1LIBEV(3)       libev - high performance full featured event loop      LIBEV(3)
2
3
4

NAME

6       libev - a high performance full-featured event loop written in C
7

SYNOPSIS

9          #include <ev.h>
10
11   EXAMPLE PROGRAM
12          // a single header file is required
13          #include <ev.h>
14
15          #include <stdio.h> // for puts
16
17          // every watcher type has its own typedef'd struct
18          // with the name ev_TYPE
19          ev_io stdin_watcher;
20          ev_timer timeout_watcher;
21
22          // all watcher callbacks have a similar signature
23          // this callback is called when data is readable on stdin
24          static void
25          stdin_cb (EV_P_ ev_io *w, int revents)
26          {
27            puts ("stdin ready");
28            // for one-shot events, one must manually stop the watcher
29            // with its corresponding stop function.
30            ev_io_stop (EV_A_ w);
31
32            // this causes all nested ev_run's to stop iterating
33            ev_break (EV_A_ EVBREAK_ALL);
34          }
35
36          // another callback, this time for a time-out
37          static void
38          timeout_cb (EV_P_ ev_timer *w, int revents)
39          {
40            puts ("timeout");
41            // this causes the innermost ev_run to stop iterating
42            ev_break (EV_A_ EVBREAK_ONE);
43          }
44
45          int
46          main (void)
47          {
48            // use the default event loop unless you have special needs
49            struct ev_loop *loop = EV_DEFAULT;
50
51            // initialise an io watcher, then start it
52            // this one will watch for stdin to become readable
53            ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
54            ev_io_start (loop, &stdin_watcher);
55
56            // initialise a timer watcher, then start it
57            // simple non-repeating 5.5 second timeout
58            ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
59            ev_timer_start (loop, &timeout_watcher);
60
61            // now wait for events to arrive
62            ev_run (loop, 0);
63
64            // break was called, so exit
65            return 0;
66          }
67

ABOUT THIS DOCUMENT

69       This document documents the libev software package.
70
71       The newest version of this document is also available as an html-
72       formatted web page you might find easier to navigate when reading it
73       for the first time:
74       <http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
75
76       While this document tries to be as complete as possible in documenting
77       libev, its usage and the rationale behind its design, it is not a
78       tutorial on event-based programming, nor will it introduce event-based
79       programming with libev.
80
81       Familiarity with event based programming techniques in general is
82       assumed throughout this document.
83

WHAT TO READ WHEN IN A HURRY

85       This manual tries to be very detailed, but unfortunately, this also
86       makes it very long. If you just want to know the basics of libev, I
87       suggest reading "ANATOMY OF A WATCHER", then the "EXAMPLE PROGRAM"
88       above and look up the missing functions in "GLOBAL FUNCTIONS" and the
89       "ev_io" and "ev_timer" sections in "WATCHER TYPES".
90

ABOUT LIBEV

92       Libev is an event loop: you register interest in certain events (such
93       as a file descriptor being readable or a timeout occurring), and it
94       will manage these event sources and provide your program with events.
95
96       To do this, it must take more or less complete control over your
97       process (or thread) by executing the event loop handler, and will then
98       communicate events via a callback mechanism.
99
100       You register interest in certain events by registering so-called event
101       watchers, which are relatively small C structures you initialise with
102       the details of the event, and then hand it over to libev by starting
103       the watcher.
104
105   FEATURES
106       Libev supports "select", "poll", the Linux-specific "epoll", the BSD-
107       specific "kqueue" and the Solaris-specific event port mechanisms for
108       file descriptor events ("ev_io"), the Linux "inotify" interface (for
109       "ev_stat"), Linux eventfd/signalfd (for faster and cleaner inter-thread
110       wakeup ("ev_async")/signal handling ("ev_signal")) relative timers
111       ("ev_timer"), absolute timers with customised rescheduling
112       ("ev_periodic"), synchronous signals ("ev_signal"), process status
113       change events ("ev_child"), and event watchers dealing with the event
114       loop mechanism itself ("ev_idle", "ev_embed", "ev_prepare" and
115       "ev_check" watchers) as well as file watchers ("ev_stat") and even
116       limited support for fork events ("ev_fork").
117
118       It also is quite fast (see this benchmark
119       <http://libev.schmorp.de/bench.html> comparing it to libevent for
120       example).
121
122   CONVENTIONS
123       Libev is very configurable. In this manual the default (and most
124       common) configuration will be described, which supports multiple event
125       loops. For more info about various configuration options please have a
126       look at EMBED section in this manual. If libev was configured without
127       support for multiple event loops, then all functions taking an initial
128       argument of name "loop" (which is always of type "struct ev_loop *")
129       will not have this argument.
130
131   TIME REPRESENTATION
132       Libev represents time as a single floating point number, representing
133       the (fractional) number of seconds since the (POSIX) epoch (in practice
134       somewhere near the beginning of 1970, details are complicated, don't
135       ask). This type is called "ev_tstamp", which is what you should use
136       too. It usually aliases to the "double" type in C. When you need to do
137       any calculations on it, you should treat it as some floating point
138       value.
139
140       Unlike the name component "stamp" might indicate, it is also used for
141       time differences (e.g. delays) throughout libev.
142

ERROR HANDLING

144       Libev knows three classes of errors: operating system errors, usage
145       errors and internal errors (bugs).
146
147       When libev catches an operating system error it cannot handle (for
148       example a system call indicating a condition libev cannot fix), it
149       calls the callback set via "ev_set_syserr_cb", which is supposed to fix
150       the problem or abort. The default is to print a diagnostic message and
151       to call "abort ()".
152
153       When libev detects a usage error such as a negative timer interval,
154       then it will print a diagnostic message and abort (via the "assert"
155       mechanism, so "NDEBUG" will disable this checking): these are
156       programming errors in the libev caller and need to be fixed there.
157
158       Libev also has a few internal error-checking "assert"ions, and also has
159       extensive consistency checking code. These do not trigger under normal
160       circumstances, as they indicate either a bug in libev or worse.
161

GLOBAL FUNCTIONS

163       These functions can be called anytime, even before initialising the
164       library in any way.
165
166       ev_tstamp ev_time ()
167           Returns the current time as libev would use it. Please note that
168           the "ev_now" function is usually faster and also often returns the
169           timestamp you actually want to know. Also interesting is the
170           combination of "ev_now_update" and "ev_now".
171
172       ev_sleep (ev_tstamp interval)
173           Sleep for the given interval: The current thread will be blocked
174           until either it is interrupted or the given time interval has
175           passed (approximately - it might return a bit earlier even if not
176           interrupted). Returns immediately if "interval <= 0".
177
178           Basically this is a sub-second-resolution "sleep ()".
179
180           The range of the "interval" is limited - libev only guarantees to
181           work with sleep times of up to one day ("interval <= 86400").
182
183       int ev_version_major ()
184       int ev_version_minor ()
185           You can find out the major and minor ABI version numbers of the
186           library you linked against by calling the functions
187           "ev_version_major" and "ev_version_minor". If you want, you can
188           compare against the global symbols "EV_VERSION_MAJOR" and
189           "EV_VERSION_MINOR", which specify the version of the library your
190           program was compiled against.
191
192           These version numbers refer to the ABI version of the library, not
193           the release version.
194
195           Usually, it's a good idea to terminate if the major versions
196           mismatch, as this indicates an incompatible change. Minor versions
197           are usually compatible to older versions, so a larger minor version
198           alone is usually not a problem.
199
200           Example: Make sure we haven't accidentally been linked against the
201           wrong version (note, however, that this will not detect other ABI
202           mismatches, such as LFS or reentrancy).
203
204              assert (("libev version mismatch",
205                       ev_version_major () == EV_VERSION_MAJOR
206                       && ev_version_minor () >= EV_VERSION_MINOR));
207
208       unsigned int ev_supported_backends ()
209           Return the set of all backends (i.e. their corresponding
210           "EV_BACKEND_*" value) compiled into this binary of libev
211           (independent of their availability on the system you are running
212           on). See "ev_default_loop" for a description of the set values.
213
214           Example: make sure we have the epoll method, because yeah this is
215           cool and a must have and can we have a torrent of it please!!!11
216
217              assert (("sorry, no epoll, no sex",
218                       ev_supported_backends () & EVBACKEND_EPOLL));
219
220       unsigned int ev_recommended_backends ()
221           Return the set of all backends compiled into this binary of libev
222           and also recommended for this platform, meaning it will work for
223           most file descriptor types. This set is often smaller than the one
224           returned by "ev_supported_backends", as for example kqueue is
225           broken on most BSDs and will not be auto-detected unless you
226           explicitly request it (assuming you know what you are doing). This
227           is the set of backends that libev will probe for if you specify no
228           backends explicitly.
229
230       unsigned int ev_embeddable_backends ()
231           Returns the set of backends that are embeddable in other event
232           loops. This value is platform-specific but can include backends not
233           available on the current system. To find which embeddable backends
234           might be supported on the current system, you would need to look at
235           "ev_embeddable_backends () & ev_supported_backends ()", likewise
236           for recommended ones.
237
238           See the description of "ev_embed" watchers for more info.
239
240       ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
241           Sets the allocation function to use (the prototype is similar - the
242           semantics are identical to the "realloc" C89/SuS/POSIX function).
243           It is used to allocate and free memory (no surprises here). If it
244           returns zero when memory needs to be allocated ("size != 0"), the
245           library might abort or take some potentially destructive action.
246
247           Since some systems (at least OpenBSD and Darwin) fail to implement
248           correct "realloc" semantics, libev will use a wrapper around the
249           system "realloc" and "free" functions by default.
250
251           You could override this function in high-availability programs to,
252           say, free some memory if it cannot allocate memory, to use a
253           special allocator, or even to sleep a while and retry until some
254           memory is available.
255
256           Example: Replace the libev allocator with one that waits a bit and
257           then retries (example requires a standards-compliant "realloc").
258
259              static void *
260              persistent_realloc (void *ptr, size_t size)
261              {
262                for (;;)
263                  {
264                    void *newptr = realloc (ptr, size);
265
266                    if (newptr)
267                      return newptr;
268
269                    sleep (60);
270                  }
271              }
272
273              ...
274              ev_set_allocator (persistent_realloc);
275
276       ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
277           Set the callback function to call on a retryable system call error
278           (such as failed select, poll, epoll_wait). The message is a
279           printable string indicating the system call or subsystem causing
280           the problem. If this callback is set, then libev will expect it to
281           remedy the situation, no matter what, when it returns. That is,
282           libev will generally retry the requested operation, or, if the
283           condition doesn't go away, do bad stuff (such as abort).
284
285           Example: This is basically the same thing that libev does
286           internally, too.
287
288              static void
289              fatal_error (const char *msg)
290              {
291                perror (msg);
292                abort ();
293              }
294
295              ...
296              ev_set_syserr_cb (fatal_error);
297
298       ev_feed_signal (int signum)
299           This function can be used to "simulate" a signal receive. It is
300           completely safe to call this function at any time, from any
301           context, including signal handlers or random threads.
302
303           Its main use is to customise signal handling in your process,
304           especially in the presence of threads. For example, you could block
305           signals by default in all threads (and specifying
306           "EVFLAG_NOSIGMASK" when creating any loops), and in one thread, use
307           "sigwait" or any other mechanism to wait for signals, then
308           "deliver" them to libev by calling "ev_feed_signal".
309

FUNCTIONS CONTROLLING EVENT LOOPS

311       An event loop is described by a "struct ev_loop *" (the "struct" is not
312       optional in this case unless libev 3 compatibility is disabled, as
313       libev 3 had an "ev_loop" function colliding with the struct name).
314
315       The library knows two types of such loops, the default loop, which
316       supports child process events, and dynamically created event loops
317       which do not.
318
319       struct ev_loop *ev_default_loop (unsigned int flags)
320           This returns the "default" event loop object, which is what you
321           should normally use when you just need "the event loop". Event loop
322           objects and the "flags" parameter are described in more detail in
323           the entry for "ev_loop_new".
324
325           If the default loop is already initialised then this function
326           simply returns it (and ignores the flags. If that is troubling you,
327           check "ev_backend ()" afterwards). Otherwise it will create it with
328           the given flags, which should almost always be 0, unless the caller
329           is also the one calling "ev_run" or otherwise qualifies as "the
330           main program".
331
332           If you don't know what event loop to use, use the one returned from
333           this function (or via the "EV_DEFAULT" macro).
334
335           Note that this function is not thread-safe, so if you want to use
336           it from multiple threads, you have to employ some kind of mutex
337           (note also that this case is unlikely, as loops cannot be shared
338           easily between threads anyway).
339
340           The default loop is the only loop that can handle "ev_child"
341           watchers, and to do this, it always registers a handler for
342           "SIGCHLD". If this is a problem for your application you can either
343           create a dynamic loop with "ev_loop_new" which doesn't do that, or
344           you can simply overwrite the "SIGCHLD" signal handler after calling
345           "ev_default_init".
346
347           Example: This is the most typical usage.
348
349              if (!ev_default_loop (0))
350                fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
351
352           Example: Restrict libev to the select and poll backends, and do not
353           allow environment settings to be taken into account:
354
355              ev_default_loop (EVBACKEND_POLL | EVBACKEND_SELECT | EVFLAG_NOENV);
356
357       struct ev_loop *ev_loop_new (unsigned int flags)
358           This will create and initialise a new event loop object. If the
359           loop could not be initialised, returns false.
360
361           This function is thread-safe, and one common way to use libev with
362           threads is indeed to create one loop per thread, and using the
363           default loop in the "main" or "initial" thread.
364
365           The flags argument can be used to specify special behaviour or
366           specific backends to use, and is usually specified as 0 (or
367           "EVFLAG_AUTO").
368
369           The following flags are supported:
370
371           "EVFLAG_AUTO"
372               The default flags value. Use this if you have no clue (it's the
373               right thing, believe me).
374
375           "EVFLAG_NOENV"
376               If this flag bit is or'ed into the flag value (or the program
377               runs setuid or setgid) then libev will not look at the
378               environment variable "LIBEV_FLAGS". Otherwise (the default),
379               this environment variable will override the flags completely if
380               it is found in the environment. This is useful to try out
381               specific backends to test their performance, to work around
382               bugs, or to make libev threadsafe (accessing environment
383               variables cannot be done in a threadsafe way, but usually it
384               works if no other thread modifies them).
385
386           "EVFLAG_FORKCHECK"
387               Instead of calling "ev_loop_fork" manually after a fork, you
388               can also make libev check for a fork in each iteration by
389               enabling this flag.
390
391               This works by calling "getpid ()" on every iteration of the
392               loop, and thus this might slow down your event loop if you do a
393               lot of loop iterations and little real work, but is usually not
394               noticeable (on my GNU/Linux system for example, "getpid" is
395               actually a simple 5-insn sequence without a system call and
396               thus very fast, but my GNU/Linux system also has
397               "pthread_atfork" which is even faster). (Update: glibc versions
398               2.25 apparently removed the "getpid" optimisation again).
399
400               The big advantage of this flag is that you can forget about
401               fork (and forget about forgetting to tell libev about forking,
402               although you still have to ignore "SIGPIPE") when you use this
403               flag.
404
405               This flag setting cannot be overridden or specified in the
406               "LIBEV_FLAGS" environment variable.
407
408           "EVFLAG_NOINOTIFY"
409               When this flag is specified, then libev will not attempt to use
410               the inotify API for its "ev_stat" watchers. Apart from
411               debugging and testing, this flag can be useful to conserve
412               inotify file descriptors, as otherwise each loop using
413               "ev_stat" watchers consumes one inotify handle.
414
415           "EVFLAG_SIGNALFD"
416               When this flag is specified, then libev will attempt to use the
417               signalfd API for its "ev_signal" (and "ev_child") watchers.
418               This API delivers signals synchronously, which makes it both
419               faster and might make it possible to get the queued signal
420               data. It can also simplify signal handling with threads, as
421               long as you properly block signals in your threads that are not
422               interested in handling them.
423
424               Signalfd will not be used by default as this changes your
425               signal mask, and there are a lot of shoddy libraries and
426               programs (glib's threadpool for example) that can't properly
427               initialise their signal masks.
428
429           "EVFLAG_NOSIGMASK"
430               When this flag is specified, then libev will avoid to modify
431               the signal mask. Specifically, this means you have to make sure
432               signals are unblocked when you want to receive them.
433
434               This behaviour is useful when you want to do your own signal
435               handling, or want to handle signals only in specific threads
436               and want to avoid libev unblocking the signals.
437
438               It's also required by POSIX in a threaded program, as libev
439               calls "sigprocmask", whose behaviour is officially unspecified.
440
441               This flag's behaviour will become the default in future
442               versions of libev.
443
444           "EVBACKEND_SELECT"  (value 1, portable select backend)
445               This is your standard select(2) backend. Not completely
446               standard, as libev tries to roll its own fd_set with no limits
447               on the number of fds, but if that fails, expect a fairly low
448               limit on the number of fds when using this backend. It doesn't
449               scale too well (O(highest_fd)), but its usually the fastest
450               backend for a low number of (low-numbered :) fds.
451
452               To get good performance out of this backend you need a high
453               amount of parallelism (most of the file descriptors should be
454               busy). If you are writing a server, you should "accept ()" in a
455               loop to accept as many connections as possible during one
456               iteration. You might also want to have a look at
457               "ev_set_io_collect_interval ()" to increase the amount of
458               readiness notifications you get per iteration.
459
460               This backend maps "EV_READ" to the "readfds" set and "EV_WRITE"
461               to the "writefds" set (and to work around Microsoft Windows
462               bugs, also onto the "exceptfds" set on that platform).
463
464           "EVBACKEND_POLL"    (value 2, poll backend, available everywhere
465           except on windows)
466               And this is your standard poll(2) backend. It's more
467               complicated than select, but handles sparse fds better and has
468               no artificial limit on the number of fds you can use (except it
469               will slow down considerably with a lot of inactive fds). It
470               scales similarly to select, i.e. O(total_fds). See the entry
471               for "EVBACKEND_SELECT", above, for performance tips.
472
473               This backend maps "EV_READ" to "POLLIN | POLLERR | POLLHUP",
474               and "EV_WRITE" to "POLLOUT | POLLERR | POLLHUP".
475
476           "EVBACKEND_EPOLL"   (value 4, Linux)
477               Use the linux-specific epoll(7) interface (for both pre- and
478               post-2.6.9 kernels).
479
480               For few fds, this backend is a bit little slower than poll and
481               select, but it scales phenomenally better. While poll and
482               select usually scale like O(total_fds) where total_fds is the
483               total number of fds (or the highest fd), epoll scales either
484               O(1) or O(active_fds).
485
486               The epoll mechanism deserves honorable mention as the most
487               misdesigned of the more advanced event mechanisms: mere
488               annoyances include silently dropping file descriptors,
489               requiring a system call per change per file descriptor (and
490               unnecessary guessing of parameters), problems with dup,
491               returning before the timeout value, resulting in additional
492               iterations (and only giving 5ms accuracy while select on the
493               same platform gives 0.1ms) and so on. The biggest issue is fork
494               races, however - if a program forks then both parent and child
495               process have to recreate the epoll set, which can take
496               considerable time (one syscall per file descriptor) and is of
497               course hard to detect.
498
499               Epoll is also notoriously buggy - embedding epoll fds should
500               work, but of course doesn't, and epoll just loves to report
501               events for totally different file descriptors (even already
502               closed ones, so one cannot even remove them from the set) than
503               registered in the set (especially on SMP systems). Libev tries
504               to counter these spurious notifications by employing an
505               additional generation counter and comparing that against the
506               events to filter out spurious ones, recreating the set when
507               required. Epoll also erroneously rounds down timeouts, but
508               gives you no way to know when and by how much, so sometimes you
509               have to busy-wait because epoll returns immediately despite a
510               nonzero timeout. And last not least, it also refuses to work
511               with some file descriptors which work perfectly fine with
512               "select" (files, many character devices...).
513
514               Epoll is truly the train wreck among event poll mechanisms, a
515               frankenpoll, cobbled together in a hurry, no thought to design
516               or interaction with others. Oh, the pain, will it ever stop...
517
518               While stopping, setting and starting an I/O watcher in the same
519               iteration will result in some caching, there is still a system
520               call per such incident (because the same file descriptor could
521               point to a different file description now), so its best to
522               avoid that. Also, "dup ()"'ed file descriptors might not work
523               very well if you register events for both file descriptors.
524
525               Best performance from this backend is achieved by not
526               unregistering all watchers for a file descriptor until it has
527               been closed, if possible, i.e. keep at least one watcher active
528               per fd at all times. Stopping and starting a watcher (without
529               re-setting it) also usually doesn't cause extra overhead. A
530               fork can both result in spurious notifications as well as in
531               libev having to destroy and recreate the epoll object, which
532               can take considerable time and thus should be avoided.
533
534               All this means that, in practice, "EVBACKEND_SELECT" can be as
535               fast or faster than epoll for maybe up to a hundred file
536               descriptors, depending on the usage. So sad.
537
538               While nominally embeddable in other event loops, this feature
539               is broken in all kernel versions tested so far.
540
541               This backend maps "EV_READ" and "EV_WRITE" in the same way as
542               "EVBACKEND_POLL".
543
544           "EVBACKEND_KQUEUE"  (value 8, most BSD clones)
545               Kqueue deserves special mention, as at the time of this
546               writing, it was broken on all BSDs except NetBSD (usually it
547               doesn't work reliably with anything but sockets and pipes,
548               except on Darwin, where of course it's completely useless).
549               Unlike epoll, however, whose brokenness is by design, these
550               kqueue bugs can (and eventually will) be fixed without API
551               changes to existing programs. For this reason it's not being
552               "auto-detected" unless you explicitly specify it in the flags
553               (i.e. using "EVBACKEND_KQUEUE") or libev was compiled on a
554               known-to-be-good (-enough) system like NetBSD.
555
556               You still can embed kqueue into a normal poll or select backend
557               and use it only for sockets (after having made sure that
558               sockets work with kqueue on the target platform). See
559               "ev_embed" watchers for more info.
560
561               It scales in the same way as the epoll backend, but the
562               interface to the kernel is more efficient (which says nothing
563               about its actual speed, of course). While stopping, setting and
564               starting an I/O watcher does never cause an extra system call
565               as with "EVBACKEND_EPOLL", it still adds up to two event
566               changes per incident. Support for "fork ()" is very bad (you
567               might have to leak fd's on fork, but it's more sane than epoll)
568               and it drops fds silently in similarly hard-to-detect cases.
569
570               This backend usually performs well under most conditions.
571
572               While nominally embeddable in other event loops, this doesn't
573               work everywhere, so you might need to test for this. And since
574               it is broken almost everywhere, you should only use it when you
575               have a lot of sockets (for which it usually works), by
576               embedding it into another event loop (e.g. "EVBACKEND_SELECT"
577               or "EVBACKEND_POLL" (but "poll" is of course also broken on OS
578               X)) and, did I mention it, using it only for sockets.
579
580               This backend maps "EV_READ" into an "EVFILT_READ" kevent with
581               "NOTE_EOF", and "EV_WRITE" into an "EVFILT_WRITE" kevent with
582               "NOTE_EOF".
583
584           "EVBACKEND_DEVPOLL" (value 16, Solaris 8)
585               This is not implemented yet (and might never be, unless you
586               send me an implementation). According to reports, "/dev/poll"
587               only supports sockets and is not embeddable, which would limit
588               the usefulness of this backend immensely.
589
590           "EVBACKEND_PORT"    (value 32, Solaris 10)
591               This uses the Solaris 10 event port mechanism. As with
592               everything on Solaris, it's really slow, but it still scales
593               very well (O(active_fds)).
594
595               While this backend scales well, it requires one system call per
596               active file descriptor per loop iteration. For small and medium
597               numbers of file descriptors a "slow" "EVBACKEND_SELECT" or
598               "EVBACKEND_POLL" backend might perform better.
599
600               On the positive side, this backend actually performed fully to
601               specification in all tests and is fully embeddable, which is a
602               rare feat among the OS-specific backends (I vastly prefer
603               correctness over speed hacks).
604
605               On the negative side, the interface is bizarre - so bizarre
606               that even sun itself gets it wrong in their code examples: The
607               event polling function sometimes returns events to the caller
608               even though an error occurred, but with no indication whether
609               it has done so or not (yes, it's even documented that way) -
610               deadly for edge-triggered interfaces where you absolutely have
611               to know whether an event occurred or not because you have to
612               re-arm the watcher.
613
614               Fortunately libev seems to be able to work around these
615               idiocies.
616
617               This backend maps "EV_READ" and "EV_WRITE" in the same way as
618               "EVBACKEND_POLL".
619
620           "EVBACKEND_ALL"
621               Try all backends (even potentially broken ones that wouldn't be
622               tried with "EVFLAG_AUTO"). Since this is a mask, you can do
623               stuff such as "EVBACKEND_ALL & ~EVBACKEND_KQUEUE".
624
625               It is definitely not recommended to use this flag, use whatever
626               "ev_recommended_backends ()" returns, or simply do not specify
627               a backend at all.
628
629           "EVBACKEND_MASK"
630               Not a backend at all, but a mask to select all backend bits
631               from a "flags" value, in case you want to mask out any backends
632               from a flags value (e.g. when modifying the "LIBEV_FLAGS"
633               environment variable).
634
635           If one or more of the backend flags are or'ed into the flags value,
636           then only these backends will be tried (in the reverse order as
637           listed here). If none are specified, all backends in
638           "ev_recommended_backends ()" will be tried.
639
640           Example: Try to create a event loop that uses epoll and nothing
641           else.
642
643              struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
644              if (!epoller)
645                fatal ("no epoll found here, maybe it hides under your chair");
646
647           Example: Use whatever libev has to offer, but make sure that kqueue
648           is used if available.
649
650              struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
651
652       ev_loop_destroy (loop)
653           Destroys an event loop object (frees all memory and kernel state
654           etc.). None of the active event watchers will be stopped in the
655           normal sense, so e.g. "ev_is_active" might still return true. It is
656           your responsibility to either stop all watchers cleanly yourself
657           before calling this function, or cope with the fact afterwards
658           (which is usually the easiest thing, you can just ignore the
659           watchers and/or "free ()" them for example).
660
661           Note that certain global state, such as signal state (and installed
662           signal handlers), will not be freed by this function, and related
663           watchers (such as signal and child watchers) would need to be
664           stopped manually.
665
666           This function is normally used on loop objects allocated by
667           "ev_loop_new", but it can also be used on the default loop returned
668           by "ev_default_loop", in which case it is not thread-safe.
669
670           Note that it is not advisable to call this function on the default
671           loop except in the rare occasion where you really need to free its
672           resources.  If you need dynamically allocated loops it is better to
673           use "ev_loop_new" and "ev_loop_destroy".
674
675       ev_loop_fork (loop)
676           This function sets a flag that causes subsequent "ev_run"
677           iterations to reinitialise the kernel state for backends that have
678           one. Despite the name, you can call it anytime you are allowed to
679           start or stop watchers (except inside an "ev_prepare" callback),
680           but it makes most sense after forking, in the child process. You
681           must call it (or use "EVFLAG_FORKCHECK") in the child before
682           resuming or calling "ev_run".
683
684           In addition, if you want to reuse a loop (via this function or
685           "EVFLAG_FORKCHECK"), you also have to ignore "SIGPIPE".
686
687           Again, you have to call it on any loop that you want to re-use
688           after a fork, even if you do not plan to use the loop in the
689           parent. This is because some kernel interfaces *cough* kqueue
690           *cough* do funny things during fork.
691
692           On the other hand, you only need to call this function in the child
693           process if and only if you want to use the event loop in the child.
694           If you just fork+exec or create a new loop in the child, you don't
695           have to call it at all (in fact, "epoll" is so badly broken that it
696           makes a difference, but libev will usually detect this case on its
697           own and do a costly reset of the backend).
698
699           The function itself is quite fast and it's usually not a problem to
700           call it just in case after a fork.
701
702           Example: Automate calling "ev_loop_fork" on the default loop when
703           using pthreads.
704
705              static void
706              post_fork_child (void)
707              {
708                ev_loop_fork (EV_DEFAULT);
709              }
710
711              ...
712              pthread_atfork (0, 0, post_fork_child);
713
714       int ev_is_default_loop (loop)
715           Returns true when the given loop is, in fact, the default loop, and
716           false otherwise.
717
718       unsigned int ev_iteration (loop)
719           Returns the current iteration count for the event loop, which is
720           identical to the number of times libev did poll for new events. It
721           starts at 0 and happily wraps around with enough iterations.
722
723           This value can sometimes be useful as a generation counter of sorts
724           (it "ticks" the number of loop iterations), as it roughly
725           corresponds with "ev_prepare" and "ev_check" calls - and is
726           incremented between the prepare and check phases.
727
728       unsigned int ev_depth (loop)
729           Returns the number of times "ev_run" was entered minus the number
730           of times "ev_run" was exited normally, in other words, the
731           recursion depth.
732
733           Outside "ev_run", this number is zero. In a callback, this number
734           is 1, unless "ev_run" was invoked recursively (or from another
735           thread), in which case it is higher.
736
737           Leaving "ev_run" abnormally (setjmp/longjmp, cancelling the thread,
738           throwing an exception etc.), doesn't count as "exit" - consider
739           this as a hint to avoid such ungentleman-like behaviour unless it's
740           really convenient, in which case it is fully supported.
741
742       unsigned int ev_backend (loop)
743           Returns one of the "EVBACKEND_*" flags indicating the event backend
744           in use.
745
746       ev_tstamp ev_now (loop)
747           Returns the current "event loop time", which is the time the event
748           loop received events and started processing them. This timestamp
749           does not change as long as callbacks are being processed, and this
750           is also the base time used for relative timers. You can treat it as
751           the timestamp of the event occurring (or more correctly, libev
752           finding out about it).
753
754       ev_now_update (loop)
755           Establishes the current time by querying the kernel, updating the
756           time returned by "ev_now ()" in the progress. This is a costly
757           operation and is usually done automatically within "ev_run ()".
758
759           This function is rarely useful, but when some event callback runs
760           for a very long time without entering the event loop, updating
761           libev's idea of the current time is a good idea.
762
763           See also "The special problem of time updates" in the "ev_timer"
764           section.
765
766       ev_suspend (loop)
767       ev_resume (loop)
768           These two functions suspend and resume an event loop, for use when
769           the loop is not used for a while and timeouts should not be
770           processed.
771
772           A typical use case would be an interactive program such as a game:
773           When the user presses "^Z" to suspend the game and resumes it an
774           hour later it would be best to handle timeouts as if no time had
775           actually passed while the program was suspended. This can be
776           achieved by calling "ev_suspend" in your "SIGTSTP" handler, sending
777           yourself a "SIGSTOP" and calling "ev_resume" directly afterwards to
778           resume timer processing.
779
780           Effectively, all "ev_timer" watchers will be delayed by the time
781           spend between "ev_suspend" and "ev_resume", and all "ev_periodic"
782           watchers will be rescheduled (that is, they will lose any events
783           that would have occurred while suspended).
784
785           After calling "ev_suspend" you must not call any function on the
786           given loop other than "ev_resume", and you must not call
787           "ev_resume" without a previous call to "ev_suspend".
788
789           Calling "ev_suspend"/"ev_resume" has the side effect of updating
790           the event loop time (see "ev_now_update").
791
792       bool ev_run (loop, int flags)
793           Finally, this is it, the event handler. This function usually is
794           called after you have initialised all your watchers and you want to
795           start handling events. It will ask the operating system for any new
796           events, call the watcher callbacks, and then repeat the whole
797           process indefinitely: This is why event loops are called loops.
798
799           If the flags argument is specified as 0, it will keep handling
800           events until either no event watchers are active anymore or
801           "ev_break" was called.
802
803           The return value is false if there are no more active watchers
804           (which usually means "all jobs done" or "deadlock"), and true in
805           all other cases (which usually means " you should call "ev_run"
806           again").
807
808           Please note that an explicit "ev_break" is usually better than
809           relying on all watchers to be stopped when deciding when a program
810           has finished (especially in interactive programs), but having a
811           program that automatically loops as long as it has to and no longer
812           by virtue of relying on its watchers stopping correctly, that is
813           truly a thing of beauty.
814
815           This function is mostly exception-safe - you can break out of a
816           "ev_run" call by calling "longjmp" in a callback, throwing a C++
817           exception and so on. This does not decrement the "ev_depth" value,
818           nor will it clear any outstanding "EVBREAK_ONE" breaks.
819
820           A flags value of "EVRUN_NOWAIT" will look for new events, will
821           handle those events and any already outstanding ones, but will not
822           wait and block your process in case there are no events and will
823           return after one iteration of the loop. This is sometimes useful to
824           poll and handle new events while doing lengthy calculations, to
825           keep the program responsive.
826
827           A flags value of "EVRUN_ONCE" will look for new events (waiting if
828           necessary) and will handle those and any already outstanding ones.
829           It will block your process until at least one new event arrives
830           (which could be an event internal to libev itself, so there is no
831           guarantee that a user-registered callback will be called), and will
832           return after one iteration of the loop.
833
834           This is useful if you are waiting for some external event in
835           conjunction with something not expressible using other libev
836           watchers (i.e. "roll your own "ev_run""). However, a pair of
837           "ev_prepare"/"ev_check" watchers is usually a better approach for
838           this kind of thing.
839
840           Here are the gory details of what "ev_run" does (this is for your
841           understanding, not a guarantee that things will work exactly like
842           this in future versions):
843
844              - Increment loop depth.
845              - Reset the ev_break status.
846              - Before the first iteration, call any pending watchers.
847              LOOP:
848              - If EVFLAG_FORKCHECK was used, check for a fork.
849              - If a fork was detected (by any means), queue and call all fork watchers.
850              - Queue and call all prepare watchers.
851              - If ev_break was called, goto FINISH.
852              - If we have been forked, detach and recreate the kernel state
853                as to not disturb the other process.
854              - Update the kernel state with all outstanding changes.
855              - Update the "event loop time" (ev_now ()).
856              - Calculate for how long to sleep or block, if at all
857                (active idle watchers, EVRUN_NOWAIT or not having
858                any active watchers at all will result in not sleeping).
859              - Sleep if the I/O and timer collect interval say so.
860              - Increment loop iteration counter.
861              - Block the process, waiting for any events.
862              - Queue all outstanding I/O (fd) events.
863              - Update the "event loop time" (ev_now ()), and do time jump adjustments.
864              - Queue all expired timers.
865              - Queue all expired periodics.
866              - Queue all idle watchers with priority higher than that of pending events.
867              - Queue all check watchers.
868              - Call all queued watchers in reverse order (i.e. check watchers first).
869                Signals and child watchers are implemented as I/O watchers, and will
870                be handled here by queueing them when their watcher gets executed.
871              - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
872                were used, or there are no active watchers, goto FINISH, otherwise
873                continue with step LOOP.
874              FINISH:
875              - Reset the ev_break status iff it was EVBREAK_ONE.
876              - Decrement the loop depth.
877              - Return.
878
879           Example: Queue some jobs and then loop until no events are
880           outstanding anymore.
881
882              ... queue jobs here, make sure they register event watchers as long
883              ... as they still have work to do (even an idle watcher will do..)
884              ev_run (my_loop, 0);
885              ... jobs done or somebody called break. yeah!
886
887       ev_break (loop, how)
888           Can be used to make a call to "ev_run" return early (but only after
889           it has processed all outstanding events). The "how" argument must
890           be either "EVBREAK_ONE", which will make the innermost "ev_run"
891           call return, or "EVBREAK_ALL", which will make all nested "ev_run"
892           calls return.
893
894           This "break state" will be cleared on the next call to "ev_run".
895
896           It is safe to call "ev_break" from outside any "ev_run" calls, too,
897           in which case it will have no effect.
898
899       ev_ref (loop)
900       ev_unref (loop)
901           Ref/unref can be used to add or remove a reference count on the
902           event loop: Every watcher keeps one reference, and as long as the
903           reference count is nonzero, "ev_run" will not return on its own.
904
905           This is useful when you have a watcher that you never intend to
906           unregister, but that nevertheless should not keep "ev_run" from
907           returning. In such a case, call "ev_unref" after starting, and
908           "ev_ref" before stopping it.
909
910           As an example, libev itself uses this for its internal signal pipe:
911           It is not visible to the libev user and should not keep "ev_run"
912           from exiting if no event watchers registered by it are active. It
913           is also an excellent way to do this for generic recurring timers or
914           from within third-party libraries. Just remember to unref after
915           start and ref before stop (but only if the watcher wasn't active
916           before, or was active before, respectively. Note also that libev
917           might stop watchers itself (e.g. non-repeating timers) in which
918           case you have to "ev_ref" in the callback).
919
920           Example: Create a signal watcher, but keep it from keeping "ev_run"
921           running when nothing else is active.
922
923              ev_signal exitsig;
924              ev_signal_init (&exitsig, sig_cb, SIGINT);
925              ev_signal_start (loop, &exitsig);
926              ev_unref (loop);
927
928           Example: For some weird reason, unregister the above signal handler
929           again.
930
931              ev_ref (loop);
932              ev_signal_stop (loop, &exitsig);
933
934       ev_set_io_collect_interval (loop, ev_tstamp interval)
935       ev_set_timeout_collect_interval (loop, ev_tstamp interval)
936           These advanced functions influence the time that libev will spend
937           waiting for events. Both time intervals are by default 0, meaning
938           that libev will try to invoke timer/periodic callbacks and I/O
939           callbacks with minimum latency.
940
941           Setting these to a higher value (the "interval" must be >= 0)
942           allows libev to delay invocation of I/O and timer/periodic
943           callbacks to increase efficiency of loop iterations (or to increase
944           power-saving opportunities).
945
946           The idea is that sometimes your program runs just fast enough to
947           handle one (or very few) event(s) per loop iteration. While this
948           makes the program responsive, it also wastes a lot of CPU time to
949           poll for new events, especially with backends like "select ()"
950           which have a high overhead for the actual polling but can deliver
951           many events at once.
952
953           By setting a higher io collect interval you allow libev to spend
954           more time collecting I/O events, so you can handle more events per
955           iteration, at the cost of increasing latency. Timeouts (both
956           "ev_periodic" and "ev_timer") will not be affected. Setting this to
957           a non-null value will introduce an additional "ev_sleep ()" call
958           into most loop iterations. The sleep time ensures that libev will
959           not poll for I/O events more often then once per this interval, on
960           average (as long as the host time resolution is good enough).
961
962           Likewise, by setting a higher timeout collect interval you allow
963           libev to spend more time collecting timeouts, at the expense of
964           increased latency/jitter/inexactness (the watcher callback will be
965           called later). "ev_io" watchers will not be affected. Setting this
966           to a non-null value will not introduce any overhead in libev.
967
968           Many (busy) programs can usually benefit by setting the I/O collect
969           interval to a value near 0.1 or so, which is often enough for
970           interactive servers (of course not for games), likewise for
971           timeouts. It usually doesn't make much sense to set it to a lower
972           value than 0.01, as this approaches the timing granularity of most
973           systems. Note that if you do transactions with the outside world
974           and you can't increase the parallelity, then this setting will
975           limit your transaction rate (if you need to poll once per
976           transaction and the I/O collect interval is 0.01, then you can't do
977           more than 100 transactions per second).
978
979           Setting the timeout collect interval can improve the opportunity
980           for saving power, as the program will "bundle" timer callback
981           invocations that are "near" in time together, by delaying some,
982           thus reducing the number of times the process sleeps and wakes up
983           again. Another useful technique to reduce iterations/wake-ups is to
984           use "ev_periodic" watchers and make sure they fire on, say, one-
985           second boundaries only.
986
987           Example: we only need 0.1s timeout granularity, and we wish not to
988           poll more often than 100 times per second:
989
990              ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
991              ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
992
993       ev_invoke_pending (loop)
994           This call will simply invoke all pending watchers while resetting
995           their pending state. Normally, "ev_run" does this automatically
996           when required, but when overriding the invoke callback this call
997           comes handy. This function can be invoked from a watcher - this can
998           be useful for example when you want to do some lengthy calculation
999           and want to pass further event handling to another thread (you
1000           still have to make sure only one thread executes within
1001           "ev_invoke_pending" or "ev_run" of course).
1002
1003       int ev_pending_count (loop)
1004           Returns the number of pending watchers - zero indicates that no
1005           watchers are pending.
1006
1007       ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
1008           This overrides the invoke pending functionality of the loop:
1009           Instead of invoking all pending watchers when there are any,
1010           "ev_run" will call this callback instead. This is useful, for
1011           example, when you want to invoke the actual watchers inside another
1012           context (another thread etc.).
1013
1014           If you want to reset the callback, use "ev_invoke_pending" as new
1015           callback.
1016
1017       ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void
1018       (*acquire)(EV_P) throw ())
1019           Sometimes you want to share the same loop between multiple threads.
1020           This can be done relatively simply by putting mutex_lock/unlock
1021           calls around each call to a libev function.
1022
1023           However, "ev_run" can run an indefinite time, so it is not feasible
1024           to wait for it to return. One way around this is to wake up the
1025           event loop via "ev_break" and "ev_async_send", another way is to
1026           set these release and acquire callbacks on the loop.
1027
1028           When set, then "release" will be called just before the thread is
1029           suspended waiting for new events, and "acquire" is called just
1030           afterwards.
1031
1032           Ideally, "release" will just call your mutex_unlock function, and
1033           "acquire" will just call the mutex_lock function again.
1034
1035           While event loop modifications are allowed between invocations of
1036           "release" and "acquire" (that's their only purpose after all), no
1037           modifications done will affect the event loop, i.e. adding watchers
1038           will have no effect on the set of file descriptors being watched,
1039           or the time waited. Use an "ev_async" watcher to wake up "ev_run"
1040           when you want it to take note of any changes you made.
1041
1042           In theory, threads executing "ev_run" will be async-cancel safe
1043           between invocations of "release" and "acquire".
1044
1045           See also the locking example in the "THREADS" section later in this
1046           document.
1047
1048       ev_set_userdata (loop, void *data)
1049       void *ev_userdata (loop)
1050           Set and retrieve a single "void *" associated with a loop. When
1051           "ev_set_userdata" has never been called, then "ev_userdata" returns
1052           0.
1053
1054           These two functions can be used to associate arbitrary data with a
1055           loop, and are intended solely for the "invoke_pending_cb",
1056           "release" and "acquire" callbacks described above, but of course
1057           can be (ab-)used for any other purpose as well.
1058
1059       ev_verify (loop)
1060           This function only does something when "EV_VERIFY" support has been
1061           compiled in, which is the default for non-minimal builds. It tries
1062           to go through all internal structures and checks them for validity.
1063           If anything is found to be inconsistent, it will print an error
1064           message to standard error and call "abort ()".
1065
1066           This can be used to catch bugs inside libev itself: under normal
1067           circumstances, this function will never abort as of course libev
1068           keeps its data structures consistent.
1069

ANATOMY OF A WATCHER

1071       In the following description, uppercase "TYPE" in names stands for the
1072       watcher type, e.g. "ev_TYPE_start" can mean "ev_timer_start" for timer
1073       watchers and "ev_io_start" for I/O watchers.
1074
1075       A watcher is an opaque structure that you allocate and register to
1076       record your interest in some event. To make a concrete example, imagine
1077       you want to wait for STDIN to become readable, you would create an
1078       "ev_io" watcher for that:
1079
1080          static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
1081          {
1082            ev_io_stop (w);
1083            ev_break (loop, EVBREAK_ALL);
1084          }
1085
1086          struct ev_loop *loop = ev_default_loop (0);
1087
1088          ev_io stdin_watcher;
1089
1090          ev_init (&stdin_watcher, my_cb);
1091          ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
1092          ev_io_start (loop, &stdin_watcher);
1093
1094          ev_run (loop, 0);
1095
1096       As you can see, you are responsible for allocating the memory for your
1097       watcher structures (and it is usually a bad idea to do this on the
1098       stack).
1099
1100       Each watcher has an associated watcher structure (called "struct
1101       ev_TYPE" or simply "ev_TYPE", as typedefs are provided for all watcher
1102       structs).
1103
1104       Each watcher structure must be initialised by a call to "ev_init
1105       (watcher *, callback)", which expects a callback to be provided. This
1106       callback is invoked each time the event occurs (or, in the case of I/O
1107       watchers, each time the event loop detects that the file descriptor
1108       given is readable and/or writable).
1109
1110       Each watcher type further has its own "ev_TYPE_set (watcher *, ...)"
1111       macro to configure it, with arguments specific to the watcher type.
1112       There is also a macro to combine initialisation and setting in one
1113       call: "ev_TYPE_init (watcher *, callback, ...)".
1114
1115       To make the watcher actually watch out for events, you have to start it
1116       with a watcher-specific start function ("ev_TYPE_start (loop, watcher
1117       *)"), and you can stop watching for events at any time by calling the
1118       corresponding stop function ("ev_TYPE_stop (loop, watcher *)".
1119
1120       As long as your watcher is active (has been started but not stopped)
1121       you must not touch the values stored in it. Most specifically you must
1122       never reinitialise it or call its "ev_TYPE_set" macro.
1123
1124       Each and every callback receives the event loop pointer as first, the
1125       registered watcher structure as second, and a bitset of received events
1126       as third argument.
1127
1128       The received events usually include a single bit per event type
1129       received (you can receive multiple events at the same time). The
1130       possible bit masks are:
1131
1132       "EV_READ"
1133       "EV_WRITE"
1134           The file descriptor in the "ev_io" watcher has become readable
1135           and/or writable.
1136
1137       "EV_TIMER"
1138           The "ev_timer" watcher has timed out.
1139
1140       "EV_PERIODIC"
1141           The "ev_periodic" watcher has timed out.
1142
1143       "EV_SIGNAL"
1144           The signal specified in the "ev_signal" watcher has been received
1145           by a thread.
1146
1147       "EV_CHILD"
1148           The pid specified in the "ev_child" watcher has received a status
1149           change.
1150
1151       "EV_STAT"
1152           The path specified in the "ev_stat" watcher changed its attributes
1153           somehow.
1154
1155       "EV_IDLE"
1156           The "ev_idle" watcher has determined that you have nothing better
1157           to do.
1158
1159       "EV_PREPARE"
1160       "EV_CHECK"
1161           All "ev_prepare" watchers are invoked just before "ev_run" starts
1162           to gather new events, and all "ev_check" watchers are queued (not
1163           invoked) just after "ev_run" has gathered them, but before it
1164           queues any callbacks for any received events. That means
1165           "ev_prepare" watchers are the last watchers invoked before the
1166           event loop sleeps or polls for new events, and "ev_check" watchers
1167           will be invoked before any other watchers of the same or lower
1168           priority within an event loop iteration.
1169
1170           Callbacks of both watcher types can start and stop as many watchers
1171           as they want, and all of them will be taken into account (for
1172           example, a "ev_prepare" watcher might start an idle watcher to keep
1173           "ev_run" from blocking).
1174
1175       "EV_EMBED"
1176           The embedded event loop specified in the "ev_embed" watcher needs
1177           attention.
1178
1179       "EV_FORK"
1180           The event loop has been resumed in the child process after fork
1181           (see "ev_fork").
1182
1183       "EV_CLEANUP"
1184           The event loop is about to be destroyed (see "ev_cleanup").
1185
1186       "EV_ASYNC"
1187           The given async watcher has been asynchronously notified (see
1188           "ev_async").
1189
1190       "EV_CUSTOM"
1191           Not ever sent (or otherwise used) by libev itself, but can be
1192           freely used by libev users to signal watchers (e.g. via
1193           "ev_feed_event").
1194
1195       "EV_ERROR"
1196           An unspecified error has occurred, the watcher has been stopped.
1197           This might happen because the watcher could not be properly started
1198           because libev ran out of memory, a file descriptor was found to be
1199           closed or any other problem. Libev considers these application
1200           bugs.
1201
1202           You best act on it by reporting the problem and somehow coping with
1203           the watcher being stopped. Note that well-written programs should
1204           not receive an error ever, so when your watcher receives it, this
1205           usually indicates a bug in your program.
1206
1207           Libev will usually signal a few "dummy" events together with an
1208           error, for example it might indicate that a fd is readable or
1209           writable, and if your callbacks is well-written it can just attempt
1210           the operation and cope with the error from read() or write(). This
1211           will not work in multi-threaded programs, though, as the fd could
1212           already be closed and reused for another thing, so beware.
1213
1214   GENERIC WATCHER FUNCTIONS
1215       "ev_init" (ev_TYPE *watcher, callback)
1216           This macro initialises the generic portion of a watcher. The
1217           contents of the watcher object can be arbitrary (so "malloc" will
1218           do). Only the generic parts of the watcher are initialised, you
1219           need to call the type-specific "ev_TYPE_set" macro afterwards to
1220           initialise the type-specific parts. For each type there is also a
1221           "ev_TYPE_init" macro which rolls both calls into one.
1222
1223           You can reinitialise a watcher at any time as long as it has been
1224           stopped (or never started) and there are no pending events
1225           outstanding.
1226
1227           The callback is always of type "void (*)(struct ev_loop *loop,
1228           ev_TYPE *watcher, int revents)".
1229
1230           Example: Initialise an "ev_io" watcher in two steps.
1231
1232              ev_io w;
1233              ev_init (&w, my_cb);
1234              ev_io_set (&w, STDIN_FILENO, EV_READ);
1235
1236       "ev_TYPE_set" (ev_TYPE *watcher, [args])
1237           This macro initialises the type-specific parts of a watcher. You
1238           need to call "ev_init" at least once before you call this macro,
1239           but you can call "ev_TYPE_set" any number of times. You must not,
1240           however, call this macro on a watcher that is active (it can be
1241           pending, however, which is a difference to the "ev_init" macro).
1242
1243           Although some watcher types do not have type-specific arguments
1244           (e.g. "ev_prepare") you still need to call its "set" macro.
1245
1246           See "ev_init", above, for an example.
1247
1248       "ev_TYPE_init" (ev_TYPE *watcher, callback, [args])
1249           This convenience macro rolls both "ev_init" and "ev_TYPE_set" macro
1250           calls into a single call. This is the most convenient method to
1251           initialise a watcher. The same limitations apply, of course.
1252
1253           Example: Initialise and set an "ev_io" watcher in one step.
1254
1255              ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
1256
1257       "ev_TYPE_start" (loop, ev_TYPE *watcher)
1258           Starts (activates) the given watcher. Only active watchers will
1259           receive events. If the watcher is already active nothing will
1260           happen.
1261
1262           Example: Start the "ev_io" watcher that is being abused as example
1263           in this whole section.
1264
1265              ev_io_start (EV_DEFAULT_UC, &w);
1266
1267       "ev_TYPE_stop" (loop, ev_TYPE *watcher)
1268           Stops the given watcher if active, and clears the pending status
1269           (whether the watcher was active or not).
1270
1271           It is possible that stopped watchers are pending - for example,
1272           non-repeating timers are being stopped when they become pending -
1273           but calling "ev_TYPE_stop" ensures that the watcher is neither
1274           active nor pending. If you want to free or reuse the memory used by
1275           the watcher it is therefore a good idea to always call its
1276           "ev_TYPE_stop" function.
1277
1278       bool ev_is_active (ev_TYPE *watcher)
1279           Returns a true value iff the watcher is active (i.e. it has been
1280           started and not yet been stopped). As long as a watcher is active
1281           you must not modify it.
1282
1283       bool ev_is_pending (ev_TYPE *watcher)
1284           Returns a true value iff the watcher is pending, (i.e. it has
1285           outstanding events but its callback has not yet been invoked). As
1286           long as a watcher is pending (but not active) you must not call an
1287           init function on it (but "ev_TYPE_set" is safe), you must not
1288           change its priority, and you must make sure the watcher is
1289           available to libev (e.g. you cannot "free ()" it).
1290
1291       callback ev_cb (ev_TYPE *watcher)
1292           Returns the callback currently set on the watcher.
1293
1294       ev_set_cb (ev_TYPE *watcher, callback)
1295           Change the callback. You can change the callback at virtually any
1296           time (modulo threads).
1297
1298       ev_set_priority (ev_TYPE *watcher, int priority)
1299       int ev_priority (ev_TYPE *watcher)
1300           Set and query the priority of the watcher. The priority is a small
1301           integer between "EV_MAXPRI" (default: 2) and "EV_MINPRI" (default:
1302           "-2"). Pending watchers with higher priority will be invoked before
1303           watchers with lower priority, but priority will not keep watchers
1304           from being executed (except for "ev_idle" watchers).
1305
1306           If you need to suppress invocation when higher priority events are
1307           pending you need to look at "ev_idle" watchers, which provide this
1308           functionality.
1309
1310           You must not change the priority of a watcher as long as it is
1311           active or pending.
1312
1313           Setting a priority outside the range of "EV_MINPRI" to "EV_MAXPRI"
1314           is fine, as long as you do not mind that the priority value you
1315           query might or might not have been clamped to the valid range.
1316
1317           The default priority used by watchers when no priority has been set
1318           is always 0, which is supposed to not be too high and not be too
1319           low :).
1320
1321           See "WATCHER PRIORITY MODELS", below, for a more thorough treatment
1322           of priorities.
1323
1324       ev_invoke (loop, ev_TYPE *watcher, int revents)
1325           Invoke the "watcher" with the given "loop" and "revents". Neither
1326           "loop" nor "revents" need to be valid as long as the watcher
1327           callback can deal with that fact, as both are simply passed through
1328           to the callback.
1329
1330       int ev_clear_pending (loop, ev_TYPE *watcher)
1331           If the watcher is pending, this function clears its pending status
1332           and returns its "revents" bitset (as if its callback was invoked).
1333           If the watcher isn't pending it does nothing and returns 0.
1334
1335           Sometimes it can be useful to "poll" a watcher instead of waiting
1336           for its callback to be invoked, which can be accomplished with this
1337           function.
1338
1339       ev_feed_event (loop, ev_TYPE *watcher, int revents)
1340           Feeds the given event set into the event loop, as if the specified
1341           event had happened for the specified watcher (which must be a
1342           pointer to an initialised but not necessarily started event
1343           watcher). Obviously you must not free the watcher as long as it has
1344           pending events.
1345
1346           Stopping the watcher, letting libev invoke it, or calling
1347           "ev_clear_pending" will clear the pending event, even if the
1348           watcher was not started in the first place.
1349
1350           See also "ev_feed_fd_event" and "ev_feed_signal_event" for related
1351           functions that do not need a watcher.
1352
1353       See also the "ASSOCIATING CUSTOM DATA WITH A WATCHER" and "BUILDING
1354       YOUR OWN COMPOSITE WATCHERS" idioms.
1355
1356   WATCHER STATES
1357       There are various watcher states mentioned throughout this manual -
1358       active, pending and so on. In this section these states and the rules
1359       to transition between them will be described in more detail - and while
1360       these rules might look complicated, they usually do "the right thing".
1361
1362       initialised
1363           Before a watcher can be registered with the event loop it has to be
1364           initialised. This can be done with a call to "ev_TYPE_init", or
1365           calls to "ev_init" followed by the watcher-specific "ev_TYPE_set"
1366           function.
1367
1368           In this state it is simply some block of memory that is suitable
1369           for use in an event loop. It can be moved around, freed, reused
1370           etc. at will - as long as you either keep the memory contents
1371           intact, or call "ev_TYPE_init" again.
1372
1373       started/running/active
1374           Once a watcher has been started with a call to "ev_TYPE_start" it
1375           becomes property of the event loop, and is actively waiting for
1376           events. While in this state it cannot be accessed (except in a few
1377           documented ways), moved, freed or anything else - the only legal
1378           thing is to keep a pointer to it, and call libev functions on it
1379           that are documented to work on active watchers.
1380
1381       pending
1382           If a watcher is active and libev determines that an event it is
1383           interested in has occurred (such as a timer expiring), it will
1384           become pending. It will stay in this pending state until either it
1385           is stopped or its callback is about to be invoked, so it is not
1386           normally pending inside the watcher callback.
1387
1388           The watcher might or might not be active while it is pending (for
1389           example, an expired non-repeating timer can be pending but no
1390           longer active). If it is stopped, it can be freely accessed (e.g.
1391           by calling "ev_TYPE_set"), but it is still property of the event
1392           loop at this time, so cannot be moved, freed or reused. And if it
1393           is active the rules described in the previous item still apply.
1394
1395           It is also possible to feed an event on a watcher that is not
1396           active (e.g.  via "ev_feed_event"), in which case it becomes
1397           pending without being active.
1398
1399       stopped
1400           A watcher can be stopped implicitly by libev (in which case it
1401           might still be pending), or explicitly by calling its
1402           "ev_TYPE_stop" function. The latter will clear any pending state
1403           the watcher might be in, regardless of whether it was active or
1404           not, so stopping a watcher explicitly before freeing it is often a
1405           good idea.
1406
1407           While stopped (and not pending) the watcher is essentially in the
1408           initialised state, that is, it can be reused, moved, modified in
1409           any way you wish (but when you trash the memory block, you need to
1410           "ev_TYPE_init" it again).
1411
1412   WATCHER PRIORITY MODELS
1413       Many event loops support watcher priorities, which are usually small
1414       integers that influence the ordering of event callback invocation
1415       between watchers in some way, all else being equal.
1416
1417       In libev, Watcher priorities can be set using "ev_set_priority". See
1418       its description for the more technical details such as the actual
1419       priority range.
1420
1421       There are two common ways how these these priorities are being
1422       interpreted by event loops:
1423
1424       In the more common lock-out model, higher priorities "lock out"
1425       invocation of lower priority watchers, which means as long as higher
1426       priority watchers receive events, lower priority watchers are not being
1427       invoked.
1428
1429       The less common only-for-ordering model uses priorities solely to order
1430       callback invocation within a single event loop iteration: Higher
1431       priority watchers are invoked before lower priority ones, but they all
1432       get invoked before polling for new events.
1433
1434       Libev uses the second (only-for-ordering) model for all its watchers
1435       except for idle watchers (which use the lock-out model).
1436
1437       The rationale behind this is that implementing the lock-out model for
1438       watchers is not well supported by most kernel interfaces, and most
1439       event libraries will just poll for the same events again and again as
1440       long as their callbacks have not been executed, which is very
1441       inefficient in the common case of one high-priority watcher locking out
1442       a mass of lower priority ones.
1443
1444       Static (ordering) priorities are most useful when you have two or more
1445       watchers handling the same resource: a typical usage example is having
1446       an "ev_io" watcher to receive data, and an associated "ev_timer" to
1447       handle timeouts. Under load, data might be received while the program
1448       handles other jobs, but since timers normally get invoked first, the
1449       timeout handler will be executed before checking for data. In that
1450       case, giving the timer a lower priority than the I/O watcher ensures
1451       that I/O will be handled first even under adverse conditions (which is
1452       usually, but not always, what you want).
1453
1454       Since idle watchers use the "lock-out" model, meaning that idle
1455       watchers will only be executed when no same or higher priority watchers
1456       have received events, they can be used to implement the "lock-out"
1457       model when required.
1458
1459       For example, to emulate how many other event libraries handle
1460       priorities, you can associate an "ev_idle" watcher to each such
1461       watcher, and in the normal watcher callback, you just start the idle
1462       watcher. The real processing is done in the idle watcher callback. This
1463       causes libev to continuously poll and process kernel event data for the
1464       watcher, but when the lock-out case is known to be rare (which in turn
1465       is rare :), this is workable.
1466
1467       Usually, however, the lock-out model implemented that way will perform
1468       miserably under the type of load it was designed to handle. In that
1469       case, it might be preferable to stop the real watcher before starting
1470       the idle watcher, so the kernel will not have to process the event in
1471       case the actual processing will be delayed for considerable time.
1472
1473       Here is an example of an I/O watcher that should run at a strictly
1474       lower priority than the default, and which should only process data
1475       when no other events are pending:
1476
1477          ev_idle idle; // actual processing watcher
1478          ev_io io;     // actual event watcher
1479
1480          static void
1481          io_cb (EV_P_ ev_io *w, int revents)
1482          {
1483            // stop the I/O watcher, we received the event, but
1484            // are not yet ready to handle it.
1485            ev_io_stop (EV_A_ w);
1486
1487            // start the idle watcher to handle the actual event.
1488            // it will not be executed as long as other watchers
1489            // with the default priority are receiving events.
1490            ev_idle_start (EV_A_ &idle);
1491          }
1492
1493          static void
1494          idle_cb (EV_P_ ev_idle *w, int revents)
1495          {
1496            // actual processing
1497            read (STDIN_FILENO, ...);
1498
1499            // have to start the I/O watcher again, as
1500            // we have handled the event
1501            ev_io_start (EV_P_ &io);
1502          }
1503
1504          // initialisation
1505          ev_idle_init (&idle, idle_cb);
1506          ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
1507          ev_io_start (EV_DEFAULT_ &io);
1508
1509       In the "real" world, it might also be beneficial to start a timer, so
1510       that low-priority connections can not be locked out forever under load.
1511       This enables your program to keep a lower latency for important
1512       connections during short periods of high load, while not completely
1513       locking out less important ones.
1514

WATCHER TYPES

1516       This section describes each watcher in detail, but will not repeat
1517       information given in the last section. Any initialisation/set macros,
1518       functions and members specific to the watcher type are explained.
1519
1520       Members are additionally marked with either [read-only], meaning that,
1521       while the watcher is active, you can look at the member and expect some
1522       sensible content, but you must not modify it (you can modify it while
1523       the watcher is stopped to your hearts content), or [read-write], which
1524       means you can expect it to have some sensible content while the watcher
1525       is active, but you can also modify it. Modifying it may not do
1526       something sensible or take immediate effect (or do anything at all),
1527       but libev will not crash or malfunction in any way.
1528
1529   "ev_io" - is this file descriptor readable or writable?
1530       I/O watchers check whether a file descriptor is readable or writable in
1531       each iteration of the event loop, or, more precisely, when reading
1532       would not block the process and writing would at least be able to write
1533       some data. This behaviour is called level-triggering because you keep
1534       receiving events as long as the condition persists. Remember you can
1535       stop the watcher if you don't want to act on the event and neither want
1536       to receive future events.
1537
1538       In general you can register as many read and/or write event watchers
1539       per fd as you want (as long as you don't confuse yourself). Setting all
1540       file descriptors to non-blocking mode is also usually a good idea (but
1541       not required if you know what you are doing).
1542
1543       Another thing you have to watch out for is that it is quite easy to
1544       receive "spurious" readiness notifications, that is, your callback
1545       might be called with "EV_READ" but a subsequent "read"(2) will actually
1546       block because there is no data. It is very easy to get into this
1547       situation even with a relatively standard program structure. Thus it is
1548       best to always use non-blocking I/O: An extra "read"(2) returning
1549       "EAGAIN" is far preferable to a program hanging until some data
1550       arrives.
1551
1552       If you cannot run the fd in non-blocking mode (for example you should
1553       not play around with an Xlib connection), then you have to separately
1554       re-test whether a file descriptor is really ready with a known-to-be
1555       good interface such as poll (fortunately in the case of Xlib, it
1556       already does this on its own, so its quite safe to use). Some people
1557       additionally use "SIGALRM" and an interval timer, just to be sure you
1558       won't block indefinitely.
1559
1560       But really, best use non-blocking mode.
1561
1562       The special problem of disappearing file descriptors
1563
1564       Some backends (e.g. kqueue, epoll) need to be told about closing a file
1565       descriptor (either due to calling "close" explicitly or any other
1566       means, such as "dup2"). The reason is that you register interest in
1567       some file descriptor, but when it goes away, the operating system will
1568       silently drop this interest. If another file descriptor with the same
1569       number then is registered with libev, there is no efficient way to see
1570       that this is, in fact, a different file descriptor.
1571
1572       To avoid having to explicitly tell libev about such cases, libev
1573       follows the following policy:  Each time "ev_io_set" is being called,
1574       libev will assume that this is potentially a new file descriptor,
1575       otherwise it is assumed that the file descriptor stays the same. That
1576       means that you have to call "ev_io_set" (or "ev_io_init") when you
1577       change the descriptor even if the file descriptor number itself did not
1578       change.
1579
1580       This is how one would do it normally anyway, the important point is
1581       that the libev application should not optimise around libev but should
1582       leave optimisations to libev.
1583
1584       The special problem of dup'ed file descriptors
1585
1586       Some backends (e.g. epoll), cannot register events for file
1587       descriptors, but only events for the underlying file descriptions. That
1588       means when you have "dup ()"'ed file descriptors or weirder
1589       constellations, and register events for them, only one file descriptor
1590       might actually receive events.
1591
1592       There is no workaround possible except not registering events for
1593       potentially "dup ()"'ed file descriptors, or to resort to
1594       "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1595
1596       The special problem of files
1597
1598       Many people try to use "select" (or libev) on file descriptors
1599       representing files, and expect it to become ready when their program
1600       doesn't block on disk accesses (which can take a long time on their
1601       own).
1602
1603       However, this cannot ever work in the "expected" way - you get a
1604       readiness notification as soon as the kernel knows whether and how much
1605       data is there, and in the case of open files, that's always the case,
1606       so you always get a readiness notification instantly, and your read (or
1607       possibly write) will still block on the disk I/O.
1608
1609       Another way to view it is that in the case of sockets, pipes, character
1610       devices and so on, there is another party (the sender) that delivers
1611       data on its own, but in the case of files, there is no such thing: the
1612       disk will not send data on its own, simply because it doesn't know what
1613       you wish to read - you would first have to request some data.
1614
1615       Since files are typically not-so-well supported by advanced
1616       notification mechanism, libev tries hard to emulate POSIX behaviour
1617       with respect to files, even though you should not use it. The reason
1618       for this is convenience: sometimes you want to watch STDIN or STDOUT,
1619       which is usually a tty, often a pipe, but also sometimes files or
1620       special devices (for example, "epoll" on Linux works with /dev/random
1621       but not with /dev/urandom), and even though the file might better be
1622       served with asynchronous I/O instead of with non-blocking I/O, it is
1623       still useful when it "just works" instead of freezing.
1624
1625       So avoid file descriptors pointing to files when you know it (e.g. use
1626       libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
1627       when you rarely read from a file instead of from a socket, and want to
1628       reuse the same code path.
1629
1630       The special problem of fork
1631
1632       Some backends (epoll, kqueue) do not support "fork ()" at all or
1633       exhibit useless behaviour. Libev fully supports fork, but needs to be
1634       told about it in the child if you want to continue to use it in the
1635       child.
1636
1637       To support fork in your child processes, you have to call "ev_loop_fork
1638       ()" after a fork in the child, enable "EVFLAG_FORKCHECK", or resort to
1639       "EVBACKEND_SELECT" or "EVBACKEND_POLL".
1640
1641       The special problem of SIGPIPE
1642
1643       While not really specific to libev, it is easy to forget about
1644       "SIGPIPE": when writing to a pipe whose other end has been closed, your
1645       program gets sent a SIGPIPE, which, by default, aborts your program.
1646       For most programs this is sensible behaviour, for daemons, this is
1647       usually undesirable.
1648
1649       So when you encounter spurious, unexplained daemon exits, make sure you
1650       ignore SIGPIPE (and maybe make sure you log the exit status of your
1651       daemon somewhere, as that would have given you a big clue).
1652
1653       The special problem of accept()ing when you can't
1654
1655       Many implementations of the POSIX "accept" function (for example, found
1656       in post-2004 Linux) have the peculiar behaviour of not removing a
1657       connection from the pending queue in all error cases.
1658
1659       For example, larger servers often run out of file descriptors (because
1660       of resource limits), causing "accept" to fail with "ENFILE" but not
1661       rejecting the connection, leading to libev signalling readiness on the
1662       next iteration again (the connection still exists after all), and
1663       typically causing the program to loop at 100% CPU usage.
1664
1665       Unfortunately, the set of errors that cause this issue differs between
1666       operating systems, there is usually little the app can do to remedy the
1667       situation, and no known thread-safe method of removing the connection
1668       to cope with overload is known (to me).
1669
1670       One of the easiest ways to handle this situation is to just ignore it -
1671       when the program encounters an overload, it will just loop until the
1672       situation is over. While this is a form of busy waiting, no OS offers
1673       an event-based way to handle this situation, so it's the best one can
1674       do.
1675
1676       A better way to handle the situation is to log any errors other than
1677       "EAGAIN" and "EWOULDBLOCK", making sure not to flood the log with such
1678       messages, and continue as usual, which at least gives the user an idea
1679       of what could be wrong ("raise the ulimit!"). For extra points one
1680       could stop the "ev_io" watcher on the listening fd "for a while", which
1681       reduces CPU usage.
1682
1683       If your program is single-threaded, then you could also keep a dummy
1684       file descriptor for overload situations (e.g. by opening /dev/null),
1685       and when you run into "ENFILE" or "EMFILE", close it, run "accept",
1686       close that fd, and create a new dummy fd. This will gracefully refuse
1687       clients under typical overload conditions.
1688
1689       The last way to handle it is to simply log the error and "exit", as is
1690       often done with "malloc" failures, but this results in an easy
1691       opportunity for a DoS attack.
1692
1693       Watcher-Specific Functions
1694
1695       ev_io_init (ev_io *, callback, int fd, int events)
1696       ev_io_set (ev_io *, int fd, int events)
1697           Configures an "ev_io" watcher. The "fd" is the file descriptor to
1698           receive events for and "events" is either "EV_READ", "EV_WRITE" or
1699           "EV_READ | EV_WRITE", to express the desire to receive the given
1700           events.
1701
1702       int fd [read-only]
1703           The file descriptor being watched.
1704
1705       int events [read-only]
1706           The events being watched.
1707
1708       Examples
1709
1710       Example: Call "stdin_readable_cb" when STDIN_FILENO has become, well
1711       readable, but only once. Since it is likely line-buffered, you could
1712       attempt to read a whole line in the callback.
1713
1714          static void
1715          stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
1716          {
1717             ev_io_stop (loop, w);
1718            .. read from stdin here (or from w->fd) and handle any I/O errors
1719          }
1720
1721          ...
1722          struct ev_loop *loop = ev_default_init (0);
1723          ev_io stdin_readable;
1724          ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
1725          ev_io_start (loop, &stdin_readable);
1726          ev_run (loop, 0);
1727
1728   "ev_timer" - relative and optionally repeating timeouts
1729       Timer watchers are simple relative timers that generate an event after
1730       a given time, and optionally repeating in regular intervals after that.
1731
1732       The timers are based on real time, that is, if you register an event
1733       that times out after an hour and you reset your system clock to January
1734       last year, it will still time out after (roughly) one hour. "Roughly"
1735       because detecting time jumps is hard, and some inaccuracies are
1736       unavoidable (the monotonic clock option helps a lot here).
1737
1738       The callback is guaranteed to be invoked only after its timeout has
1739       passed (not at, so on systems with very low-resolution clocks this
1740       might introduce a small delay, see "the special problem of being too
1741       early", below). If multiple timers become ready during the same loop
1742       iteration then the ones with earlier time-out values are invoked before
1743       ones of the same priority with later time-out values (but this is no
1744       longer true when a callback calls "ev_run" recursively).
1745
1746       Be smart about timeouts
1747
1748       Many real-world problems involve some kind of timeout, usually for
1749       error recovery. A typical example is an HTTP request - if the other
1750       side hangs, you want to raise some error after a while.
1751
1752       What follows are some ways to handle this problem, from obvious and
1753       inefficient to smart and efficient.
1754
1755       In the following, a 60 second activity timeout is assumed - a timeout
1756       that gets reset to 60 seconds each time there is activity (e.g. each
1757       time some data or other life sign was received).
1758
1759       1. Use a timer and stop, reinitialise and start it on activity.
1760           This is the most obvious, but not the most simple way: In the
1761           beginning, start the watcher:
1762
1763              ev_timer_init (timer, callback, 60., 0.);
1764              ev_timer_start (loop, timer);
1765
1766           Then, each time there is some activity, "ev_timer_stop" it,
1767           initialise it and start it again:
1768
1769              ev_timer_stop (loop, timer);
1770              ev_timer_set (timer, 60., 0.);
1771              ev_timer_start (loop, timer);
1772
1773           This is relatively simple to implement, but means that each time
1774           there is some activity, libev will first have to remove the timer
1775           from its internal data structure and then add it again. Libev tries
1776           to be fast, but it's still not a constant-time operation.
1777
1778       2. Use a timer and re-start it with "ev_timer_again" inactivity.
1779           This is the easiest way, and involves using "ev_timer_again"
1780           instead of "ev_timer_start".
1781
1782           To implement this, configure an "ev_timer" with a "repeat" value of
1783           60 and then call "ev_timer_again" at start and each time you
1784           successfully read or write some data. If you go into an idle state
1785           where you do not expect data to travel on the socket, you can
1786           "ev_timer_stop" the timer, and "ev_timer_again" will automatically
1787           restart it if need be.
1788
1789           That means you can ignore both the "ev_timer_start" function and
1790           the "after" argument to "ev_timer_set", and only ever use the
1791           "repeat" member and "ev_timer_again".
1792
1793           At start:
1794
1795              ev_init (timer, callback);
1796              timer->repeat = 60.;
1797              ev_timer_again (loop, timer);
1798
1799           Each time there is some activity:
1800
1801              ev_timer_again (loop, timer);
1802
1803           It is even possible to change the time-out on the fly, regardless
1804           of whether the watcher is active or not:
1805
1806              timer->repeat = 30.;
1807              ev_timer_again (loop, timer);
1808
1809           This is slightly more efficient then stopping/starting the timer
1810           each time you want to modify its timeout value, as libev does not
1811           have to completely remove and re-insert the timer from/into its
1812           internal data structure.
1813
1814           It is, however, even simpler than the "obvious" way to do it.
1815
1816       3. Let the timer time out, but then re-arm it as required.
1817           This method is more tricky, but usually most efficient: Most
1818           timeouts are relatively long compared to the intervals between
1819           other activity - in our example, within 60 seconds, there are
1820           usually many I/O events with associated activity resets.
1821
1822           In this case, it would be more efficient to leave the "ev_timer"
1823           alone, but remember the time of last activity, and check for a real
1824           timeout only within the callback:
1825
1826              ev_tstamp timeout = 60.;
1827              ev_tstamp last_activity; // time of last activity
1828              ev_timer timer;
1829
1830              static void
1831              callback (EV_P_ ev_timer *w, int revents)
1832              {
1833                // calculate when the timeout would happen
1834                ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
1835
1836                // if negative, it means we the timeout already occurred
1837                if (after < 0.)
1838                  {
1839                    // timeout occurred, take action
1840                  }
1841                else
1842                  {
1843                    // callback was invoked, but there was some recent
1844                    // activity. simply restart the timer to time out
1845                    // after "after" seconds, which is the earliest time
1846                    // the timeout can occur.
1847                    ev_timer_set (w, after, 0.);
1848                    ev_timer_start (EV_A_ w);
1849                  }
1850              }
1851
1852           To summarise the callback: first calculate in how many seconds the
1853           timeout will occur (by calculating the absolute time when it would
1854           occur, "last_activity + timeout", and subtracting the current time,
1855           "ev_now (EV_A)" from that).
1856
1857           If this value is negative, then we are already past the timeout,
1858           i.e. we timed out, and need to do whatever is needed in this case.
1859
1860           Otherwise, we now the earliest time at which the timeout would
1861           trigger, and simply start the timer with this timeout value.
1862
1863           In other words, each time the callback is invoked it will check
1864           whether the timeout occurred. If not, it will simply reschedule
1865           itself to check again at the earliest time it could time out.
1866           Rinse. Repeat.
1867
1868           This scheme causes more callback invocations (about one every 60
1869           seconds minus half the average time between activity), but
1870           virtually no calls to libev to change the timeout.
1871
1872           To start the machinery, simply initialise the watcher and set
1873           "last_activity" to the current time (meaning there was some
1874           activity just now), then call the callback, which will "do the
1875           right thing" and start the timer:
1876
1877              last_activity = ev_now (EV_A);
1878              ev_init (&timer, callback);
1879              callback (EV_A_ &timer, 0);
1880
1881           When there is some activity, simply store the current time in
1882           "last_activity", no libev calls at all:
1883
1884              if (activity detected)
1885                last_activity = ev_now (EV_A);
1886
1887           When your timeout value changes, then the timeout can be changed by
1888           simply providing a new value, stopping the timer and calling the
1889           callback, which will again do the right thing (for example, time
1890           out immediately :).
1891
1892              timeout = new_value;
1893              ev_timer_stop (EV_A_ &timer);
1894              callback (EV_A_ &timer, 0);
1895
1896           This technique is slightly more complex, but in most cases where
1897           the time-out is unlikely to be triggered, much more efficient.
1898
1899       4. Wee, just use a double-linked list for your timeouts.
1900           If there is not one request, but many thousands (millions...), all
1901           employing some kind of timeout with the same timeout value, then
1902           one can do even better:
1903
1904           When starting the timeout, calculate the timeout value and put the
1905           timeout at the end of the list.
1906
1907           Then use an "ev_timer" to fire when the timeout at the beginning of
1908           the list is expected to fire (for example, using the technique #3).
1909
1910           When there is some activity, remove the timer from the list,
1911           recalculate the timeout, append it to the end of the list again,
1912           and make sure to update the "ev_timer" if it was taken from the
1913           beginning of the list.
1914
1915           This way, one can manage an unlimited number of timeouts in O(1)
1916           time for starting, stopping and updating the timers, at the expense
1917           of a major complication, and having to use a constant timeout. The
1918           constant timeout ensures that the list stays sorted.
1919
1920       So which method the best?
1921
1922       Method #2 is a simple no-brain-required solution that is adequate in
1923       most situations. Method #3 requires a bit more thinking, but handles
1924       many cases better, and isn't very complicated either. In most case,
1925       choosing either one is fine, with #3 being better in typical
1926       situations.
1927
1928       Method #1 is almost always a bad idea, and buys you nothing. Method #4
1929       is rather complicated, but extremely efficient, something that really
1930       pays off after the first million or so of active timers, i.e. it's
1931       usually overkill :)
1932
1933       The special problem of being too early
1934
1935       If you ask a timer to call your callback after three seconds, then you
1936       expect it to be invoked after three seconds - but of course, this
1937       cannot be guaranteed to infinite precision. Less obviously, it cannot
1938       be guaranteed to any precision by libev - imagine somebody suspending
1939       the process with a STOP signal for a few hours for example.
1940
1941       So, libev tries to invoke your callback as soon as possible after the
1942       delay has occurred, but cannot guarantee this.
1943
1944       A less obvious failure mode is calling your callback too early: many
1945       event loops compare timestamps with a "elapsed delay >= requested
1946       delay", but this can cause your callback to be invoked much earlier
1947       than you would expect.
1948
1949       To see why, imagine a system with a clock that only offers full second
1950       resolution (think windows if you can't come up with a broken enough OS
1951       yourself). If you schedule a one-second timer at the time 500.9, then
1952       the event loop will schedule your timeout to elapse at a system time of
1953       500 (500.9 truncated to the resolution) + 1, or 501.
1954
1955       If an event library looks at the timeout 0.1s later, it will see "501
1956       >= 501" and invoke the callback 0.1s after it was started, even though
1957       a one-second delay was requested - this is being "too early", despite
1958       best intentions.
1959
1960       This is the reason why libev will never invoke the callback if the
1961       elapsed delay equals the requested delay, but only when the elapsed
1962       delay is larger than the requested delay. In the example above, libev
1963       would only invoke the callback at system time 502, or 1.1s after the
1964       timer was started.
1965
1966       So, while libev cannot guarantee that your callback will be invoked
1967       exactly when requested, it can and does guarantee that the requested
1968       delay has actually elapsed, or in other words, it always errs on the
1969       "too late" side of things.
1970
1971       The special problem of time updates
1972
1973       Establishing the current time is a costly operation (it usually takes
1974       at least one system call): EV therefore updates its idea of the current
1975       time only before and after "ev_run" collects new events, which causes a
1976       growing difference between "ev_now ()" and "ev_time ()" when handling
1977       lots of events in one iteration.
1978
1979       The relative timeouts are calculated relative to the "ev_now ()" time.
1980       This is usually the right thing as this timestamp refers to the time of
1981       the event triggering whatever timeout you are modifying/starting. If
1982       you suspect event processing to be delayed and you need to base the
1983       timeout on the current time, use something like the following to adjust
1984       for it:
1985
1986          ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
1987
1988       If the event loop is suspended for a long time, you can also force an
1989       update of the time returned by "ev_now ()" by calling "ev_now_update
1990       ()", although that will push the event time of all outstanding events
1991       further into the future.
1992
1993       The special problem of unsynchronised clocks
1994
1995       Modern systems have a variety of clocks - libev itself uses the normal
1996       "wall clock" clock and, if available, the monotonic clock (to avoid
1997       time jumps).
1998
1999       Neither of these clocks is synchronised with each other or any other
2000       clock on the system, so "ev_time ()" might return a considerably
2001       different time than "gettimeofday ()" or "time ()". On a GNU/Linux
2002       system, for example, a call to "gettimeofday" might return a second
2003       count that is one higher than a directly following call to "time".
2004
2005       The moral of this is to only compare libev-related timestamps with
2006       "ev_time ()" and "ev_now ()", at least if you want better precision
2007       than a second or so.
2008
2009       One more problem arises due to this lack of synchronisation: if libev
2010       uses the system monotonic clock and you compare timestamps from
2011       "ev_time" or "ev_now" from when you started your timer and when your
2012       callback is invoked, you will find that sometimes the callback is a bit
2013       "early".
2014
2015       This is because "ev_timer"s work in real time, not wall clock time, so
2016       libev makes sure your callback is not invoked before the delay
2017       happened, measured according to the real time, not the system clock.
2018
2019       If your timeouts are based on a physical timescale (e.g. "time out this
2020       connection after 100 seconds") then this shouldn't bother you as it is
2021       exactly the right behaviour.
2022
2023       If you want to compare wall clock/system timestamps to your timers,
2024       then you need to use "ev_periodic"s, as these are based on the wall
2025       clock time, where your comparisons will always generate correct
2026       results.
2027
2028       The special problems of suspended animation
2029
2030       When you leave the server world it is quite customary to hit machines
2031       that can suspend/hibernate - what happens to the clocks during such a
2032       suspend?
2033
2034       Some quick tests made with a Linux 2.6.28 indicate that a suspend
2035       freezes all processes, while the clocks ("times", "CLOCK_MONOTONIC")
2036       continue to run until the system is suspended, but they will not
2037       advance while the system is suspended. That means, on resume, it will
2038       be as if the program was frozen for a few seconds, but the suspend time
2039       will not be counted towards "ev_timer" when a monotonic clock source is
2040       used. The real time clock advanced as expected, but if it is used as
2041       sole clocksource, then a long suspend would be detected as a time jump
2042       by libev, and timers would be adjusted accordingly.
2043
2044       I would not be surprised to see different behaviour in different
2045       between operating systems, OS versions or even different hardware.
2046
2047       The other form of suspend (job control, or sending a SIGSTOP) will see
2048       a time jump in the monotonic clocks and the realtime clock. If the
2049       program is suspended for a very long time, and monotonic clock sources
2050       are in use, then you can expect "ev_timer"s to expire as the full
2051       suspension time will be counted towards the timers. When no monotonic
2052       clock source is in use, then libev will again assume a timejump and
2053       adjust accordingly.
2054
2055       It might be beneficial for this latter case to call "ev_suspend" and
2056       "ev_resume" in code that handles "SIGTSTP", to at least get
2057       deterministic behaviour in this case (you can do nothing against
2058       "SIGSTOP").
2059
2060       Watcher-Specific Functions and Data Members
2061
2062       ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
2063       ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
2064           Configure the timer to trigger after "after" seconds (fractional
2065           and negative values are supported). If "repeat" is 0., then it will
2066           automatically be stopped once the timeout is reached. If it is
2067           positive, then the timer will automatically be configured to
2068           trigger again "repeat" seconds later, again, and again, until
2069           stopped manually.
2070
2071           The timer itself will do a best-effort at avoiding drift, that is,
2072           if you configure a timer to trigger every 10 seconds, then it will
2073           normally trigger at exactly 10 second intervals. If, however, your
2074           program cannot keep up with the timer (because it takes longer than
2075           those 10 seconds to do stuff) the timer will not fire more than
2076           once per event loop iteration.
2077
2078       ev_timer_again (loop, ev_timer *)
2079           This will act as if the timer timed out, and restarts it again if
2080           it is repeating. It basically works like calling "ev_timer_stop",
2081           updating the timeout to the "repeat" value and calling
2082           "ev_timer_start".
2083
2084           The exact semantics are as in the following rules, all of which
2085           will be applied to the watcher:
2086
2087           If the timer is pending, the pending status is always cleared.
2088           If the timer is started but non-repeating, stop it (as if it timed
2089           out, without invoking it).
2090           If the timer is repeating, make the "repeat" value the new timeout
2091           and start the timer, if necessary.
2092
2093           This sounds a bit complicated, see "Be smart about timeouts",
2094           above, for a usage example.
2095
2096       ev_tstamp ev_timer_remaining (loop, ev_timer *)
2097           Returns the remaining time until a timer fires. If the timer is
2098           active, then this time is relative to the current event loop time,
2099           otherwise it's the timeout value currently configured.
2100
2101           That is, after an "ev_timer_set (w, 5, 7)", "ev_timer_remaining"
2102           returns 5. When the timer is started and one second passes,
2103           "ev_timer_remaining" will return 4. When the timer expires and is
2104           restarted, it will return roughly 7 (likely slightly less as
2105           callback invocation takes some time, too), and so on.
2106
2107       ev_tstamp repeat [read-write]
2108           The current "repeat" value. Will be used each time the watcher
2109           times out or "ev_timer_again" is called, and determines the next
2110           timeout (if any), which is also when any modifications are taken
2111           into account.
2112
2113       Examples
2114
2115       Example: Create a timer that fires after 60 seconds.
2116
2117          static void
2118          one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
2119          {
2120            .. one minute over, w is actually stopped right here
2121          }
2122
2123          ev_timer mytimer;
2124          ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
2125          ev_timer_start (loop, &mytimer);
2126
2127       Example: Create a timeout timer that times out after 10 seconds of
2128       inactivity.
2129
2130          static void
2131          timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
2132          {
2133            .. ten seconds without any activity
2134          }
2135
2136          ev_timer mytimer;
2137          ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
2138          ev_timer_again (&mytimer); /* start timer */
2139          ev_run (loop, 0);
2140
2141          // and in some piece of code that gets executed on any "activity":
2142          // reset the timeout to start ticking again at 10 seconds
2143          ev_timer_again (&mytimer);
2144
2145   "ev_periodic" - to cron or not to cron?
2146       Periodic watchers are also timers of a kind, but they are very
2147       versatile (and unfortunately a bit complex).
2148
2149       Unlike "ev_timer", periodic watchers are not based on real time (or
2150       relative time, the physical time that passes) but on wall clock time
2151       (absolute time, the thing you can read on your calendar or clock). The
2152       difference is that wall clock time can run faster or slower than real
2153       time, and time jumps are not uncommon (e.g. when you adjust your wrist-
2154       watch).
2155
2156       You can tell a periodic watcher to trigger after some specific point in
2157       time: for example, if you tell a periodic watcher to trigger "in 10
2158       seconds" (by specifying e.g. "ev_now () + 10.", that is, an absolute
2159       time not a delay) and then reset your system clock to January of the
2160       previous year, then it will take a year or more to trigger the event
2161       (unlike an "ev_timer", which would still trigger roughly 10 seconds
2162       after starting it, as it uses a relative timeout).
2163
2164       "ev_periodic" watchers can also be used to implement vastly more
2165       complex timers, such as triggering an event on each "midnight, local
2166       time", or other complicated rules. This cannot easily be done with
2167       "ev_timer" watchers, as those cannot react to time jumps.
2168
2169       As with timers, the callback is guaranteed to be invoked only when the
2170       point in time where it is supposed to trigger has passed. If multiple
2171       timers become ready during the same loop iteration then the ones with
2172       earlier time-out values are invoked before ones with later time-out
2173       values (but this is no longer true when a callback calls "ev_run"
2174       recursively).
2175
2176       Watcher-Specific Functions and Data Members
2177
2178       ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp
2179       interval, reschedule_cb)
2180       ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval,
2181       reschedule_cb)
2182           Lots of arguments, let's sort it out... There are basically three
2183           modes of operation, and we will explain them from simplest to most
2184           complex:
2185
2186           ·   absolute timer (offset = absolute time, interval = 0,
2187               reschedule_cb = 0)
2188
2189               In this configuration the watcher triggers an event after the
2190               wall clock time "offset" has passed. It will not repeat and
2191               will not adjust when a time jump occurs, that is, if it is to
2192               be run at January 1st 2011 then it will be stopped and invoked
2193               when the system clock reaches or surpasses this point in time.
2194
2195           ·   repeating interval timer (offset = offset within interval,
2196               interval > 0, reschedule_cb = 0)
2197
2198               In this mode the watcher will always be scheduled to time out
2199               at the next "offset + N * interval" time (for some integer N,
2200               which can also be negative) and then repeat, regardless of any
2201               time jumps. The "offset" argument is merely an offset into the
2202               "interval" periods.
2203
2204               This can be used to create timers that do not drift with
2205               respect to the system clock, for example, here is an
2206               "ev_periodic" that triggers each hour, on the hour (with
2207               respect to UTC):
2208
2209                  ev_periodic_set (&periodic, 0., 3600., 0);
2210
2211               This doesn't mean there will always be 3600 seconds in between
2212               triggers, but only that the callback will be called when the
2213               system time shows a full hour (UTC), or more correctly, when
2214               the system time is evenly divisible by 3600.
2215
2216               Another way to think about it (for the mathematically inclined)
2217               is that "ev_periodic" will try to run the callback in this mode
2218               at the next possible time where "time = offset (mod interval)",
2219               regardless of any time jumps.
2220
2221               The "interval" MUST be positive, and for numerical stability,
2222               the interval value should be higher than "1/8192" (which is
2223               around 100 microseconds) and "offset" should be higher than 0
2224               and should have at most a similar magnitude as the current time
2225               (say, within a factor of ten). Typical values for offset are,
2226               in fact, 0 or something between 0 and "interval", which is also
2227               the recommended range.
2228
2229               Note also that there is an upper limit to how often a timer can
2230               fire (CPU speed for example), so if "interval" is very small
2231               then timing stability will of course deteriorate. Libev itself
2232               tries to be exact to be about one millisecond (if the OS
2233               supports it and the machine is fast enough).
2234
2235           ·   manual reschedule mode (offset ignored, interval ignored,
2236               reschedule_cb = callback)
2237
2238               In this mode the values for "interval" and "offset" are both
2239               being ignored. Instead, each time the periodic watcher gets
2240               scheduled, the reschedule callback will be called with the
2241               watcher as first, and the current time as second argument.
2242
2243               NOTE: This callback MUST NOT stop or destroy any periodic
2244               watcher, ever, or make ANY other event loop modifications
2245               whatsoever, unless explicitly allowed by documentation here.
2246
2247               If you need to stop it, return "now + 1e30" (or so, fudge
2248               fudge) and stop it afterwards (e.g. by starting an "ev_prepare"
2249               watcher, which is the only event loop modification you are
2250               allowed to do).
2251
2252               The callback prototype is "ev_tstamp
2253               (*reschedule_cb)(ev_periodic *w, ev_tstamp now)", e.g.:
2254
2255                  static ev_tstamp
2256                  my_rescheduler (ev_periodic *w, ev_tstamp now)
2257                  {
2258                    return now + 60.;
2259                  }
2260
2261               It must return the next time to trigger, based on the passed
2262               time value (that is, the lowest time value larger than to the
2263               second argument). It will usually be called just before the
2264               callback will be triggered, but might be called at other times,
2265               too.
2266
2267               NOTE: This callback must always return a time that is higher
2268               than or equal to the passed "now" value.
2269
2270               This can be used to create very complex timers, such as a timer
2271               that triggers on "next midnight, local time". To do this, you
2272               would calculate the next midnight after "now" and return the
2273               timestamp value for this. Here is a (completely untested, no
2274               error checking) example on how to do this:
2275
2276                  #include <time.h>
2277
2278                  static ev_tstamp
2279                  my_rescheduler (ev_periodic *w, ev_tstamp now)
2280                  {
2281                    time_t tnow = (time_t)now;
2282                    struct tm tm;
2283                    localtime_r (&tnow, &tm);
2284
2285                    tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
2286                    ++tm.tm_mday; // midnight next day
2287
2288                    return mktime (&tm);
2289                  }
2290
2291               Note: this code might run into trouble on days that have more
2292               then two midnights (beginning and end).
2293
2294       ev_periodic_again (loop, ev_periodic *)
2295           Simply stops and restarts the periodic watcher again. This is only
2296           useful when you changed some parameters or the reschedule callback
2297           would return a different time than the last time it was called
2298           (e.g. in a crond like program when the crontabs have changed).
2299
2300       ev_tstamp ev_periodic_at (ev_periodic *)
2301           When active, returns the absolute time that the watcher is supposed
2302           to trigger next. This is not the same as the "offset" argument to
2303           "ev_periodic_set", but indeed works even in interval and manual
2304           rescheduling modes.
2305
2306       ev_tstamp offset [read-write]
2307           When repeating, this contains the offset value, otherwise this is
2308           the absolute point in time (the "offset" value passed to
2309           "ev_periodic_set", although libev might modify this value for
2310           better numerical stability).
2311
2312           Can be modified any time, but changes only take effect when the
2313           periodic timer fires or "ev_periodic_again" is being called.
2314
2315       ev_tstamp interval [read-write]
2316           The current interval value. Can be modified any time, but changes
2317           only take effect when the periodic timer fires or
2318           "ev_periodic_again" is being called.
2319
2320       ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
2321           The current reschedule callback, or 0, if this functionality is
2322           switched off. Can be changed any time, but changes only take effect
2323           when the periodic timer fires or "ev_periodic_again" is being
2324           called.
2325
2326       Examples
2327
2328       Example: Call a callback every hour, or, more precisely, whenever the
2329       system time is divisible by 3600. The callback invocation times have
2330       potentially a lot of jitter, but good long-term stability.
2331
2332          static void
2333          clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
2334          {
2335            ... its now a full hour (UTC, or TAI or whatever your clock follows)
2336          }
2337
2338          ev_periodic hourly_tick;
2339          ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
2340          ev_periodic_start (loop, &hourly_tick);
2341
2342       Example: The same as above, but use a reschedule callback to do it:
2343
2344          #include <math.h>
2345
2346          static ev_tstamp
2347          my_scheduler_cb (ev_periodic *w, ev_tstamp now)
2348          {
2349            return now + (3600. - fmod (now, 3600.));
2350          }
2351
2352          ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
2353
2354       Example: Call a callback every hour, starting now:
2355
2356          ev_periodic hourly_tick;
2357          ev_periodic_init (&hourly_tick, clock_cb,
2358                            fmod (ev_now (loop), 3600.), 3600., 0);
2359          ev_periodic_start (loop, &hourly_tick);
2360
2361   "ev_signal" - signal me when a signal gets signalled!
2362       Signal watchers will trigger an event when the process receives a
2363       specific signal one or more times. Even though signals are very
2364       asynchronous, libev will try its best to deliver signals synchronously,
2365       i.e. as part of the normal event processing, like any other event.
2366
2367       If you want signals to be delivered truly asynchronously, just use
2368       "sigaction" as you would do without libev and forget about sharing the
2369       signal. You can even use "ev_async" from a signal handler to
2370       synchronously wake up an event loop.
2371
2372       You can configure as many watchers as you like for the same signal, but
2373       only within the same loop, i.e. you can watch for "SIGINT" in your
2374       default loop and for "SIGIO" in another loop, but you cannot watch for
2375       "SIGINT" in both the default loop and another loop at the same time. At
2376       the moment, "SIGCHLD" is permanently tied to the default loop.
2377
2378       Only after the first watcher for a signal is started will libev
2379       actually register something with the kernel. It thus coexists with your
2380       own signal handlers as long as you don't register any with libev for
2381       the same signal.
2382
2383       If possible and supported, libev will install its handlers with
2384       "SA_RESTART" (or equivalent) behaviour enabled, so system calls should
2385       not be unduly interrupted. If you have a problem with system calls
2386       getting interrupted by signals you can block all signals in an
2387       "ev_check" watcher and unblock them in an "ev_prepare" watcher.
2388
2389       The special problem of inheritance over fork/execve/pthread_create
2390
2391       Both the signal mask ("sigprocmask") and the signal disposition
2392       ("sigaction") are unspecified after starting a signal watcher (and
2393       after stopping it again), that is, libev might or might not block the
2394       signal, and might or might not set or restore the installed signal
2395       handler (but see "EVFLAG_NOSIGMASK").
2396
2397       While this does not matter for the signal disposition (libev never sets
2398       signals to "SIG_IGN", so handlers will be reset to "SIG_DFL" on
2399       "execve"), this matters for the signal mask: many programs do not
2400       expect certain signals to be blocked.
2401
2402       This means that before calling "exec" (from the child) you should reset
2403       the signal mask to whatever "default" you expect (all clear is a good
2404       choice usually).
2405
2406       The simplest way to ensure that the signal mask is reset in the child
2407       is to install a fork handler with "pthread_atfork" that resets it. That
2408       will catch fork calls done by libraries (such as the libc) as well.
2409
2410       In current versions of libev, the signal will not be blocked
2411       indefinitely unless you use the "signalfd" API ("EV_SIGNALFD"). While
2412       this reduces the window of opportunity for problems, it will not go
2413       away, as libev has to modify the signal mask, at least temporarily.
2414
2415       So I can't stress this enough: If you do not reset your signal mask
2416       when you expect it to be empty, you have a race condition in your code.
2417       This is not a libev-specific thing, this is true for most event
2418       libraries.
2419
2420       The special problem of threads signal handling
2421
2422       POSIX threads has problematic signal handling semantics, specifically,
2423       a lot of functionality (sigfd, sigwait etc.) only really works if all
2424       threads in a process block signals, which is hard to achieve.
2425
2426       When you want to use sigwait (or mix libev signal handling with your
2427       own for the same signals), you can tackle this problem by globally
2428       blocking all signals before creating any threads (or creating them with
2429       a fully set sigprocmask) and also specifying the "EVFLAG_NOSIGMASK"
2430       when creating loops. Then designate one thread as "signal receiver
2431       thread" which handles these signals. You can pass on any signals that
2432       libev might be interested in by calling "ev_feed_signal".
2433
2434       Watcher-Specific Functions and Data Members
2435
2436       ev_signal_init (ev_signal *, callback, int signum)
2437       ev_signal_set (ev_signal *, int signum)
2438           Configures the watcher to trigger on the given signal number
2439           (usually one of the "SIGxxx" constants).
2440
2441       int signum [read-only]
2442           The signal the watcher watches out for.
2443
2444       Examples
2445
2446       Example: Try to exit cleanly on SIGINT.
2447
2448          static void
2449          sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
2450          {
2451            ev_break (loop, EVBREAK_ALL);
2452          }
2453
2454          ev_signal signal_watcher;
2455          ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
2456          ev_signal_start (loop, &signal_watcher);
2457
2458   "ev_child" - watch out for process status changes
2459       Child watchers trigger when your process receives a SIGCHLD in response
2460       to some child status changes (most typically when a child of yours dies
2461       or exits). It is permissible to install a child watcher after the child
2462       has been forked (which implies it might have already exited), as long
2463       as the event loop isn't entered (or is continued from a watcher), i.e.,
2464       forking and then immediately registering a watcher for the child is
2465       fine, but forking and registering a watcher a few event loop iterations
2466       later or in the next callback invocation is not.
2467
2468       Only the default event loop is capable of handling signals, and
2469       therefore you can only register child watchers in the default event
2470       loop.
2471
2472       Due to some design glitches inside libev, child watchers will always be
2473       handled at maximum priority (their priority is set to "EV_MAXPRI" by
2474       libev)
2475
2476       Process Interaction
2477
2478       Libev grabs "SIGCHLD" as soon as the default event loop is initialised.
2479       This is necessary to guarantee proper behaviour even if the first child
2480       watcher is started after the child exits. The occurrence of "SIGCHLD"
2481       is recorded asynchronously, but child reaping is done synchronously as
2482       part of the event loop processing. Libev always reaps all children,
2483       even ones not watched.
2484
2485       Overriding the Built-In Processing
2486
2487       Libev offers no special support for overriding the built-in child
2488       processing, but if your application collides with libev's default child
2489       handler, you can override it easily by installing your own handler for
2490       "SIGCHLD" after initialising the default loop, and making sure the
2491       default loop never gets destroyed. You are encouraged, however, to use
2492       an event-based approach to child reaping and thus use libev's support
2493       for that, so other libev users can use "ev_child" watchers freely.
2494
2495       Stopping the Child Watcher
2496
2497       Currently, the child watcher never gets stopped, even when the child
2498       terminates, so normally one needs to stop the watcher in the callback.
2499       Future versions of libev might stop the watcher automatically when a
2500       child exit is detected (calling "ev_child_stop" twice is not a
2501       problem).
2502
2503       Watcher-Specific Functions and Data Members
2504
2505       ev_child_init (ev_child *, callback, int pid, int trace)
2506       ev_child_set (ev_child *, int pid, int trace)
2507           Configures the watcher to wait for status changes of process "pid"
2508           (or any process if "pid" is specified as 0). The callback can look
2509           at the "rstatus" member of the "ev_child" watcher structure to see
2510           the status word (use the macros from "sys/wait.h" and see your
2511           systems "waitpid" documentation). The "rpid" member contains the
2512           pid of the process causing the status change. "trace" must be
2513           either 0 (only activate the watcher when the process terminates) or
2514           1 (additionally activate the watcher when the process is stopped or
2515           continued).
2516
2517       int pid [read-only]
2518           The process id this watcher watches out for, or 0, meaning any
2519           process id.
2520
2521       int rpid [read-write]
2522           The process id that detected a status change.
2523
2524       int rstatus [read-write]
2525           The process exit/trace status caused by "rpid" (see your systems
2526           "waitpid" and "sys/wait.h" documentation for details).
2527
2528       Examples
2529
2530       Example: "fork()" a new process and install a child handler to wait for
2531       its completion.
2532
2533          ev_child cw;
2534
2535          static void
2536          child_cb (EV_P_ ev_child *w, int revents)
2537          {
2538            ev_child_stop (EV_A_ w);
2539            printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
2540          }
2541
2542          pid_t pid = fork ();
2543
2544          if (pid < 0)
2545            // error
2546          else if (pid == 0)
2547            {
2548              // the forked child executes here
2549              exit (1);
2550            }
2551          else
2552            {
2553              ev_child_init (&cw, child_cb, pid, 0);
2554              ev_child_start (EV_DEFAULT_ &cw);
2555            }
2556
2557   "ev_stat" - did the file attributes just change?
2558       This watches a file system path for attribute changes. That is, it
2559       calls "stat" on that path in regular intervals (or when the OS says it
2560       changed) and sees if it changed compared to the last time, invoking the
2561       callback if it did. Starting the watcher "stat"'s the file, so only
2562       changes that happen after the watcher has been started will be
2563       reported.
2564
2565       The path does not need to exist: changing from "path exists" to "path
2566       does not exist" is a status change like any other. The condition "path
2567       does not exist" (or more correctly "path cannot be stat'ed") is
2568       signified by the "st_nlink" field being zero (which is otherwise always
2569       forced to be at least one) and all the other fields of the stat buffer
2570       having unspecified contents.
2571
2572       The path must not end in a slash or contain special components such as
2573       "." or "..". The path should be absolute: If it is relative and your
2574       working directory changes, then the behaviour is undefined.
2575
2576       Since there is no portable change notification interface available, the
2577       portable implementation simply calls stat(2) regularly on the path to
2578       see if it changed somehow. You can specify a recommended polling
2579       interval for this case. If you specify a polling interval of 0 (highly
2580       recommended!) then a suitable, unspecified default value will be used
2581       (which you can expect to be around five seconds, although this might
2582       change dynamically). Libev will also impose a minimum interval which is
2583       currently around 0.1, but that's usually overkill.
2584
2585       This watcher type is not meant for massive numbers of stat watchers, as
2586       even with OS-supported change notifications, this can be resource-
2587       intensive.
2588
2589       At the time of this writing, the only OS-specific interface implemented
2590       is the Linux inotify interface (implementing kqueue support is left as
2591       an exercise for the reader. Note, however, that the author sees no way
2592       of implementing "ev_stat" semantics with kqueue, except as a hint).
2593
2594       ABI Issues (Largefile Support)
2595
2596       Libev by default (unless the user overrides this) uses the default
2597       compilation environment, which means that on systems with large file
2598       support disabled by default, you get the 32 bit version of the stat
2599       structure. When using the library from programs that change the ABI to
2600       use 64 bit file offsets the programs will fail. In that case you have
2601       to compile libev with the same flags to get binary compatibility. This
2602       is obviously the case with any flags that change the ABI, but the
2603       problem is most noticeably displayed with ev_stat and large file
2604       support.
2605
2606       The solution for this is to lobby your distribution maker to make large
2607       file interfaces available by default (as e.g. FreeBSD does) and not
2608       optional. Libev cannot simply switch on large file support because it
2609       has to exchange stat structures with application programs compiled
2610       using the default compilation environment.
2611
2612       Inotify and Kqueue
2613
2614       When "inotify (7)" support has been compiled into libev and present at
2615       runtime, it will be used to speed up change detection where possible.
2616       The inotify descriptor will be created lazily when the first "ev_stat"
2617       watcher is being started.
2618
2619       Inotify presence does not change the semantics of "ev_stat" watchers
2620       except that changes might be detected earlier, and in some cases, to
2621       avoid making regular "stat" calls. Even in the presence of inotify
2622       support there are many cases where libev has to resort to regular
2623       "stat" polling, but as long as kernel 2.6.25 or newer is used (2.6.24
2624       and older have too many bugs), the path exists (i.e. stat succeeds),
2625       and the path resides on a local filesystem (libev currently assumes
2626       only ext2/3, jfs, reiserfs and xfs are fully working) libev usually
2627       gets away without polling.
2628
2629       There is no support for kqueue, as apparently it cannot be used to
2630       implement this functionality, due to the requirement of having a file
2631       descriptor open on the object at all times, and detecting renames,
2632       unlinks etc. is difficult.
2633
2634       "stat ()" is a synchronous operation
2635
2636       Libev doesn't normally do any kind of I/O itself, and so is not
2637       blocking the process. The exception are "ev_stat" watchers - those call
2638       "stat ()", which is a synchronous operation.
2639
2640       For local paths, this usually doesn't matter: unless the system is very
2641       busy or the intervals between stat's are large, a stat call will be
2642       fast, as the path data is usually in memory already (except when
2643       starting the watcher).
2644
2645       For networked file systems, calling "stat ()" can block an indefinite
2646       time due to network issues, and even under good conditions, a stat call
2647       often takes multiple milliseconds.
2648
2649       Therefore, it is best to avoid using "ev_stat" watchers on networked
2650       paths, although this is fully supported by libev.
2651
2652       The special problem of stat time resolution
2653
2654       The "stat ()" system call only supports full-second resolution
2655       portably, and even on systems where the resolution is higher, most file
2656       systems still only support whole seconds.
2657
2658       That means that, if the time is the only thing that changes, you can
2659       easily miss updates: on the first update, "ev_stat" detects a change
2660       and calls your callback, which does something. When there is another
2661       update within the same second, "ev_stat" will be unable to detect
2662       unless the stat data does change in other ways (e.g. file size).
2663
2664       The solution to this is to delay acting on a change for slightly more
2665       than a second (or till slightly after the next full second boundary),
2666       using a roughly one-second-delay "ev_timer" (e.g. "ev_timer_set (w, 0.,
2667       1.02); ev_timer_again (loop, w)").
2668
2669       The .02 offset is added to work around small timing inconsistencies of
2670       some operating systems (where the second counter of the current time
2671       might be be delayed. One such system is the Linux kernel, where a call
2672       to "gettimeofday" might return a timestamp with a full second later
2673       than a subsequent "time" call - if the equivalent of "time ()" is used
2674       to update file times then there will be a small window where the kernel
2675       uses the previous second to update file times but libev might already
2676       execute the timer callback).
2677
2678       Watcher-Specific Functions and Data Members
2679
2680       ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp
2681       interval)
2682       ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
2683           Configures the watcher to wait for status changes of the given
2684           "path". The "interval" is a hint on how quickly a change is
2685           expected to be detected and should normally be specified as 0 to
2686           let libev choose a suitable value. The memory pointed to by "path"
2687           must point to the same path for as long as the watcher is active.
2688
2689           The callback will receive an "EV_STAT" event when a change was
2690           detected, relative to the attributes at the time the watcher was
2691           started (or the last change was detected).
2692
2693       ev_stat_stat (loop, ev_stat *)
2694           Updates the stat buffer immediately with new values. If you change
2695           the watched path in your callback, you could call this function to
2696           avoid detecting this change (while introducing a race condition if
2697           you are not the only one changing the path). Can also be useful
2698           simply to find out the new values.
2699
2700       ev_statdata attr [read-only]
2701           The most-recently detected attributes of the file. Although the
2702           type is "ev_statdata", this is usually the (or one of the) "struct
2703           stat" types suitable for your system, but you can only rely on the
2704           POSIX-standardised members to be present. If the "st_nlink" member
2705           is 0, then there was some error while "stat"ing the file.
2706
2707       ev_statdata prev [read-only]
2708           The previous attributes of the file. The callback gets invoked
2709           whenever "prev" != "attr", or, more precisely, one or more of these
2710           members differ: "st_dev", "st_ino", "st_mode", "st_nlink",
2711           "st_uid", "st_gid", "st_rdev", "st_size", "st_atime", "st_mtime",
2712           "st_ctime".
2713
2714       ev_tstamp interval [read-only]
2715           The specified interval.
2716
2717       const char *path [read-only]
2718           The file system path that is being watched.
2719
2720       Examples
2721
2722       Example: Watch "/etc/passwd" for attribute changes.
2723
2724          static void
2725          passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
2726          {
2727            /* /etc/passwd changed in some way */
2728            if (w->attr.st_nlink)
2729              {
2730                printf ("passwd current size  %ld\n", (long)w->attr.st_size);
2731                printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
2732                printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
2733              }
2734            else
2735              /* you shalt not abuse printf for puts */
2736              puts ("wow, /etc/passwd is not there, expect problems. "
2737                    "if this is windows, they already arrived\n");
2738          }
2739
2740          ...
2741          ev_stat passwd;
2742
2743          ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
2744          ev_stat_start (loop, &passwd);
2745
2746       Example: Like above, but additionally use a one-second delay so we do
2747       not miss updates (however, frequent updates will delay processing, too,
2748       so one might do the work both on "ev_stat" callback invocation and on
2749       "ev_timer" callback invocation).
2750
2751          static ev_stat passwd;
2752          static ev_timer timer;
2753
2754          static void
2755          timer_cb (EV_P_ ev_timer *w, int revents)
2756          {
2757            ev_timer_stop (EV_A_ w);
2758
2759            /* now it's one second after the most recent passwd change */
2760          }
2761
2762          static void
2763          stat_cb (EV_P_ ev_stat *w, int revents)
2764          {
2765            /* reset the one-second timer */
2766            ev_timer_again (EV_A_ &timer);
2767          }
2768
2769          ...
2770          ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
2771          ev_stat_start (loop, &passwd);
2772          ev_timer_init (&timer, timer_cb, 0., 1.02);
2773
2774   "ev_idle" - when you've got nothing better to do...
2775       Idle watchers trigger events when no other events of the same or higher
2776       priority are pending (prepare, check and other idle watchers do not
2777       count as receiving "events").
2778
2779       That is, as long as your process is busy handling sockets or timeouts
2780       (or even signals, imagine) of the same or higher priority it will not
2781       be triggered. But when your process is idle (or only lower-priority
2782       watchers are pending), the idle watchers are being called once per
2783       event loop iteration - until stopped, that is, or your process receives
2784       more events and becomes busy again with higher priority stuff.
2785
2786       The most noteworthy effect is that as long as any idle watchers are
2787       active, the process will not block when waiting for new events.
2788
2789       Apart from keeping your process non-blocking (which is a useful effect
2790       on its own sometimes), idle watchers are a good place to do "pseudo-
2791       background processing", or delay processing stuff to after the event
2792       loop has handled all outstanding events.
2793
2794       Abusing an "ev_idle" watcher for its side-effect
2795
2796       As long as there is at least one active idle watcher, libev will never
2797       sleep unnecessarily. Or in other words, it will loop as fast as
2798       possible.  For this to work, the idle watcher doesn't need to be
2799       invoked at all - the lowest priority will do.
2800
2801       This mode of operation can be useful together with an "ev_check"
2802       watcher, to do something on each event loop iteration - for example to
2803       balance load between different connections.
2804
2805       See "Abusing an ev_check watcher for its side-effect" for a longer
2806       example.
2807
2808       Watcher-Specific Functions and Data Members
2809
2810       ev_idle_init (ev_idle *, callback)
2811           Initialises and configures the idle watcher - it has no parameters
2812           of any kind. There is a "ev_idle_set" macro, but using it is
2813           utterly pointless, believe me.
2814
2815       Examples
2816
2817       Example: Dynamically allocate an "ev_idle" watcher, start it, and in
2818       the callback, free it. Also, use no error checking, as usual.
2819
2820          static void
2821          idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
2822          {
2823            // stop the watcher
2824            ev_idle_stop (loop, w);
2825
2826            // now we can free it
2827            free (w);
2828
2829            // now do something you wanted to do when the program has
2830            // no longer anything immediate to do.
2831          }
2832
2833          ev_idle *idle_watcher = malloc (sizeof (ev_idle));
2834          ev_idle_init (idle_watcher, idle_cb);
2835          ev_idle_start (loop, idle_watcher);
2836
2837   "ev_prepare" and "ev_check" - customise your event loop!
2838       Prepare and check watchers are often (but not always) used in pairs:
2839       prepare watchers get invoked before the process blocks and check
2840       watchers afterwards.
2841
2842       You must not call "ev_run" (or similar functions that enter the current
2843       event loop) or "ev_loop_fork" from either "ev_prepare" or "ev_check"
2844       watchers. Other loops than the current one are fine, however. The
2845       rationale behind this is that you do not need to check for recursion in
2846       those watchers, i.e. the sequence will always be "ev_prepare",
2847       blocking, "ev_check" so if you have one watcher of each kind they will
2848       always be called in pairs bracketing the blocking call.
2849
2850       Their main purpose is to integrate other event mechanisms into libev
2851       and their use is somewhat advanced. They could be used, for example, to
2852       track variable changes, implement your own watchers, integrate net-snmp
2853       or a coroutine library and lots more. They are also occasionally useful
2854       if you cache some data and want to flush it before blocking (for
2855       example, in X programs you might want to do an "XFlush ()" in an
2856       "ev_prepare" watcher).
2857
2858       This is done by examining in each prepare call which file descriptors
2859       need to be watched by the other library, registering "ev_io" watchers
2860       for them and starting an "ev_timer" watcher for any timeouts (many
2861       libraries provide exactly this functionality). Then, in the check
2862       watcher, you check for any events that occurred (by checking the
2863       pending status of all watchers and stopping them) and call back into
2864       the library. The I/O and timer callbacks will never actually be called
2865       (but must be valid nevertheless, because you never know, you know?).
2866
2867       As another example, the Perl Coro module uses these hooks to integrate
2868       coroutines into libev programs, by yielding to other active coroutines
2869       during each prepare and only letting the process block if no coroutines
2870       are ready to run (it's actually more complicated: it only runs
2871       coroutines with priority higher than or equal to the event loop and one
2872       coroutine of lower priority, but only once, using idle watchers to keep
2873       the event loop from blocking if lower-priority coroutines are active,
2874       thus mapping low-priority coroutines to idle/background tasks).
2875
2876       When used for this purpose, it is recommended to give "ev_check"
2877       watchers highest ("EV_MAXPRI") priority, to ensure that they are being
2878       run before any other watchers after the poll (this doesn't matter for
2879       "ev_prepare" watchers).
2880
2881       Also, "ev_check" watchers (and "ev_prepare" watchers, too) should not
2882       activate ("feed") events into libev. While libev fully supports this,
2883       they might get executed before other "ev_check" watchers did their job.
2884       As "ev_check" watchers are often used to embed other (non-libev) event
2885       loops those other event loops might be in an unusable state until their
2886       "ev_check" watcher ran (always remind yourself to coexist peacefully
2887       with others).
2888
2889       Abusing an "ev_check" watcher for its side-effect
2890
2891       "ev_check" (and less often also "ev_prepare") watchers can also be
2892       useful because they are called once per event loop iteration. For
2893       example, if you want to handle a large number of connections fairly,
2894       you normally only do a bit of work for each active connection, and if
2895       there is more work to do, you wait for the next event loop iteration,
2896       so other connections have a chance of making progress.
2897
2898       Using an "ev_check" watcher is almost enough: it will be called on the
2899       next event loop iteration. However, that isn't as soon as possible -
2900       without external events, your "ev_check" watcher will not be invoked.
2901
2902       This is where "ev_idle" watchers come in handy - all you need is a
2903       single global idle watcher that is active as long as you have one
2904       active "ev_check" watcher. The "ev_idle" watcher makes sure the event
2905       loop will not sleep, and the "ev_check" watcher makes sure a callback
2906       gets invoked. Neither watcher alone can do that.
2907
2908       Watcher-Specific Functions and Data Members
2909
2910       ev_prepare_init (ev_prepare *, callback)
2911       ev_check_init (ev_check *, callback)
2912           Initialises and configures the prepare or check watcher - they have
2913           no parameters of any kind. There are "ev_prepare_set" and
2914           "ev_check_set" macros, but using them is utterly, utterly, utterly
2915           and completely pointless.
2916
2917       Examples
2918
2919       There are a number of principal ways to embed other event loops or
2920       modules into libev. Here are some ideas on how to include libadns into
2921       libev (there is a Perl module named "EV::ADNS" that does this, which
2922       you could use as a working example. Another Perl module named
2923       "EV::Glib" embeds a Glib main context into libev, and finally,
2924       "Glib::EV" embeds EV into the Glib event loop).
2925
2926       Method 1: Add IO watchers and a timeout watcher in a prepare handler,
2927       and in a check watcher, destroy them and call into libadns. What
2928       follows is pseudo-code only of course. This requires you to either use
2929       a low priority for the check watcher or use "ev_clear_pending"
2930       explicitly, as the callbacks for the IO/timeout watchers might not have
2931       been called yet.
2932
2933          static ev_io iow [nfd];
2934          static ev_timer tw;
2935
2936          static void
2937          io_cb (struct ev_loop *loop, ev_io *w, int revents)
2938          {
2939          }
2940
2941          // create io watchers for each fd and a timer before blocking
2942          static void
2943          adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
2944          {
2945            int timeout = 3600000;
2946            struct pollfd fds [nfd];
2947            // actual code will need to loop here and realloc etc.
2948            adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
2949
2950            /* the callback is illegal, but won't be called as we stop during check */
2951            ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
2952            ev_timer_start (loop, &tw);
2953
2954            // create one ev_io per pollfd
2955            for (int i = 0; i < nfd; ++i)
2956              {
2957                ev_io_init (iow + i, io_cb, fds [i].fd,
2958                  ((fds [i].events & POLLIN ? EV_READ : 0)
2959                   | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
2960
2961                fds [i].revents = 0;
2962                ev_io_start (loop, iow + i);
2963              }
2964          }
2965
2966          // stop all watchers after blocking
2967          static void
2968          adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
2969          {
2970            ev_timer_stop (loop, &tw);
2971
2972            for (int i = 0; i < nfd; ++i)
2973              {
2974                // set the relevant poll flags
2975                // could also call adns_processreadable etc. here
2976                struct pollfd *fd = fds + i;
2977                int revents = ev_clear_pending (iow + i);
2978                if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
2979                if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
2980
2981                // now stop the watcher
2982                ev_io_stop (loop, iow + i);
2983              }
2984
2985            adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
2986          }
2987
2988       Method 2: This would be just like method 1, but you run
2989       "adns_afterpoll" in the prepare watcher and would dispose of the check
2990       watcher.
2991
2992       Method 3: If the module to be embedded supports explicit event
2993       notification (libadns does), you can also make use of the actual
2994       watcher callbacks, and only destroy/create the watchers in the prepare
2995       watcher.
2996
2997          static void
2998          timer_cb (EV_P_ ev_timer *w, int revents)
2999          {
3000            adns_state ads = (adns_state)w->data;
3001            update_now (EV_A);
3002
3003            adns_processtimeouts (ads, &tv_now);
3004          }
3005
3006          static void
3007          io_cb (EV_P_ ev_io *w, int revents)
3008          {
3009            adns_state ads = (adns_state)w->data;
3010            update_now (EV_A);
3011
3012            if (revents & EV_READ ) adns_processreadable  (ads, w->fd, &tv_now);
3013            if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
3014          }
3015
3016          // do not ever call adns_afterpoll
3017
3018       Method 4: Do not use a prepare or check watcher because the module you
3019       want to embed is not flexible enough to support it. Instead, you can
3020       override their poll function. The drawback with this solution is that
3021       the main loop is now no longer controllable by EV. The "Glib::EV"
3022       module uses this approach, effectively embedding EV as a client into
3023       the horrible libglib event loop.
3024
3025          static gint
3026          event_poll_func (GPollFD *fds, guint nfds, gint timeout)
3027          {
3028            int got_events = 0;
3029
3030            for (n = 0; n < nfds; ++n)
3031              // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
3032
3033            if (timeout >= 0)
3034              // create/start timer
3035
3036            // poll
3037            ev_run (EV_A_ 0);
3038
3039            // stop timer again
3040            if (timeout >= 0)
3041              ev_timer_stop (EV_A_ &to);
3042
3043            // stop io watchers again - their callbacks should have set
3044            for (n = 0; n < nfds; ++n)
3045              ev_io_stop (EV_A_ iow [n]);
3046
3047            return got_events;
3048          }
3049
3050   "ev_embed" - when one backend isn't enough...
3051       This is a rather advanced watcher type that lets you embed one event
3052       loop into another (currently only "ev_io" events are supported in the
3053       embedded loop, other types of watchers might be handled in a delayed or
3054       incorrect fashion and must not be used).
3055
3056       There are primarily two reasons you would want that: work around bugs
3057       and prioritise I/O.
3058
3059       As an example for a bug workaround, the kqueue backend might only
3060       support sockets on some platform, so it is unusable as generic backend,
3061       but you still want to make use of it because you have many sockets and
3062       it scales so nicely. In this case, you would create a kqueue-based loop
3063       and embed it into your default loop (which might use e.g. poll).
3064       Overall operation will be a bit slower because first libev has to call
3065       "poll" and then "kevent", but at least you can use both mechanisms for
3066       what they are best: "kqueue" for scalable sockets and "poll" if you
3067       want it to work :)
3068
3069       As for prioritising I/O: under rare circumstances you have the case
3070       where some fds have to be watched and handled very quickly (with low
3071       latency), and even priorities and idle watchers might have too much
3072       overhead. In this case you would put all the high priority stuff in one
3073       loop and all the rest in a second one, and embed the second one in the
3074       first.
3075
3076       As long as the watcher is active, the callback will be invoked every
3077       time there might be events pending in the embedded loop. The callback
3078       must then call "ev_embed_sweep (mainloop, watcher)" to make a single
3079       sweep and invoke their callbacks (the callback doesn't need to invoke
3080       the "ev_embed_sweep" function directly, it could also start an idle
3081       watcher to give the embedded loop strictly lower priority for example).
3082
3083       You can also set the callback to 0, in which case the embed watcher
3084       will automatically execute the embedded loop sweep whenever necessary.
3085
3086       Fork detection will be handled transparently while the "ev_embed"
3087       watcher is active, i.e., the embedded loop will automatically be forked
3088       when the embedding loop forks. In other cases, the user is responsible
3089       for calling "ev_loop_fork" on the embedded loop.
3090
3091       Unfortunately, not all backends are embeddable: only the ones returned
3092       by "ev_embeddable_backends" are, which, unfortunately, does not include
3093       any portable one.
3094
3095       So when you want to use this feature you will always have to be
3096       prepared that you cannot get an embeddable loop. The recommended way to
3097       get around this is to have a separate variables for your embeddable
3098       loop, try to create it, and if that fails, use the normal loop for
3099       everything.
3100
3101       "ev_embed" and fork
3102
3103       While the "ev_embed" watcher is running, forks in the embedding loop
3104       will automatically be applied to the embedded loop as well, so no
3105       special fork handling is required in that case. When the watcher is not
3106       running, however, it is still the task of the libev user to call
3107       "ev_loop_fork ()" as applicable.
3108
3109       Watcher-Specific Functions and Data Members
3110
3111       ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
3112       ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
3113           Configures the watcher to embed the given loop, which must be
3114           embeddable. If the callback is 0, then "ev_embed_sweep" will be
3115           invoked automatically, otherwise it is the responsibility of the
3116           callback to invoke it (it will continue to be called until the
3117           sweep has been done, if you do not want that, you need to
3118           temporarily stop the embed watcher).
3119
3120       ev_embed_sweep (loop, ev_embed *)
3121           Make a single, non-blocking sweep over the embedded loop. This
3122           works similarly to "ev_run (embedded_loop, EVRUN_NOWAIT)", but in
3123           the most appropriate way for embedded loops.
3124
3125       struct ev_loop *other [read-only]
3126           The embedded event loop.
3127
3128       Examples
3129
3130       Example: Try to get an embeddable event loop and embed it into the
3131       default event loop. If that is not possible, use the default loop. The
3132       default loop is stored in "loop_hi", while the embeddable loop is
3133       stored in "loop_lo" (which is "loop_hi" in the case no embeddable loop
3134       can be used).
3135
3136          struct ev_loop *loop_hi = ev_default_init (0);
3137          struct ev_loop *loop_lo = 0;
3138          ev_embed embed;
3139
3140          // see if there is a chance of getting one that works
3141          // (remember that a flags value of 0 means autodetection)
3142          loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
3143            ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
3144            : 0;
3145
3146          // if we got one, then embed it, otherwise default to loop_hi
3147          if (loop_lo)
3148            {
3149              ev_embed_init (&embed, 0, loop_lo);
3150              ev_embed_start (loop_hi, &embed);
3151            }
3152          else
3153            loop_lo = loop_hi;
3154
3155       Example: Check if kqueue is available but not recommended and create a
3156       kqueue backend for use with sockets (which usually work with any kqueue
3157       implementation). Store the kqueue/socket-only event loop in
3158       "loop_socket". (One might optionally use "EVFLAG_NOENV", too).
3159
3160          struct ev_loop *loop = ev_default_init (0);
3161          struct ev_loop *loop_socket = 0;
3162          ev_embed embed;
3163
3164          if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
3165            if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
3166              {
3167                ev_embed_init (&embed, 0, loop_socket);
3168                ev_embed_start (loop, &embed);
3169              }
3170
3171          if (!loop_socket)
3172            loop_socket = loop;
3173
3174          // now use loop_socket for all sockets, and loop for everything else
3175
3176   "ev_fork" - the audacity to resume the event loop after a fork
3177       Fork watchers are called when a "fork ()" was detected (usually because
3178       whoever is a good citizen cared to tell libev about it by calling
3179       "ev_loop_fork"). The invocation is done before the event loop blocks
3180       next and before "ev_check" watchers are being called, and only in the
3181       child after the fork. If whoever good citizen calling "ev_default_fork"
3182       cheats and calls it in the wrong process, the fork handlers will be
3183       invoked, too, of course.
3184
3185       The special problem of life after fork - how is it possible?
3186
3187       Most uses of "fork ()" consist of forking, then some simple calls to
3188       set up/change the process environment, followed by a call to "exec()".
3189       This sequence should be handled by libev without any problems.
3190
3191       This changes when the application actually wants to do event handling
3192       in the child, or both parent in child, in effect "continuing" after the
3193       fork.
3194
3195       The default mode of operation (for libev, with application help to
3196       detect forks) is to duplicate all the state in the child, as would be
3197       expected when either the parent or the child process continues.
3198
3199       When both processes want to continue using libev, then this is usually
3200       the wrong result. In that case, usually one process (typically the
3201       parent) is supposed to continue with all watchers in place as before,
3202       while the other process typically wants to start fresh, i.e. without
3203       any active watchers.
3204
3205       The cleanest and most efficient way to achieve that with libev is to
3206       simply create a new event loop, which of course will be "empty", and
3207       use that for new watchers. This has the advantage of not touching more
3208       memory than necessary, and thus avoiding the copy-on-write, and the
3209       disadvantage of having to use multiple event loops (which do not
3210       support signal watchers).
3211
3212       When this is not possible, or you want to use the default loop for
3213       other reasons, then in the process that wants to start "fresh", call
3214       "ev_loop_destroy (EV_DEFAULT)" followed by "ev_default_loop (...)".
3215       Destroying the default loop will "orphan" (not stop) all registered
3216       watchers, so you have to be careful not to execute code that modifies
3217       those watchers. Note also that in that case, you have to re-register
3218       any signal watchers.
3219
3220       Watcher-Specific Functions and Data Members
3221
3222       ev_fork_init (ev_fork *, callback)
3223           Initialises and configures the fork watcher - it has no parameters
3224           of any kind. There is a "ev_fork_set" macro, but using it is
3225           utterly pointless, really.
3226
3227   "ev_cleanup" - even the best things end
3228       Cleanup watchers are called just before the event loop is being
3229       destroyed by a call to "ev_loop_destroy".
3230
3231       While there is no guarantee that the event loop gets destroyed, cleanup
3232       watchers provide a convenient method to install cleanup hooks for your
3233       program, worker threads and so on - you just to make sure to destroy
3234       the loop when you want them to be invoked.
3235
3236       Cleanup watchers are invoked in the same way as any other watcher.
3237       Unlike all other watchers, they do not keep a reference to the event
3238       loop (which makes a lot of sense if you think about it). Like all other
3239       watchers, you can call libev functions in the callback, except
3240       "ev_cleanup_start".
3241
3242       Watcher-Specific Functions and Data Members
3243
3244       ev_cleanup_init (ev_cleanup *, callback)
3245           Initialises and configures the cleanup watcher - it has no
3246           parameters of any kind. There is a "ev_cleanup_set" macro, but
3247           using it is utterly pointless, I assure you.
3248
3249       Example: Register an atexit handler to destroy the default loop, so any
3250       cleanup functions are called.
3251
3252          static void
3253          program_exits (void)
3254          {
3255            ev_loop_destroy (EV_DEFAULT_UC);
3256          }
3257
3258          ...
3259          atexit (program_exits);
3260
3261   "ev_async" - how to wake up an event loop
3262       In general, you cannot use an "ev_loop" from multiple threads or other
3263       asynchronous sources such as signal handlers (as opposed to multiple
3264       event loops - those are of course safe to use in different threads).
3265
3266       Sometimes, however, you need to wake up an event loop you do not
3267       control, for example because it belongs to another thread. This is what
3268       "ev_async" watchers do: as long as the "ev_async" watcher is active,
3269       you can signal it by calling "ev_async_send", which is thread- and
3270       signal safe.
3271
3272       This functionality is very similar to "ev_signal" watchers, as signals,
3273       too, are asynchronous in nature, and signals, too, will be compressed
3274       (i.e. the number of callback invocations may be less than the number of
3275       "ev_async_send" calls). In fact, you could use signal watchers as a
3276       kind of "global async watchers" by using a watcher on an otherwise
3277       unused signal, and "ev_feed_signal" to signal this watcher from another
3278       thread, even without knowing which loop owns the signal.
3279
3280       Queueing
3281
3282       "ev_async" does not support queueing of data in any way. The reason is
3283       that the author does not know of a simple (or any) algorithm for a
3284       multiple-writer-single-reader queue that works in all cases and doesn't
3285       need elaborate support such as pthreads or unportable memory access
3286       semantics.
3287
3288       That means that if you want to queue data, you have to provide your own
3289       queue. But at least I can tell you how to implement locking around your
3290       queue:
3291
3292       queueing from a signal handler context
3293           To implement race-free queueing, you simply add to the queue in the
3294           signal handler but you block the signal handler in the watcher
3295           callback. Here is an example that does that for some fictitious
3296           SIGUSR1 handler:
3297
3298              static ev_async mysig;
3299
3300              static void
3301              sigusr1_handler (void)
3302              {
3303                sometype data;
3304
3305                // no locking etc.
3306                queue_put (data);
3307                ev_async_send (EV_DEFAULT_ &mysig);
3308              }
3309
3310              static void
3311              mysig_cb (EV_P_ ev_async *w, int revents)
3312              {
3313                sometype data;
3314                sigset_t block, prev;
3315
3316                sigemptyset (&block);
3317                sigaddset (&block, SIGUSR1);
3318                sigprocmask (SIG_BLOCK, &block, &prev);
3319
3320                while (queue_get (&data))
3321                  process (data);
3322
3323                if (sigismember (&prev, SIGUSR1)
3324                  sigprocmask (SIG_UNBLOCK, &block, 0);
3325              }
3326
3327           (Note: pthreads in theory requires you to use "pthread_setmask"
3328           instead of "sigprocmask" when you use threads, but libev doesn't do
3329           it either...).
3330
3331       queueing from a thread context
3332           The strategy for threads is different, as you cannot (easily) block
3333           threads but you can easily preempt them, so to queue safely you
3334           need to employ a traditional mutex lock, such as in this pthread
3335           example:
3336
3337              static ev_async mysig;
3338              static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
3339
3340              static void
3341              otherthread (void)
3342              {
3343                // only need to lock the actual queueing operation
3344                pthread_mutex_lock (&mymutex);
3345                queue_put (data);
3346                pthread_mutex_unlock (&mymutex);
3347
3348                ev_async_send (EV_DEFAULT_ &mysig);
3349              }
3350
3351              static void
3352              mysig_cb (EV_P_ ev_async *w, int revents)
3353              {
3354                pthread_mutex_lock (&mymutex);
3355
3356                while (queue_get (&data))
3357                  process (data);
3358
3359                pthread_mutex_unlock (&mymutex);
3360              }
3361
3362       Watcher-Specific Functions and Data Members
3363
3364       ev_async_init (ev_async *, callback)
3365           Initialises and configures the async watcher - it has no parameters
3366           of any kind. There is a "ev_async_set" macro, but using it is
3367           utterly pointless, trust me.
3368
3369       ev_async_send (loop, ev_async *)
3370           Sends/signals/activates the given "ev_async" watcher, that is,
3371           feeds an "EV_ASYNC" event on the watcher into the event loop, and
3372           instantly returns.
3373
3374           Unlike "ev_feed_event", this call is safe to do from other threads,
3375           signal or similar contexts (see the discussion of "EV_ATOMIC_T" in
3376           the embedding section below on what exactly this means).
3377
3378           Note that, as with other watchers in libev, multiple events might
3379           get compressed into a single callback invocation (another way to
3380           look at this is that "ev_async" watchers are level-triggered: they
3381           are set on "ev_async_send", reset when the event loop detects
3382           that).
3383
3384           This call incurs the overhead of at most one extra system call per
3385           event loop iteration, if the event loop is blocked, and no syscall
3386           at all if the event loop (or your program) is processing events.
3387           That means that repeated calls are basically free (there is no need
3388           to avoid calls for performance reasons) and that the overhead
3389           becomes smaller (typically zero) under load.
3390
3391       bool = ev_async_pending (ev_async *)
3392           Returns a non-zero value when "ev_async_send" has been called on
3393           the watcher but the event has not yet been processed (or even
3394           noted) by the event loop.
3395
3396           "ev_async_send" sets a flag in the watcher and wakes up the loop.
3397           When the loop iterates next and checks for the watcher to have
3398           become active, it will reset the flag again. "ev_async_pending" can
3399           be used to very quickly check whether invoking the loop might be a
3400           good idea.
3401
3402           Not that this does not check whether the watcher itself is pending,
3403           only whether it has been requested to make this watcher pending:
3404           there is a time window between the event loop checking and
3405           resetting the async notification, and the callback being invoked.
3406

OTHER FUNCTIONS

3408       There are some other functions of possible interest. Described. Here.
3409       Now.
3410
3411       ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
3412           This function combines a simple timer and an I/O watcher, calls
3413           your callback on whichever event happens first and automatically
3414           stops both watchers. This is useful if you want to wait for a
3415           single event on an fd or timeout without having to
3416           allocate/configure/start/stop/free one or more watchers yourself.
3417
3418           If "fd" is less than 0, then no I/O watcher will be started and the
3419           "events" argument is being ignored. Otherwise, an "ev_io" watcher
3420           for the given "fd" and "events" set will be created and started.
3421
3422           If "timeout" is less than 0, then no timeout watcher will be
3423           started. Otherwise an "ev_timer" watcher with after = "timeout"
3424           (and repeat = 0) will be started. 0 is a valid timeout.
3425
3426           The callback has the type "void (*cb)(int revents, void *arg)" and
3427           is passed an "revents" set like normal event callbacks (a
3428           combination of "EV_ERROR", "EV_READ", "EV_WRITE" or "EV_TIMER") and
3429           the "arg" value passed to "ev_once". Note that it is possible to
3430           receive both a timeout and an io event at the same time - you
3431           probably should give io events precedence.
3432
3433           Example: wait up to ten seconds for data to appear on STDIN_FILENO.
3434
3435              static void stdin_ready (int revents, void *arg)
3436              {
3437                if (revents & EV_READ)
3438                  /* stdin might have data for us, joy! */;
3439                else if (revents & EV_TIMER)
3440                  /* doh, nothing entered */;
3441              }
3442
3443              ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
3444
3445       ev_feed_fd_event (loop, int fd, int revents)
3446           Feed an event on the given fd, as if a file descriptor backend
3447           detected the given events.
3448
3449       ev_feed_signal_event (loop, int signum)
3450           Feed an event as if the given signal occurred. See also
3451           "ev_feed_signal", which is async-safe.
3452

COMMON OR USEFUL IDIOMS (OR BOTH)

3454       This section explains some common idioms that are not immediately
3455       obvious. Note that examples are sprinkled over the whole manual, and
3456       this section only contains stuff that wouldn't fit anywhere else.
3457
3458   ASSOCIATING CUSTOM DATA WITH A WATCHER
3459       Each watcher has, by default, a "void *data" member that you can read
3460       or modify at any time: libev will completely ignore it. This can be
3461       used to associate arbitrary data with your watcher. If you need more
3462       data and don't want to allocate memory separately and store a pointer
3463       to it in that data member, you can also "subclass" the watcher type and
3464       provide your own data:
3465
3466          struct my_io
3467          {
3468            ev_io io;
3469            int otherfd;
3470            void *somedata;
3471            struct whatever *mostinteresting;
3472          };
3473
3474          ...
3475          struct my_io w;
3476          ev_io_init (&w.io, my_cb, fd, EV_READ);
3477
3478       And since your callback will be called with a pointer to the watcher,
3479       you can cast it back to your own type:
3480
3481          static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
3482          {
3483            struct my_io *w = (struct my_io *)w_;
3484            ...
3485          }
3486
3487       More interesting and less C-conformant ways of casting your callback
3488       function type instead have been omitted.
3489
3490   BUILDING YOUR OWN COMPOSITE WATCHERS
3491       Another common scenario is to use some data structure with multiple
3492       embedded watchers, in effect creating your own watcher that combines
3493       multiple libev event sources into one "super-watcher":
3494
3495          struct my_biggy
3496          {
3497            int some_data;
3498            ev_timer t1;
3499            ev_timer t2;
3500          }
3501
3502       In this case getting the pointer to "my_biggy" is a bit more
3503       complicated: Either you store the address of your "my_biggy" struct in
3504       the "data" member of the watcher (for woozies or C++ coders), or you
3505       need to use some pointer arithmetic using "offsetof" inside your
3506       watchers (for real programmers):
3507
3508          #include <stddef.h>
3509
3510          static void
3511          t1_cb (EV_P_ ev_timer *w, int revents)
3512          {
3513            struct my_biggy big = (struct my_biggy *)
3514              (((char *)w) - offsetof (struct my_biggy, t1));
3515          }
3516
3517          static void
3518          t2_cb (EV_P_ ev_timer *w, int revents)
3519          {
3520            struct my_biggy big = (struct my_biggy *)
3521              (((char *)w) - offsetof (struct my_biggy, t2));
3522          }
3523
3524   AVOIDING FINISHING BEFORE RETURNING
3525       Often you have structures like this in event-based programs:
3526
3527         callback ()
3528         {
3529           free (request);
3530         }
3531
3532         request = start_new_request (..., callback);
3533
3534       The intent is to start some "lengthy" operation. The "request" could be
3535       used to cancel the operation, or do other things with it.
3536
3537       It's not uncommon to have code paths in "start_new_request" that
3538       immediately invoke the callback, for example, to report errors. Or you
3539       add some caching layer that finds that it can skip the lengthy aspects
3540       of the operation and simply invoke the callback with the result.
3541
3542       The problem here is that this will happen before "start_new_request"
3543       has returned, so "request" is not set.
3544
3545       Even if you pass the request by some safer means to the callback, you
3546       might want to do something to the request after starting it, such as
3547       canceling it, which probably isn't working so well when the callback
3548       has already been invoked.
3549
3550       A common way around all these issues is to make sure that
3551       "start_new_request" always returns before the callback is invoked. If
3552       "start_new_request" immediately knows the result, it can artificially
3553       delay invoking the callback by using a "prepare" or "idle" watcher for
3554       example, or more sneakily, by reusing an existing (stopped) watcher and
3555       pushing it into the pending queue:
3556
3557          ev_set_cb (watcher, callback);
3558          ev_feed_event (EV_A_ watcher, 0);
3559
3560       This way, "start_new_request" can safely return before the callback is
3561       invoked, while not delaying callback invocation too much.
3562
3563   MODEL/NESTED EVENT LOOP INVOCATIONS AND EXIT CONDITIONS
3564       Often (especially in GUI toolkits) there are places where you have
3565       modal interaction, which is most easily implemented by recursively
3566       invoking "ev_run".
3567
3568       This brings the problem of exiting - a callback might want to finish
3569       the main "ev_run" call, but not the nested one (e.g. user clicked
3570       "Quit", but a modal "Are you sure?" dialog is still waiting), or just
3571       the nested one and not the main one (e.g. user clocked "Ok" in a modal
3572       dialog), or some other combination: In these cases, a simple "ev_break"
3573       will not work.
3574
3575       The solution is to maintain "break this loop" variable for each
3576       "ev_run" invocation, and use a loop around "ev_run" until the condition
3577       is triggered, using "EVRUN_ONCE":
3578
3579          // main loop
3580          int exit_main_loop = 0;
3581
3582          while (!exit_main_loop)
3583            ev_run (EV_DEFAULT_ EVRUN_ONCE);
3584
3585          // in a modal watcher
3586          int exit_nested_loop = 0;
3587
3588          while (!exit_nested_loop)
3589            ev_run (EV_A_ EVRUN_ONCE);
3590
3591       To exit from any of these loops, just set the corresponding exit
3592       variable:
3593
3594          // exit modal loop
3595          exit_nested_loop = 1;
3596
3597          // exit main program, after modal loop is finished
3598          exit_main_loop = 1;
3599
3600          // exit both
3601          exit_main_loop = exit_nested_loop = 1;
3602
3603   THREAD LOCKING EXAMPLE
3604       Here is a fictitious example of how to run an event loop in a different
3605       thread from where callbacks are being invoked and watchers are
3606       created/added/removed.
3607
3608       For a real-world example, see the "EV::Loop::Async" perl module, which
3609       uses exactly this technique (which is suited for many high-level
3610       languages).
3611
3612       The example uses a pthread mutex to protect the loop data, a condition
3613       variable to wait for callback invocations, an async watcher to notify
3614       the event loop thread and an unspecified mechanism to wake up the main
3615       thread.
3616
3617       First, you need to associate some data with the event loop:
3618
3619          typedef struct {
3620            mutex_t lock; /* global loop lock */
3621            ev_async async_w;
3622            thread_t tid;
3623            cond_t invoke_cv;
3624          } userdata;
3625
3626          void prepare_loop (EV_P)
3627          {
3628             // for simplicity, we use a static userdata struct.
3629             static userdata u;
3630
3631             ev_async_init (&u->async_w, async_cb);
3632             ev_async_start (EV_A_ &u->async_w);
3633
3634             pthread_mutex_init (&u->lock, 0);
3635             pthread_cond_init (&u->invoke_cv, 0);
3636
3637             // now associate this with the loop
3638             ev_set_userdata (EV_A_ u);
3639             ev_set_invoke_pending_cb (EV_A_ l_invoke);
3640             ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
3641
3642             // then create the thread running ev_run
3643             pthread_create (&u->tid, 0, l_run, EV_A);
3644          }
3645
3646       The callback for the "ev_async" watcher does nothing: the watcher is
3647       used solely to wake up the event loop so it takes notice of any new
3648       watchers that might have been added:
3649
3650          static void
3651          async_cb (EV_P_ ev_async *w, int revents)
3652          {
3653             // just used for the side effects
3654          }
3655
3656       The "l_release" and "l_acquire" callbacks simply unlock/lock the mutex
3657       protecting the loop data, respectively.
3658
3659          static void
3660          l_release (EV_P)
3661          {
3662            userdata *u = ev_userdata (EV_A);
3663            pthread_mutex_unlock (&u->lock);
3664          }
3665
3666          static void
3667          l_acquire (EV_P)
3668          {
3669            userdata *u = ev_userdata (EV_A);
3670            pthread_mutex_lock (&u->lock);
3671          }
3672
3673       The event loop thread first acquires the mutex, and then jumps straight
3674       into "ev_run":
3675
3676          void *
3677          l_run (void *thr_arg)
3678          {
3679            struct ev_loop *loop = (struct ev_loop *)thr_arg;
3680
3681            l_acquire (EV_A);
3682            pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
3683            ev_run (EV_A_ 0);
3684            l_release (EV_A);
3685
3686            return 0;
3687          }
3688
3689       Instead of invoking all pending watchers, the "l_invoke" callback will
3690       signal the main thread via some unspecified mechanism (signals? pipe
3691       writes? "Async::Interrupt"?) and then waits until all pending watchers
3692       have been called (in a while loop because a) spurious wakeups are
3693       possible and b) skipping inter-thread-communication when there are no
3694       pending watchers is very beneficial):
3695
3696          static void
3697          l_invoke (EV_P)
3698          {
3699            userdata *u = ev_userdata (EV_A);
3700
3701            while (ev_pending_count (EV_A))
3702              {
3703                wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
3704                pthread_cond_wait (&u->invoke_cv, &u->lock);
3705              }
3706          }
3707
3708       Now, whenever the main thread gets told to invoke pending watchers, it
3709       will grab the lock, call "ev_invoke_pending" and then signal the loop
3710       thread to continue:
3711
3712          static void
3713          real_invoke_pending (EV_P)
3714          {
3715            userdata *u = ev_userdata (EV_A);
3716
3717            pthread_mutex_lock (&u->lock);
3718            ev_invoke_pending (EV_A);
3719            pthread_cond_signal (&u->invoke_cv);
3720            pthread_mutex_unlock (&u->lock);
3721          }
3722
3723       Whenever you want to start/stop a watcher or do other modifications to
3724       an event loop, you will now have to lock:
3725
3726          ev_timer timeout_watcher;
3727          userdata *u = ev_userdata (EV_A);
3728
3729          ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
3730
3731          pthread_mutex_lock (&u->lock);
3732          ev_timer_start (EV_A_ &timeout_watcher);
3733          ev_async_send (EV_A_ &u->async_w);
3734          pthread_mutex_unlock (&u->lock);
3735
3736       Note that sending the "ev_async" watcher is required because otherwise
3737       an event loop currently blocking in the kernel will have no knowledge
3738       about the newly added timer. By waking up the loop it will pick up any
3739       new watchers in the next event loop iteration.
3740
3741   THREADS, COROUTINES, CONTINUATIONS, QUEUES... INSTEAD OF CALLBACKS
3742       While the overhead of a callback that e.g. schedules a thread is small,
3743       it is still an overhead. If you embed libev, and your main usage is
3744       with some kind of threads or coroutines, you might want to customise
3745       libev so that doesn't need callbacks anymore.
3746
3747       Imagine you have coroutines that you can switch to using a function
3748       "switch_to (coro)", that libev runs in a coroutine called "libev_coro"
3749       and that due to some magic, the currently active coroutine is stored in
3750       a global called "current_coro". Then you can build your own "wait for
3751       libev event" primitive by changing "EV_CB_DECLARE" and "EV_CB_INVOKE"
3752       (note the differing ";" conventions):
3753
3754          #define EV_CB_DECLARE(type)   struct my_coro *cb;
3755          #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3756
3757       That means instead of having a C callback function, you store the
3758       coroutine to switch to in each watcher, and instead of having libev
3759       call your callback, you instead have it switch to that coroutine.
3760
3761       A coroutine might now wait for an event with a function called
3762       "wait_for_event". (the watcher needs to be started, as always, but it
3763       doesn't matter when, or whether the watcher is active or not when this
3764       function is called):
3765
3766          void
3767          wait_for_event (ev_watcher *w)
3768          {
3769            ev_set_cb (w, current_coro);
3770            switch_to (libev_coro);
3771          }
3772
3773       That basically suspends the coroutine inside "wait_for_event" and
3774       continues the libev coroutine, which, when appropriate, switches back
3775       to this or any other coroutine.
3776
3777       You can do similar tricks if you have, say, threads with an event queue
3778       - instead of storing a coroutine, you store the queue object and
3779       instead of switching to a coroutine, you push the watcher onto the
3780       queue and notify any waiters.
3781
3782       To embed libev, see "EMBEDDING", but in short, it's easiest to create
3783       two files, my_ev.h and my_ev.c that include the respective libev files:
3784
3785          // my_ev.h
3786          #define EV_CB_DECLARE(type)   struct my_coro *cb;
3787          #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
3788          #include "../libev/ev.h"
3789
3790          // my_ev.c
3791          #define EV_H "my_ev.h"
3792          #include "../libev/ev.c"
3793
3794       And then use my_ev.h when you would normally use ev.h, and compile
3795       my_ev.c into your project. When properly specifying include paths, you
3796       can even use ev.h as header file name directly.
3797

LIBEVENT EMULATION

3799       Libev offers a compatibility emulation layer for libevent. It cannot
3800       emulate the internals of libevent, so here are some usage hints:
3801
3802       ·   Only the libevent-1.4.1-beta API is being emulated.
3803
3804           This was the newest libevent version available when libev was
3805           implemented, and is still mostly unchanged in 2010.
3806
3807       ·   Use it by including <event.h>, as usual.
3808
3809       ·   The following members are fully supported: ev_base, ev_callback,
3810           ev_arg, ev_fd, ev_res, ev_events.
3811
3812       ·   Avoid using ev_flags and the EVLIST_*-macros, while it is
3813           maintained by libev, it does not work exactly the same way as in
3814           libevent (consider it a private API).
3815
3816       ·   Priorities are not currently supported. Initialising priorities
3817           will fail and all watchers will have the same priority, even though
3818           there is an ev_pri field.
3819
3820       ·   In libevent, the last base created gets the signals, in libev, the
3821           base that registered the signal gets the signals.
3822
3823       ·   Other members are not supported.
3824
3825       ·   The libev emulation is not ABI compatible to libevent, you need to
3826           use the libev header file and library.
3827

C++ SUPPORT

3829   C API
3830       The normal C API should work fine when used from C++: both ev.h and the
3831       libev sources can be compiled as C++. Therefore, code that uses the C
3832       API will work fine.
3833
3834       Proper exception specifications might have to be added to callbacks
3835       passed to libev: exceptions may be thrown only from watcher callbacks,
3836       all other callbacks (allocator, syserr, loop acquire/release and
3837       periodic reschedule callbacks) must not throw exceptions, and might
3838       need a "noexcept" specification. If you have code that needs to be
3839       compiled as both C and C++ you can use the "EV_NOEXCEPT" macro for
3840       this:
3841
3842          static void
3843          fatal_error (const char *msg) EV_NOEXCEPT
3844          {
3845            perror (msg);
3846            abort ();
3847          }
3848
3849          ...
3850          ev_set_syserr_cb (fatal_error);
3851
3852       The only API functions that can currently throw exceptions are
3853       "ev_run", "ev_invoke", "ev_invoke_pending" and "ev_loop_destroy" (the
3854       latter because it runs cleanup watchers).
3855
3856       Throwing exceptions in watcher callbacks is only supported if libev
3857       itself is compiled with a C++ compiler or your C and C++ environments
3858       allow throwing exceptions through C libraries (most do).
3859
3860   C++ API
3861       Libev comes with some simplistic wrapper classes for C++ that mainly
3862       allow you to use some convenience methods to start/stop watchers and
3863       also change the callback model to a model using method callbacks on
3864       objects.
3865
3866       To use it,
3867
3868          #include <ev++.h>
3869
3870       This automatically includes ev.h and puts all of its definitions (many
3871       of them macros) into the global namespace. All C++ specific things are
3872       put into the "ev" namespace. It should support all the same embedding
3873       options as ev.h, most notably "EV_MULTIPLICITY".
3874
3875       Care has been taken to keep the overhead low. The only data member the
3876       C++ classes add (compared to plain C-style watchers) is the event loop
3877       pointer that the watcher is associated with (or no additional members
3878       at all if you disable "EV_MULTIPLICITY" when embedding libev).
3879
3880       Currently, functions, static and non-static member functions and
3881       classes with "operator ()" can be used as callbacks. Other types should
3882       be easy to add as long as they only need one additional pointer for
3883       context. If you need support for other types of functors please contact
3884       the author (preferably after implementing it).
3885
3886       For all this to work, your C++ compiler either has to use the same
3887       calling conventions as your C compiler (for static member functions),
3888       or you have to embed libev and compile libev itself as C++.
3889
3890       Here is a list of things available in the "ev" namespace:
3891
3892       "ev::READ", "ev::WRITE" etc.
3893           These are just enum values with the same values as the "EV_READ"
3894           etc.  macros from ev.h.
3895
3896       "ev::tstamp", "ev::now"
3897           Aliases to the same types/functions as with the "ev_" prefix.
3898
3899       "ev::io", "ev::timer", "ev::periodic", "ev::idle", "ev::sig" etc.
3900           For each "ev_TYPE" watcher in ev.h there is a corresponding class
3901           of the same name in the "ev" namespace, with the exception of
3902           "ev_signal" which is called "ev::sig" to avoid clashes with the
3903           "signal" macro defined by many implementations.
3904
3905           All of those classes have these methods:
3906
3907           ev::TYPE::TYPE ()
3908           ev::TYPE::TYPE (loop)
3909           ev::TYPE::~TYPE
3910               The constructor (optionally) takes an event loop to associate
3911               the watcher with. If it is omitted, it will use "EV_DEFAULT".
3912
3913               The constructor calls "ev_init" for you, which means you have
3914               to call the "set" method before starting it.
3915
3916               It will not set a callback, however: You have to call the
3917               templated "set" method to set a callback before you can start
3918               the watcher.
3919
3920               (The reason why you have to use a method is a limitation in C++
3921               which does not allow explicit template arguments for
3922               constructors).
3923
3924               The destructor automatically stops the watcher if it is active.
3925
3926           w->set<class, &class::method> (object *)
3927               This method sets the callback method to call. The method has to
3928               have a signature of "void (*)(ev_TYPE &, int)", it receives the
3929               watcher as first argument and the "revents" as second. The
3930               object must be given as parameter and is stored in the "data"
3931               member of the watcher.
3932
3933               This method synthesizes efficient thunking code to call your
3934               method from the C callback that libev requires. If your
3935               compiler can inline your callback (i.e. it is visible to it at
3936               the place of the "set" call and your compiler is good :), then
3937               the method will be fully inlined into the thunking function,
3938               making it as fast as a direct C callback.
3939
3940               Example: simple class declaration and watcher initialisation
3941
3942                  struct myclass
3943                  {
3944                    void io_cb (ev::io &w, int revents) { }
3945                  }
3946
3947                  myclass obj;
3948                  ev::io iow;
3949                  iow.set <myclass, &myclass::io_cb> (&obj);
3950
3951           w->set (object *)
3952               This is a variation of a method callback - leaving out the
3953               method to call will default the method to "operator ()", which
3954               makes it possible to use functor objects without having to
3955               manually specify the "operator ()" all the time. Incidentally,
3956               you can then also leave out the template argument list.
3957
3958               The "operator ()" method prototype must be "void operator
3959               ()(watcher &w, int revents)".
3960
3961               See the method-"set" above for more details.
3962
3963               Example: use a functor object as callback.
3964
3965                  struct myfunctor
3966                  {
3967                    void operator() (ev::io &w, int revents)
3968                    {
3969                      ...
3970                    }
3971                  }
3972
3973                  myfunctor f;
3974
3975                  ev::io w;
3976                  w.set (&f);
3977
3978           w->set<function> (void *data = 0)
3979               Also sets a callback, but uses a static method or plain
3980               function as callback. The optional "data" argument will be
3981               stored in the watcher's "data" member and is free for you to
3982               use.
3983
3984               The prototype of the "function" must be "void (*)(ev::TYPE &w,
3985               int)".
3986
3987               See the method-"set" above for more details.
3988
3989               Example: Use a plain function as callback.
3990
3991                  static void io_cb (ev::io &w, int revents) { }
3992                  iow.set <io_cb> ();
3993
3994           w->set (loop)
3995               Associates a different "struct ev_loop" with this watcher. You
3996               can only do this when the watcher is inactive (and not pending
3997               either).
3998
3999           w->set ([arguments])
4000               Basically the same as "ev_TYPE_set" (except for "ev::embed"
4001               watchers>), with the same arguments. Either this method or a
4002               suitable start method must be called at least once. Unlike the
4003               C counterpart, an active watcher gets automatically stopped and
4004               restarted when reconfiguring it with this method.
4005
4006               For "ev::embed" watchers this method is called "set_embed", to
4007               avoid clashing with the "set (loop)" method.
4008
4009           w->start ()
4010               Starts the watcher. Note that there is no "loop" argument, as
4011               the constructor already stores the event loop.
4012
4013           w->start ([arguments])
4014               Instead of calling "set" and "start" methods separately, it is
4015               often convenient to wrap them in one call. Uses the same type
4016               of arguments as the configure "set" method of the watcher.
4017
4018           w->stop ()
4019               Stops the watcher if it is active. Again, no "loop" argument.
4020
4021           w->again () ("ev::timer", "ev::periodic" only)
4022               For "ev::timer" and "ev::periodic", this invokes the
4023               corresponding "ev_TYPE_again" function.
4024
4025           w->sweep () ("ev::embed" only)
4026               Invokes "ev_embed_sweep".
4027
4028           w->update () ("ev::stat" only)
4029               Invokes "ev_stat_stat".
4030
4031       Example: Define a class with two I/O and idle watchers, start the I/O
4032       watchers in the constructor.
4033
4034          class myclass
4035          {
4036            ev::io   io  ; void io_cb   (ev::io   &w, int revents);
4037            ev::io   io2 ; void io2_cb  (ev::io   &w, int revents);
4038            ev::idle idle; void idle_cb (ev::idle &w, int revents);
4039
4040            myclass (int fd)
4041            {
4042              io  .set <myclass, &myclass::io_cb  > (this);
4043              io2 .set <myclass, &myclass::io2_cb > (this);
4044              idle.set <myclass, &myclass::idle_cb> (this);
4045
4046              io.set (fd, ev::WRITE); // configure the watcher
4047              io.start ();            // start it whenever convenient
4048
4049              io2.start (fd, ev::READ); // set + start in one call
4050            }
4051          };
4052

OTHER LANGUAGE BINDINGS

4054       Libev does not offer other language bindings itself, but bindings for a
4055       number of languages exist in the form of third-party packages. If you
4056       know any interesting language binding in addition to the ones listed
4057       here, drop me a note.
4058
4059       Perl
4060           The EV module implements the full libev API and is actually used to
4061           test libev. EV is developed together with libev. Apart from the EV
4062           core module, there are additional modules that implement libev-
4063           compatible interfaces to "libadns" ("EV::ADNS", but "AnyEvent::DNS"
4064           is preferred nowadays), "Net::SNMP" ("Net::SNMP::EV") and the
4065           "libglib" event core ("Glib::EV" and "EV::Glib").
4066
4067           It can be found and installed via CPAN, its homepage is at
4068           <http://software.schmorp.de/pkg/EV>.
4069
4070       Python
4071           Python bindings can be found at <http://code.google.com/p/pyev/>.
4072           It seems to be quite complete and well-documented.
4073
4074       Ruby
4075           Tony Arcieri has written a ruby extension that offers access to a
4076           subset of the libev API and adds file handle abstractions,
4077           asynchronous DNS and more on top of it. It can be found via gem
4078           servers. Its homepage is at <http://rev.rubyforge.org/>.
4079
4080           Roger Pack reports that using the link order "-lws2_32
4081           -lmsvcrt-ruby-190" makes rev work even on mingw.
4082
4083       Haskell
4084           A haskell binding to libev is available at
4085           <http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
4086
4087       D   Leandro Lucarella has written a D language binding (ev.d) for
4088           libev, to be found at
4089           <http://www.llucax.com.ar/proj/ev.d/index.html>.
4090
4091       Ocaml
4092           Erkki Seppala has written Ocaml bindings for libev, to be found at
4093           <http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
4094
4095       Lua Brian Maher has written a partial interface to libev for lua (at
4096           the time of this writing, only "ev_io" and "ev_timer"), to be found
4097           at <http://github.com/brimworks/lua-ev>.
4098
4099       Javascript
4100           Node.js (<http://nodejs.org>) uses libev as the underlying event
4101           library.
4102
4103       Others
4104           There are others, and I stopped counting.
4105

MACRO MAGIC

4107       Libev can be compiled with a variety of options, the most fundamental
4108       of which is "EV_MULTIPLICITY". This option determines whether (most)
4109       functions and callbacks have an initial "struct ev_loop *" argument.
4110
4111       To make it easier to write programs that cope with either variant, the
4112       following macros are defined:
4113
4114       "EV_A", "EV_A_"
4115           This provides the loop argument for functions, if one is required
4116           ("ev loop argument"). The "EV_A" form is used when this is the sole
4117           argument, "EV_A_" is used when other arguments are following.
4118           Example:
4119
4120              ev_unref (EV_A);
4121              ev_timer_add (EV_A_ watcher);
4122              ev_run (EV_A_ 0);
4123
4124           It assumes the variable "loop" of type "struct ev_loop *" is in
4125           scope, which is often provided by the following macro.
4126
4127       "EV_P", "EV_P_"
4128           This provides the loop parameter for functions, if one is required
4129           ("ev loop parameter"). The "EV_P" form is used when this is the
4130           sole parameter, "EV_P_" is used when other parameters are
4131           following. Example:
4132
4133              // this is how ev_unref is being declared
4134              static void ev_unref (EV_P);
4135
4136              // this is how you can declare your typical callback
4137              static void cb (EV_P_ ev_timer *w, int revents)
4138
4139           It declares a parameter "loop" of type "struct ev_loop *", quite
4140           suitable for use with "EV_A".
4141
4142       "EV_DEFAULT", "EV_DEFAULT_"
4143           Similar to the other two macros, this gives you the value of the
4144           default loop, if multiple loops are supported ("ev loop default").
4145           The default loop will be initialised if it isn't already
4146           initialised.
4147
4148           For non-multiplicity builds, these macros do nothing, so you always
4149           have to initialise the loop somewhere.
4150
4151       "EV_DEFAULT_UC", "EV_DEFAULT_UC_"
4152           Usage identical to "EV_DEFAULT" and "EV_DEFAULT_", but requires
4153           that the default loop has been initialised ("UC" == unchecked).
4154           Their behaviour is undefined when the default loop has not been
4155           initialised by a previous execution of "EV_DEFAULT", "EV_DEFAULT_"
4156           or "ev_default_init (...)".
4157
4158           It is often prudent to use "EV_DEFAULT" when initialising the first
4159           watcher in a function but use "EV_DEFAULT_UC" afterwards.
4160
4161       Example: Declare and initialise a check watcher, utilising the above
4162       macros so it will work regardless of whether multiple loops are
4163       supported or not.
4164
4165          static void
4166          check_cb (EV_P_ ev_timer *w, int revents)
4167          {
4168            ev_check_stop (EV_A_ w);
4169          }
4170
4171          ev_check check;
4172          ev_check_init (&check, check_cb);
4173          ev_check_start (EV_DEFAULT_ &check);
4174          ev_run (EV_DEFAULT_ 0);
4175

EMBEDDING

4177       Libev can (and often is) directly embedded into host applications.
4178       Examples of applications that embed it include the Deliantra Game
4179       Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe) and
4180       rxvt-unicode.
4181
4182       The goal is to enable you to just copy the necessary files into your
4183       source directory without having to change even a single line in them,
4184       so you can easily upgrade by simply copying (or having a checked-out
4185       copy of libev somewhere in your source tree).
4186
4187   FILESETS
4188       Depending on what features you need you need to include one or more
4189       sets of files in your application.
4190
4191       CORE EVENT LOOP
4192
4193       To include only the libev core (all the "ev_*" functions), with manual
4194       configuration (no autoconf):
4195
4196          #define EV_STANDALONE 1
4197          #include "ev.c"
4198
4199       This will automatically include ev.h, too, and should be done in a
4200       single C source file only to provide the function implementations. To
4201       use it, do the same for ev.h in all files wishing to use this API (best
4202       done by writing a wrapper around ev.h that you can include instead and
4203       where you can put other configuration options):
4204
4205          #define EV_STANDALONE 1
4206          #include "ev.h"
4207
4208       Both header files and implementation files can be compiled with a C++
4209       compiler (at least, that's a stated goal, and breakage will be treated
4210       as a bug).
4211
4212       You need the following files in your source tree, or in a directory in
4213       your include path (e.g. in libev/ when using -Ilibev):
4214
4215          ev.h
4216          ev.c
4217          ev_vars.h
4218          ev_wrap.h
4219
4220          ev_win32.c      required on win32 platforms only
4221
4222          ev_select.c     only when select backend is enabled
4223          ev_poll.c       only when poll backend is enabled
4224          ev_epoll.c      only when the epoll backend is enabled
4225          ev_kqueue.c     only when the kqueue backend is enabled
4226          ev_port.c       only when the solaris port backend is enabled
4227
4228       ev.c includes the backend files directly when enabled, so you only need
4229       to compile this single file.
4230
4231       LIBEVENT COMPATIBILITY API
4232
4233       To include the libevent compatibility API, also include:
4234
4235          #include "event.c"
4236
4237       in the file including ev.c, and:
4238
4239          #include "event.h"
4240
4241       in the files that want to use the libevent API. This also includes
4242       ev.h.
4243
4244       You need the following additional files for this:
4245
4246          event.h
4247          event.c
4248
4249       AUTOCONF SUPPORT
4250
4251       Instead of using "EV_STANDALONE=1" and providing your configuration in
4252       whatever way you want, you can also "m4_include([libev.m4])" in your
4253       configure.ac and leave "EV_STANDALONE" undefined. ev.c will then
4254       include config.h and configure itself accordingly.
4255
4256       For this of course you need the m4 file:
4257
4258          libev.m4
4259
4260   PREPROCESSOR SYMBOLS/MACROS
4261       Libev can be configured via a variety of preprocessor symbols you have
4262       to define before including (or compiling) any of its files. The default
4263       in the absence of autoconf is documented for every option.
4264
4265       Symbols marked with "(h)" do not change the ABI, and can have different
4266       values when compiling libev vs. including ev.h, so it is permissible to
4267       redefine them before including ev.h without breaking compatibility to a
4268       compiled library. All other symbols change the ABI, which means all
4269       users of libev and the libev code itself must be compiled with
4270       compatible settings.
4271
4272       EV_COMPAT3 (h)
4273           Backwards compatibility is a major concern for libev. This is why
4274           this release of libev comes with wrappers for the functions and
4275           symbols that have been renamed between libev version 3 and 4.
4276
4277           You can disable these wrappers (to test compatibility with future
4278           versions) by defining "EV_COMPAT3" to 0 when compiling your
4279           sources. This has the additional advantage that you can drop the
4280           "struct" from "struct ev_loop" declarations, as libev will provide
4281           an "ev_loop" typedef in that case.
4282
4283           In some future version, the default for "EV_COMPAT3" will become 0,
4284           and in some even more future version the compatibility code will be
4285           removed completely.
4286
4287       EV_STANDALONE (h)
4288           Must always be 1 if you do not use autoconf configuration, which
4289           keeps libev from including config.h, and it also defines dummy
4290           implementations for some libevent functions (such as logging, which
4291           is not supported). It will also not define any of the structs
4292           usually found in event.h that are not directly supported by the
4293           libev core alone.
4294
4295           In standalone mode, libev will still try to automatically deduce
4296           the configuration, but has to be more conservative.
4297
4298       EV_USE_FLOOR
4299           If defined to be 1, libev will use the "floor ()" function for its
4300           periodic reschedule calculations, otherwise libev will fall back on
4301           a portable (slower) implementation. If you enable this, you usually
4302           have to link against libm or something equivalent. Enabling this
4303           when the "floor" function is not available will fail, so the safe
4304           default is to not enable this.
4305
4306       EV_USE_MONOTONIC
4307           If defined to be 1, libev will try to detect the availability of
4308           the monotonic clock option at both compile time and runtime.
4309           Otherwise no use of the monotonic clock option will be attempted.
4310           If you enable this, you usually have to link against librt or
4311           something similar. Enabling it when the functionality isn't
4312           available is safe, though, although you have to make sure you link
4313           against any libraries where the "clock_gettime" function is hiding
4314           in (often -lrt). See also "EV_USE_CLOCK_SYSCALL".
4315
4316       EV_USE_REALTIME
4317           If defined to be 1, libev will try to detect the availability of
4318           the real-time clock option at compile time (and assume its
4319           availability at runtime if successful). Otherwise no use of the
4320           real-time clock option will be attempted. This effectively replaces
4321           "gettimeofday" by "clock_get (CLOCK_REALTIME, ...)" and will not
4322           normally affect correctness. See the note about libraries in the
4323           description of "EV_USE_MONOTONIC", though. Defaults to the opposite
4324           value of "EV_USE_CLOCK_SYSCALL".
4325
4326       EV_USE_CLOCK_SYSCALL
4327           If defined to be 1, libev will try to use a direct syscall instead
4328           of calling the system-provided "clock_gettime" function. This
4329           option exists because on GNU/Linux, "clock_gettime" is in "librt",
4330           but "librt" unconditionally pulls in "libpthread", slowing down
4331           single-threaded programs needlessly. Using a direct syscall is
4332           slightly slower (in theory), because no optimised vdso
4333           implementation can be used, but avoids the pthread dependency.
4334           Defaults to 1 on GNU/Linux with glibc 2.x or higher, as it
4335           simplifies linking (no need for "-lrt").
4336
4337       EV_USE_NANOSLEEP
4338           If defined to be 1, libev will assume that "nanosleep ()" is
4339           available and will use it for delays. Otherwise it will use "select
4340           ()".
4341
4342       EV_USE_EVENTFD
4343           If defined to be 1, then libev will assume that "eventfd ()" is
4344           available and will probe for kernel support at runtime. This will
4345           improve "ev_signal" and "ev_async" performance and reduce resource
4346           consumption.  If undefined, it will be enabled if the headers
4347           indicate GNU/Linux + Glibc 2.7 or newer, otherwise disabled.
4348
4349       EV_USE_SELECT
4350           If undefined or defined to be 1, libev will compile in support for
4351           the "select"(2) backend. No attempt at auto-detection will be done:
4352           if no other method takes over, select will be it. Otherwise the
4353           select backend will not be compiled in.
4354
4355       EV_SELECT_USE_FD_SET
4356           If defined to 1, then the select backend will use the system
4357           "fd_set" structure. This is useful if libev doesn't compile due to
4358           a missing "NFDBITS" or "fd_mask" definition or it mis-guesses the
4359           bitset layout on exotic systems. This usually limits the range of
4360           file descriptors to some low limit such as 1024 or might have other
4361           limitations (winsocket only allows 64 sockets). The "FD_SETSIZE"
4362           macro, set before compilation, configures the maximum size of the
4363           "fd_set".
4364
4365       EV_SELECT_IS_WINSOCKET
4366           When defined to 1, the select backend will assume that
4367           select/socket/connect etc. don't understand file descriptors but
4368           wants osf handles on win32 (this is the case when the select to be
4369           used is the winsock select). This means that it will call
4370           "_get_osfhandle" on the fd to convert it to an OS handle.
4371           Otherwise, it is assumed that all these functions actually work on
4372           fds, even on win32. Should not be defined on non-win32 platforms.
4373
4374       EV_FD_TO_WIN32_HANDLE(fd)
4375           If "EV_SELECT_IS_WINSOCKET" is enabled, then libev needs a way to
4376           map file descriptors to socket handles. When not defining this
4377           symbol (the default), then libev will call "_get_osfhandle", which
4378           is usually correct. In some cases, programs use their own file
4379           descriptor management, in which case they can provide this function
4380           to map fds to socket handles.
4381
4382       EV_WIN32_HANDLE_TO_FD(handle)
4383           If "EV_SELECT_IS_WINSOCKET" then libev maps handles to file
4384           descriptors using the standard "_open_osfhandle" function. For
4385           programs implementing their own fd to handle mapping, overwriting
4386           this function makes it easier to do so. This can be done by
4387           defining this macro to an appropriate value.
4388
4389       EV_WIN32_CLOSE_FD(fd)
4390           If programs implement their own fd to handle mapping on win32, then
4391           this macro can be used to override the "close" function, useful to
4392           unregister file descriptors again. Note that the replacement
4393           function has to close the underlying OS handle.
4394
4395       EV_USE_WSASOCKET
4396           If defined to be 1, libev will use "WSASocket" to create its
4397           internal communication socket, which works better in some
4398           environments. Otherwise, the normal "socket" function will be used,
4399           which works better in other environments.
4400
4401       EV_USE_POLL
4402           If defined to be 1, libev will compile in support for the "poll"(2)
4403           backend. Otherwise it will be enabled on non-win32 platforms. It
4404           takes precedence over select.
4405
4406       EV_USE_EPOLL
4407           If defined to be 1, libev will compile in support for the Linux
4408           "epoll"(7) backend. Its availability will be detected at runtime,
4409           otherwise another method will be used as fallback. This is the
4410           preferred backend for GNU/Linux systems. If undefined, it will be
4411           enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4412           otherwise disabled.
4413
4414       EV_USE_KQUEUE
4415           If defined to be 1, libev will compile in support for the BSD style
4416           "kqueue"(2) backend. Its actual availability will be detected at
4417           runtime, otherwise another method will be used as fallback. This is
4418           the preferred backend for BSD and BSD-like systems, although on
4419           most BSDs kqueue only supports some types of fds correctly (the
4420           only platform we found that supports ptys for example was NetBSD),
4421           so kqueue might be compiled in, but not be used unless explicitly
4422           requested. The best way to use it is to find out whether kqueue
4423           supports your type of fd properly and use an embedded kqueue loop.
4424
4425       EV_USE_PORT
4426           If defined to be 1, libev will compile in support for the Solaris
4427           10 port style backend. Its availability will be detected at
4428           runtime, otherwise another method will be used as fallback. This is
4429           the preferred backend for Solaris 10 systems.
4430
4431       EV_USE_DEVPOLL
4432           Reserved for future expansion, works like the USE symbols above.
4433
4434       EV_USE_INOTIFY
4435           If defined to be 1, libev will compile in support for the Linux
4436           inotify interface to speed up "ev_stat" watchers. Its actual
4437           availability will be detected at runtime. If undefined, it will be
4438           enabled if the headers indicate GNU/Linux + Glibc 2.4 or newer,
4439           otherwise disabled.
4440
4441       EV_NO_SMP
4442           If defined to be 1, libev will assume that memory is always
4443           coherent between threads, that is, threads can be used, but threads
4444           never run on different cpus (or different cpu cores). This reduces
4445           dependencies and makes libev faster.
4446
4447       EV_NO_THREADS
4448           If defined to be 1, libev will assume that it will never be called
4449           from different threads (that includes signal handlers), which is a
4450           stronger assumption than "EV_NO_SMP", above. This reduces
4451           dependencies and makes libev faster.
4452
4453       EV_ATOMIC_T
4454           Libev requires an integer type (suitable for storing 0 or 1) whose
4455           access is atomic with respect to other threads or signal contexts.
4456           No such type is easily found in the C language, so you can provide
4457           your own type that you know is safe for your purposes. It is used
4458           both for signal handler "locking" as well as for signal and thread
4459           safety in "ev_async" watchers.
4460
4461           In the absence of this define, libev will use "sig_atomic_t
4462           volatile" (from signal.h), which is usually good enough on most
4463           platforms.
4464
4465       EV_H (h)
4466           The name of the ev.h header file used to include it. The default if
4467           undefined is "ev.h" in event.h, ev.c and ev++.h. This can be used
4468           to virtually rename the ev.h header file in case of conflicts.
4469
4470       EV_CONFIG_H (h)
4471           If "EV_STANDALONE" isn't 1, this variable can be used to override
4472           ev.c's idea of where to find the config.h file, similarly to
4473           "EV_H", above.
4474
4475       EV_EVENT_H (h)
4476           Similarly to "EV_H", this macro can be used to override event.c's
4477           idea of how the event.h header can be found, the default is
4478           "event.h".
4479
4480       EV_PROTOTYPES (h)
4481           If defined to be 0, then ev.h will not define any function
4482           prototypes, but still define all the structs and other symbols.
4483           This is occasionally useful if you want to provide your own wrapper
4484           functions around libev functions.
4485
4486       EV_MULTIPLICITY
4487           If undefined or defined to 1, then all event-loop-specific
4488           functions will have the "struct ev_loop *" as first argument, and
4489           you can create additional independent event loops. Otherwise there
4490           will be no support for multiple event loops and there is no first
4491           event loop pointer argument. Instead, all functions act on the
4492           single default loop.
4493
4494           Note that "EV_DEFAULT" and "EV_DEFAULT_" will no longer provide a
4495           default loop when multiplicity is switched off - you always have to
4496           initialise the loop manually in this case.
4497
4498       EV_MINPRI
4499       EV_MAXPRI
4500           The range of allowed priorities. "EV_MINPRI" must be smaller or
4501           equal to "EV_MAXPRI", but otherwise there are no non-obvious
4502           limitations. You can provide for more priorities by overriding
4503           those symbols (usually defined to be "-2" and 2, respectively).
4504
4505           When doing priority-based operations, libev usually has to linearly
4506           search all the priorities, so having many of them (hundreds) uses a
4507           lot of space and time, so using the defaults of five priorities (-2
4508           .. +2) is usually fine.
4509
4510           If your embedding application does not need any priorities,
4511           defining these both to 0 will save some memory and CPU.
4512
4513       EV_PERIODIC_ENABLE, EV_IDLE_ENABLE, EV_EMBED_ENABLE, EV_STAT_ENABLE,
4514       EV_PREPARE_