4463
4464The model used to describe how an event loop handles and processes
4465watchers and events.
4466
4467=item pending
4468
4469A watcher is pending as soon as the corresponding event has been detected,
4470and stops being pending as soon as the watcher will be invoked or its
4471pending status is explicitly cleared by the application.
4472
4473A watcher can be pending, but not active. Stopping a watcher also clears
4474its pending status.
4475
4476=item real time
4477
4478The physical time that is observed. It is apparently strictly monotonic :)
4479
4480=item wall-clock time
4481
4482The time and date as shown on clocks. Unlike real time, it can actually
4483be wrong and jump forwards and backwards, e.g. when the you adjust your
4484clock.
4485
4486=item watcher
4487
4488A data structure that describes interest in certain events. Watchers need
4489to be started (attached to an event loop) before they can receive events.
4490
4491=item watcher invocation
4492
4493The act of calling the callback associated with a watcher.
4494
4495=back
3166 4496
3167=head1 AUTHOR 4497=head1 AUTHOR
3168 4498
3169Marc Lehmann <libev@schmorp.de>. 4499Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
3170 4500

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines