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

Comparing libev/ev.pod (file contents):
Revision 1.33 by root, Fri Nov 23 15:26:08 2007 UTC vs.
Revision 1.265 by root, Wed Aug 26 17:11:42 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=head1 DESCRIPTION 9=head2 EXAMPLE PROGRAM
10
11 // a single header file is required
12 #include <ev.h>
13
14 #include <stdio.h> // for puts
15
16 // every watcher type has its own typedef'd struct
17 // with the name ev_TYPE
18 ev_io stdin_watcher;
19 ev_timer timeout_watcher;
20
21 // all watcher callbacks have a similar signature
22 // this callback is called when data is readable on stdin
23 static void
24 stdin_cb (EV_P_ ev_io *w, int revents)
25 {
26 puts ("stdin ready");
27 // for one-shot events, one must manually stop the watcher
28 // with its corresponding stop function.
29 ev_io_stop (EV_A_ w);
30
31 // this causes all nested ev_loop's to stop iterating
32 ev_unloop (EV_A_ EVUNLOOP_ALL);
33 }
34
35 // another callback, this time for a time-out
36 static void
37 timeout_cb (EV_P_ ev_timer *w, int revents)
38 {
39 puts ("timeout");
40 // this causes the innermost ev_loop to stop iterating
41 ev_unloop (EV_A_ EVUNLOOP_ONE);
42 }
43
44 int
45 main (void)
46 {
47 // use the default event loop unless you have special needs
48 struct ev_loop *loop = ev_default_loop (0);
49
50 // initialise an io watcher, then start it
51 // this one will watch for stdin to become readable
52 ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
53 ev_io_start (loop, &stdin_watcher);
54
55 // initialise a timer watcher, then start it
56 // simple non-repeating 5.5 second timeout
57 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
58 ev_timer_start (loop, &timeout_watcher);
59
60 // now wait for events to arrive
61 ev_loop (loop, 0);
62
63 // unloop was called, so exit
64 return 0;
65 }
66
67=head1 ABOUT THIS DOCUMENT
68
69This document documents the libev software package.
70
71The newest version of this document is also available as an html-formatted
72web page you might find easier to navigate when reading it for the first
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
10 84
11Libev 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
12file descriptor being readable or a timeout occuring), and it will manage 86file descriptor being readable or a timeout occurring), and it will manage
13these event sources and provide your program with events. 87these event sources and provide your program with events.
14 88
15To do this, it must take more or less complete control over your process 89To do this, it must take more or less complete control over your process
16(or thread) by executing the I<event loop> handler, and will then 90(or thread) by executing the I<event loop> handler, and will then
17communicate events via a callback mechanism. 91communicate events via a callback mechanism.
19You register interest in certain events by registering so-called I<event 93You register interest in certain events by registering so-called I<event
20watchers>, which are relatively small C structures you initialise with the 94watchers>, which are relatively small C structures you initialise with the
21details of the event, and then hand it over to libev by I<starting> the 95details of the event, and then hand it over to libev by I<starting> the
22watcher. 96watcher.
23 97
24=head1 FEATURES 98=head2 FEATURES
25 99
26Libev supports select, poll, the linux-specific epoll and the bsd-specific 100Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
27kqueue mechanisms for file descriptor events, relative timers, absolute 101BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
28timers with customised rescheduling, signal events, process status change 102for file descriptor events (C<ev_io>), the Linux C<inotify> interface
103(for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
104inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
105timers (C<ev_timer>), absolute timers with customised rescheduling
106(C<ev_periodic>), synchronous signals (C<ev_signal>), process status
29events (related to SIGCHLD), and event watchers dealing with the event 107change events (C<ev_child>), and event watchers dealing with the event
30loop mechanism itself (idle, prepare and check watchers). It also is quite 108loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
109C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
110limited support for fork events (C<ev_fork>).
111
112It also is quite fast (see this
31fast (see this L<benchmark|http://libev.schmorp.de/bench.html> comparing 113L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
32it to libevent for example). 114for example).
33 115
34=head1 CONVENTIONS 116=head2 CONVENTIONS
35 117
36Libev is very configurable. In this manual the default configuration 118Libev is very configurable. In this manual the default (and most common)
37will be described, which supports multiple event loops. For more info 119configuration will be described, which supports multiple event loops. For
38about various configuration options please have a look at the file 120more info about various configuration options please have a look at
39F<README.embed> in the libev distribution. If libev was configured without 121B<EMBED> section in this manual. If libev was configured without support
40support for multiple event loops, then all functions taking an initial 122for multiple event loops, then all functions taking an initial argument of
41argument of name C<loop> (which is always of type C<struct ev_loop *>) 123name C<loop> (which is always of type C<ev_loop *>) will not have
42will not have this argument. 124this argument.
43 125
44=head1 TIME REPRESENTATION 126=head2 TIME REPRESENTATION
45 127
46Libev represents time as a single floating point number, representing the 128Libev represents time as a single floating point number, representing
47(fractional) number of seconds since the (POSIX) epoch (somewhere near 129the (fractional) number of seconds since the (POSIX) epoch (somewhere
48the beginning of 1970, details are complicated, don't ask). This type is 130near the beginning of 1970, details are complicated, don't ask). This
49called C<ev_tstamp>, which is what you should use too. It usually aliases 131type is called C<ev_tstamp>, which is what you should use too. It usually
50to the double type in C. 132aliases to the C<double> type in C. When you need to do any calculations
133on it, you should treat it as some floating point value. Unlike the name
134component C<stamp> might indicate, it is also used for time differences
135throughout libev.
136
137=head1 ERROR HANDLING
138
139Libev knows three classes of errors: operating system errors, usage errors
140and internal errors (bugs).
141
142When libev catches an operating system error it cannot handle (for example
143a system call indicating a condition libev cannot fix), it calls the callback
144set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
145abort. The default is to print a diagnostic message and to call C<abort
146()>.
147
148When libev detects a usage error such as a negative timer interval, then
149it will print a diagnostic message and abort (via the C<assert> mechanism,
150so C<NDEBUG> will disable this checking): these are programming errors in
151the libev caller and need to be fixed there.
152
153Libev also has a few internal error-checking C<assert>ions, and also has
154extensive consistency checking code. These do not trigger under normal
155circumstances, as they indicate either a bug in libev or worse.
156
51 157
52=head1 GLOBAL FUNCTIONS 158=head1 GLOBAL FUNCTIONS
53 159
54These functions can be called anytime, even before initialising the 160These functions can be called anytime, even before initialising the
55library in any way. 161library in any way.
60 166
61Returns the current time as libev would use it. Please note that the 167Returns the current time as libev would use it. Please note that the
62C<ev_now> function is usually faster and also often returns the timestamp 168C<ev_now> function is usually faster and also often returns the timestamp
63you actually want to know. 169you actually want to know.
64 170
171=item ev_sleep (ev_tstamp interval)
172
173Sleep for the given interval: The current thread will be blocked until
174either it is interrupted or the given time interval has passed. Basically
175this is a sub-second-resolution C<sleep ()>.
176
65=item int ev_version_major () 177=item int ev_version_major ()
66 178
67=item int ev_version_minor () 179=item int ev_version_minor ()
68 180
69You can find out the major and minor version numbers of the library 181You can find out the major and minor ABI version numbers of the library
70you linked against by calling the functions C<ev_version_major> and 182you linked against by calling the functions C<ev_version_major> and
71C<ev_version_minor>. If you want, you can compare against the global 183C<ev_version_minor>. If you want, you can compare against the global
72symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the 184symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
73version of the library your program was compiled against. 185version of the library your program was compiled against.
74 186
187These version numbers refer to the ABI version of the library, not the
188release version.
189
75Usually, it's a good idea to terminate if the major versions mismatch, 190Usually, it's a good idea to terminate if the major versions mismatch,
76as this indicates an incompatible change. Minor versions are usually 191as this indicates an incompatible change. Minor versions are usually
77compatible to older versions, so a larger minor version alone is usually 192compatible to older versions, so a larger minor version alone is usually
78not a problem. 193not a problem.
194
195Example: Make sure we haven't accidentally been linked against the wrong
196version.
197
198 assert (("libev version mismatch",
199 ev_version_major () == EV_VERSION_MAJOR
200 && ev_version_minor () >= EV_VERSION_MINOR));
79 201
80=item unsigned int ev_supported_backends () 202=item unsigned int ev_supported_backends ()
81 203
82Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*> 204Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
83value) compiled into this binary of libev (independent of their 205value) compiled into this binary of libev (independent of their
84availability on the system you are running on). See C<ev_default_loop> for 206availability on the system you are running on). See C<ev_default_loop> for
85a description of the set values. 207a description of the set values.
86 208
209Example: make sure we have the epoll method, because yeah this is cool and
210a must have and can we have a torrent of it please!!!11
211
212 assert (("sorry, no epoll, no sex",
213 ev_supported_backends () & EVBACKEND_EPOLL));
214
87=item unsigned int ev_recommended_backends () 215=item unsigned int ev_recommended_backends ()
88 216
89Return the set of all backends compiled into this binary of libev and also 217Return the set of all backends compiled into this binary of libev and also
90recommended for this platform. This set is often smaller than the one 218recommended for this platform. This set is often smaller than the one
91returned by C<ev_supported_backends>, as for example kqueue is broken on 219returned by C<ev_supported_backends>, as for example kqueue is broken on
92most BSDs and will not be autodetected unless you explicitly request it 220most BSDs and will not be auto-detected unless you explicitly request it
93(assuming you know what you are doing). This is the set of backends that 221(assuming you know what you are doing). This is the set of backends that
94libev will probe for if you specify no backends explicitly. 222libev will probe for if you specify no backends explicitly.
95 223
224=item unsigned int ev_embeddable_backends ()
225
226Returns the set of backends that are embeddable in other event loops. This
227is the theoretical, all-platform, value. To find which backends
228might be supported on the current system, you would need to look at
229C<ev_embeddable_backends () & ev_supported_backends ()>, likewise for
230recommended ones.
231
232See the description of C<ev_embed> watchers for more info.
233
96=item ev_set_allocator (void *(*cb)(void *ptr, long size)) 234=item ev_set_allocator (void *(*cb)(void *ptr, long size)) [NOT REENTRANT]
97 235
98Sets the allocation function to use (the prototype is similar to the 236Sets the allocation function to use (the prototype is similar - the
99realloc C function, the semantics are identical). It is used to allocate 237semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
100and free memory (no surprises here). If it returns zero when memory 238used to allocate and free memory (no surprises here). If it returns zero
101needs to be allocated, the library might abort or take some potentially 239when memory needs to be allocated (C<size != 0>), the library might abort
102destructive action. The default is your system realloc function. 240or take some potentially destructive action.
241
242Since some systems (at least OpenBSD and Darwin) fail to implement
243correct C<realloc> semantics, libev will use a wrapper around the system
244C<realloc> and C<free> functions by default.
103 245
104You could override this function in high-availability programs to, say, 246You could override this function in high-availability programs to, say,
105free some memory if it cannot allocate memory, to use a special allocator, 247free some memory if it cannot allocate memory, to use a special allocator,
106or even to sleep a while and retry until some memory is available. 248or even to sleep a while and retry until some memory is available.
107 249
250Example: Replace the libev allocator with one that waits a bit and then
251retries (example requires a standards-compliant C<realloc>).
252
253 static void *
254 persistent_realloc (void *ptr, size_t size)
255 {
256 for (;;)
257 {
258 void *newptr = realloc (ptr, size);
259
260 if (newptr)
261 return newptr;
262
263 sleep (60);
264 }
265 }
266
267 ...
268 ev_set_allocator (persistent_realloc);
269
108=item ev_set_syserr_cb (void (*cb)(const char *msg)); 270=item ev_set_syserr_cb (void (*cb)(const char *msg)); [NOT REENTRANT]
109 271
110Set the callback function to call on a retryable syscall error (such 272Set the callback function to call on a retryable system call error (such
111as failed select, poll, epoll_wait). The message is a printable string 273as failed select, poll, epoll_wait). The message is a printable string
112indicating the system call or subsystem causing the problem. If this 274indicating the system call or subsystem causing the problem. If this
113callback is set, then libev will expect it to remedy the sitution, no 275callback is set, then libev will expect it to remedy the situation, no
114matter what, when it returns. That is, libev will generally retry the 276matter what, when it returns. That is, libev will generally retry the
115requested operation, or, if the condition doesn't go away, do bad stuff 277requested operation, or, if the condition doesn't go away, do bad stuff
116(such as abort). 278(such as abort).
117 279
280Example: This is basically the same thing that libev does internally, too.
281
282 static void
283 fatal_error (const char *msg)
284 {
285 perror (msg);
286 abort ();
287 }
288
289 ...
290 ev_set_syserr_cb (fatal_error);
291
118=back 292=back
119 293
120=head1 FUNCTIONS CONTROLLING THE EVENT LOOP 294=head1 FUNCTIONS CONTROLLING THE EVENT LOOP
121 295
122An event loop is described by a C<struct ev_loop *>. The library knows two 296An event loop is described by a C<struct ev_loop *> (the C<struct>
123types of such loops, the I<default> loop, which supports signals and child 297is I<not> optional in this case, as there is also an C<ev_loop>
124events, and dynamically created loops which do not. 298I<function>).
125 299
126If you use threads, a common model is to run the default event loop 300The library knows two types of such loops, the I<default> loop, which
127in your main thread (or in a separate thread) and for each thread you 301supports signals and child events, and dynamically created loops which do
128create, you also create another event loop. Libev itself does no locking 302not.
129whatsoever, so if you mix calls to the same event loop in different
130threads, make sure you lock (this is usually a bad idea, though, even if
131done correctly, because it's hideous and inefficient).
132 303
133=over 4 304=over 4
134 305
135=item struct ev_loop *ev_default_loop (unsigned int flags) 306=item struct ev_loop *ev_default_loop (unsigned int flags)
136 307
140flags. If that is troubling you, check C<ev_backend ()> afterwards). 311flags. If that is troubling you, check C<ev_backend ()> afterwards).
141 312
142If you don't know what event loop to use, use the one returned from this 313If you don't know what event loop to use, use the one returned from this
143function. 314function.
144 315
316Note that this function is I<not> thread-safe, so if you want to use it
317from multiple threads, you have to lock (note also that this is unlikely,
318as loops cannot be shared easily between threads anyway).
319
320The default loop is the only loop that can handle C<ev_signal> and
321C<ev_child> watchers, and to do this, it always registers a handler
322for C<SIGCHLD>. If this is a problem for your application you can either
323create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
324can simply overwrite the C<SIGCHLD> signal handler I<after> calling
325C<ev_default_init>.
326
145The flags argument can be used to specify special behaviour or specific 327The flags argument can be used to specify special behaviour or specific
146backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>). 328backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
147 329
148The following flags are supported: 330The following flags are supported:
149 331
154The default flags value. Use this if you have no clue (it's the right 336The default flags value. Use this if you have no clue (it's the right
155thing, believe me). 337thing, believe me).
156 338
157=item C<EVFLAG_NOENV> 339=item C<EVFLAG_NOENV>
158 340
159If this flag bit is ored into the flag value (or the program runs setuid 341If this flag bit is or'ed into the flag value (or the program runs setuid
160or setgid) then libev will I<not> look at the environment variable 342or setgid) then libev will I<not> look at the environment variable
161C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will 343C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
162override the flags completely if it is found in the environment. This is 344override the flags completely if it is found in the environment. This is
163useful to try out specific backends to test their performance, or to work 345useful to try out specific backends to test their performance, or to work
164around bugs. 346around bugs.
165 347
348=item C<EVFLAG_FORKCHECK>
349
350Instead of calling C<ev_default_fork> or C<ev_loop_fork> manually after
351a fork, you can also make libev check for a fork in each iteration by
352enabling this flag.
353
354This works by calling C<getpid ()> on every iteration of the loop,
355and thus this might slow down your event loop if you do a lot of loop
356iterations and little real work, but is usually not noticeable (on my
357GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
358without a system call and thus I<very> fast, but my GNU/Linux system also has
359C<pthread_atfork> which is even faster).
360
361The big advantage of this flag is that you can forget about fork (and
362forget about forgetting to tell libev about forking) when you use this
363flag.
364
365This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
366environment variable.
367
368=item C<EVFLAG_NOINOTIFY>
369
370When this flag is specified, then libev will not attempt to use the
371I<inotify> API for it's C<ev_stat> watchers. Apart from debugging and
372testing, this flag can be useful to conserve inotify file descriptors, as
373otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
374
375=item C<EVFLAG_NOSIGNALFD>
376
377When this flag is specified, then libev will not attempt to use the
378I<signalfd> API for it's C<ev_signal> (and C<ev_child>) watchers. This is
379probably only useful to work around any bugs in libev. Consequently, this
380flag might go away once the signalfd functionality is considered stable,
381so it's useful mostly in environment variables and not in program code.
382
166=item C<EVBACKEND_SELECT> (value 1, portable select backend) 383=item C<EVBACKEND_SELECT> (value 1, portable select backend)
167 384
168This is your standard select(2) backend. Not I<completely> standard, as 385This is your standard select(2) backend. Not I<completely> standard, as
169libev tries to roll its own fd_set with no limits on the number of fds, 386libev tries to roll its own fd_set with no limits on the number of fds,
170but if that fails, expect a fairly low limit on the number of fds when 387but if that fails, expect a fairly low limit on the number of fds when
171using this backend. It doesn't scale too well (O(highest_fd)), but its usually 388using this backend. It doesn't scale too well (O(highest_fd)), but its
172the fastest backend for a low number of fds. 389usually the fastest backend for a low number of (low-numbered :) fds.
390
391To get good performance out of this backend you need a high amount of
392parallelism (most of the file descriptors should be busy). If you are
393writing a server, you should C<accept ()> in a loop to accept as many
394connections as possible during one iteration. You might also want to have
395a look at C<ev_set_io_collect_interval ()> to increase the amount of
396readiness notifications you get per iteration.
397
398This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
399C<writefds> set (and to work around Microsoft Windows bugs, also onto the
400C<exceptfds> set on that platform).
173 401
174=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows) 402=item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
175 403
176And this is your standard poll(2) backend. It's more complicated than 404And this is your standard poll(2) backend. It's more complicated
177select, but handles sparse fds better and has no artificial limit on the 405than select, but handles sparse fds better and has no artificial
178number of fds you can use (except it will slow down considerably with a 406limit on the number of fds you can use (except it will slow down
179lot of inactive fds). It scales similarly to select, i.e. O(total_fds). 407considerably with a lot of inactive fds). It scales similarly to select,
408i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
409performance tips.
410
411This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
412C<EV_WRITE> to C<POLLOUT | POLLERR | POLLHUP>.
180 413
181=item C<EVBACKEND_EPOLL> (value 4, Linux) 414=item C<EVBACKEND_EPOLL> (value 4, Linux)
182 415
183For few fds, this backend is a bit little slower than poll and select, 416For few fds, this backend is a bit little slower than poll and select,
184but it scales phenomenally better. While poll and select usually scale like 417but it scales phenomenally better. While poll and select usually scale
185O(total_fds) where n is the total number of fds (or the highest fd), epoll scales 418like O(total_fds) where n is the total number of fds (or the highest fd),
186either O(1) or O(active_fds). 419epoll scales either O(1) or O(active_fds).
187 420
421The epoll mechanism deserves honorable mention as the most misdesigned
422of the more advanced event mechanisms: mere annoyances include silently
423dropping file descriptors, requiring a system call per change per file
424descriptor (and unnecessary guessing of parameters), problems with dup and
425so on. The biggest issue is fork races, however - if a program forks then
426I<both> parent and child process have to recreate the epoll set, which can
427take considerable time (one syscall per file descriptor) and is of course
428hard to detect.
429
430Epoll is also notoriously buggy - embedding epoll fds I<should> work, but
431of course I<doesn't>, and epoll just loves to report events for totally
432I<different> file descriptors (even already closed ones, so one cannot
433even remove them from the set) than registered in the set (especially
434on SMP systems). Libev tries to counter these spurious notifications by
435employing an additional generation counter and comparing that against the
436events to filter out spurious ones, recreating the set when required.
437
188While stopping and starting an I/O watcher in the same iteration will 438While stopping, setting and starting an I/O watcher in the same iteration
189result in some caching, there is still a syscall per such incident 439will result in some caching, there is still a system call per such
190(because the fd could point to a different file description now), so its 440incident (because the same I<file descriptor> could point to a different
191best to avoid that. Also, dup()ed file descriptors might not work very 441I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
192well if you register events for both fds. 442file descriptors might not work very well if you register events for both
443file descriptors.
193 444
194Please note that epoll sometimes generates spurious notifications, so you 445Best performance from this backend is achieved by not unregistering all
195need to use non-blocking I/O or other means to avoid blocking when no data 446watchers for a file descriptor until it has been closed, if possible,
196(or space) is available. 447i.e. keep at least one watcher active per fd at all times. Stopping and
448starting a watcher (without re-setting it) also usually doesn't cause
449extra overhead. A fork can both result in spurious notifications as well
450as in libev having to destroy and recreate the epoll object, which can
451take considerable time and thus should be avoided.
452
453All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
454faster than epoll for maybe up to a hundred file descriptors, depending on
455the usage. So sad.
456
457While nominally embeddable in other event loops, this feature is broken in
458all kernel versions tested so far.
459
460This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
461C<EVBACKEND_POLL>.
197 462
198=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones) 463=item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
199 464
200Kqueue deserves special mention, as at the time of this writing, it 465Kqueue deserves special mention, as at the time of this writing, it
201was broken on all BSDs except NetBSD (usually it doesn't work with 466was broken on all BSDs except NetBSD (usually it doesn't work reliably
202anything but sockets and pipes, except on Darwin, where of course its 467with anything but sockets and pipes, except on Darwin, where of course
203completely useless). For this reason its not being "autodetected" 468it's completely useless). Unlike epoll, however, whose brokenness
469is by design, these kqueue bugs can (and eventually will) be fixed
470without API changes to existing programs. For this reason it's not being
204unless you explicitly specify it explicitly in the flags (i.e. using 471"auto-detected" unless you explicitly specify it in the flags (i.e. using
205C<EVBACKEND_KQUEUE>). 472C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
473system like NetBSD.
474
475You still can embed kqueue into a normal poll or select backend and use it
476only for sockets (after having made sure that sockets work with kqueue on
477the target platform). See C<ev_embed> watchers for more info.
206 478
207It scales in the same way as the epoll backend, but the interface to the 479It scales in the same way as the epoll backend, but the interface to the
208kernel is more efficient (which says nothing about its actual speed, of 480kernel is more efficient (which says nothing about its actual speed, of
209course). While starting and stopping an I/O watcher does not cause an 481course). While stopping, setting and starting an I/O watcher does never
210extra syscall as with epoll, it still adds up to four event changes per 482cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
211incident, so its best to avoid that. 483two event changes per incident. Support for C<fork ()> is very bad (but
484sane, unlike epoll) and it drops fds silently in similarly hard-to-detect
485cases
486
487This backend usually performs well under most conditions.
488
489While nominally embeddable in other event loops, this doesn't work
490everywhere, so you might need to test for this. And since it is broken
491almost everywhere, you should only use it when you have a lot of sockets
492(for which it usually works), by embedding it into another event loop
493(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
494also broken on OS X)) and, did I mention it, using it only for sockets.
495
496This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
497C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
498C<NOTE_EOF>.
212 499
213=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8) 500=item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
214 501
215This is not implemented yet (and might never be). 502This is not implemented yet (and might never be, unless you send me an
503implementation). According to reports, C</dev/poll> only supports sockets
504and is not embeddable, which would limit the usefulness of this backend
505immensely.
216 506
217=item C<EVBACKEND_PORT> (value 32, Solaris 10) 507=item C<EVBACKEND_PORT> (value 32, Solaris 10)
218 508
219This uses the Solaris 10 port mechanism. As with everything on Solaris, 509This uses the Solaris 10 event port mechanism. As with everything on Solaris,
220it's really slow, but it still scales very well (O(active_fds)). 510it's really slow, but it still scales very well (O(active_fds)).
221 511
222Please note that solaris ports can result in a lot of spurious 512Please note that Solaris event ports can deliver a lot of spurious
223notifications, so you need to use non-blocking I/O or other means to avoid 513notifications, so you need to use non-blocking I/O or other means to avoid
224blocking when no data (or space) is available. 514blocking when no data (or space) is available.
515
516While this backend scales well, it requires one system call per active
517file descriptor per loop iteration. For small and medium numbers of file
518descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
519might perform better.
520
521On the positive side, with the exception of the spurious readiness
522notifications, this backend actually performed fully to specification
523in all tests and is fully embeddable, which is a rare feat among the
524OS-specific backends (I vastly prefer correctness over speed hacks).
525
526This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
527C<EVBACKEND_POLL>.
225 528
226=item C<EVBACKEND_ALL> 529=item C<EVBACKEND_ALL>
227 530
228Try all backends (even potentially broken ones that wouldn't be tried 531Try all backends (even potentially broken ones that wouldn't be tried
229with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as 532with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
230C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>. 533C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
231 534
535It is definitely not recommended to use this flag.
536
232=back 537=back
233 538
234If one or more of these are ored into the flags value, then only these 539If one or more of the backend flags are or'ed into the flags value,
235backends will be tried (in the reverse order as given here). If none are 540then only these backends will be tried (in the reverse order as listed
236specified, most compiled-in backend will be tried, usually in reverse 541here). If none are specified, all backends in C<ev_recommended_backends
237order of their flag values :) 542()> will be tried.
238 543
239The most typical usage is like this: 544Example: This is the most typical usage.
240 545
241 if (!ev_default_loop (0)) 546 if (!ev_default_loop (0))
242 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?"); 547 fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
243 548
244Restrict libev to the select and poll backends, and do not allow 549Example: Restrict libev to the select and poll backends, and do not allow
245environment settings to be taken into account: 550environment settings to be taken into account:
246 551
247 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV); 552 ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
248 553
249Use whatever libev has to offer, but make sure that kqueue is used if 554Example: Use whatever libev has to offer, but make sure that kqueue is
250available (warning, breaks stuff, best use only with your own private 555used if available (warning, breaks stuff, best use only with your own
251event loop and only if you know the OS supports your types of fds): 556private event loop and only if you know the OS supports your types of
557fds):
252 558
253 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE); 559 ev_default_loop (ev_recommended_backends () | EVBACKEND_KQUEUE);
254 560
255=item struct ev_loop *ev_loop_new (unsigned int flags) 561=item struct ev_loop *ev_loop_new (unsigned int flags)
256 562
257Similar to C<ev_default_loop>, but always creates a new event loop that is 563Similar to C<ev_default_loop>, but always creates a new event loop that is
258always distinct from the default loop. Unlike the default loop, it cannot 564always distinct from the default loop. Unlike the default loop, it cannot
259handle signal and child watchers, and attempts to do so will be greeted by 565handle signal and child watchers, and attempts to do so will be greeted by
260undefined behaviour (or a failed assertion if assertions are enabled). 566undefined behaviour (or a failed assertion if assertions are enabled).
261 567
568Note that this function I<is> thread-safe, and the recommended way to use
569libev with threads is indeed to create one loop per thread, and using the
570default loop in the "main" or "initial" thread.
571
572Example: Try to create a event loop that uses epoll and nothing else.
573
574 struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
575 if (!epoller)
576 fatal ("no epoll found here, maybe it hides under your chair");
577
262=item ev_default_destroy () 578=item ev_default_destroy ()
263 579
264Destroys the default loop again (frees all memory and kernel state 580Destroys the default loop again (frees all memory and kernel state
265etc.). This stops all registered event watchers (by not touching them in 581etc.). None of the active event watchers will be stopped in the normal
266any way whatsoever, although you cannot rely on this :). 582sense, so e.g. C<ev_is_active> might still return true. It is your
583responsibility to either stop all watchers cleanly yourself I<before>
584calling this function, or cope with the fact afterwards (which is usually
585the easiest thing, you can just ignore the watchers and/or C<free ()> them
586for example).
587
588Note that certain global state, such as signal state (and installed signal
589handlers), will not be freed by this function, and related watchers (such
590as signal and child watchers) would need to be stopped manually.
591
592In general it is not advisable to call this function except in the
593rare occasion where you really need to free e.g. the signal handling
594pipe fds. If you need dynamically allocated loops it is better to use
595C<ev_loop_new> and C<ev_loop_destroy>).
267 596
268=item ev_loop_destroy (loop) 597=item ev_loop_destroy (loop)
269 598
270Like C<ev_default_destroy>, but destroys an event loop created by an 599Like C<ev_default_destroy>, but destroys an event loop created by an
271earlier call to C<ev_loop_new>. 600earlier call to C<ev_loop_new>.
272 601
273=item ev_default_fork () 602=item ev_default_fork ()
274 603
604This function sets a flag that causes subsequent C<ev_loop> iterations
275This function reinitialises the kernel state for backends that have 605to reinitialise the kernel state for backends that have one. Despite the
276one. Despite the name, you can call it anytime, but it makes most sense 606name, you can call it anytime, but it makes most sense after forking, in
277after forking, in either the parent or child process (or both, but that 607the child process (or both child and parent, but that again makes little
278again makes little sense). 608sense). You I<must> call it in the child before using any of the libev
609functions, and it will only take effect at the next C<ev_loop> iteration.
279 610
280You I<must> call this function in the child process after forking if and 611On the other hand, you only need to call this function in the child
281only if you want to use the event library in both processes. If you just 612process if and only if you want to use the event library in the child. If
282fork+exec, you don't have to call it. 613you just fork+exec, you don't have to call it at all.
283 614
284The function itself is quite fast and it's usually not a problem to call 615The function itself is quite fast and it's usually not a problem to call
285it just in case after a fork. To make this easy, the function will fit in 616it just in case after a fork. To make this easy, the function will fit in
286quite nicely into a call to C<pthread_atfork>: 617quite nicely into a call to C<pthread_atfork>:
287 618
288 pthread_atfork (0, 0, ev_default_fork); 619 pthread_atfork (0, 0, ev_default_fork);
289 620
290At the moment, C<EVBACKEND_SELECT> and C<EVBACKEND_POLL> are safe to use
291without calling this function, so if you force one of those backends you
292do not need to care.
293
294=item ev_loop_fork (loop) 621=item ev_loop_fork (loop)
295 622
296Like C<ev_default_fork>, but acts on an event loop created by 623Like C<ev_default_fork>, but acts on an event loop created by
297C<ev_loop_new>. Yes, you have to call this on every allocated event loop 624C<ev_loop_new>. Yes, you have to call this on every allocated event loop
298after fork, and how you do this is entirely your own problem. 625after fork that you want to re-use in the child, and how you do this is
626entirely your own problem.
627
628=item int ev_is_default_loop (loop)
629
630Returns true when the given loop is, in fact, the default loop, and false
631otherwise.
632
633=item unsigned int ev_loop_count (loop)
634
635Returns the count of loop iterations for the loop, which is identical to
636the number of times libev did poll for new events. It starts at C<0> and
637happily wraps around with enough iterations.
638
639This value can sometimes be useful as a generation counter of sorts (it
640"ticks" the number of loop iterations), as it roughly corresponds with
641C<ev_prepare> and C<ev_check> calls.
642
643=item unsigned int ev_loop_depth (loop)
644
645Returns the number of times C<ev_loop> was entered minus the number of
646times C<ev_loop> was exited, in other words, the recursion depth.
647
648Outside C<ev_loop>, this number is zero. In a callback, this number is
649C<1>, unless C<ev_loop> was invoked recursively (or from another thread),
650in which case it is higher.
651
652Leaving C<ev_loop> abnormally (setjmp/longjmp, cancelling the thread
653etc.), doesn't count as exit.
299 654
300=item unsigned int ev_backend (loop) 655=item unsigned int ev_backend (loop)
301 656
302Returns one of the C<EVBACKEND_*> flags indicating the event backend in 657Returns one of the C<EVBACKEND_*> flags indicating the event backend in
303use. 658use.
304 659
305=item ev_tstamp ev_now (loop) 660=item ev_tstamp ev_now (loop)
306 661
307Returns the current "event loop time", which is the time the event loop 662Returns the current "event loop time", which is the time the event loop
308got events and started processing them. This timestamp does not change 663received events and started processing them. This timestamp does not
309as long as callbacks are being processed, and this is also the base time 664change as long as callbacks are being processed, and this is also the base
310used for relative timers. You can treat it as the timestamp of the event 665time used for relative timers. You can treat it as the timestamp of the
311occuring (or more correctly, the mainloop finding out about it). 666event occurring (or more correctly, libev finding out about it).
667
668=item ev_now_update (loop)
669
670Establishes the current time by querying the kernel, updating the time
671returned by C<ev_now ()> in the progress. This is a costly operation and
672is usually done automatically within C<ev_loop ()>.
673
674This function is rarely useful, but when some event callback runs for a
675very long time without entering the event loop, updating libev's idea of
676the current time is a good idea.
677
678See also L<The special problem of time updates> in the C<ev_timer> section.
679
680=item ev_suspend (loop)
681
682=item ev_resume (loop)
683
684These two functions suspend and resume a loop, for use when the loop is
685not used for a while and timeouts should not be processed.
686
687A typical use case would be an interactive program such as a game: When
688the user presses C<^Z> to suspend the game and resumes it an hour later it
689would be best to handle timeouts as if no time had actually passed while
690the program was suspended. This can be achieved by calling C<ev_suspend>
691in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
692C<ev_resume> directly afterwards to resume timer processing.
693
694Effectively, all C<ev_timer> watchers will be delayed by the time spend
695between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
696will be rescheduled (that is, they will lose any events that would have
697occured while suspended).
698
699After calling C<ev_suspend> you B<must not> call I<any> function on the
700given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
701without a previous call to C<ev_suspend>.
702
703Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
704event loop time (see C<ev_now_update>).
312 705
313=item ev_loop (loop, int flags) 706=item ev_loop (loop, int flags)
314 707
315Finally, this is it, the event handler. This function usually is called 708Finally, this is it, the event handler. This function usually is called
316after you initialised all your watchers and you want to start handling 709after you initialised all your watchers and you want to start handling
317events. 710events.
318 711
319If the flags argument is specified as C<0>, it will not return until 712If the flags argument is specified as C<0>, it will not return until
320either no event watchers are active anymore or C<ev_unloop> was called. 713either no event watchers are active anymore or C<ev_unloop> was called.
321 714
715Please note that an explicit C<ev_unloop> is usually better than
716relying on all watchers to be stopped when deciding when a program has
717finished (especially in interactive programs), but having a program
718that automatically loops as long as it has to and no longer by virtue
719of relying on its watchers stopping correctly, that is truly a thing of
720beauty.
721
322A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle 722A flags value of C<EVLOOP_NONBLOCK> will look for new events, will handle
323those events and any outstanding ones, but will not block your process in 723those events and any already outstanding ones, but will not block your
324case there are no events and will return after one iteration of the loop. 724process in case there are no events and will return after one iteration of
725the loop.
325 726
326A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if 727A flags value of C<EVLOOP_ONESHOT> will look for new events (waiting if
327neccessary) and will handle those and any outstanding ones. It will block 728necessary) and will handle those and any already outstanding ones. It
328your process until at least one new event arrives, and will return after 729will block your process until at least one new event arrives (which could
329one iteration of the loop. This is useful if you are waiting for some 730be an event internal to libev itself, so there is no guarantee that a
330external event in conjunction with something not expressible using other 731user-registered callback will be called), and will return after one
732iteration of the loop.
733
734This is useful if you are waiting for some external event in conjunction
735with something not expressible using other libev watchers (i.e. "roll your
331libev watchers. However, a pair of C<ev_prepare>/C<ev_check> watchers is 736own C<ev_loop>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
332usually a better approach for this kind of thing. 737usually a better approach for this kind of thing.
333 738
334Here are the gory details of what C<ev_loop> does: 739Here are the gory details of what C<ev_loop> does:
335 740
336 * If there are no active watchers (reference count is zero), return. 741 - Before the first iteration, call any pending watchers.
337 - Queue prepare watchers and then call all outstanding watchers. 742 * If EVFLAG_FORKCHECK was used, check for a fork.
743 - If a fork was detected (by any means), queue and call all fork watchers.
744 - Queue and call all prepare watchers.
338 - If we have been forked, recreate the kernel state. 745 - If we have been forked, detach and recreate the kernel state
746 as to not disturb the other process.
339 - Update the kernel state with all outstanding changes. 747 - Update the kernel state with all outstanding changes.
340 - Update the "event loop time". 748 - Update the "event loop time" (ev_now ()).
341 - Calculate for how long to block. 749 - Calculate for how long to sleep or block, if at all
750 (active idle watchers, EVLOOP_NONBLOCK or not having
751 any active watchers at all will result in not sleeping).
752 - Sleep if the I/O and timer collect interval say so.
342 - Block the process, waiting for any events. 753 - Block the process, waiting for any events.
343 - Queue all outstanding I/O (fd) events. 754 - Queue all outstanding I/O (fd) events.
344 - Update the "event loop time" and do time jump handling. 755 - Update the "event loop time" (ev_now ()), and do time jump adjustments.
345 - Queue all outstanding timers. 756 - Queue all expired timers.
346 - Queue all outstanding periodics. 757 - Queue all expired periodics.
347 - If no events are pending now, queue all idle watchers. 758 - Unless any events are pending now, queue all idle watchers.
348 - Queue all check watchers. 759 - Queue all check watchers.
349 - Call all queued watchers in reverse order (i.e. check watchers first). 760 - Call all queued watchers in reverse order (i.e. check watchers first).
350 Signals and child watchers are implemented as I/O watchers, and will 761 Signals and child watchers are implemented as I/O watchers, and will
351 be handled here by queueing them when their watcher gets executed. 762 be handled here by queueing them when their watcher gets executed.
352 - If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK 763 - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
353 were used, return, otherwise continue with step *. 764 were used, or there are no active watchers, return, otherwise
765 continue with step *.
766
767Example: Queue some jobs and then loop until no events are outstanding
768anymore.
769
770 ... queue jobs here, make sure they register event watchers as long
771 ... as they still have work to do (even an idle watcher will do..)
772 ev_loop (my_loop, 0);
773 ... jobs done or somebody called unloop. yeah!
354 774
355=item ev_unloop (loop, how) 775=item ev_unloop (loop, how)
356 776
357Can be used to make a call to C<ev_loop> return early (but only after it 777Can be used to make a call to C<ev_loop> return early (but only after it
358has processed all outstanding events). The C<how> argument must be either 778has processed all outstanding events). The C<how> argument must be either
359C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or 779C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
360C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return. 780C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
361 781
782This "unloop state" will be cleared when entering C<ev_loop> again.
783
784It is safe to call C<ev_unloop> from otuside any C<ev_loop> calls.
785
362=item ev_ref (loop) 786=item ev_ref (loop)
363 787
364=item ev_unref (loop) 788=item ev_unref (loop)
365 789
366Ref/unref can be used to add or remove a reference count on the event 790Ref/unref can be used to add or remove a reference count on the event
367loop: Every watcher keeps one reference, and as long as the reference 791loop: Every watcher keeps one reference, and as long as the reference
368count is nonzero, C<ev_loop> will not return on its own. If you have 792count is nonzero, C<ev_loop> will not return on its own.
793
369a watcher you never unregister that should not keep C<ev_loop> from 794If you have a watcher you never unregister that should not keep C<ev_loop>
370returning, ev_unref() after starting, and ev_ref() before stopping it. For 795from returning, call ev_unref() after starting, and ev_ref() before
796stopping it.
797
371example, libev itself uses this for its internal signal pipe: It is not 798As an example, libev itself uses this for its internal signal pipe: It
372visible to the libev user and should not keep C<ev_loop> from exiting if 799is not visible to the libev user and should not keep C<ev_loop> from
373no event watchers registered by it are active. It is also an excellent 800exiting if no event watchers registered by it are active. It is also an
374way to do this for generic recurring timers or from within third-party 801excellent way to do this for generic recurring timers or from within
375libraries. Just remember to I<unref after start> and I<ref before stop>. 802third-party libraries. Just remember to I<unref after start> and I<ref
803before stop> (but only if the watcher wasn't active before, or was active
804before, respectively. Note also that libev might stop watchers itself
805(e.g. non-repeating timers) in which case you have to C<ev_ref>
806in the callback).
807
808Example: Create a signal watcher, but keep it from keeping C<ev_loop>
809running when nothing else is active.
810
811 ev_signal exitsig;
812 ev_signal_init (&exitsig, sig_cb, SIGINT);
813 ev_signal_start (loop, &exitsig);
814 evf_unref (loop);
815
816Example: For some weird reason, unregister the above signal handler again.
817
818 ev_ref (loop);
819 ev_signal_stop (loop, &exitsig);
820
821=item ev_set_io_collect_interval (loop, ev_tstamp interval)
822
823=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
824
825These advanced functions influence the time that libev will spend waiting
826for events. Both time intervals are by default C<0>, meaning that libev
827will try to invoke timer/periodic callbacks and I/O callbacks with minimum
828latency.
829
830Setting these to a higher value (the C<interval> I<must> be >= C<0>)
831allows libev to delay invocation of I/O and timer/periodic callbacks
832to increase efficiency of loop iterations (or to increase power-saving
833opportunities).
834
835The idea is that sometimes your program runs just fast enough to handle
836one (or very few) event(s) per loop iteration. While this makes the
837program responsive, it also wastes a lot of CPU time to poll for new
838events, especially with backends like C<select ()> which have a high
839overhead for the actual polling but can deliver many events at once.
840
841By setting a higher I<io collect interval> you allow libev to spend more
842time collecting I/O events, so you can handle more events per iteration,
843at the cost of increasing latency. Timeouts (both C<ev_periodic> and
844C<ev_timer>) will be not affected. Setting this to a non-null value will
845introduce an additional C<ev_sleep ()> call into most loop iterations. The
846sleep time ensures that libev will not poll for I/O events more often then
847once per this interval, on average.
848
849Likewise, by setting a higher I<timeout collect interval> you allow libev
850to spend more time collecting timeouts, at the expense of increased
851latency/jitter/inexactness (the watcher callback will be called
852later). C<ev_io> watchers will not be affected. Setting this to a non-null
853value will not introduce any overhead in libev.
854
855Many (busy) programs can usually benefit by setting the I/O collect
856interval to a value near C<0.1> or so, which is often enough for
857interactive servers (of course not for games), likewise for timeouts. It
858usually doesn't make much sense to set it to a lower value than C<0.01>,
859as this approaches the timing granularity of most systems. Note that if
860you do transactions with the outside world and you can't increase the
861parallelity, then this setting will limit your transaction rate (if you
862need to poll once per transaction and the I/O collect interval is 0.01,
863then you can't do more than 100 transations per second).
864
865Setting the I<timeout collect interval> can improve the opportunity for
866saving power, as the program will "bundle" timer callback invocations that
867are "near" in time together, by delaying some, thus reducing the number of
868times the process sleeps and wakes up again. Another useful technique to
869reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
870they fire on, say, one-second boundaries only.
871
872Example: we only need 0.1s timeout granularity, and we wish not to poll
873more often than 100 times per second:
874
875 ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
876 ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
877
878=item ev_invoke_pending (loop)
879
880This call will simply invoke all pending watchers while resetting their
881pending state. Normally, C<ev_loop> does this automatically when required,
882but when overriding the invoke callback this call comes handy.
883
884=item int ev_pending_count (loop)
885
886Returns the number of pending watchers - zero indicates that no watchers
887are pending.
888
889=item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
890
891This overrides the invoke pending functionality of the loop: Instead of
892invoking all pending watchers when there are any, C<ev_loop> will call
893this callback instead. This is useful, for example, when you want to
894invoke the actual watchers inside another context (another thread etc.).
895
896If you want to reset the callback, use C<ev_invoke_pending> as new
897callback.
898
899=item ev_set_loop_release_cb (loop, void (*release)(EV_P), void (*acquire)(EV_P))
900
901Sometimes you want to share the same loop between multiple threads. This
902can be done relatively simply by putting mutex_lock/unlock calls around
903each call to a libev function.
904
905However, C<ev_loop> can run an indefinite time, so it is not feasible to
906wait for it to return. One way around this is to wake up the loop via
907C<ev_unloop> and C<av_async_send>, another way is to set these I<release>
908and I<acquire> callbacks on the loop.
909
910When set, then C<release> will be called just before the thread is
911suspended waiting for new events, and C<acquire> is called just
912afterwards.
913
914Ideally, C<release> will just call your mutex_unlock function, and
915C<acquire> will just call the mutex_lock function again.
916
917While event loop modifications are allowed between invocations of
918C<release> and C<acquire> (that's their only purpose after all), no
919modifications done will affect the event loop, i.e. adding watchers will
920have no effect on the set of file descriptors being watched, or the time
921waited. USe an C<ev_async> watcher to wake up C<ev_loop> when you want it
922to take note of any changes you made.
923
924In theory, threads executing C<ev_loop> will be async-cancel safe between
925invocations of C<release> and C<acquire>.
926
927See also the locking example in the C<THREADS> section later in this
928document.
929
930=item ev_set_userdata (loop, void *data)
931
932=item ev_userdata (loop)
933
934Set and retrieve a single C<void *> associated with a loop. When
935C<ev_set_userdata> has never been called, then C<ev_userdata> returns
936C<0.>
937
938These two functions can be used to associate arbitrary data with a loop,
939and are intended solely for the C<invoke_pending_cb>, C<release> and
940C<acquire> callbacks described above, but of course can be (ab-)used for
941any other purpose as well.
942
943=item ev_loop_verify (loop)
944
945This function only does something when C<EV_VERIFY> support has been
946compiled in, which is the default for non-minimal builds. It tries to go
947through all internal structures and checks them for validity. If anything
948is found to be inconsistent, it will print an error message to standard
949error and call C<abort ()>.
950
951This can be used to catch bugs inside libev itself: under normal
952circumstances, this function will never abort as of course libev keeps its
953data structures consistent.
376 954
377=back 955=back
378 956
957
379=head1 ANATOMY OF A WATCHER 958=head1 ANATOMY OF A WATCHER
959
960In the following description, uppercase C<TYPE> in names stands for the
961watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
962watchers and C<ev_io_start> for I/O watchers.
380 963
381A watcher is a structure that you create and register to record your 964A watcher is a structure that you create and register to record your
382interest in some event. For instance, if you want to wait for STDIN to 965interest in some event. For instance, if you want to wait for STDIN to
383become readable, you would create an C<ev_io> watcher for that: 966become readable, you would create an C<ev_io> watcher for that:
384 967
385 static void my_cb (struct ev_loop *loop, struct ev_io *w, int revents) 968 static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
386 { 969 {
387 ev_io_stop (w); 970 ev_io_stop (w);
388 ev_unloop (loop, EVUNLOOP_ALL); 971 ev_unloop (loop, EVUNLOOP_ALL);
389 } 972 }
390 973
391 struct ev_loop *loop = ev_default_loop (0); 974 struct ev_loop *loop = ev_default_loop (0);
975
392 struct ev_io stdin_watcher; 976 ev_io stdin_watcher;
977
393 ev_init (&stdin_watcher, my_cb); 978 ev_init (&stdin_watcher, my_cb);
394 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ); 979 ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
395 ev_io_start (loop, &stdin_watcher); 980 ev_io_start (loop, &stdin_watcher);
981
396 ev_loop (loop, 0); 982 ev_loop (loop, 0);
397 983
398As you can see, you are responsible for allocating the memory for your 984As you can see, you are responsible for allocating the memory for your
399watcher structures (and it is usually a bad idea to do this on the stack, 985watcher structures (and it is I<usually> a bad idea to do this on the
400although this can sometimes be quite valid). 986stack).
987
988Each watcher has an associated watcher structure (called C<struct ev_TYPE>
989or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
401 990
402Each watcher structure must be initialised by a call to C<ev_init 991Each watcher structure must be initialised by a call to C<ev_init
403(watcher *, callback)>, which expects a callback to be provided. This 992(watcher *, callback)>, which expects a callback to be provided. This
404callback gets invoked each time the event occurs (or, in the case of io 993callback gets invoked each time the event occurs (or, in the case of I/O
405watchers, each time the event loop detects that the file descriptor given 994watchers, each time the event loop detects that the file descriptor given
406is readable and/or writable). 995is readable and/or writable).
407 996
408Each watcher type has its own C<< ev_<type>_set (watcher *, ...) >> macro 997Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
409with arguments specific to this watcher type. There is also a macro 998macro to configure it, with arguments specific to the watcher type. There
410to combine initialisation and setting in one call: C<< ev_<type>_init 999is also a macro to combine initialisation and setting in one call: C<<
411(watcher *, callback, ...) >>. 1000ev_TYPE_init (watcher *, callback, ...) >>.
412 1001
413To make the watcher actually watch out for events, you have to start it 1002To make the watcher actually watch out for events, you have to start it
414with a watcher-specific start function (C<< ev_<type>_start (loop, watcher 1003with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
415*) >>), and you can stop watching for events at any time by calling the 1004*) >>), and you can stop watching for events at any time by calling the
416corresponding stop function (C<< ev_<type>_stop (loop, watcher *) >>. 1005corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
417 1006
418As long as your watcher is active (has been started but not stopped) you 1007As long as your watcher is active (has been started but not stopped) you
419must not touch the values stored in it. Most specifically you must never 1008must not touch the values stored in it. Most specifically you must never
420reinitialise it or call its set macro. 1009reinitialise it or call its C<ev_TYPE_set> macro.
421
422You can check whether an event is active by calling the C<ev_is_active
423(watcher *)> macro. To see whether an event is outstanding (but the
424callback for it has not been called yet) you can use the C<ev_is_pending
425(watcher *)> macro.
426 1010
427Each and every callback receives the event loop pointer as first, the 1011Each and every callback receives the event loop pointer as first, the
428registered watcher structure as second, and a bitset of received events as 1012registered watcher structure as second, and a bitset of received events as
429third argument. 1013third argument.
430 1014
454The signal specified in the C<ev_signal> watcher has been received by a thread. 1038The signal specified in the C<ev_signal> watcher has been received by a thread.
455 1039
456=item C<EV_CHILD> 1040=item C<EV_CHILD>
457 1041
458The pid specified in the C<ev_child> watcher has received a status change. 1042The pid specified in the C<ev_child> watcher has received a status change.
1043
1044=item C<EV_STAT>
1045
1046The path specified in the C<ev_stat> watcher changed its attributes somehow.
459 1047
460=item C<EV_IDLE> 1048=item C<EV_IDLE>
461 1049
462The C<ev_idle> watcher has determined that you have nothing better to do. 1050The C<ev_idle> watcher has determined that you have nothing better to do.
463 1051
471received events. Callbacks of both watcher types can start and stop as 1059received events. Callbacks of both watcher types can start and stop as
472many watchers as they want, and all of them will be taken into account 1060many watchers as they want, and all of them will be taken into account
473(for example, a C<ev_prepare> watcher might start an idle watcher to keep 1061(for example, a C<ev_prepare> watcher might start an idle watcher to keep
474C<ev_loop> from blocking). 1062C<ev_loop> from blocking).
475 1063
1064=item C<EV_EMBED>
1065
1066The embedded event loop specified in the C<ev_embed> watcher needs attention.
1067
1068=item C<EV_FORK>
1069
1070The event loop has been resumed in the child process after fork (see
1071C<ev_fork>).
1072
1073=item C<EV_ASYNC>
1074
1075The given async watcher has been asynchronously notified (see C<ev_async>).
1076
1077=item C<EV_CUSTOM>
1078
1079Not ever sent (or otherwise used) by libev itself, but can be freely used
1080by libev users to signal watchers (e.g. via C<ev_feed_event>).
1081
476=item C<EV_ERROR> 1082=item C<EV_ERROR>
477 1083
478An unspecified error has occured, the watcher has been stopped. This might 1084An unspecified error has occurred, the watcher has been stopped. This might
479happen because the watcher could not be properly started because libev 1085happen because the watcher could not be properly started because libev
480ran out of memory, a file descriptor was found to be closed or any other 1086ran out of memory, a file descriptor was found to be closed or any other
1087problem. Libev considers these application bugs.
1088
481problem. You best act on it by reporting the problem and somehow coping 1089You best act on it by reporting the problem and somehow coping with the
482with the watcher being stopped. 1090watcher being stopped. Note that well-written programs should not receive
1091an error ever, so when your watcher receives it, this usually indicates a
1092bug in your program.
483 1093
484Libev will usually signal a few "dummy" events together with an error, 1094Libev will usually signal a few "dummy" events together with an error, for
485for example it might indicate that a fd is readable or writable, and if 1095example it might indicate that a fd is readable or writable, and if your
486your callbacks is well-written it can just attempt the operation and cope 1096callbacks is well-written it can just attempt the operation and cope with
487with the error from read() or write(). This will not work in multithreaded 1097the error from read() or write(). This will not work in multi-threaded
488programs, though, so beware. 1098programs, though, as the fd could already be closed and reused for another
1099thing, so beware.
489 1100
490=back 1101=back
491 1102
1103=head2 GENERIC WATCHER FUNCTIONS
1104
1105=over 4
1106
1107=item C<ev_init> (ev_TYPE *watcher, callback)
1108
1109This macro initialises the generic portion of a watcher. The contents
1110of the watcher object can be arbitrary (so C<malloc> will do). Only
1111the generic parts of the watcher are initialised, you I<need> to call
1112the type-specific C<ev_TYPE_set> macro afterwards to initialise the
1113type-specific parts. For each type there is also a C<ev_TYPE_init> macro
1114which rolls both calls into one.
1115
1116You can reinitialise a watcher at any time as long as it has been stopped
1117(or never started) and there are no pending events outstanding.
1118
1119The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
1120int revents)>.
1121
1122Example: Initialise an C<ev_io> watcher in two steps.
1123
1124 ev_io w;
1125 ev_init (&w, my_cb);
1126 ev_io_set (&w, STDIN_FILENO, EV_READ);
1127
1128=item C<ev_TYPE_set> (ev_TYPE *, [args])
1129
1130This macro initialises the type-specific parts of a watcher. You need to
1131call C<ev_init> at least once before you call this macro, but you can
1132call C<ev_TYPE_set> any number of times. You must not, however, call this
1133macro on a watcher that is active (it can be pending, however, which is a
1134difference to the C<ev_init> macro).
1135
1136Although some watcher types do not have type-specific arguments
1137(e.g. C<ev_prepare>) you still need to call its C<set> macro.
1138
1139See C<ev_init>, above, for an example.
1140
1141=item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
1142
1143This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
1144calls into a single call. This is the most convenient method to initialise
1145a watcher. The same limitations apply, of course.
1146
1147Example: Initialise and set an C<ev_io> watcher in one step.
1148
1149 ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1150
1151=item C<ev_TYPE_start> (loop *, ev_TYPE *watcher)
1152
1153Starts (activates) the given watcher. Only active watchers will receive
1154events. If the watcher is already active nothing will happen.
1155
1156Example: Start the C<ev_io> watcher that is being abused as example in this
1157whole section.
1158
1159 ev_io_start (EV_DEFAULT_UC, &w);
1160
1161=item C<ev_TYPE_stop> (loop *, ev_TYPE *watcher)
1162
1163Stops the given watcher if active, and clears the pending status (whether
1164the watcher was active or not).
1165
1166It is possible that stopped watchers are pending - for example,
1167non-repeating timers are being stopped when they become pending - but
1168calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
1169pending. If you want to free or reuse the memory used by the watcher it is
1170therefore a good idea to always call its C<ev_TYPE_stop> function.
1171
1172=item bool ev_is_active (ev_TYPE *watcher)
1173
1174Returns a true value iff the watcher is active (i.e. it has been started
1175and not yet been stopped). As long as a watcher is active you must not modify
1176it.
1177
1178=item bool ev_is_pending (ev_TYPE *watcher)
1179
1180Returns a true value iff the watcher is pending, (i.e. it has outstanding
1181events but its callback has not yet been invoked). As long as a watcher
1182is pending (but not active) you must not call an init function on it (but
1183C<ev_TYPE_set> is safe), you must not change its priority, and you must
1184make sure the watcher is available to libev (e.g. you cannot C<free ()>
1185it).
1186
1187=item callback ev_cb (ev_TYPE *watcher)
1188
1189Returns the callback currently set on the watcher.
1190
1191=item ev_cb_set (ev_TYPE *watcher, callback)
1192
1193Change the callback. You can change the callback at virtually any time
1194(modulo threads).
1195
1196=item ev_set_priority (ev_TYPE *watcher, priority)
1197
1198=item int ev_priority (ev_TYPE *watcher)
1199
1200Set and query the priority of the watcher. The priority is a small
1201integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
1202(default: C<-2>). Pending watchers with higher priority will be invoked
1203before watchers with lower priority, but priority will not keep watchers
1204from being executed (except for C<ev_idle> watchers).
1205
1206If you need to suppress invocation when higher priority events are pending
1207you need to look at C<ev_idle> watchers, which provide this functionality.
1208
1209You I<must not> change the priority of a watcher as long as it is active or
1210pending.
1211
1212Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
1213fine, as long as you do not mind that the priority value you query might
1214or might not have been clamped to the valid range.
1215
1216The default priority used by watchers when no priority has been set is
1217always C<0>, which is supposed to not be too high and not be too low :).
1218
1219See L<WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
1220priorities.
1221
1222=item ev_invoke (loop, ev_TYPE *watcher, int revents)
1223
1224Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
1225C<loop> nor C<revents> need to be valid as long as the watcher callback
1226can deal with that fact, as both are simply passed through to the
1227callback.
1228
1229=item int ev_clear_pending (loop, ev_TYPE *watcher)
1230
1231If the watcher is pending, this function clears its pending status and
1232returns its C<revents> bitset (as if its callback was invoked). If the
1233watcher isn't pending it does nothing and returns C<0>.
1234
1235Sometimes it can be useful to "poll" a watcher instead of waiting for its
1236callback to be invoked, which can be accomplished with this function.
1237
1238=back
1239
1240
492=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER 1241=head2 ASSOCIATING CUSTOM DATA WITH A WATCHER
493 1242
494Each watcher has, by default, a member C<void *data> that you can change 1243Each watcher has, by default, a member C<void *data> that you can change
495and read at any time, libev will completely ignore it. This can be used 1244and read at any time: libev will completely ignore it. This can be used
496to associate arbitrary data with your watcher. If you need more data and 1245to associate arbitrary data with your watcher. If you need more data and
497don't want to allocate memory and store a pointer to it in that data 1246don't want to allocate memory and store a pointer to it in that data
498member, you can also "subclass" the watcher type and provide your own 1247member, you can also "subclass" the watcher type and provide your own
499data: 1248data:
500 1249
501 struct my_io 1250 struct my_io
502 { 1251 {
503 struct ev_io io; 1252 ev_io io;
504 int otherfd; 1253 int otherfd;
505 void *somedata; 1254 void *somedata;
506 struct whatever *mostinteresting; 1255 struct whatever *mostinteresting;
507 } 1256 };
1257
1258 ...
1259 struct my_io w;
1260 ev_io_init (&w.io, my_cb, fd, EV_READ);
508 1261
509And since your callback will be called with a pointer to the watcher, you 1262And since your callback will be called with a pointer to the watcher, you
510can cast it back to your own type: 1263can cast it back to your own type:
511 1264
512 static void my_cb (struct ev_loop *loop, struct ev_io *w_, int revents) 1265 static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
513 { 1266 {
514 struct my_io *w = (struct my_io *)w_; 1267 struct my_io *w = (struct my_io *)w_;
515 ... 1268 ...
516 } 1269 }
517 1270
518More interesting and less C-conformant ways of catsing your callback type 1271More interesting and less C-conformant ways of casting your callback type
519have been omitted.... 1272instead have been omitted.
1273
1274Another common scenario is to use some data structure with multiple
1275embedded watchers:
1276
1277 struct my_biggy
1278 {
1279 int some_data;
1280 ev_timer t1;
1281 ev_timer t2;
1282 }
1283
1284In this case getting the pointer to C<my_biggy> is a bit more
1285complicated: Either you store the address of your C<my_biggy> struct
1286in the C<data> member of the watcher (for woozies), or you need to use
1287some pointer arithmetic using C<offsetof> inside your watchers (for real
1288programmers):
1289
1290 #include <stddef.h>
1291
1292 static void
1293 t1_cb (EV_P_ ev_timer *w, int revents)
1294 {
1295 struct my_biggy big = (struct my_biggy *)
1296 (((char *)w) - offsetof (struct my_biggy, t1));
1297 }
1298
1299 static void
1300 t2_cb (EV_P_ ev_timer *w, int revents)
1301 {
1302 struct my_biggy big = (struct my_biggy *)
1303 (((char *)w) - offsetof (struct my_biggy, t2));
1304 }
1305
1306=head2 WATCHER PRIORITY MODELS
1307
1308Many event loops support I<watcher priorities>, which are usually small
1309integers that influence the ordering of event callback invocation
1310between watchers in some way, all else being equal.
1311
1312In libev, Watcher priorities can be set using C<ev_set_priority>. See its
1313description for the more technical details such as the actual priority
1314range.
1315
1316There are two common ways how these these priorities are being interpreted
1317by event loops:
1318
1319In the more common lock-out model, higher priorities "lock out" invocation
1320of lower priority watchers, which means as long as higher priority
1321watchers receive events, lower priority watchers are not being invoked.
1322
1323The less common only-for-ordering model uses priorities solely to order
1324callback invocation within a single event loop iteration: Higher priority
1325watchers are invoked before lower priority ones, but they all get invoked
1326before polling for new events.
1327
1328Libev uses the second (only-for-ordering) model for all its watchers
1329except for idle watchers (which use the lock-out model).
1330
1331The rationale behind this is that implementing the lock-out model for
1332watchers is not well supported by most kernel interfaces, and most event
1333libraries will just poll for the same events again and again as long as
1334their callbacks have not been executed, which is very inefficient in the
1335common case of one high-priority watcher locking out a mass of lower
1336priority ones.
1337
1338Static (ordering) priorities are most useful when you have two or more
1339watchers handling the same resource: a typical usage example is having an
1340C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
1341timeouts. Under load, data might be received while the program handles
1342other jobs, but since timers normally get invoked first, the timeout
1343handler will be executed before checking for data. In that case, giving
1344the timer a lower priority than the I/O watcher ensures that I/O will be
1345handled first even under adverse conditions (which is usually, but not
1346always, what you want).
1347
1348Since idle watchers use the "lock-out" model, meaning that idle watchers
1349will only be executed when no same or higher priority watchers have
1350received events, they can be used to implement the "lock-out" model when
1351required.
1352
1353For example, to emulate how many other event libraries handle priorities,
1354you can associate an C<ev_idle> watcher to each such watcher, and in
1355the normal watcher callback, you just start the idle watcher. The real
1356processing is done in the idle watcher callback. This causes libev to
1357continously poll and process kernel event data for the watcher, but when
1358the lock-out case is known to be rare (which in turn is rare :), this is
1359workable.
1360
1361Usually, however, the lock-out model implemented that way will perform
1362miserably under the type of load it was designed to handle. In that case,
1363it might be preferable to stop the real watcher before starting the
1364idle watcher, so the kernel will not have to process the event in case
1365the actual processing will be delayed for considerable time.
1366
1367Here is an example of an I/O watcher that should run at a strictly lower
1368priority than the default, and which should only process data when no
1369other events are pending:
1370
1371 ev_idle idle; // actual processing watcher
1372 ev_io io; // actual event watcher
1373
1374 static void
1375 io_cb (EV_P_ ev_io *w, int revents)
1376 {
1377 // stop the I/O watcher, we received the event, but
1378 // are not yet ready to handle it.
1379 ev_io_stop (EV_A_ w);
1380
1381 // start the idle watcher to ahndle the actual event.
1382 // it will not be executed as long as other watchers
1383 // with the default priority are receiving events.
1384 ev_idle_start (EV_A_ &idle);
1385 }
1386
1387 static void
1388 idle_cb (EV_P_ ev_idle *w, int revents)
1389 {
1390 // actual processing
1391 read (STDIN_FILENO, ...);
1392
1393 // have to start the I/O watcher again, as
1394 // we have handled the event
1395 ev_io_start (EV_P_ &io);
1396 }
1397
1398 // initialisation
1399 ev_idle_init (&idle, idle_cb);
1400 ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1401 ev_io_start (EV_DEFAULT_ &io);
1402
1403In the "real" world, it might also be beneficial to start a timer, so that
1404low-priority connections can not be locked out forever under load. This
1405enables your program to keep a lower latency for important connections
1406during short periods of high load, while not completely locking out less
1407important ones.
520 1408
521 1409
522=head1 WATCHER TYPES 1410=head1 WATCHER TYPES
523 1411
524This section describes each watcher in detail, but will not repeat 1412This section describes each watcher in detail, but will not repeat
525information given in the last section. 1413information given in the last section. Any initialisation/set macros,
1414functions and members specific to the watcher type are explained.
526 1415
1416Members are additionally marked with either I<[read-only]>, meaning that,
1417while the watcher is active, you can look at the member and expect some
1418sensible content, but you must not modify it (you can modify it while the
1419watcher is stopped to your hearts content), or I<[read-write]>, which
1420means you can expect it to have some sensible content while the watcher
1421is active, but you can also modify it. Modifying it may not do something
1422sensible or take immediate effect (or do anything at all), but libev will
1423not crash or malfunction in any way.
1424
1425
527=head2 C<ev_io> - is this file descriptor readable or writable 1426=head2 C<ev_io> - is this file descriptor readable or writable?
528 1427
529I/O watchers check whether a file descriptor is readable or writable 1428I/O watchers check whether a file descriptor is readable or writable
530in each iteration of the event loop (This behaviour is called 1429in each iteration of the event loop, or, more precisely, when reading
531level-triggering because you keep receiving events as long as the 1430would not block the process and writing would at least be able to write
532condition persists. Remember you can stop the watcher if you don't want to 1431some data. This behaviour is called level-triggering because you keep
533act on the event and neither want to receive future events). 1432receiving events as long as the condition persists. Remember you can stop
1433the watcher if you don't want to act on the event and neither want to
1434receive future events.
534 1435
535In general you can register as many read and/or write event watchers per 1436In general you can register as many read and/or write event watchers per
536fd as you want (as long as you don't confuse yourself). Setting all file 1437fd as you want (as long as you don't confuse yourself). Setting all file
537descriptors to non-blocking mode is also usually a good idea (but not 1438descriptors to non-blocking mode is also usually a good idea (but not
538required if you know what you are doing). 1439required if you know what you are doing).
539 1440
540You have to be careful with dup'ed file descriptors, though. Some backends 1441If you cannot use non-blocking mode, then force the use of a
541(the linux epoll backend is a notable example) cannot handle dup'ed file 1442known-to-be-good backend (at the time of this writing, this includes only
542descriptors correctly if you register interest in two or more fds pointing 1443C<EVBACKEND_SELECT> and C<EVBACKEND_POLL>). The same applies to file
543to the same underlying file/socket etc. description (that is, they share 1444descriptors for which non-blocking operation makes no sense (such as
544the same underlying "file open"). 1445files) - libev doesn't guarentee any specific behaviour in that case.
545 1446
546If you must do this, then force the use of a known-to-be-good backend 1447Another thing you have to watch out for is that it is quite easy to
547(at the time of this writing, this includes only C<EVBACKEND_SELECT> and 1448receive "spurious" readiness notifications, that is your callback might
1449be called with C<EV_READ> but a subsequent C<read>(2) will actually block
1450because there is no data. Not only are some backends known to create a
1451lot of those (for example Solaris ports), it is very easy to get into
1452this situation even with a relatively standard program structure. Thus
1453it is best to always use non-blocking I/O: An extra C<read>(2) returning
1454C<EAGAIN> is far preferable to a program hanging until some data arrives.
1455
1456If you cannot run the fd in non-blocking mode (for example you should
1457not play around with an Xlib connection), then you have to separately
1458re-test whether a file descriptor is really ready with a known-to-be good
1459interface such as poll (fortunately in our Xlib example, Xlib already
1460does this on its own, so its quite safe to use). Some people additionally
1461use C<SIGALRM> and an interval timer, just to be sure you won't block
1462indefinitely.
1463
1464But really, best use non-blocking mode.
1465
1466=head3 The special problem of disappearing file descriptors
1467
1468Some backends (e.g. kqueue, epoll) need to be told about closing a file
1469descriptor (either due to calling C<close> explicitly or any other means,
1470such as C<dup2>). The reason is that you register interest in some file
1471descriptor, but when it goes away, the operating system will silently drop
1472this interest. If another file descriptor with the same number then is
1473registered with libev, there is no efficient way to see that this is, in
1474fact, a different file descriptor.
1475
1476To avoid having to explicitly tell libev about such cases, libev follows
1477the following policy: Each time C<ev_io_set> is being called, libev
1478will assume that this is potentially a new file descriptor, otherwise
1479it is assumed that the file descriptor stays the same. That means that
1480you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
1481descriptor even if the file descriptor number itself did not change.
1482
1483This is how one would do it normally anyway, the important point is that
1484the libev application should not optimise around libev but should leave
1485optimisations to libev.
1486
1487=head3 The special problem of dup'ed file descriptors
1488
1489Some backends (e.g. epoll), cannot register events for file descriptors,
1490but only events for the underlying file descriptions. That means when you
1491have C<dup ()>'ed file descriptors or weirder constellations, and register
1492events for them, only one file descriptor might actually receive events.
1493
1494There is no workaround possible except not registering events
1495for potentially C<dup ()>'ed file descriptors, or to resort to
1496C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
1497
1498=head3 The special problem of fork
1499
1500Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
1501useless behaviour. Libev fully supports fork, but needs to be told about
1502it in the child.
1503
1504To support fork in your programs, you either have to call
1505C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
1506enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
548C<EVBACKEND_POLL>). 1507C<EVBACKEND_POLL>.
1508
1509=head3 The special problem of SIGPIPE
1510
1511While not really specific to libev, it is easy to forget about C<SIGPIPE>:
1512when writing to a pipe whose other end has been closed, your program gets
1513sent a SIGPIPE, which, by default, aborts your program. For most programs
1514this is sensible behaviour, for daemons, this is usually undesirable.
1515
1516So when you encounter spurious, unexplained daemon exits, make sure you
1517ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
1518somewhere, as that would have given you a big clue).
1519
1520
1521=head3 Watcher-Specific Functions
549 1522
550=over 4 1523=over 4
551 1524
552=item ev_io_init (ev_io *, callback, int fd, int events) 1525=item ev_io_init (ev_io *, callback, int fd, int events)
553 1526
554=item ev_io_set (ev_io *, int fd, int events) 1527=item ev_io_set (ev_io *, int fd, int events)
555 1528
556Configures an C<ev_io> watcher. The fd is the file descriptor to rceeive 1529Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
557events for and events is either C<EV_READ>, C<EV_WRITE> or C<EV_READ | 1530receive events for and C<events> is either C<EV_READ>, C<EV_WRITE> or
558EV_WRITE> to receive the given events. 1531C<EV_READ | EV_WRITE>, to express the desire to receive the given events.
559 1532
560Please note that most of the more scalable backend mechanisms (for example 1533=item int fd [read-only]
561epoll and solaris ports) can result in spurious readyness notifications 1534
562for file descriptors, so you practically need to use non-blocking I/O (and 1535The file descriptor being watched.
563treat callback invocation as hint only), or retest separately with a safe 1536
564interface before doing I/O (XLib can do this), or force the use of either 1537=item int events [read-only]
565C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>, which don't suffer from this 1538
566problem. Also note that it is quite easy to have your callback invoked 1539The events being watched.
567when the readyness condition is no longer valid even when employing
568typical ways of handling events, so its a good idea to use non-blocking
569I/O unconditionally.
570 1540
571=back 1541=back
572 1542
1543=head3 Examples
1544
1545Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
1546readable, but only once. Since it is likely line-buffered, you could
1547attempt to read a whole line in the callback.
1548
1549 static void
1550 stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1551 {
1552 ev_io_stop (loop, w);
1553 .. read from stdin here (or from w->fd) and handle any I/O errors
1554 }
1555
1556 ...
1557 struct ev_loop *loop = ev_default_init (0);
1558 ev_io stdin_readable;
1559 ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1560 ev_io_start (loop, &stdin_readable);
1561 ev_loop (loop, 0);
1562
1563
573=head2 C<ev_timer> - relative and optionally recurring timeouts 1564=head2 C<ev_timer> - relative and optionally repeating timeouts
574 1565
575Timer watchers are simple relative timers that generate an event after a 1566Timer watchers are simple relative timers that generate an event after a
576given time, and optionally repeating in regular intervals after that. 1567given time, and optionally repeating in regular intervals after that.
577 1568
578The timers are based on real time, that is, if you register an event that 1569The timers are based on real time, that is, if you register an event that
579times out after an hour and you reset your system clock to last years 1570times out after an hour and you reset your system clock to January last
580time, it will still time out after (roughly) and hour. "Roughly" because 1571year, it will still time out after (roughly) one hour. "Roughly" because
581detecting time jumps is hard, and some inaccuracies are unavoidable (the 1572detecting time jumps is hard, and some inaccuracies are unavoidable (the
582monotonic clock option helps a lot here). 1573monotonic clock option helps a lot here).
1574
1575The callback is guaranteed to be invoked only I<after> its timeout has
1576passed (not I<at>, so on systems with very low-resolution clocks this
1577might introduce a small delay). If multiple timers become ready during the
1578same loop iteration then the ones with earlier time-out values are invoked
1579before ones of the same priority with later time-out values (but this is
1580no longer true when a callback calls C<ev_loop> recursively).
1581
1582=head3 Be smart about timeouts
1583
1584Many real-world problems involve some kind of timeout, usually for error
1585recovery. A typical example is an HTTP request - if the other side hangs,
1586you want to raise some error after a while.
1587
1588What follows are some ways to handle this problem, from obvious and
1589inefficient to smart and efficient.
1590
1591In the following, a 60 second activity timeout is assumed - a timeout that
1592gets reset to 60 seconds each time there is activity (e.g. each time some
1593data or other life sign was received).
1594
1595=over 4
1596
1597=item 1. Use a timer and stop, reinitialise and start it on activity.
1598
1599This is the most obvious, but not the most simple way: In the beginning,
1600start the watcher:
1601
1602 ev_timer_init (timer, callback, 60., 0.);
1603 ev_timer_start (loop, timer);
1604
1605Then, each time there is some activity, C<ev_timer_stop> it, initialise it
1606and start it again:
1607
1608 ev_timer_stop (loop, timer);
1609 ev_timer_set (timer, 60., 0.);
1610 ev_timer_start (loop, timer);
1611
1612This is relatively simple to implement, but means that each time there is
1613some activity, libev will first have to remove the timer from its internal
1614data structure and then add it again. Libev tries to be fast, but it's
1615still not a constant-time operation.
1616
1617=item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
1618
1619This is the easiest way, and involves using C<ev_timer_again> instead of
1620C<ev_timer_start>.
1621
1622To implement this, configure an C<ev_timer> with a C<repeat> value
1623of C<60> and then call C<ev_timer_again> at start and each time you
1624successfully read or write some data. If you go into an idle state where
1625you do not expect data to travel on the socket, you can C<ev_timer_stop>
1626the timer, and C<ev_timer_again> will automatically restart it if need be.
1627
1628That means you can ignore both the C<ev_timer_start> function and the
1629C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
1630member and C<ev_timer_again>.
1631
1632At start:
1633
1634 ev_init (timer, callback);
1635 timer->repeat = 60.;
1636 ev_timer_again (loop, timer);
1637
1638Each time there is some activity:
1639
1640 ev_timer_again (loop, timer);
1641
1642It is even possible to change the time-out on the fly, regardless of
1643whether the watcher is active or not:
1644
1645 timer->repeat = 30.;
1646 ev_timer_again (loop, timer);
1647
1648This is slightly more efficient then stopping/starting the timer each time
1649you want to modify its timeout value, as libev does not have to completely
1650remove and re-insert the timer from/into its internal data structure.
1651
1652It is, however, even simpler than the "obvious" way to do it.
1653
1654=item 3. Let the timer time out, but then re-arm it as required.
1655
1656This method is more tricky, but usually most efficient: Most timeouts are
1657relatively long compared to the intervals between other activity - in
1658our example, within 60 seconds, there are usually many I/O events with
1659associated activity resets.
1660
1661In this case, it would be more efficient to leave the C<ev_timer> alone,
1662but remember the time of last activity, and check for a real timeout only
1663within the callback:
1664
1665 ev_tstamp last_activity; // time of last activity
1666
1667 static void
1668 callback (EV_P_ ev_timer *w, int revents)
1669 {
1670 ev_tstamp now = ev_now (EV_A);
1671 ev_tstamp timeout = last_activity + 60.;
1672
1673 // if last_activity + 60. is older than now, we did time out
1674 if (timeout < now)
1675 {
1676 // timeout occured, take action
1677 }
1678 else
1679 {
1680 // callback was invoked, but there was some activity, re-arm
1681 // the watcher to fire in last_activity + 60, which is
1682 // guaranteed to be in the future, so "again" is positive:
1683 w->repeat = timeout - now;
1684 ev_timer_again (EV_A_ w);
1685 }
1686 }
1687
1688To summarise the callback: first calculate the real timeout (defined
1689as "60 seconds after the last activity"), then check if that time has
1690been reached, which means something I<did>, in fact, time out. Otherwise
1691the callback was invoked too early (C<timeout> is in the future), so
1692re-schedule the timer to fire at that future time, to see if maybe we have
1693a timeout then.
1694
1695Note how C<ev_timer_again> is used, taking advantage of the
1696C<ev_timer_again> optimisation when the timer is already running.
1697
1698This scheme causes more callback invocations (about one every 60 seconds
1699minus half the average time between activity), but virtually no calls to
1700libev to change the timeout.
1701
1702To start the timer, simply initialise the watcher and set C<last_activity>
1703to the current time (meaning we just have some activity :), then call the
1704callback, which will "do the right thing" and start the timer:
1705
1706 ev_init (timer, callback);
1707 last_activity = ev_now (loop);
1708 callback (loop, timer, EV_TIMEOUT);
1709
1710And when there is some activity, simply store the current time in
1711C<last_activity>, no libev calls at all:
1712
1713 last_actiivty = ev_now (loop);
1714
1715This technique is slightly more complex, but in most cases where the
1716time-out is unlikely to be triggered, much more efficient.
1717
1718Changing the timeout is trivial as well (if it isn't hard-coded in the
1719callback :) - just change the timeout and invoke the callback, which will
1720fix things for you.
1721
1722=item 4. Wee, just use a double-linked list for your timeouts.
1723
1724If there is not one request, but many thousands (millions...), all
1725employing some kind of timeout with the same timeout value, then one can
1726do even better:
1727
1728When starting the timeout, calculate the timeout value and put the timeout
1729at the I<end> of the list.
1730
1731Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
1732the list is expected to fire (for example, using the technique #3).
1733
1734When there is some activity, remove the timer from the list, recalculate
1735the timeout, append it to the end of the list again, and make sure to
1736update the C<ev_timer> if it was taken from the beginning of the list.
1737
1738This way, one can manage an unlimited number of timeouts in O(1) time for
1739starting, stopping and updating the timers, at the expense of a major
1740complication, and having to use a constant timeout. The constant timeout
1741ensures that the list stays sorted.
1742
1743=back
1744
1745So which method the best?
1746
1747Method #2 is a simple no-brain-required solution that is adequate in most
1748situations. Method #3 requires a bit more thinking, but handles many cases
1749better, and isn't very complicated either. In most case, choosing either
1750one is fine, with #3 being better in typical situations.
1751
1752Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
1753rather complicated, but extremely efficient, something that really pays
1754off after the first million or so of active timers, i.e. it's usually
1755overkill :)
1756
1757=head3 The special problem of time updates
1758
1759Establishing the current time is a costly operation (it usually takes at
1760least two system calls): EV therefore updates its idea of the current
1761time only before and after C<ev_loop> collects new events, which causes a
1762growing difference between C<ev_now ()> and C<ev_time ()> when handling
1763lots of events in one iteration.
583 1764
584The relative timeouts are calculated relative to the C<ev_now ()> 1765The relative timeouts are calculated relative to the C<ev_now ()>
585time. This is usually the right thing as this timestamp refers to the time 1766time. This is usually the right thing as this timestamp refers to the time
586of the event triggering whatever timeout you are modifying/starting. If 1767of the event triggering whatever timeout you are modifying/starting. If
587you suspect event processing to be delayed and you I<need> to base the timeout 1768you suspect event processing to be delayed and you I<need> to base the
588on the current time, use something like this to adjust for this: 1769timeout on the current time, use something like this to adjust for this:
589 1770
590 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.); 1771 ev_timer_set (&timer, after + ev_now () - ev_time (), 0.);
591 1772
592The callback is guarenteed to be invoked only when its timeout has passed, 1773If the event loop is suspended for a long time, you can also force an
593but if multiple timers become ready during the same loop iteration then 1774update of the time returned by C<ev_now ()> by calling C<ev_now_update
594order of execution is undefined. 1775()>.
1776
1777=head3 The special problems of suspended animation
1778
1779When you leave the server world it is quite customary to hit machines that
1780can suspend/hibernate - what happens to the clocks during such a suspend?
1781
1782Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
1783all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
1784to run until the system is suspended, but they will not advance while the
1785system is suspended. That means, on resume, it will be as if the program
1786was frozen for a few seconds, but the suspend time will not be counted
1787towards C<ev_timer> when a monotonic clock source is used. The real time
1788clock advanced as expected, but if it is used as sole clocksource, then a
1789long suspend would be detected as a time jump by libev, and timers would
1790be adjusted accordingly.
1791
1792I would not be surprised to see different behaviour in different between
1793operating systems, OS versions or even different hardware.
1794
1795The other form of suspend (job control, or sending a SIGSTOP) will see a
1796time jump in the monotonic clocks and the realtime clock. If the program
1797is suspended for a very long time, and monotonic clock sources are in use,
1798then you can expect C<ev_timer>s to expire as the full suspension time
1799will be counted towards the timers. When no monotonic clock source is in
1800use, then libev will again assume a timejump and adjust accordingly.
1801
1802It might be beneficial for this latter case to call C<ev_suspend>
1803and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
1804deterministic behaviour in this case (you can do nothing against
1805C<SIGSTOP>).
1806
1807=head3 Watcher-Specific Functions and Data Members
595 1808
596=over 4 1809=over 4
597 1810
598=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat) 1811=item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
599 1812
600=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat) 1813=item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
601 1814
602Configure the timer to trigger after C<after> seconds. If C<repeat> is 1815Configure the timer to trigger after C<after> seconds. If C<repeat>
603C<0.>, then it will automatically be stopped. If it is positive, then the 1816is C<0.>, then it will automatically be stopped once the timeout is
604timer will automatically be configured to trigger again C<repeat> seconds 1817reached. If it is positive, then the timer will automatically be
605later, again, and again, until stopped manually. 1818configured to trigger again C<repeat> seconds later, again, and again,
1819until stopped manually.
606 1820
607The timer itself will do a best-effort at avoiding drift, that is, if you 1821The timer itself will do a best-effort at avoiding drift, that is, if
608configure a timer to trigger every 10 seconds, then it will trigger at 1822you configure a timer to trigger every 10 seconds, then it will normally
609exactly 10 second intervals. If, however, your program cannot keep up with 1823trigger at exactly 10 second intervals. If, however, your program cannot
610the timer (because it takes longer than those 10 seconds to do stuff) the 1824keep up with the timer (because it takes longer than those 10 seconds to
611timer will not fire more than once per event loop iteration. 1825do stuff) the timer will not fire more than once per event loop iteration.
612 1826
613=item ev_timer_again (loop) 1827=item ev_timer_again (loop, ev_timer *)
614 1828
615This will act as if the timer timed out and restart it again if it is 1829This will act as if the timer timed out and restart it again if it is
616repeating. The exact semantics are: 1830repeating. The exact semantics are:
617 1831
1832If the timer is pending, its pending status is cleared.
1833
618If the timer is started but nonrepeating, stop it. 1834If the timer is started but non-repeating, stop it (as if it timed out).
619 1835
620If the timer is repeating, either start it if necessary (with the repeat 1836If the timer is repeating, either start it if necessary (with the
621value), or reset the running timer to the repeat value. 1837C<repeat> value), or reset the running timer to the C<repeat> value.
622 1838
623This sounds a bit complicated, but here is a useful and typical 1839This sounds a bit complicated, see L<Be smart about timeouts>, above, for a
624example: Imagine you have a tcp connection and you want a so-called idle 1840usage example.
625timeout, that is, you want to be called when there have been, say, 60 1841
626seconds of inactivity on the socket. The easiest way to do this is to 1842=item ev_timer_remaining (loop, ev_timer *)
627configure an C<ev_timer> with after=repeat=60 and calling ev_timer_again each 1843
628time you successfully read or write some data. If you go into an idle 1844Returns the remaining time until a timer fires. If the timer is active,
629state where you do not expect data to travel on the socket, you can stop 1845then this time is relative to the current event loop time, otherwise it's
630the timer, and again will automatically restart it if need be. 1846the timeout value currently configured.
1847
1848That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
1849C<5>. When the timer is started and one second passes, C<ev_timer_remain>
1850will return C<4>. When the timer expires and is restarted, it will return
1851roughly C<7> (likely slightly less as callback invocation takes some time,
1852too), and so on.
1853
1854=item ev_tstamp repeat [read-write]
1855
1856The current C<repeat> value. Will be used each time the watcher times out
1857or C<ev_timer_again> is called, and determines the next timeout (if any),
1858which is also when any modifications are taken into account.
631 1859
632=back 1860=back
633 1861
1862=head3 Examples
1863
1864Example: Create a timer that fires after 60 seconds.
1865
1866 static void
1867 one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
1868 {
1869 .. one minute over, w is actually stopped right here
1870 }
1871
1872 ev_timer mytimer;
1873 ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
1874 ev_timer_start (loop, &mytimer);
1875
1876Example: Create a timeout timer that times out after 10 seconds of
1877inactivity.
1878
1879 static void
1880 timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
1881 {
1882 .. ten seconds without any activity
1883 }
1884
1885 ev_timer mytimer;
1886 ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
1887 ev_timer_again (&mytimer); /* start timer */
1888 ev_loop (loop, 0);
1889
1890 // and in some piece of code that gets executed on any "activity":
1891 // reset the timeout to start ticking again at 10 seconds
1892 ev_timer_again (&mytimer);
1893
1894
634=head2 C<ev_periodic> - to cron or not to cron 1895=head2 C<ev_periodic> - to cron or not to cron?
635 1896
636Periodic watchers are also timers of a kind, but they are very versatile 1897Periodic watchers are also timers of a kind, but they are very versatile
637(and unfortunately a bit complex). 1898(and unfortunately a bit complex).
638 1899
639Unlike C<ev_timer>'s, they are not based on real time (or relative time) 1900Unlike C<ev_timer>, periodic watchers are not based on real time (or
640but on wallclock time (absolute time). You can tell a periodic watcher 1901relative time, the physical time that passes) but on wall clock time
641to trigger "at" some specific point in time. For example, if you tell a 1902(absolute time, the thing you can read on your calender or clock). The
642periodic watcher to trigger in 10 seconds (by specifiying e.g. c<ev_now () 1903difference is that wall clock time can run faster or slower than real
643+ 10.>) and then reset your system clock to the last year, then it will 1904time, and time jumps are not uncommon (e.g. when you adjust your
644take a year to trigger the event (unlike an C<ev_timer>, which would trigger 1905wrist-watch).
645roughly 10 seconds later and of course not if you reset your system time
646again).
647 1906
648They can also be used to implement vastly more complex timers, such as 1907You can tell a periodic watcher to trigger after some specific point
1908in time: for example, if you tell a periodic watcher to trigger "in 10
1909seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
1910not a delay) and then reset your system clock to January of the previous
1911year, then it will take a year or more to trigger the event (unlike an
1912C<ev_timer>, which would still trigger roughly 10 seconds after starting
1913it, as it uses a relative timeout).
1914
1915C<ev_periodic> watchers can also be used to implement vastly more complex
649triggering an event on eahc midnight, local time. 1916timers, such as triggering an event on each "midnight, local time", or
1917other complicated rules. This cannot be done with C<ev_timer> watchers, as
1918those cannot react to time jumps.
650 1919
651As with timers, the callback is guarenteed to be invoked only when the 1920As with timers, the callback is guaranteed to be invoked only when the
652time (C<at>) has been passed, but if multiple periodic timers become ready 1921point in time where it is supposed to trigger has passed. If multiple
653during the same loop iteration then order of execution is undefined. 1922timers become ready during the same loop iteration then the ones with
1923earlier time-out values are invoked before ones with later time-out values
1924(but this is no longer true when a callback calls C<ev_loop> recursively).
1925
1926=head3 Watcher-Specific Functions and Data Members
654 1927
655=over 4 1928=over 4
656 1929
657=item ev_periodic_init (ev_periodic *, callback, ev_tstamp at, ev_tstamp interval, reschedule_cb) 1930=item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
658 1931
659=item ev_periodic_set (ev_periodic *, ev_tstamp after, ev_tstamp repeat, reschedule_cb) 1932=item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
660 1933
661Lots of arguments, lets sort it out... There are basically three modes of 1934Lots of arguments, let's sort it out... There are basically three modes of
662operation, and we will explain them from simplest to complex: 1935operation, and we will explain them from simplest to most complex:
663 1936
664=over 4 1937=over 4
665 1938
666=item * absolute timer (interval = reschedule_cb = 0) 1939=item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
667 1940
668In this configuration the watcher triggers an event at the wallclock time 1941In this configuration the watcher triggers an event after the wall clock
669C<at> and doesn't repeat. It will not adjust when a time jump occurs, 1942time C<offset> has passed. It will not repeat and will not adjust when a
670that is, if it is to be run at January 1st 2011 then it will run when the 1943time jump occurs, that is, if it is to be run at January 1st 2011 then it
671system time reaches or surpasses this time. 1944will be stopped and invoked when the system clock reaches or surpasses
1945this point in time.
672 1946
673=item * non-repeating interval timer (interval > 0, reschedule_cb = 0) 1947=item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
674 1948
675In this mode the watcher will always be scheduled to time out at the next 1949In this mode the watcher will always be scheduled to time out at the next
676C<at + N * interval> time (for some integer N) and then repeat, regardless 1950C<offset + N * interval> time (for some integer N, which can also be
677of any time jumps. 1951negative) and then repeat, regardless of any time jumps. The C<offset>
1952argument is merely an offset into the C<interval> periods.
678 1953
679This can be used to create timers that do not drift with respect to system 1954This can be used to create timers that do not drift with respect to the
680time: 1955system clock, for example, here is an C<ev_periodic> that triggers each
1956hour, on the hour (with respect to UTC):
681 1957
682 ev_periodic_set (&periodic, 0., 3600., 0); 1958 ev_periodic_set (&periodic, 0., 3600., 0);
683 1959
684This doesn't mean there will always be 3600 seconds in between triggers, 1960This doesn't mean there will always be 3600 seconds in between triggers,
685but only that the the callback will be called when the system time shows a 1961but only that the callback will be called when the system time shows a
686full hour (UTC), or more correctly, when the system time is evenly divisible 1962full hour (UTC), or more correctly, when the system time is evenly divisible
687by 3600. 1963by 3600.
688 1964
689Another way to think about it (for the mathematically inclined) is that 1965Another way to think about it (for the mathematically inclined) is that
690C<ev_periodic> will try to run the callback in this mode at the next possible 1966C<ev_periodic> will try to run the callback in this mode at the next possible
691time where C<time = at (mod interval)>, regardless of any time jumps. 1967time where C<time = offset (mod interval)>, regardless of any time jumps.
692 1968
1969For numerical stability it is preferable that the C<offset> value is near
1970C<ev_now ()> (the current time), but there is no range requirement for
1971this value, and in fact is often specified as zero.
1972
1973Note also that there is an upper limit to how often a timer can fire (CPU
1974speed for example), so if C<interval> is very small then timing stability
1975will of course deteriorate. Libev itself tries to be exact to be about one
1976millisecond (if the OS supports it and the machine is fast enough).
1977
693=item * manual reschedule mode (reschedule_cb = callback) 1978=item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
694 1979
695In this mode the values for C<interval> and C<at> are both being 1980In this mode the values for C<interval> and C<offset> are both being
696ignored. Instead, each time the periodic watcher gets scheduled, the 1981ignored. Instead, each time the periodic watcher gets scheduled, the
697reschedule callback will be called with the watcher as first, and the 1982reschedule callback will be called with the watcher as first, and the
698current time as second argument. 1983current time as second argument.
699 1984
700NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, 1985NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
701ever, or make any event loop modifications>. If you need to stop it, 1986or make ANY other event loop modifications whatsoever, unless explicitly
702return C<now + 1e30> (or so, fudge fudge) and stop it afterwards (e.g. by 1987allowed by documentation here>.
703starting a prepare watcher).
704 1988
1989If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
1990it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
1991only event loop modification you are allowed to do).
1992
705Its prototype is C<ev_tstamp (*reschedule_cb)(struct ev_periodic *w, 1993The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
706ev_tstamp now)>, e.g.: 1994*w, ev_tstamp now)>, e.g.:
707 1995
1996 static ev_tstamp
708 static ev_tstamp my_rescheduler (struct ev_periodic *w, ev_tstamp now) 1997 my_rescheduler (ev_periodic *w, ev_tstamp now)
709 { 1998 {
710 return now + 60.; 1999 return now + 60.;
711 } 2000 }
712 2001
713It must return the next time to trigger, based on the passed time value 2002It must return the next time to trigger, based on the passed time value
714(that is, the lowest time value larger than to the second argument). It 2003(that is, the lowest time value larger than to the second argument). It
715will usually be called just before the callback will be triggered, but 2004will usually be called just before the callback will be triggered, but
716might be called at other times, too. 2005might be called at other times, too.
717 2006
718NOTE: I<< This callback must always return a time that is later than the 2007NOTE: I<< This callback must always return a time that is higher than or
719passed C<now> value >>. Not even C<now> itself will do, it I<must> be larger. 2008equal to the passed C<now> value >>.
720 2009
721This can be used to create very complex timers, such as a timer that 2010This can be used to create very complex timers, such as a timer that
722triggers on each midnight, local time. To do this, you would calculate the 2011triggers on "next midnight, local time". To do this, you would calculate the
723next midnight after C<now> and return the timestamp value for this. How 2012next midnight after C<now> and return the timestamp value for this. How
724you do this is, again, up to you (but it is not trivial, which is the main 2013you do this is, again, up to you (but it is not trivial, which is the main
725reason I omitted it as an example). 2014reason I omitted it as an example).
726 2015
727=back 2016=back
731Simply stops and restarts the periodic watcher again. This is only useful 2020Simply stops and restarts the periodic watcher again. This is only useful
732when you changed some parameters or the reschedule callback would return 2021when you changed some parameters or the reschedule callback would return
733a different time than the last time it was called (e.g. in a crond like 2022a different time than the last time it was called (e.g. in a crond like
734program when the crontabs have changed). 2023program when the crontabs have changed).
735 2024
2025=item ev_tstamp ev_periodic_at (ev_periodic *)
2026
2027When active, returns the absolute time that the watcher is supposed
2028to trigger next. This is not the same as the C<offset> argument to
2029C<ev_periodic_set>, but indeed works even in interval and manual
2030rescheduling modes.
2031
2032=item ev_tstamp offset [read-write]
2033
2034When repeating, this contains the offset value, otherwise this is the
2035absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
2036although libev might modify this value for better numerical stability).
2037
2038Can be modified any time, but changes only take effect when the periodic
2039timer fires or C<ev_periodic_again> is being called.
2040
2041=item ev_tstamp interval [read-write]
2042
2043The current interval value. Can be modified any time, but changes only
2044take effect when the periodic timer fires or C<ev_periodic_again> is being
2045called.
2046
2047=item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2048
2049The current reschedule callback, or C<0>, if this functionality is
2050switched off. Can be changed any time, but changes only take effect when
2051the periodic timer fires or C<ev_periodic_again> is being called.
2052
736=back 2053=back
737 2054
2055=head3 Examples
2056
2057Example: Call a callback every hour, or, more precisely, whenever the
2058system time is divisible by 3600. The callback invocation times have
2059potentially a lot of jitter, but good long-term stability.
2060
2061 static void
2062 clock_cb (struct ev_loop *loop, ev_io *w, int revents)
2063 {
2064 ... its now a full hour (UTC, or TAI or whatever your clock follows)
2065 }
2066
2067 ev_periodic hourly_tick;
2068 ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2069 ev_periodic_start (loop, &hourly_tick);
2070
2071Example: The same as above, but use a reschedule callback to do it:
2072
2073 #include <math.h>
2074
2075 static ev_tstamp
2076 my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2077 {
2078 return now + (3600. - fmod (now, 3600.));
2079 }
2080
2081 ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2082
2083Example: Call a callback every hour, starting now:
2084
2085 ev_periodic hourly_tick;
2086 ev_periodic_init (&hourly_tick, clock_cb,
2087 fmod (ev_now (loop), 3600.), 3600., 0);
2088 ev_periodic_start (loop, &hourly_tick);
2089
2090
738=head2 C<ev_signal> - signal me when a signal gets signalled 2091=head2 C<ev_signal> - signal me when a signal gets signalled!
739 2092
740Signal watchers will trigger an event when the process receives a specific 2093Signal watchers will trigger an event when the process receives a specific
741signal one or more times. Even though signals are very asynchronous, libev 2094signal one or more times. Even though signals are very asynchronous, libev
742will try it's best to deliver signals synchronously, i.e. as part of the 2095will try it's best to deliver signals synchronously, i.e. as part of the
743normal event processing, like any other event. 2096normal event processing, like any other event.
744 2097
2098If you want signals to be delivered truly asynchronously, just use
2099C<sigaction> as you would do without libev and forget about sharing
2100the signal. You can even use C<ev_async> from a signal handler to
2101synchronously wake up an event loop.
2102
745You can configure as many watchers as you like per signal. Only when the 2103You can configure as many watchers as you like for the same signal, but
2104only within the same loop, i.e. you can watch for C<SIGINT> in your
2105default loop and for C<SIGIO> in another loop, but you cannot watch for
2106C<SIGINT> in both the default loop and another loop at the same time. At
2107the moment, C<SIGCHLD> is permanently tied to the default loop.
2108
746first watcher gets started will libev actually register a signal watcher 2109When the first watcher gets started will libev actually register something
747with the kernel (thus it coexists with your own signal handlers as long 2110with the kernel (thus it coexists with your own signal handlers as long as
748as you don't register any with libev). Similarly, when the last signal 2111you don't register any with libev for the same signal).
749watcher for a signal is stopped libev will reset the signal handler to 2112
750SIG_DFL (regardless of what it was set to before). 2113If possible and supported, libev will install its handlers with
2114C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
2115not be unduly interrupted. If you have a problem with system calls getting
2116interrupted by signals you can block all signals in an C<ev_check> watcher
2117and unblock them in an C<ev_prepare> watcher.
2118
2119=head3 The special problem of inheritance over execve
2120
2121Both the signal mask (C<sigprocmask>) and the signal disposition
2122(C<sigaction>) are unspecified after starting a signal watcher (and after
2123stopping it again), that is, libev might or might not block the signal,
2124and might or might not set or restore the installed signal handler.
2125
2126While this does not matter for the signal disposition (libev never
2127sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
2128C<execve>), this matters for the signal mask: many programs do not expect
2129many signals to be blocked.
2130
2131This means that before calling C<exec> (from the child) you should reset
2132the signal mask to whatever "default" you expect (all clear is a good
2133choice usually).
2134
2135=head3 Watcher-Specific Functions and Data Members
751 2136
752=over 4 2137=over 4
753 2138
754=item ev_signal_init (ev_signal *, callback, int signum) 2139=item ev_signal_init (ev_signal *, callback, int signum)
755 2140
756=item ev_signal_set (ev_signal *, int signum) 2141=item ev_signal_set (ev_signal *, int signum)
757 2142
758Configures the watcher to trigger on the given signal number (usually one 2143Configures the watcher to trigger on the given signal number (usually one
759of the C<SIGxxx> constants). 2144of the C<SIGxxx> constants).
760 2145
2146=item int signum [read-only]
2147
2148The signal the watcher watches out for.
2149
761=back 2150=back
762 2151
2152=head3 Examples
2153
2154Example: Try to exit cleanly on SIGINT.
2155
2156 static void
2157 sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2158 {
2159 ev_unloop (loop, EVUNLOOP_ALL);
2160 }
2161
2162 ev_signal signal_watcher;
2163 ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2164 ev_signal_start (loop, &signal_watcher);
2165
2166
763=head2 C<ev_child> - wait for pid status changes 2167=head2 C<ev_child> - watch out for process status changes
764 2168
765Child watchers trigger when your process receives a SIGCHLD in response to 2169Child watchers trigger when your process receives a SIGCHLD in response to
766some child status changes (most typically when a child of yours dies). 2170some child status changes (most typically when a child of yours dies or
2171exits). It is permissible to install a child watcher I<after> the child
2172has been forked (which implies it might have already exited), as long
2173as the event loop isn't entered (or is continued from a watcher), i.e.,
2174forking and then immediately registering a watcher for the child is fine,
2175but forking and registering a watcher a few event loop iterations later or
2176in the next callback invocation is not.
2177
2178Only the default event loop is capable of handling signals, and therefore
2179you can only register child watchers in the default event loop.
2180
2181Due to some design glitches inside libev, child watchers will always be
2182handled at maximum priority (their priority is set to C<EV_MAXPRI> by
2183libev)
2184
2185=head3 Process Interaction
2186
2187Libev grabs C<SIGCHLD> as soon as the default event loop is
2188initialised. This is necessary to guarantee proper behaviour even if the
2189first child watcher is started after the child exits. The occurrence
2190of C<SIGCHLD> is recorded asynchronously, but child reaping is done
2191synchronously as part of the event loop processing. Libev always reaps all
2192children, even ones not watched.
2193
2194=head3 Overriding the Built-In Processing
2195
2196Libev offers no special support for overriding the built-in child
2197processing, but if your application collides with libev's default child
2198handler, you can override it easily by installing your own handler for
2199C<SIGCHLD> after initialising the default loop, and making sure the
2200default loop never gets destroyed. You are encouraged, however, to use an
2201event-based approach to child reaping and thus use libev's support for
2202that, so other libev users can use C<ev_child> watchers freely.
2203
2204=head3 Stopping the Child Watcher
2205
2206Currently, the child watcher never gets stopped, even when the
2207child terminates, so normally one needs to stop the watcher in the
2208callback. Future versions of libev might stop the watcher automatically
2209when a child exit is detected (calling C<ev_child_stop> twice is not a
2210problem).
2211
2212=head3 Watcher-Specific Functions and Data Members
767 2213
768=over 4 2214=over 4
769 2215
770=item ev_child_init (ev_child *, callback, int pid) 2216=item ev_child_init (ev_child *, callback, int pid, int trace)
771 2217
772=item ev_child_set (ev_child *, int pid) 2218=item ev_child_set (ev_child *, int pid, int trace)
773 2219
774Configures the watcher to wait for status changes of process C<pid> (or 2220Configures the watcher to wait for status changes of process C<pid> (or
775I<any> process if C<pid> is specified as C<0>). The callback can look 2221I<any> process if C<pid> is specified as C<0>). The callback can look
776at the C<rstatus> member of the C<ev_child> watcher structure to see 2222at the C<rstatus> member of the C<ev_child> watcher structure to see
777the status word (use the macros from C<sys/wait.h> and see your systems 2223the status word (use the macros from C<sys/wait.h> and see your systems
778C<waitpid> documentation). The C<rpid> member contains the pid of the 2224C<waitpid> documentation). The C<rpid> member contains the pid of the
779process causing the status change. 2225process causing the status change. C<trace> must be either C<0> (only
2226activate the watcher when the process terminates) or C<1> (additionally
2227activate the watcher when the process is stopped or continued).
2228
2229=item int pid [read-only]
2230
2231The process id this watcher watches out for, or C<0>, meaning any process id.
2232
2233=item int rpid [read-write]
2234
2235The process id that detected a status change.
2236
2237=item int rstatus [read-write]
2238
2239The process exit/trace status caused by C<rpid> (see your systems
2240C<waitpid> and C<sys/wait.h> documentation for details).
780 2241
781=back 2242=back
782 2243
2244=head3 Examples
2245
2246Example: C<fork()> a new process and install a child handler to wait for
2247its completion.
2248
2249 ev_child cw;
2250
2251 static void
2252 child_cb (EV_P_ ev_child *w, int revents)
2253 {
2254 ev_child_stop (EV_A_ w);
2255 printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2256 }
2257
2258 pid_t pid = fork ();
2259
2260 if (pid < 0)
2261 // error
2262 else if (pid == 0)
2263 {
2264 // the forked child executes here
2265 exit (1);
2266 }
2267 else
2268 {
2269 ev_child_init (&cw, child_cb, pid, 0);
2270 ev_child_start (EV_DEFAULT_ &cw);
2271 }
2272
2273
2274=head2 C<ev_stat> - did the file attributes just change?
2275
2276This watches a file system path for attribute changes. That is, it calls
2277C<stat> on that path in regular intervals (or when the OS says it changed)
2278and sees if it changed compared to the last time, invoking the callback if
2279it did.
2280
2281The path does not need to exist: changing from "path exists" to "path does
2282not exist" is a status change like any other. The condition "path does not
2283exist" (or more correctly "path cannot be stat'ed") is signified by the
2284C<st_nlink> field being zero (which is otherwise always forced to be at
2285least one) and all the other fields of the stat buffer having unspecified
2286contents.
2287
2288The path I<must not> end in a slash or contain special components such as
2289C<.> or C<..>. The path I<should> be absolute: If it is relative and
2290your working directory changes, then the behaviour is undefined.
2291
2292Since there is no portable change notification interface available, the
2293portable implementation simply calls C<stat(2)> regularly on the path
2294to see if it changed somehow. You can specify a recommended polling
2295interval for this case. If you specify a polling interval of C<0> (highly
2296recommended!) then a I<suitable, unspecified default> value will be used
2297(which you can expect to be around five seconds, although this might
2298change dynamically). Libev will also impose a minimum interval which is
2299currently around C<0.1>, but that's usually overkill.
2300
2301This watcher type is not meant for massive numbers of stat watchers,
2302as even with OS-supported change notifications, this can be
2303resource-intensive.
2304
2305At the time of this writing, the only OS-specific interface implemented
2306is the Linux inotify interface (implementing kqueue support is left as an
2307exercise for the reader. Note, however, that the author sees no way of
2308implementing C<ev_stat> semantics with kqueue, except as a hint).
2309
2310=head3 ABI Issues (Largefile Support)
2311
2312Libev by default (unless the user overrides this) uses the default
2313compilation environment, which means that on systems with large file
2314support disabled by default, you get the 32 bit version of the stat
2315structure. When using the library from programs that change the ABI to
2316use 64 bit file offsets the programs will fail. In that case you have to
2317compile libev with the same flags to get binary compatibility. This is
2318obviously the case with any flags that change the ABI, but the problem is
2319most noticeably displayed with ev_stat and large file support.
2320
2321The solution for this is to lobby your distribution maker to make large
2322file interfaces available by default (as e.g. FreeBSD does) and not
2323optional. Libev cannot simply switch on large file support because it has
2324to exchange stat structures with application programs compiled using the
2325default compilation environment.
2326
2327=head3 Inotify and Kqueue
2328
2329When C<inotify (7)> support has been compiled into libev and present at
2330runtime, it will be used to speed up change detection where possible. The
2331inotify descriptor will be created lazily when the first C<ev_stat>
2332watcher is being started.
2333
2334Inotify presence does not change the semantics of C<ev_stat> watchers
2335except that changes might be detected earlier, and in some cases, to avoid
2336making regular C<stat> calls. Even in the presence of inotify support
2337there are many cases where libev has to resort to regular C<stat> polling,
2338but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
2339many bugs), the path exists (i.e. stat succeeds), and the path resides on
2340a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
2341xfs are fully working) libev usually gets away without polling.
2342
2343There is no support for kqueue, as apparently it cannot be used to
2344implement this functionality, due to the requirement of having a file
2345descriptor open on the object at all times, and detecting renames, unlinks
2346etc. is difficult.
2347
2348=head3 C<stat ()> is a synchronous operation
2349
2350Libev doesn't normally do any kind of I/O itself, and so is not blocking
2351the process. The exception are C<ev_stat> watchers - those call C<stat
2352()>, which is a synchronous operation.
2353
2354For local paths, this usually doesn't matter: unless the system is very
2355busy or the intervals between stat's are large, a stat call will be fast,
2356as the path data is usually in memory already (except when starting the
2357watcher).
2358
2359For networked file systems, calling C<stat ()> can block an indefinite
2360time due to network issues, and even under good conditions, a stat call
2361often takes multiple milliseconds.
2362
2363Therefore, it is best to avoid using C<ev_stat> watchers on networked
2364paths, although this is fully supported by libev.
2365
2366=head3 The special problem of stat time resolution
2367
2368The C<stat ()> system call only supports full-second resolution portably,
2369and even on systems where the resolution is higher, most file systems
2370still only support whole seconds.
2371
2372That means that, if the time is the only thing that changes, you can
2373easily miss updates: on the first update, C<ev_stat> detects a change and
2374calls your callback, which does something. When there is another update
2375within the same second, C<ev_stat> will be unable to detect unless the
2376stat data does change in other ways (e.g. file size).
2377
2378The solution to this is to delay acting on a change for slightly more
2379than a second (or till slightly after the next full second boundary), using
2380a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
2381ev_timer_again (loop, w)>).
2382
2383The C<.02> offset is added to work around small timing inconsistencies
2384of some operating systems (where the second counter of the current time
2385might be be delayed. One such system is the Linux kernel, where a call to
2386C<gettimeofday> might return a timestamp with a full second later than
2387a subsequent C<time> call - if the equivalent of C<time ()> is used to
2388update file times then there will be a small window where the kernel uses
2389the previous second to update file times but libev might already execute
2390the timer callback).
2391
2392=head3 Watcher-Specific Functions and Data Members
2393
2394=over 4
2395
2396=item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
2397
2398=item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2399
2400Configures the watcher to wait for status changes of the given
2401C<path>. The C<interval> is a hint on how quickly a change is expected to
2402be detected and should normally be specified as C<0> to let libev choose
2403a suitable value. The memory pointed to by C<path> must point to the same
2404path for as long as the watcher is active.
2405
2406The callback will receive an C<EV_STAT> event when a change was detected,
2407relative to the attributes at the time the watcher was started (or the
2408last change was detected).
2409
2410=item ev_stat_stat (loop, ev_stat *)
2411
2412Updates the stat buffer immediately with new values. If you change the
2413watched path in your callback, you could call this function to avoid
2414detecting this change (while introducing a race condition if you are not
2415the only one changing the path). Can also be useful simply to find out the
2416new values.
2417
2418=item ev_statdata attr [read-only]
2419
2420The most-recently detected attributes of the file. Although the type is
2421C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
2422suitable for your system, but you can only rely on the POSIX-standardised
2423members to be present. If the C<st_nlink> member is C<0>, then there was
2424some error while C<stat>ing the file.
2425
2426=item ev_statdata prev [read-only]
2427
2428The previous attributes of the file. The callback gets invoked whenever
2429C<prev> != C<attr>, or, more precisely, one or more of these members
2430differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
2431C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
2432
2433=item ev_tstamp interval [read-only]
2434
2435The specified interval.
2436
2437=item const char *path [read-only]
2438
2439The file system path that is being watched.
2440
2441=back
2442
2443=head3 Examples
2444
2445Example: Watch C</etc/passwd> for attribute changes.
2446
2447 static void
2448 passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2449 {
2450 /* /etc/passwd changed in some way */
2451 if (w->attr.st_nlink)
2452 {
2453 printf ("passwd current size %ld\n", (long)w->attr.st_size);
2454 printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2455 printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2456 }
2457 else
2458 /* you shalt not abuse printf for puts */
2459 puts ("wow, /etc/passwd is not there, expect problems. "
2460 "if this is windows, they already arrived\n");
2461 }
2462
2463 ...
2464 ev_stat passwd;
2465
2466 ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2467 ev_stat_start (loop, &passwd);
2468
2469Example: Like above, but additionally use a one-second delay so we do not
2470miss updates (however, frequent updates will delay processing, too, so
2471one might do the work both on C<ev_stat> callback invocation I<and> on
2472C<ev_timer> callback invocation).
2473
2474 static ev_stat passwd;
2475 static ev_timer timer;
2476
2477 static void
2478 timer_cb (EV_P_ ev_timer *w, int revents)
2479 {
2480 ev_timer_stop (EV_A_ w);
2481
2482 /* now it's one second after the most recent passwd change */
2483 }
2484
2485 static void
2486 stat_cb (EV_P_ ev_stat *w, int revents)
2487 {
2488 /* reset the one-second timer */
2489 ev_timer_again (EV_A_ &timer);
2490 }
2491
2492 ...
2493 ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2494 ev_stat_start (loop, &passwd);
2495 ev_timer_init (&timer, timer_cb, 0., 1.02);
2496
2497
783=head2 C<ev_idle> - when you've got nothing better to do 2498=head2 C<ev_idle> - when you've got nothing better to do...
784 2499
785Idle watchers trigger events when there are no other events are pending 2500Idle watchers trigger events when no other events of the same or higher
786(prepare, check and other idle watchers do not count). That is, as long 2501priority are pending (prepare, check and other idle watchers do not count
787as your process is busy handling sockets or timeouts (or even signals, 2502as receiving "events").
788imagine) it will not be triggered. But when your process is idle all idle 2503
789watchers are being called again and again, once per event loop iteration - 2504That is, as long as your process is busy handling sockets or timeouts
2505(or even signals, imagine) of the same or higher priority it will not be
2506triggered. But when your process is idle (or only lower-priority watchers
2507are pending), the idle watchers are being called once per event loop
790until stopped, that is, or your process receives more events and becomes 2508iteration - until stopped, that is, or your process receives more events
791busy. 2509and becomes busy again with higher priority stuff.
792 2510
793The most noteworthy effect is that as long as any idle watchers are 2511The most noteworthy effect is that as long as any idle watchers are
794active, the process will not block when waiting for new events. 2512active, the process will not block when waiting for new events.
795 2513
796Apart from keeping your process non-blocking (which is a useful 2514Apart from keeping your process non-blocking (which is a useful
797effect on its own sometimes), idle watchers are a good place to do 2515effect on its own sometimes), idle watchers are a good place to do
798"pseudo-background processing", or delay processing stuff to after the 2516"pseudo-background processing", or delay processing stuff to after the
799event loop has handled all outstanding events. 2517event loop has handled all outstanding events.
800 2518
2519=head3 Watcher-Specific Functions and Data Members
2520
801=over 4 2521=over 4
802 2522
803=item ev_idle_init (ev_signal *, callback) 2523=item ev_idle_init (ev_idle *, callback)
804 2524
805Initialises and configures the idle watcher - it has no parameters of any 2525Initialises and configures the idle watcher - it has no parameters of any
806kind. There is a C<ev_idle_set> macro, but using it is utterly pointless, 2526kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
807believe me. 2527believe me.
808 2528
809=back 2529=back
810 2530
2531=head3 Examples
2532
2533Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
2534callback, free it. Also, use no error checking, as usual.
2535
2536 static void
2537 idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2538 {
2539 free (w);
2540 // now do something you wanted to do when the program has
2541 // no longer anything immediate to do.
2542 }
2543
2544 ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2545 ev_idle_init (idle_watcher, idle_cb);
2546 ev_idle_start (loop, idle_watcher);
2547
2548
811=head2 C<ev_prepare> and C<ev_check> - customise your event loop 2549=head2 C<ev_prepare> and C<ev_check> - customise your event loop!
812 2550
813Prepare and check watchers are usually (but not always) used in tandem: 2551Prepare and check watchers are usually (but not always) used in pairs:
814prepare watchers get invoked before the process blocks and check watchers 2552prepare watchers get invoked before the process blocks and check watchers
815afterwards. 2553afterwards.
816 2554
2555You I<must not> call C<ev_loop> or similar functions that enter
2556the current event loop from either C<ev_prepare> or C<ev_check>
2557watchers. Other loops than the current one are fine, however. The
2558rationale behind this is that you do not need to check for recursion in
2559those watchers, i.e. the sequence will always be C<ev_prepare>, blocking,
2560C<ev_check> so if you have one watcher of each kind they will always be
2561called in pairs bracketing the blocking call.
2562
817Their main purpose is to integrate other event mechanisms into libev. This 2563Their main purpose is to integrate other event mechanisms into libev and
818could be used, for example, to track variable changes, implement your own 2564their use is somewhat advanced. They could be used, for example, to track
819watchers, integrate net-snmp or a coroutine library and lots more. 2565variable changes, implement your own watchers, integrate net-snmp or a
2566coroutine library and lots more. They are also occasionally useful if
2567you cache some data and want to flush it before blocking (for example,
2568in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
2569watcher).
820 2570
821This is done by examining in each prepare call which file descriptors need 2571This is done by examining in each prepare call which file descriptors
822to be watched by the other library, registering C<ev_io> watchers for 2572need to be watched by the other library, registering C<ev_io> watchers
823them and starting an C<ev_timer> watcher for any timeouts (many libraries 2573for them and starting an C<ev_timer> watcher for any timeouts (many
824provide just this functionality). Then, in the check watcher you check for 2574libraries provide exactly this functionality). Then, in the check watcher,
825any events that occured (by checking the pending status of all watchers 2575you check for any events that occurred (by checking the pending status
826and stopping them) and call back into the library. The I/O and timer 2576of all watchers and stopping them) and call back into the library. The
827callbacks will never actually be called (but must be valid nevertheless, 2577I/O and timer callbacks will never actually be called (but must be valid
828because you never know, you know?). 2578nevertheless, because you never know, you know?).
829 2579
830As another example, the Perl Coro module uses these hooks to integrate 2580As another example, the Perl Coro module uses these hooks to integrate
831coroutines into libev programs, by yielding to other active coroutines 2581coroutines into libev programs, by yielding to other active coroutines
832during each prepare and only letting the process block if no coroutines 2582during each prepare and only letting the process block if no coroutines
833are ready to run (it's actually more complicated: it only runs coroutines 2583are ready to run (it's actually more complicated: it only runs coroutines
834with priority higher than or equal to the event loop and one coroutine 2584with priority higher than or equal to the event loop and one coroutine
835of lower priority, but only once, using idle watchers to keep the event 2585of lower priority, but only once, using idle watchers to keep the event
836loop from blocking if lower-priority coroutines are active, thus mapping 2586loop from blocking if lower-priority coroutines are active, thus mapping
837low-priority coroutines to idle/background tasks). 2587low-priority coroutines to idle/background tasks).
838 2588
2589It is recommended to give C<ev_check> watchers highest (C<EV_MAXPRI>)
2590priority, to ensure that they are being run before any other watchers
2591after the poll (this doesn't matter for C<ev_prepare> watchers).
2592
2593Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
2594activate ("feed") events into libev. While libev fully supports this, they
2595might get executed before other C<ev_check> watchers did their job. As
2596C<ev_check> watchers are often used to embed other (non-libev) event
2597loops those other event loops might be in an unusable state until their
2598C<ev_check> watcher ran (always remind yourself to coexist peacefully with
2599others).
2600
2601=head3 Watcher-Specific Functions and Data Members
2602
839=over 4 2603=over 4
840 2604
841=item ev_prepare_init (ev_prepare *, callback) 2605=item ev_prepare_init (ev_prepare *, callback)
842 2606
843=item ev_check_init (ev_check *, callback) 2607=item ev_check_init (ev_check *, callback)
844 2608
845Initialises and configures the prepare or check watcher - they have no 2609Initialises and configures the prepare or check watcher - they have no
846parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set> 2610parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
847macros, but using them is utterly, utterly and completely pointless. 2611macros, but using them is utterly, utterly, utterly and completely
2612pointless.
848 2613
849=back 2614=back
850 2615
2616=head3 Examples
2617
2618There are a number of principal ways to embed other event loops or modules
2619into libev. Here are some ideas on how to include libadns into libev
2620(there is a Perl module named C<EV::ADNS> that does this, which you could
2621use as a working example. Another Perl module named C<EV::Glib> embeds a
2622Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
2623Glib event loop).
2624
2625Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2626and in a check watcher, destroy them and call into libadns. What follows
2627is pseudo-code only of course. This requires you to either use a low
2628priority for the check watcher or use C<ev_clear_pending> explicitly, as
2629the callbacks for the IO/timeout watchers might not have been called yet.
2630
2631 static ev_io iow [nfd];
2632 static ev_timer tw;
2633
2634 static void
2635 io_cb (struct ev_loop *loop, ev_io *w, int revents)
2636 {
2637 }
2638
2639 // create io watchers for each fd and a timer before blocking
2640 static void
2641 adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2642 {
2643 int timeout = 3600000;
2644 struct pollfd fds [nfd];
2645 // actual code will need to loop here and realloc etc.
2646 adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2647
2648 /* the callback is illegal, but won't be called as we stop during check */
2649 ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2650 ev_timer_start (loop, &tw);
2651
2652 // create one ev_io per pollfd
2653 for (int i = 0; i < nfd; ++i)
2654 {
2655 ev_io_init (iow + i, io_cb, fds [i].fd,
2656 ((fds [i].events & POLLIN ? EV_READ : 0)
2657 | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2658
2659 fds [i].revents = 0;
2660 ev_io_start (loop, iow + i);
2661 }
2662 }
2663
2664 // stop all watchers after blocking
2665 static void
2666 adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2667 {
2668 ev_timer_stop (loop, &tw);
2669
2670 for (int i = 0; i < nfd; ++i)
2671 {
2672 // set the relevant poll flags
2673 // could also call adns_processreadable etc. here
2674 struct pollfd *fd = fds + i;
2675 int revents = ev_clear_pending (iow + i);
2676 if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2677 if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2678
2679 // now stop the watcher
2680 ev_io_stop (loop, iow + i);
2681 }
2682
2683 adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2684 }
2685
2686Method 2: This would be just like method 1, but you run C<adns_afterpoll>
2687in the prepare watcher and would dispose of the check watcher.
2688
2689Method 3: If the module to be embedded supports explicit event
2690notification (libadns does), you can also make use of the actual watcher
2691callbacks, and only destroy/create the watchers in the prepare watcher.
2692
2693 static void
2694 timer_cb (EV_P_ ev_timer *w, int revents)
2695 {
2696 adns_state ads = (adns_state)w->data;
2697 update_now (EV_A);
2698
2699 adns_processtimeouts (ads, &tv_now);
2700 }
2701
2702 static void
2703 io_cb (EV_P_ ev_io *w, int revents)
2704 {
2705 adns_state ads = (adns_state)w->data;
2706 update_now (EV_A);
2707
2708 if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
2709 if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
2710 }
2711
2712 // do not ever call adns_afterpoll
2713
2714Method 4: Do not use a prepare or check watcher because the module you
2715want to embed is not flexible enough to support it. Instead, you can
2716override their poll function. The drawback with this solution is that the
2717main loop is now no longer controllable by EV. The C<Glib::EV> module uses
2718this approach, effectively embedding EV as a client into the horrible
2719libglib event loop.
2720
2721 static gint
2722 event_poll_func (GPollFD *fds, guint nfds, gint timeout)
2723 {
2724 int got_events = 0;
2725
2726 for (n = 0; n < nfds; ++n)
2727 // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
2728
2729 if (timeout >= 0)
2730 // create/start timer
2731
2732 // poll
2733 ev_loop (EV_A_ 0);
2734
2735 // stop timer again
2736 if (timeout >= 0)
2737 ev_timer_stop (EV_A_ &to);
2738
2739 // stop io watchers again - their callbacks should have set
2740 for (n = 0; n < nfds; ++n)
2741 ev_io_stop (EV_A_ iow [n]);
2742
2743 return got_events;
2744 }
2745
2746
2747=head2 C<ev_embed> - when one backend isn't enough...
2748
2749This is a rather advanced watcher type that lets you embed one event loop
2750into another (currently only C<ev_io> events are supported in the embedded
2751loop, other types of watchers might be handled in a delayed or incorrect
2752fashion and must not be used).
2753
2754There are primarily two reasons you would want that: work around bugs and
2755prioritise I/O.
2756
2757As an example for a bug workaround, the kqueue backend might only support
2758sockets on some platform, so it is unusable as generic backend, but you
2759still want to make use of it because you have many sockets and it scales
2760so nicely. In this case, you would create a kqueue-based loop and embed
2761it into your default loop (which might use e.g. poll). Overall operation
2762will be a bit slower because first libev has to call C<poll> and then
2763C<kevent>, but at least you can use both mechanisms for what they are
2764best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
2765
2766As for prioritising I/O: under rare circumstances you have the case where
2767some fds have to be watched and handled very quickly (with low latency),
2768and even priorities and idle watchers might have too much overhead. In
2769this case you would put all the high priority stuff in one loop and all
2770the rest in a second one, and embed the second one in the first.
2771
2772As long as the watcher is active, the callback will be invoked every
2773time there might be events pending in the embedded loop. The callback
2774must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
2775sweep and invoke their callbacks (the callback doesn't need to invoke the
2776C<ev_embed_sweep> function directly, it could also start an idle watcher
2777to give the embedded loop strictly lower priority for example).
2778
2779You can also set the callback to C<0>, in which case the embed watcher
2780will automatically execute the embedded loop sweep whenever necessary.
2781
2782Fork detection will be handled transparently while the C<ev_embed> watcher
2783is active, i.e., the embedded loop will automatically be forked when the
2784embedding loop forks. In other cases, the user is responsible for calling
2785C<ev_loop_fork> on the embedded loop.
2786
2787Unfortunately, not all backends are embeddable: only the ones returned by
2788C<ev_embeddable_backends> are, which, unfortunately, does not include any
2789portable one.
2790
2791So when you want to use this feature you will always have to be prepared
2792that you cannot get an embeddable loop. The recommended way to get around
2793this is to have a separate variables for your embeddable loop, try to
2794create it, and if that fails, use the normal loop for everything.
2795
2796=head3 C<ev_embed> and fork
2797
2798While the C<ev_embed> watcher is running, forks in the embedding loop will
2799automatically be applied to the embedded loop as well, so no special
2800fork handling is required in that case. When the watcher is not running,
2801however, it is still the task of the libev user to call C<ev_loop_fork ()>
2802as applicable.
2803
2804=head3 Watcher-Specific Functions and Data Members
2805
2806=over 4
2807
2808=item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
2809
2810=item ev_embed_set (ev_embed *, callback, struct ev_loop *embedded_loop)
2811
2812Configures the watcher to embed the given loop, which must be
2813embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
2814invoked automatically, otherwise it is the responsibility of the callback
2815to invoke it (it will continue to be called until the sweep has been done,
2816if you do not want that, you need to temporarily stop the embed watcher).
2817
2818=item ev_embed_sweep (loop, ev_embed *)
2819
2820Make a single, non-blocking sweep over the embedded loop. This works
2821similarly to C<ev_loop (embedded_loop, EVLOOP_NONBLOCK)>, but in the most
2822appropriate way for embedded loops.
2823
2824=item struct ev_loop *other [read-only]
2825
2826The embedded event loop.
2827
2828=back
2829
2830=head3 Examples
2831
2832Example: Try to get an embeddable event loop and embed it into the default
2833event loop. If that is not possible, use the default loop. The default
2834loop is stored in C<loop_hi>, while the embeddable loop is stored in
2835C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
2836used).
2837
2838 struct ev_loop *loop_hi = ev_default_init (0);
2839 struct ev_loop *loop_lo = 0;
2840 ev_embed embed;
2841
2842 // see if there is a chance of getting one that works
2843 // (remember that a flags value of 0 means autodetection)
2844 loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
2845 ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
2846 : 0;
2847
2848 // if we got one, then embed it, otherwise default to loop_hi
2849 if (loop_lo)
2850 {
2851 ev_embed_init (&embed, 0, loop_lo);
2852 ev_embed_start (loop_hi, &embed);
2853 }
2854 else
2855 loop_lo = loop_hi;
2856
2857Example: Check if kqueue is available but not recommended and create
2858a kqueue backend for use with sockets (which usually work with any
2859kqueue implementation). Store the kqueue/socket-only event loop in
2860C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
2861
2862 struct ev_loop *loop = ev_default_init (0);
2863 struct ev_loop *loop_socket = 0;
2864 ev_embed embed;
2865
2866 if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
2867 if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
2868 {
2869 ev_embed_init (&embed, 0, loop_socket);
2870 ev_embed_start (loop, &embed);
2871 }
2872
2873 if (!loop_socket)
2874 loop_socket = loop;
2875
2876 // now use loop_socket for all sockets, and loop for everything else
2877
2878
2879=head2 C<ev_fork> - the audacity to resume the event loop after a fork
2880
2881Fork watchers are called when a C<fork ()> was detected (usually because
2882whoever is a good citizen cared to tell libev about it by calling
2883C<ev_default_fork> or C<ev_loop_fork>). The invocation is done before the
2884event loop blocks next and before C<ev_check> watchers are being called,
2885and only in the child after the fork. If whoever good citizen calling
2886C<ev_default_fork> cheats and calls it in the wrong process, the fork
2887handlers will be invoked, too, of course.
2888
2889=head3 The special problem of life after fork - how is it possible?
2890
2891Most uses of C<fork()> consist of forking, then some simple calls to ste
2892up/change the process environment, followed by a call to C<exec()>. This
2893sequence should be handled by libev without any problems.
2894
2895This changes when the application actually wants to do event handling
2896in the child, or both parent in child, in effect "continuing" after the
2897fork.
2898
2899The default mode of operation (for libev, with application help to detect
2900forks) is to duplicate all the state in the child, as would be expected
2901when I<either> the parent I<or> the child process continues.
2902
2903When both processes want to continue using libev, then this is usually the
2904wrong result. In that case, usually one process (typically the parent) is
2905supposed to continue with all watchers in place as before, while the other
2906process typically wants to start fresh, i.e. without any active watchers.
2907
2908The cleanest and most efficient way to achieve that with libev is to
2909simply create a new event loop, which of course will be "empty", and
2910use that for new watchers. This has the advantage of not touching more
2911memory than necessary, and thus avoiding the copy-on-write, and the
2912disadvantage of having to use multiple event loops (which do not support
2913signal watchers).
2914
2915When this is not possible, or you want to use the default loop for
2916other reasons, then in the process that wants to start "fresh", call
2917C<ev_default_destroy ()> followed by C<ev_default_loop (...)>. Destroying
2918the default loop will "orphan" (not stop) all registered watchers, so you
2919have to be careful not to execute code that modifies those watchers. Note
2920also that in that case, you have to re-register any signal watchers.
2921
2922=head3 Watcher-Specific Functions and Data Members
2923
2924=over 4
2925
2926=item ev_fork_init (ev_signal *, callback)
2927
2928Initialises and configures the fork watcher - it has no parameters of any
2929kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
2930believe me.
2931
2932=back
2933
2934
2935=head2 C<ev_async> - how to wake up another event loop
2936
2937In general, you cannot use an C<ev_loop> from multiple threads or other
2938asynchronous sources such as signal handlers (as opposed to multiple event
2939loops - those are of course safe to use in different threads).
2940
2941Sometimes, however, you need to wake up another event loop you do not
2942control, for example because it belongs to another thread. This is what
2943C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
2944can signal it by calling C<ev_async_send>, which is thread- and signal
2945safe.
2946
2947This functionality is very similar to C<ev_signal> watchers, as signals,
2948too, are asynchronous in nature, and signals, too, will be compressed
2949(i.e. the number of callback invocations may be less than the number of
2950C<ev_async_sent> calls).
2951
2952Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
2953just the default loop.
2954
2955=head3 Queueing
2956
2957C<ev_async> does not support queueing of data in any way. The reason
2958is that the author does not know of a simple (or any) algorithm for a
2959multiple-writer-single-reader queue that works in all cases and doesn't
2960need elaborate support such as pthreads.
2961
2962That means that if you want to queue data, you have to provide your own
2963queue. But at least I can tell you how to implement locking around your
2964queue:
2965
2966=over 4
2967
2968=item queueing from a signal handler context
2969
2970To implement race-free queueing, you simply add to the queue in the signal
2971handler but you block the signal handler in the watcher callback. Here is
2972an example that does that for some fictitious SIGUSR1 handler:
2973
2974 static ev_async mysig;
2975
2976 static void
2977 sigusr1_handler (void)
2978 {
2979 sometype data;
2980
2981 // no locking etc.
2982 queue_put (data);
2983 ev_async_send (EV_DEFAULT_ &mysig);
2984 }
2985
2986 static void
2987 mysig_cb (EV_P_ ev_async *w, int revents)
2988 {
2989 sometype data;
2990 sigset_t block, prev;
2991
2992 sigemptyset (&block);
2993 sigaddset (&block, SIGUSR1);
2994 sigprocmask (SIG_BLOCK, &block, &prev);
2995
2996 while (queue_get (&data))
2997 process (data);
2998
2999 if (sigismember (&prev, SIGUSR1)
3000 sigprocmask (SIG_UNBLOCK, &block, 0);
3001 }
3002
3003(Note: pthreads in theory requires you to use C<pthread_setmask>
3004instead of C<sigprocmask> when you use threads, but libev doesn't do it
3005either...).
3006
3007=item queueing from a thread context
3008
3009The strategy for threads is different, as you cannot (easily) block
3010threads but you can easily preempt them, so to queue safely you need to
3011employ a traditional mutex lock, such as in this pthread example:
3012
3013 static ev_async mysig;
3014 static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3015
3016 static void
3017 otherthread (void)
3018 {
3019 // only need to lock the actual queueing operation
3020 pthread_mutex_lock (&mymutex);
3021 queue_put (data);
3022 pthread_mutex_unlock (&mymutex);
3023
3024 ev_async_send (EV_DEFAULT_ &mysig);
3025 }
3026
3027 static void
3028 mysig_cb (EV_P_ ev_async *w, int revents)
3029 {
3030 pthread_mutex_lock (&mymutex);
3031
3032 while (queue_get (&data))
3033 process (data);
3034
3035 pthread_mutex_unlock (&mymutex);
3036 }
3037
3038=back
3039
3040
3041=head3 Watcher-Specific Functions and Data Members
3042
3043=over 4
3044
3045=item ev_async_init (ev_async *, callback)
3046
3047Initialises and configures the async watcher - it has no parameters of any
3048kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
3049trust me.
3050
3051=item ev_async_send (loop, ev_async *)
3052
3053Sends/signals/activates the given C<ev_async> watcher, that is, feeds
3054an C<EV_ASYNC> event on the watcher into the event loop. Unlike
3055C<ev_feed_event>, this call is safe to do from other threads, signal or
3056similar contexts (see the discussion of C<EV_ATOMIC_T> in the embedding
3057section below on what exactly this means).
3058
3059Note that, as with other watchers in libev, multiple events might get
3060compressed into a single callback invocation (another way to look at this
3061is that C<ev_async> watchers are level-triggered, set on C<ev_async_send>,
3062reset when the event loop detects that).
3063
3064This call incurs the overhead of a system call only once per event loop
3065iteration, so while the overhead might be noticeable, it doesn't apply to
3066repeated calls to C<ev_async_send> for the same event loop.
3067
3068=item bool = ev_async_pending (ev_async *)
3069
3070Returns a non-zero value when C<ev_async_send> has been called on the
3071watcher but the event has not yet been processed (or even noted) by the
3072event loop.
3073
3074C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
3075the loop iterates next and checks for the watcher to have become active,
3076it will reset the flag again. C<ev_async_pending> can be used to very
3077quickly check whether invoking the loop might be a good idea.
3078
3079Not that this does I<not> check whether the watcher itself is pending,
3080only whether it has been requested to make this watcher pending: there
3081is a time window between the event loop checking and resetting the async
3082notification, and the callback being invoked.
3083
3084=back
3085
3086
851=head1 OTHER FUNCTIONS 3087=head1 OTHER FUNCTIONS
852 3088
853There are some other functions of possible interest. Described. Here. Now. 3089There are some other functions of possible interest. Described. Here. Now.
854 3090
855=over 4 3091=over 4
856 3092
857=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback) 3093=item ev_once (loop, int fd, int events, ev_tstamp timeout, callback)
858 3094
859This function combines a simple timer and an I/O watcher, calls your 3095This function combines a simple timer and an I/O watcher, calls your
860callback on whichever event happens first and automatically stop both 3096callback on whichever event happens first and automatically stops both
861watchers. This is useful if you want to wait for a single event on an fd 3097watchers. This is useful if you want to wait for a single event on an fd
862or timeout without having to allocate/configure/start/stop/free one or 3098or timeout without having to allocate/configure/start/stop/free one or
863more watchers yourself. 3099more watchers yourself.
864 3100
865If C<fd> is less than 0, then no I/O watcher will be started and events 3101If C<fd> is less than 0, then no I/O watcher will be started and the
866is being ignored. Otherwise, an C<ev_io> watcher for the given C<fd> and 3102C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
867C<events> set will be craeted and started. 3103the given C<fd> and C<events> set will be created and started.
868 3104
869If C<timeout> is less than 0, then no timeout watcher will be 3105If C<timeout> is less than 0, then no timeout watcher will be
870started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and 3106started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
871repeat = 0) will be started. While C<0> is a valid timeout, it is of 3107repeat = 0) will be started. C<0> is a valid timeout.
872dubious value.
873 3108
874The callback has the type C<void (*cb)(int revents, void *arg)> and gets 3109The callback has the type C<void (*cb)(int revents, void *arg)> and gets
875passed an C<revents> set like normal event callbacks (a combination of 3110passed an C<revents> set like normal event callbacks (a combination of
876C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg> 3111C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMEOUT>) and the C<arg>
877value passed to C<ev_once>: 3112value passed to C<ev_once>. Note that it is possible to receive I<both>
3113a timeout and an io event at the same time - you probably should give io
3114events precedence.
878 3115
3116Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3117
879 static void stdin_ready (int revents, void *arg) 3118 static void stdin_ready (int revents, void *arg)
880 { 3119 {
881 if (revents & EV_TIMEOUT)
882 /* doh, nothing entered */;
883 else if (revents & EV_READ) 3120 if (revents & EV_READ)
884 /* stdin might have data for us, joy! */; 3121 /* stdin might have data for us, joy! */;
3122 else if (revents & EV_TIMEOUT)
3123 /* doh, nothing entered */;
885 } 3124 }
886 3125
887 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0); 3126 ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
888 3127
889=item ev_feed_event (loop, watcher, int events) 3128=item ev_feed_event (struct ev_loop *, watcher *, int revents)
890 3129
891Feeds the given event set into the event loop, as if the specified event 3130Feeds the given event set into the event loop, as if the specified event
892had happened for the specified watcher (which must be a pointer to an 3131had happened for the specified watcher (which must be a pointer to an
893initialised but not necessarily started event watcher). 3132initialised but not necessarily started event watcher).
894 3133
895=item ev_feed_fd_event (loop, int fd, int revents) 3134=item ev_feed_fd_event (struct ev_loop *, int fd, int revents)
896 3135
897Feed an event on the given fd, as if a file descriptor backend detected 3136Feed an event on the given fd, as if a file descriptor backend detected
898the given events it. 3137the given events it.
899 3138
900=item ev_feed_signal_event (loop, int signum) 3139=item ev_feed_signal_event (struct ev_loop *loop, int signum)
901 3140
902Feed an event as if the given signal occured (loop must be the default loop!). 3141Feed an event as if the given signal occurred (C<loop> must be the default
3142loop!).
903 3143
904=back 3144=back
3145
905 3146
906=head1 LIBEVENT EMULATION 3147=head1 LIBEVENT EMULATION
907 3148
908Libev offers a compatibility emulation layer for libevent. It cannot 3149Libev offers a compatibility emulation layer for libevent. It cannot
909emulate the internals of libevent, so here are some usage hints: 3150emulate the internals of libevent, so here are some usage hints:
921 3162
922=item * Priorities are not currently supported. Initialising priorities 3163=item * Priorities are not currently supported. Initialising priorities
923will fail and all watchers will have the same priority, even though there 3164will fail and all watchers will have the same priority, even though there
924is an ev_pri field. 3165is an ev_pri field.
925 3166
3167=item * In libevent, the last base created gets the signals, in libev, the
3168first base created (== the default loop) gets the signals.
3169
926=item * Other members are not supported. 3170=item * Other members are not supported.
927 3171
928=item * The libev emulation is I<not> ABI compatible to libevent, you need 3172=item * The libev emulation is I<not> ABI compatible to libevent, you need
929to use the libev header file and library. 3173to use the libev header file and library.
930 3174
931=back 3175=back
932 3176
933=head1 C++ SUPPORT 3177=head1 C++ SUPPORT
934 3178
935TBD. 3179Libev comes with some simplistic wrapper classes for C++ that mainly allow
3180you to use some convenience methods to start/stop watchers and also change
3181the callback model to a model using method callbacks on objects.
3182
3183To use it,
3184
3185 #include <ev++.h>
3186
3187This automatically includes F<ev.h> and puts all of its definitions (many
3188of them macros) into the global namespace. All C++ specific things are
3189put into the C<ev> namespace. It should support all the same embedding
3190options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
3191
3192Care has been taken to keep the overhead low. The only data member the C++
3193classes add (compared to plain C-style watchers) is the event loop pointer
3194that the watcher is associated with (or no additional members at all if
3195you disable C<EV_MULTIPLICITY> when embedding libev).
3196
3197Currently, functions, and static and non-static member functions can be
3198used as callbacks. Other types should be easy to add as long as they only
3199need one additional pointer for context. If you need support for other
3200types of functors please contact the author (preferably after implementing
3201it).
3202
3203Here is a list of things available in the C<ev> namespace:
3204
3205=over 4
3206
3207=item C<ev::READ>, C<ev::WRITE> etc.
3208
3209These are just enum values with the same values as the C<EV_READ> etc.
3210macros from F<ev.h>.
3211
3212=item C<ev::tstamp>, C<ev::now>
3213
3214Aliases to the same types/functions as with the C<ev_> prefix.
3215
3216=item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
3217
3218For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
3219the same name in the C<ev> namespace, with the exception of C<ev_signal>
3220which is called C<ev::sig> to avoid clashes with the C<signal> macro
3221defines by many implementations.
3222
3223All of those classes have these methods:
3224
3225=over 4
3226
3227=item ev::TYPE::TYPE ()
3228
3229=item ev::TYPE::TYPE (struct ev_loop *)
3230
3231=item ev::TYPE::~TYPE
3232
3233The constructor (optionally) takes an event loop to associate the watcher
3234with. If it is omitted, it will use C<EV_DEFAULT>.
3235
3236The constructor calls C<ev_init> for you, which means you have to call the
3237C<set> method before starting it.
3238
3239It will not set a callback, however: You have to call the templated C<set>
3240method to set a callback before you can start the watcher.
3241
3242(The reason why you have to use a method is a limitation in C++ which does
3243not allow explicit template arguments for constructors).
3244
3245The destructor automatically stops the watcher if it is active.
3246
3247=item w->set<class, &class::method> (object *)
3248
3249This method sets the callback method to call. The method has to have a
3250signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
3251first argument and the C<revents> as second. The object must be given as
3252parameter and is stored in the C<data> member of the watcher.
3253
3254This method synthesizes efficient thunking code to call your method from
3255the C callback that libev requires. If your compiler can inline your
3256callback (i.e. it is visible to it at the place of the C<set> call and
3257your compiler is good :), then the method will be fully inlined into the
3258thunking function, making it as fast as a direct C callback.
3259
3260Example: simple class declaration and watcher initialisation
3261
3262 struct myclass
3263 {
3264 void io_cb (ev::io &w, int revents) { }
3265 }
3266
3267 myclass obj;
3268 ev::io iow;
3269 iow.set <myclass, &myclass::io_cb> (&obj);
3270
3271=item w->set (object *)
3272
3273This is an B<experimental> feature that might go away in a future version.
3274
3275This is a variation of a method callback - leaving out the method to call
3276will default the method to C<operator ()>, which makes it possible to use
3277functor objects without having to manually specify the C<operator ()> all
3278the time. Incidentally, you can then also leave out the template argument
3279list.
3280
3281The C<operator ()> method prototype must be C<void operator ()(watcher &w,
3282int revents)>.
3283
3284See the method-C<set> above for more details.
3285
3286Example: use a functor object as callback.
3287
3288 struct myfunctor
3289 {
3290 void operator() (ev::io &w, int revents)
3291 {
3292 ...
3293 }
3294 }
3295
3296 myfunctor f;
3297
3298 ev::io w;
3299 w.set (&f);
3300
3301=item w->set<function> (void *data = 0)
3302
3303Also sets a callback, but uses a static method or plain function as
3304callback. The optional C<data> argument will be stored in the watcher's
3305C<data> member and is free for you to use.
3306
3307The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
3308
3309See the method-C<set> above for more details.
3310
3311Example: Use a plain function as callback.
3312
3313 static void io_cb (ev::io &w, int revents) { }
3314 iow.set <io_cb> ();
3315
3316=item w->set (struct ev_loop *)
3317
3318Associates a different C<struct ev_loop> with this watcher. You can only
3319do this when the watcher is inactive (and not pending either).
3320
3321=item w->set ([arguments])
3322
3323Basically the same as C<ev_TYPE_set>, with the same arguments. Must be
3324called at least once. Unlike the C counterpart, an active watcher gets
3325automatically stopped and restarted when reconfiguring it with this
3326method.
3327
3328=item w->start ()
3329
3330Starts the watcher. Note that there is no C<loop> argument, as the
3331constructor already stores the event loop.
3332
3333=item w->stop ()
3334
3335Stops the watcher if it is active. Again, no C<loop> argument.
3336
3337=item w->again () (C<ev::timer>, C<ev::periodic> only)
3338
3339For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
3340C<ev_TYPE_again> function.
3341
3342=item w->sweep () (C<ev::embed> only)
3343
3344Invokes C<ev_embed_sweep>.
3345
3346=item w->update () (C<ev::stat> only)
3347
3348Invokes C<ev_stat_stat>.
3349
3350=back
3351
3352=back
3353
3354Example: Define a class with an IO and idle watcher, start one of them in
3355the constructor.
3356
3357 class myclass
3358 {
3359 ev::io io ; void io_cb (ev::io &w, int revents);
3360 ev::idle idle; void idle_cb (ev::idle &w, int revents);
3361
3362 myclass (int fd)
3363 {
3364 io .set <myclass, &myclass::io_cb > (this);
3365 idle.set <myclass, &myclass::idle_cb> (this);
3366
3367 io.start (fd, ev::READ);
3368 }
3369 };
3370
3371
3372=head1 OTHER LANGUAGE BINDINGS
3373
3374Libev does not offer other language bindings itself, but bindings for a
3375number of languages exist in the form of third-party packages. If you know
3376any interesting language binding in addition to the ones listed here, drop
3377me a note.
3378
3379=over 4
3380
3381=item Perl
3382
3383The EV module implements the full libev API and is actually used to test
3384libev. EV is developed together with libev. Apart from the EV core module,
3385there are additional modules that implement libev-compatible interfaces
3386to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
3387C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
3388and C<EV::Glib>).
3389
3390It can be found and installed via CPAN, its homepage is at
3391L<http://software.schmorp.de/pkg/EV>.
3392
3393=item Python
3394
3395Python bindings can be found at L<http://code.google.com/p/pyev/>. It
3396seems to be quite complete and well-documented.
3397
3398=item Ruby
3399
3400Tony Arcieri has written a ruby extension that offers access to a subset
3401of the libev API and adds file handle abstractions, asynchronous DNS and
3402more on top of it. It can be found via gem servers. Its homepage is at
3403L<http://rev.rubyforge.org/>.
3404
3405Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
3406makes rev work even on mingw.
3407
3408=item Haskell
3409
3410A haskell binding to libev is available at
3411L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
3412
3413=item D
3414
3415Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
3416be found at L<http://proj.llucax.com.ar/wiki/evd>.
3417
3418=item Ocaml
3419
3420Erkki Seppala has written Ocaml bindings for libev, to be found at
3421L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
3422
3423=item Lua
3424
3425Brian Maher has written a partial interface to libev
3426for lua (only C<ev_io> and C<ev_timer>), to be found at
3427L<http://github.com/brimworks/lua-ev>.
3428
3429=back
3430
3431
3432=head1 MACRO MAGIC
3433
3434Libev can be compiled with a variety of options, the most fundamental
3435of which is C<EV_MULTIPLICITY>. This option determines whether (most)
3436functions and callbacks have an initial C<struct ev_loop *> argument.
3437
3438To make it easier to write programs that cope with either variant, the
3439following macros are defined:
3440
3441=over 4
3442
3443=item C<EV_A>, C<EV_A_>
3444
3445This provides the loop I<argument> for functions, if one is required ("ev
3446loop argument"). The C<EV_A> form is used when this is the sole argument,
3447C<EV_A_> is used when other arguments are following. Example:
3448
3449 ev_unref (EV_A);
3450 ev_timer_add (EV_A_ watcher);
3451 ev_loop (EV_A_ 0);
3452
3453It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
3454which is often provided by the following macro.
3455
3456=item C<EV_P>, C<EV_P_>
3457
3458This provides the loop I<parameter> for functions, if one is required ("ev
3459loop parameter"). The C<EV_P> form is used when this is the sole parameter,
3460C<EV_P_> is used when other parameters are following. Example:
3461
3462 // this is how ev_unref is being declared
3463 static void ev_unref (EV_P);
3464
3465 // this is how you can declare your typical callback
3466 static void cb (EV_P_ ev_timer *w, int revents)
3467
3468It declares a parameter C<loop> of type C<struct ev_loop *>, quite
3469suitable for use with C<EV_A>.
3470
3471=item C<EV_DEFAULT>, C<EV_DEFAULT_>
3472
3473Similar to the other two macros, this gives you the value of the default
3474loop, if multiple loops are supported ("ev loop default").
3475
3476=item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
3477
3478Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
3479default loop has been initialised (C<UC> == unchecked). Their behaviour
3480is undefined when the default loop has not been initialised by a previous
3481execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
3482
3483It is often prudent to use C<EV_DEFAULT> when initialising the first
3484watcher in a function but use C<EV_DEFAULT_UC> afterwards.
3485
3486=back
3487
3488Example: Declare and initialise a check watcher, utilising the above
3489macros so it will work regardless of whether multiple loops are supported
3490or not.
3491
3492 static void
3493 check_cb (EV_P_ ev_timer *w, int revents)
3494 {
3495 ev_check_stop (EV_A_ w);
3496 }
3497
3498 ev_check check;
3499 ev_check_init (&check, check_cb);
3500 ev_check_start (EV_DEFAULT_ &check);
3501 ev_loop (EV_DEFAULT_ 0);
3502
3503=head1 EMBEDDING
3504
3505Libev can (and often is) directly embedded into host
3506applications. Examples of applications that embed it include the Deliantra
3507Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
3508and rxvt-unicode.
3509
3510The goal is to enable you to just copy the necessary files into your
3511source directory without having to change even a single line in them, so
3512you can easily upgrade by simply copying (or having a checked-out copy of
3513libev somewhere in your source tree).
3514
3515=head2 FILESETS
3516
3517Depending on what features you need you need to include one or more sets of files
3518in your application.
3519
3520=head3 CORE EVENT LOOP
3521
3522To include only the libev core (all the C<ev_*> functions), with manual
3523configuration (no autoconf):
3524
3525 #define EV_STANDALONE 1
3526 #include "ev.c"
3527
3528This will automatically include F<ev.h>, too, and should be done in a
3529single C source file only to provide the function implementations. To use
3530it, do the same for F<ev.h> in all files wishing to use this API (best
3531done by writing a wrapper around F<ev.h> that you can include instead and
3532where you can put other configuration options):
3533
3534 #define EV_STANDALONE 1
3535 #include "ev.h"
3536
3537Both header files and implementation files can be compiled with a C++
3538compiler (at least, that's a stated goal, and breakage will be treated
3539as a bug).
3540
3541You need the following files in your source tree, or in a directory
3542in your include path (e.g. in libev/ when using -Ilibev):
3543
3544 ev.h
3545 ev.c
3546 ev_vars.h
3547 ev_wrap.h
3548
3549 ev_win32.c required on win32 platforms only
3550
3551 ev_select.c only when select backend is enabled (which is enabled by default)
3552 ev_poll.c only when poll backend is enabled (disabled by default)
3553 ev_epoll.c only when the epoll backend is enabled (disabled by default)
3554 ev_kqueue.c only when the kqueue backend is enabled (disabled by default)
3555 ev_port.c only when the solaris port backend is enabled (disabled by default)
3556
3557F<ev.c> includes the backend files directly when enabled, so you only need
3558to compile this single file.
3559
3560=head3 LIBEVENT COMPATIBILITY API
3561
3562To include the libevent compatibility API, also include:
3563
3564 #include "event.c"
3565
3566in the file including F<ev.c>, and:
3567
3568 #include "event.h"
3569
3570in the files that want to use the libevent API. This also includes F<ev.h>.
3571
3572You need the following additional files for this:
3573
3574 event.h
3575 event.c
3576
3577=head3 AUTOCONF SUPPORT
3578
3579Instead of using C<EV_STANDALONE=1> and providing your configuration in
3580whatever way you want, you can also C<m4_include([libev.m4])> in your
3581F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
3582include F<config.h> and configure itself accordingly.
3583
3584For this of course you need the m4 file:
3585
3586 libev.m4
3587
3588=head2 PREPROCESSOR SYMBOLS/MACROS
3589
3590Libev can be configured via a variety of preprocessor symbols you have to
3591define before including any of its files. The default in the absence of
3592autoconf is documented for every option.
3593
3594=over 4
3595
3596=item EV_STANDALONE
3597
3598Must always be C<1> if you do not use autoconf configuration, which
3599keeps libev from including F<config.h>, and it also defines dummy
3600implementations for some libevent functions (such as logging, which is not
3601supported). It will also not define any of the structs usually found in
3602F<event.h> that are not directly supported by the libev core alone.
3603
3604In standalone mode, libev will still try to automatically deduce the
3605configuration, but has to be more conservative.
3606
3607=item EV_USE_MONOTONIC
3608
3609If defined to be C<1>, libev will try to detect the availability of the
3610monotonic clock option at both compile time and runtime. Otherwise no
3611use of the monotonic clock option will be attempted. If you enable this,
3612you usually have to link against librt or something similar. Enabling it
3613when the functionality isn't available is safe, though, although you have
3614to make sure you link against any libraries where the C<clock_gettime>
3615function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
3616
3617=item EV_USE_REALTIME
3618
3619If defined to be C<1>, libev will try to detect the availability of the
3620real-time clock option at compile time (and assume its availability
3621at runtime if successful). Otherwise no use of the real-time clock
3622option will be attempted. This effectively replaces C<gettimeofday>
3623by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
3624correctness. See the note about libraries in the description of
3625C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
3626C<EV_USE_CLOCK_SYSCALL>.
3627
3628=item EV_USE_CLOCK_SYSCALL
3629
3630If defined to be C<1>, libev will try to use a direct syscall instead
3631of calling the system-provided C<clock_gettime> function. This option
3632exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
3633unconditionally pulls in C<libpthread>, slowing down single-threaded
3634programs needlessly. Using a direct syscall is slightly slower (in
3635theory), because no optimised vdso implementation can be used, but avoids
3636the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
3637higher, as it simplifies linking (no need for C<-lrt>).
3638
3639=item EV_USE_NANOSLEEP
3640
3641If defined to be C<1>, libev will assume that C<nanosleep ()> is available
3642and will use it for delays. Otherwise it will use C<select ()>.
3643
3644=item EV_USE_EVENTFD
3645
3646If defined to be C<1>, then libev will assume that C<eventfd ()> is
3647available and will probe for kernel support at runtime. This will improve
3648C<ev_signal> and C<ev_async> performance and reduce resource consumption.
3649If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
36502.7 or newer, otherwise disabled.
3651
3652=item EV_USE_SELECT
3653
3654If undefined or defined to be C<1>, libev will compile in support for the
3655C<select>(2) backend. No attempt at auto-detection will be done: if no
3656other method takes over, select will be it. Otherwise the select backend
3657will not be compiled in.
3658
3659=item EV_SELECT_USE_FD_SET
3660
3661If defined to C<1>, then the select backend will use the system C<fd_set>
3662structure. This is useful if libev doesn't compile due to a missing
3663C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
3664on exotic systems. This usually limits the range of file descriptors to
3665some low limit such as 1024 or might have other limitations (winsocket
3666only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
3667configures the maximum size of the C<fd_set>.
3668
3669=item EV_SELECT_IS_WINSOCKET
3670
3671When defined to C<1>, the select backend will assume that
3672select/socket/connect etc. don't understand file descriptors but
3673wants osf handles on win32 (this is the case when the select to
3674be used is the winsock select). This means that it will call
3675C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
3676it is assumed that all these functions actually work on fds, even
3677on win32. Should not be defined on non-win32 platforms.
3678
3679=item EV_FD_TO_WIN32_HANDLE(fd)
3680
3681If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
3682file descriptors to socket handles. When not defining this symbol (the
3683default), then libev will call C<_get_osfhandle>, which is usually
3684correct. In some cases, programs use their own file descriptor management,
3685in which case they can provide this function to map fds to socket handles.
3686
3687=item EV_WIN32_HANDLE_TO_FD(handle)
3688
3689If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
3690using the standard C<_open_osfhandle> function. For programs implementing
3691their own fd to handle mapping, overwriting this function makes it easier
3692to do so. This can be done by defining this macro to an appropriate value.
3693
3694=item EV_WIN32_CLOSE_FD(fd)
3695
3696If programs implement their own fd to handle mapping on win32, then this
3697macro can be used to override the C<close> function, useful to unregister
3698file descriptors again. Note that the replacement function has to close
3699the underlying OS handle.
3700
3701=item EV_USE_POLL
3702
3703If defined to be C<1>, libev will compile in support for the C<poll>(2)
3704backend. Otherwise it will be enabled on non-win32 platforms. It
3705takes precedence over select.
3706
3707=item EV_USE_EPOLL
3708
3709If defined to be C<1>, libev will compile in support for the Linux
3710C<epoll>(7) backend. Its availability will be detected at runtime,
3711otherwise another method will be used as fallback. This is the preferred
3712backend for GNU/Linux systems. If undefined, it will be enabled if the
3713headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3714
3715=item EV_USE_KQUEUE
3716
3717If defined to be C<1>, libev will compile in support for the BSD style
3718C<kqueue>(2) backend. Its actual availability will be detected at runtime,
3719otherwise another method will be used as fallback. This is the preferred
3720backend for BSD and BSD-like systems, although on most BSDs kqueue only
3721supports some types of fds correctly (the only platform we found that
3722supports ptys for example was NetBSD), so kqueue might be compiled in, but
3723not be used unless explicitly requested. The best way to use it is to find
3724out whether kqueue supports your type of fd properly and use an embedded
3725kqueue loop.
3726
3727=item EV_USE_PORT
3728
3729If defined to be C<1>, libev will compile in support for the Solaris
373010 port style backend. Its availability will be detected at runtime,
3731otherwise another method will be used as fallback. This is the preferred
3732backend for Solaris 10 systems.
3733
3734=item EV_USE_DEVPOLL
3735
3736Reserved for future expansion, works like the USE symbols above.
3737
3738=item EV_USE_INOTIFY
3739
3740If defined to be C<1>, libev will compile in support for the Linux inotify
3741interface to speed up C<ev_stat> watchers. Its actual availability will
3742be detected at runtime. If undefined, it will be enabled if the headers
3743indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
3744
3745=item EV_ATOMIC_T
3746
3747Libev requires an integer type (suitable for storing C<0> or C<1>) whose
3748access is atomic with respect to other threads or signal contexts. No such
3749type is easily found in the C language, so you can provide your own type
3750that you know is safe for your purposes. It is used both for signal handler "locking"
3751as well as for signal and thread safety in C<ev_async> watchers.
3752
3753In the absence of this define, libev will use C<sig_atomic_t volatile>
3754(from F<signal.h>), which is usually good enough on most platforms.
3755
3756=item EV_H
3757
3758The name of the F<ev.h> header file used to include it. The default if
3759undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
3760used to virtually rename the F<ev.h> header file in case of conflicts.
3761
3762=item EV_CONFIG_H
3763
3764If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
3765F<ev.c>'s idea of where to find the F<config.h> file, similarly to
3766C<EV_H>, above.
3767
3768=item EV_EVENT_H
3769
3770Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
3771of how the F<event.h> header can be found, the default is C<"event.h">.
3772
3773=item EV_PROTOTYPES
3774
3775If defined to be C<0>, then F<ev.h> will not define any function
3776prototypes, but still define all the structs and other symbols. This is
3777occasionally useful if you want to provide your own wrapper functions
3778around libev functions.
3779
3780=item EV_MULTIPLICITY
3781
3782If undefined or defined to C<1>, then all event-loop-specific functions
3783will have the C<struct ev_loop *> as first argument, and you can create
3784additional independent event loops. Otherwise there will be no support
3785for multiple event loops and there is no first event loop pointer
3786argument. Instead, all functions act on the single default loop.
3787
3788=item EV_MINPRI
3789
3790=item EV_MAXPRI
3791
3792The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
3793C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
3794provide for more priorities by overriding those symbols (usually defined
3795to be C<-2> and C<2>, respectively).
3796
3797When doing priority-based operations, libev usually has to linearly search
3798all the priorities, so having many of them (hundreds) uses a lot of space
3799and time, so using the defaults of five priorities (-2 .. +2) is usually
3800fine.
3801
3802If your embedding application does not need any priorities, defining these
3803both to C<0> will save some memory and CPU.
3804
3805=item EV_PERIODIC_ENABLE
3806
3807If undefined or defined to be C<1>, then periodic timers are supported. If
3808defined to be C<0>, then they are not. Disabling them saves a few kB of
3809code.
3810
3811=item EV_IDLE_ENABLE
3812
3813If undefined or defined to be C<1>, then idle watchers are supported. If
3814defined to be C<0>, then they are not. Disabling them saves a few kB of
3815code.
3816
3817=item EV_EMBED_ENABLE
3818
3819If undefined or defined to be C<1>, then embed watchers are supported. If
3820defined to be C<0>, then they are not. Embed watchers rely on most other
3821watcher types, which therefore must not be disabled.
3822
3823=item EV_STAT_ENABLE
3824
3825If undefined or defined to be C<1>, then stat watchers are supported. If
3826defined to be C<0>, then they are not.
3827
3828=item EV_FORK_ENABLE
3829
3830If undefined or defined to be C<1>, then fork watchers are supported. If
3831defined to be C<0>, then they are not.
3832
3833=item EV_ASYNC_ENABLE
3834
3835If undefined or defined to be C<1>, then async watchers are supported. If
3836defined to be C<0>, then they are not.
3837
3838=item EV_MINIMAL
3839
3840If you need to shave off some kilobytes of code at the expense of some
3841speed (but with the full API), define this symbol to C<1>. Currently this
3842is used to override some inlining decisions, saves roughly 30% code size
3843on amd64. It also selects a much smaller 2-heap for timer management over
3844the default 4-heap.
3845
3846You can save even more by disabling watcher types you do not need
3847and setting C<EV_MAXPRI> == C<EV_MINPRI>. Also, disabling C<assert>
3848(C<-DNDEBUG>) will usually reduce code size a lot.
3849
3850Defining C<EV_MINIMAL> to C<2> will additionally reduce the core API to
3851provide a bare-bones event library. See C<ev.h> for details on what parts
3852of the API are still available, and do not complain if this subset changes
3853over time.
3854
3855=item EV_NSIG
3856
3857The highest supported signal number, +1 (or, the number of
3858signals): Normally, libev tries to deduce the maximum number of signals
3859automatically, but sometimes this fails, in which case it can be
3860specified. Also, using a lower number than detected (C<32> should be
3861good for about any system in existance) can save some memory, as libev
3862statically allocates some 12-24 bytes per signal number.
3863
3864=item EV_PID_HASHSIZE
3865
3866C<ev_child> watchers use a small hash table to distribute workload by
3867pid. The default size is C<16> (or C<1> with C<EV_MINIMAL>), usually more
3868than enough. If you need to manage thousands of children you might want to
3869increase this value (I<must> be a power of two).
3870
3871=item EV_INOTIFY_HASHSIZE
3872
3873C<ev_stat> watchers use a small hash table to distribute workload by
3874inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
3875usually more than enough. If you need to manage thousands of C<ev_stat>
3876watchers you might want to increase this value (I<must> be a power of
3877two).
3878
3879=item EV_USE_4HEAP
3880
3881Heaps are not very cache-efficient. To improve the cache-efficiency of the
3882timer and periodics heaps, libev uses a 4-heap when this symbol is defined
3883to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
3884faster performance with many (thousands) of watchers.
3885
3886The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3887(disabled).
3888
3889=item EV_HEAP_CACHE_AT
3890
3891Heaps are not very cache-efficient. To improve the cache-efficiency of the
3892timer and periodics heaps, libev can cache the timestamp (I<at>) within
3893the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
3894which uses 8-12 bytes more per watcher and a few hundred bytes more code,
3895but avoids random read accesses on heap changes. This improves performance
3896noticeably with many (hundreds) of watchers.
3897
3898The default is C<1> unless C<EV_MINIMAL> is set in which case it is C<0>
3899(disabled).
3900
3901=item EV_VERIFY
3902
3903Controls how much internal verification (see C<ev_loop_verify ()>) will
3904be done: If set to C<0>, no internal verification code will be compiled
3905in. If set to C<1>, then verification code will be compiled in, but not
3906called. If set to C<2>, then the internal verification code will be
3907called once per loop, which can slow down libev. If set to C<3>, then the
3908verification code will be called very frequently, which will slow down
3909libev considerably.
3910
3911The default is C<1>, unless C<EV_MINIMAL> is set, in which case it will be
3912C<0>.
3913
3914=item EV_COMMON
3915
3916By default, all watchers have a C<void *data> member. By redefining
3917this macro to a something else you can include more and other types of
3918members. You have to define it each time you include one of the files,
3919though, and it must be identical each time.
3920
3921For example, the perl EV module uses something like this:
3922
3923 #define EV_COMMON \
3924 SV *self; /* contains this struct */ \
3925 SV *cb_sv, *fh /* note no trailing ";" */
3926
3927=item EV_CB_DECLARE (type)
3928
3929=item EV_CB_INVOKE (watcher, revents)
3930
3931=item ev_set_cb (ev, cb)
3932
3933Can be used to change the callback member declaration in each watcher,
3934and the way callbacks are invoked and set. Must expand to a struct member
3935definition and a statement, respectively. See the F<ev.h> header file for
3936their default definitions. One possible use for overriding these is to
3937avoid the C<struct ev_loop *> as first argument in all cases, or to use
3938method calls instead of plain function calls in C++.
3939
3940=back
3941
3942=head2 EXPORTED API SYMBOLS
3943
3944If you need to re-export the API (e.g. via a DLL) and you need a list of
3945exported symbols, you can use the provided F<Symbol.*> files which list
3946all public symbols, one per line:
3947
3948 Symbols.ev for libev proper
3949 Symbols.event for the libevent emulation
3950
3951This can also be used to rename all public symbols to avoid clashes with
3952multiple versions of libev linked together (which is obviously bad in
3953itself, but sometimes it is inconvenient to avoid this).
3954
3955A sed command like this will create wrapper C<#define>'s that you need to
3956include before including F<ev.h>:
3957
3958 <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
3959
3960This would create a file F<wrap.h> which essentially looks like this:
3961
3962 #define ev_backend myprefix_ev_backend
3963 #define ev_check_start myprefix_ev_check_start
3964 #define ev_check_stop myprefix_ev_check_stop
3965 ...
3966
3967=head2 EXAMPLES
3968
3969For a real-world example of a program the includes libev
3970verbatim, you can have a look at the EV perl module
3971(L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
3972the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
3973interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
3974will be compiled. It is pretty complex because it provides its own header
3975file.
3976
3977The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
3978that everybody includes and which overrides some configure choices:
3979
3980 #define EV_MINIMAL 1
3981 #define EV_USE_POLL 0
3982 #define EV_MULTIPLICITY 0
3983 #define EV_PERIODIC_ENABLE 0
3984 #define EV_STAT_ENABLE 0
3985 #define EV_FORK_ENABLE 0
3986 #define EV_CONFIG_H <config.h>
3987 #define EV_MINPRI 0
3988 #define EV_MAXPRI 0
3989
3990 #include "ev++.h"
3991
3992And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
3993
3994 #include "ev_cpp.h"
3995 #include "ev.c"
3996
3997=head1 INTERACTION WITH OTHER PROGRAMS OR LIBRARIES
3998
3999=head2 THREADS AND COROUTINES
4000
4001=head3 THREADS
4002
4003All libev functions are reentrant and thread-safe unless explicitly
4004documented otherwise, but libev implements no locking itself. This means
4005that you can use as many loops as you want in parallel, as long as there
4006are no concurrent calls into any libev function with the same loop
4007parameter (C<ev_default_*> calls have an implicit default loop parameter,
4008of course): libev guarantees that different event loops share no data
4009structures that need any locking.
4010
4011Or to put it differently: calls with different loop parameters can be done
4012concurrently from multiple threads, calls with the same loop parameter
4013must be done serially (but can be done from different threads, as long as
4014only one thread ever is inside a call at any point in time, e.g. by using
4015a mutex per loop).
4016
4017Specifically to support threads (and signal handlers), libev implements
4018so-called C<ev_async> watchers, which allow some limited form of
4019concurrency on the same event loop, namely waking it up "from the
4020outside".
4021
4022If you want to know which design (one loop, locking, or multiple loops
4023without or something else still) is best for your problem, then I cannot
4024help you, but here is some generic advice:
4025
4026=over 4
4027
4028=item * most applications have a main thread: use the default libev loop
4029in that thread, or create a separate thread running only the default loop.
4030
4031This helps integrating other libraries or software modules that use libev
4032themselves and don't care/know about threading.
4033
4034=item * one loop per thread is usually a good model.
4035
4036Doing this is almost never wrong, sometimes a better-performance model
4037exists, but it is always a good start.
4038
4039=item * other models exist, such as the leader/follower pattern, where one
4040loop is handed through multiple threads in a kind of round-robin fashion.
4041
4042Choosing a model is hard - look around, learn, know that usually you can do
4043better than you currently do :-)
4044
4045=item * often you need to talk to some other thread which blocks in the
4046event loop.
4047
4048C<ev_async> watchers can be used to wake them up from other threads safely
4049(or from signal contexts...).
4050
4051An example use would be to communicate signals or other events that only
4052work in the default loop by registering the signal watcher with the
4053default loop and triggering an C<ev_async> watcher from the default loop
4054watcher callback into the event loop interested in the signal.
4055
4056=back
4057
4058=head4 THREAD LOCKING EXAMPLE
4059
4060Here is a fictitious example of how to run an event loop in a different
4061thread than where callbacks are being invoked and watchers are
4062created/added/removed.
4063
4064For a real-world example, see the C<EV::Loop::Async> perl module,
4065which uses exactly this technique (which is suited for many high-level
4066languages).
4067
4068The example uses a pthread mutex to protect the loop data, a condition
4069variable to wait for callback invocations, an async watcher to notify the
4070event loop thread and an unspecified mechanism to wake up the main thread.
4071
4072First, you need to associate some data with the event loop:
4073
4074 typedef struct {
4075 mutex_t lock; /* global loop lock */
4076 ev_async async_w;
4077 thread_t tid;
4078 cond_t invoke_cv;
4079 } userdata;
4080
4081 void prepare_loop (EV_P)
4082 {
4083 // for simplicity, we use a static userdata struct.
4084 static userdata u;
4085
4086 ev_async_init (&u->async_w, async_cb);
4087 ev_async_start (EV_A_ &u->async_w);
4088
4089 pthread_mutex_init (&u->lock, 0);
4090 pthread_cond_init (&u->invoke_cv, 0);
4091
4092 // now associate this with the loop
4093 ev_set_userdata (EV_A_ u);
4094 ev_set_invoke_pending_cb (EV_A_ l_invoke);
4095 ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
4096
4097 // then create the thread running ev_loop
4098 pthread_create (&u->tid, 0, l_run, EV_A);
4099 }
4100
4101The callback for the C<ev_async> watcher does nothing: the watcher is used
4102solely to wake up the event loop so it takes notice of any new watchers
4103that might have been added:
4104
4105 static void
4106 async_cb (EV_P_ ev_async *w, int revents)
4107 {
4108 // just used for the side effects
4109 }
4110
4111The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
4112protecting the loop data, respectively.
4113
4114 static void
4115 l_release (EV_P)
4116 {
4117 userdata *u = ev_userdata (EV_A);
4118 pthread_mutex_unlock (&u->lock);
4119 }
4120
4121 static void
4122 l_acquire (EV_P)
4123 {
4124 userdata *u = ev_userdata (EV_A);
4125 pthread_mutex_lock (&u->lock);
4126 }
4127
4128The event loop thread first acquires the mutex, and then jumps straight
4129into C<ev_loop>:
4130
4131 void *
4132 l_run (void *thr_arg)
4133 {
4134 struct ev_loop *loop = (struct ev_loop *)thr_arg;
4135
4136 l_acquire (EV_A);
4137 pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
4138 ev_loop (EV_A_ 0);
4139 l_release (EV_A);
4140
4141 return 0;
4142 }
4143
4144Instead of invoking all pending watchers, the C<l_invoke> callback will
4145signal the main thread via some unspecified mechanism (signals? pipe
4146writes? C<Async::Interrupt>?) and then waits until all pending watchers
4147have been called (in a while loop because a) spurious wakeups are possible
4148and b) skipping inter-thread-communication when there are no pending
4149watchers is very beneficial):
4150
4151 static void
4152 l_invoke (EV_P)
4153 {
4154 userdata *u = ev_userdata (EV_A);
4155
4156 while (ev_pending_count (EV_A))
4157 {
4158 wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
4159 pthread_cond_wait (&u->invoke_cv, &u->lock);
4160 }
4161 }
4162
4163Now, whenever the main thread gets told to invoke pending watchers, it
4164will grab the lock, call C<ev_invoke_pending> and then signal the loop
4165thread to continue:
4166
4167 static void
4168 real_invoke_pending (EV_P)
4169 {
4170 userdata *u = ev_userdata (EV_A);
4171
4172 pthread_mutex_lock (&u->lock);
4173 ev_invoke_pending (EV_A);
4174 pthread_cond_signal (&u->invoke_cv);
4175 pthread_mutex_unlock (&u->lock);
4176 }
4177
4178Whenever you want to start/stop a watcher or do other modifications to an
4179event loop, you will now have to lock:
4180
4181 ev_timer timeout_watcher;
4182 userdata *u = ev_userdata (EV_A);
4183
4184 ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
4185
4186 pthread_mutex_lock (&u->lock);
4187 ev_timer_start (EV_A_ &timeout_watcher);
4188 ev_async_send (EV_A_ &u->async_w);
4189 pthread_mutex_unlock (&u->lock);
4190
4191Note that sending the C<ev_async> watcher is required because otherwise
4192an event loop currently blocking in the kernel will have no knowledge
4193about the newly added timer. By waking up the loop it will pick up any new
4194watchers in the next event loop iteration.
4195
4196=head3 COROUTINES
4197
4198Libev is very accommodating to coroutines ("cooperative threads"):
4199libev fully supports nesting calls to its functions from different
4200coroutines (e.g. you can call C<ev_loop> on the same loop from two
4201different coroutines, and switch freely between both coroutines running
4202the loop, as long as you don't confuse yourself). The only exception is
4203that you must not do this from C<ev_periodic> reschedule callbacks.
4204
4205Care has been taken to ensure that libev does not keep local state inside
4206C<ev_loop>, and other calls do not usually allow for coroutine switches as
4207they do not call any callbacks.
4208
4209=head2 COMPILER WARNINGS
4210
4211Depending on your compiler and compiler settings, you might get no or a
4212lot of warnings when compiling libev code. Some people are apparently
4213scared by this.
4214
4215However, these are unavoidable for many reasons. For one, each compiler
4216has different warnings, and each user has different tastes regarding
4217warning options. "Warn-free" code therefore cannot be a goal except when
4218targeting a specific compiler and compiler-version.
4219
4220Another reason is that some compiler warnings require elaborate
4221workarounds, or other changes to the code that make it less clear and less
4222maintainable.
4223
4224And of course, some compiler warnings are just plain stupid, or simply
4225wrong (because they don't actually warn about the condition their message
4226seems to warn about). For example, certain older gcc versions had some
4227warnings that resulted an extreme number of false positives. These have
4228been fixed, but some people still insist on making code warn-free with
4229such buggy versions.
4230
4231While libev is written to generate as few warnings as possible,
4232"warn-free" code is not a goal, and it is recommended not to build libev
4233with any compiler warnings enabled unless you are prepared to cope with
4234them (e.g. by ignoring them). Remember that warnings are just that:
4235warnings, not errors, or proof of bugs.
4236
4237
4238=head2 VALGRIND
4239
4240Valgrind has a special section here because it is a popular tool that is
4241highly useful. Unfortunately, valgrind reports are very hard to interpret.
4242
4243If you think you found a bug (memory leak, uninitialised data access etc.)
4244in libev, then check twice: If valgrind reports something like:
4245
4246 ==2274== definitely lost: 0 bytes in 0 blocks.
4247 ==2274== possibly lost: 0 bytes in 0 blocks.
4248 ==2274== still reachable: 256 bytes in 1 blocks.
4249
4250Then there is no memory leak, just as memory accounted to global variables
4251is not a memleak - the memory is still being referenced, and didn't leak.
4252
4253Similarly, under some circumstances, valgrind might report kernel bugs
4254as if it were a bug in libev (e.g. in realloc or in the poll backend,
4255although an acceptable workaround has been found here), or it might be
4256confused.
4257
4258Keep in mind that valgrind is a very good tool, but only a tool. Don't
4259make it into some kind of religion.
4260
4261If you are unsure about something, feel free to contact the mailing list
4262with the full valgrind report and an explanation on why you think this
4263is a bug in libev (best check the archives, too :). However, don't be
4264annoyed when you get a brisk "this is no bug" answer and take the chance
4265of learning how to interpret valgrind properly.
4266
4267If you need, for some reason, empty reports from valgrind for your project
4268I suggest using suppression lists.
4269
4270
4271=head1 PORTABILITY NOTES
4272
4273=head2 WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4274
4275Win32 doesn't support any of the standards (e.g. POSIX) that libev
4276requires, and its I/O model is fundamentally incompatible with the POSIX
4277model. Libev still offers limited functionality on this platform in
4278the form of the C<EVBACKEND_SELECT> backend, and only supports socket
4279descriptors. This only applies when using Win32 natively, not when using
4280e.g. cygwin.
4281
4282Lifting these limitations would basically require the full
4283re-implementation of the I/O system. If you are into these kinds of
4284things, then note that glib does exactly that for you in a very portable
4285way (note also that glib is the slowest event library known to man).
4286
4287There is no supported compilation method available on windows except
4288embedding it into other applications.
4289
4290Sensible signal handling is officially unsupported by Microsoft - libev
4291tries its best, but under most conditions, signals will simply not work.
4292
4293Not a libev limitation but worth mentioning: windows apparently doesn't
4294accept large writes: instead of resulting in a partial write, windows will
4295either accept everything or return C<ENOBUFS> if the buffer is too large,
4296so make sure you only write small amounts into your sockets (less than a
4297megabyte seems safe, but this apparently depends on the amount of memory
4298available).
4299
4300Due to the many, low, and arbitrary limits on the win32 platform and
4301the abysmal performance of winsockets, using a large number of sockets
4302is not recommended (and not reasonable). If your program needs to use
4303more than a hundred or so sockets, then likely it needs to use a totally
4304different implementation for windows, as libev offers the POSIX readiness
4305notification model, which cannot be implemented efficiently on windows
4306(due to Microsoft monopoly games).
4307
4308A typical way to use libev under windows is to embed it (see the embedding
4309section for details) and use the following F<evwrap.h> header file instead
4310of F<ev.h>:
4311
4312 #define EV_STANDALONE /* keeps ev from requiring config.h */
4313 #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
4314
4315 #include "ev.h"
4316
4317And compile the following F<evwrap.c> file into your project (make sure
4318you do I<not> compile the F<ev.c> or any other embedded source files!):
4319
4320 #include "evwrap.h"
4321 #include "ev.c"
4322
4323=over 4
4324
4325=item The winsocket select function
4326
4327The winsocket C<select> function doesn't follow POSIX in that it
4328requires socket I<handles> and not socket I<file descriptors> (it is
4329also extremely buggy). This makes select very inefficient, and also
4330requires a mapping from file descriptors to socket handles (the Microsoft
4331C runtime provides the function C<_open_osfhandle> for this). See the
4332discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
4333C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
4334
4335The configuration for a "naked" win32 using the Microsoft runtime
4336libraries and raw winsocket select is:
4337
4338 #define EV_USE_SELECT 1
4339 #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
4340
4341Note that winsockets handling of fd sets is O(n), so you can easily get a
4342complexity in the O(n²) range when using win32.
4343
4344=item Limited number of file descriptors
4345
4346Windows has numerous arbitrary (and low) limits on things.
4347
4348Early versions of winsocket's select only supported waiting for a maximum
4349of C<64> handles (probably owning to the fact that all windows kernels
4350can only wait for C<64> things at the same time internally; Microsoft
4351recommends spawning a chain of threads and wait for 63 handles and the
4352previous thread in each. Sounds great!).
4353
4354Newer versions support more handles, but you need to define C<FD_SETSIZE>
4355to some high number (e.g. C<2048>) before compiling the winsocket select
4356call (which might be in libev or elsewhere, for example, perl and many
4357other interpreters do their own select emulation on windows).
4358
4359Another limit is the number of file descriptors in the Microsoft runtime
4360libraries, which by default is C<64> (there must be a hidden I<64>
4361fetish or something like this inside Microsoft). You can increase this
4362by calling C<_setmaxstdio>, which can increase this limit to C<2048>
4363(another arbitrary limit), but is broken in many versions of the Microsoft
4364runtime libraries. This might get you to about C<512> or C<2048> sockets
4365(depending on windows version and/or the phase of the moon). To get more,
4366you need to wrap all I/O functions and provide your own fd management, but
4367the cost of calling select (O(n²)) will likely make this unworkable.
4368
4369=back
4370
4371=head2 PORTABILITY REQUIREMENTS
4372
4373In addition to a working ISO-C implementation and of course the
4374backend-specific APIs, libev relies on a few additional extensions:
4375
4376=over 4
4377
4378=item C<void (*)(ev_watcher_type *, int revents)> must have compatible
4379calling conventions regardless of C<ev_watcher_type *>.
4380
4381Libev assumes not only that all watcher pointers have the same internal
4382structure (guaranteed by POSIX but not by ISO C for example), but it also
4383assumes that the same (machine) code can be used to call any watcher
4384callback: The watcher callbacks have different type signatures, but libev
4385calls them using an C<ev_watcher *> internally.
4386
4387=item C<sig_atomic_t volatile> must be thread-atomic as well
4388
4389The type C<sig_atomic_t volatile> (or whatever is defined as
4390C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
4391threads. This is not part of the specification for C<sig_atomic_t>, but is
4392believed to be sufficiently portable.
4393
4394=item C<sigprocmask> must work in a threaded environment
4395
4396Libev uses C<sigprocmask> to temporarily block signals. This is not
4397allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
4398pthread implementations will either allow C<sigprocmask> in the "main
4399thread" or will block signals process-wide, both behaviours would
4400be compatible with libev. Interaction between C<sigprocmask> and
4401C<pthread_sigmask> could complicate things, however.
4402
4403The most portable way to handle signals is to block signals in all threads
4404except the initial one, and run the default loop in the initial thread as
4405well.
4406
4407=item C<long> must be large enough for common memory allocation sizes
4408
4409To improve portability and simplify its API, libev uses C<long> internally
4410instead of C<size_t> when allocating its data structures. On non-POSIX
4411systems (Microsoft...) this might be unexpectedly low, but is still at
4412least 31 bits everywhere, which is enough for hundreds of millions of
4413watchers.
4414
4415=item C<double> must hold a time value in seconds with enough accuracy
4416
4417The type C<double> is used to represent timestamps. It is required to
4418have at least 51 bits of mantissa (and 9 bits of exponent), which is good
4419enough for at least into the year 4000. This requirement is fulfilled by
4420implementations implementing IEEE 754, which is basically all existing
4421ones. With IEEE 754 doubles, you get microsecond accuracy until at least
44222200.
4423
4424=back
4425
4426If you know of other additional requirements drop me a note.
4427
4428
4429=head1 ALGORITHMIC COMPLEXITIES
4430
4431In this section the complexities of (many of) the algorithms used inside
4432libev will be documented. For complexity discussions about backends see
4433the documentation for C<ev_default_init>.
4434
4435All of the following are about amortised time: If an array needs to be
4436extended, libev needs to realloc and move the whole array, but this
4437happens asymptotically rarer with higher number of elements, so O(1) might
4438mean that libev does a lengthy realloc operation in rare cases, but on
4439average it is much faster and asymptotically approaches constant time.
4440
4441=over 4
4442
4443=item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
4444
4445This means that, when you have a watcher that triggers in one hour and
4446there are 100 watchers that would trigger before that, then inserting will
4447have to skip roughly seven (C<ld 100>) of these watchers.
4448
4449=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
4450
4451That means that changing a timer costs less than removing/adding them,
4452as only the relative motion in the event queue has to be paid for.
4453
4454=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
4455
4456These just add the watcher into an array or at the head of a list.
4457
4458=item Stopping check/prepare/idle/fork/async watchers: O(1)
4459
4460=item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
4461
4462These watchers are stored in lists, so they need to be walked to find the
4463correct watcher to remove. The lists are usually short (you don't usually
4464have many watchers waiting for the same fd or signal: one is typical, two
4465is rare).
4466
4467=item Finding the next timer in each loop iteration: O(1)
4468
4469By virtue of using a binary or 4-heap, the next timer is always found at a
4470fixed position in the storage array.
4471
4472=item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
4473
4474A change means an I/O watcher gets started or stopped, which requires
4475libev to recalculate its status (and possibly tell the kernel, depending
4476on backend and whether C<ev_io_set> was used).
4477
4478=item Activating one watcher (putting it into the pending state): O(1)
4479
4480=item Priority handling: O(number_of_priorities)
4481
4482Priorities are implemented by allocating some space for each
4483priority. When doing priority-based operations, libev usually has to
4484linearly search all the priorities, but starting/stopping and activating
4485watchers becomes O(1) with respect to priority handling.
4486
4487=item Sending an ev_async: O(1)
4488
4489=item Processing ev_async_send: O(number_of_async_watchers)
4490
4491=item Processing signals: O(max_signal_number)
4492
4493Sending involves a system call I<iff> there were no other C<ev_async_send>
4494calls in the current loop iteration. Checking for async and signal events
4495involves iterating over all running async watchers or all signal numbers.
4496
4497=back
4498
4499
4500=head1 GLOSSARY
4501
4502=over 4
4503
4504=item active
4505
4506A watcher is active as long as it has been started (has been attached to
4507an event loop) but not yet stopped (disassociated from the event loop).
4508
4509=item application
4510
4511In this document, an application is whatever is using libev.
4512
4513=item callback
4514
4515The address of a function that is called when some event has been
4516detected. Callbacks are being passed the event loop, the watcher that
4517received the event, and the actual event bitset.
4518
4519=item callback invocation
4520
4521The act of calling the callback associated with a watcher.
4522
4523=item event
4524
4525A change of state of some external event, such as data now being available
4526for reading on a file descriptor, time having passed or simply not having
4527any other events happening anymore.
4528
4529In libev, events are represented as single bits (such as C<EV_READ> or
4530C<EV_TIMEOUT>).
4531
4532=item event library
4533
4534A software package implementing an event model and loop.
4535
4536=item event loop
4537
4538An entity that handles and processes external events and converts them
4539into callback invocations.
4540
4541=item event model
4542
4543The model used to describe how an event loop handles and processes
4544watchers and events.
4545
4546=item pending
4547
4548A watcher is pending as soon as the corresponding event has been detected,
4549and stops being pending as soon as the watcher will be invoked or its
4550pending status is explicitly cleared by the application.
4551
4552A watcher can be pending, but not active. Stopping a watcher also clears
4553its pending status.
4554
4555=item real time
4556
4557The physical time that is observed. It is apparently strictly monotonic :)
4558
4559=item wall-clock time
4560
4561The time and date as shown on clocks. Unlike real time, it can actually
4562be wrong and jump forwards and backwards, e.g. when the you adjust your
4563clock.
4564
4565=item watcher
4566
4567A data structure that describes interest in certain events. Watchers need
4568to be started (attached to an event loop) before they can receive events.
4569
4570=item watcher invocation
4571
4572The act of calling the callback associated with a watcher.
4573
4574=back
936 4575
937=head1 AUTHOR 4576=head1 AUTHOR
938 4577
939Marc Lehmann <libev@schmorp.de>. 4578Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael Magnusson.
940 4579

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines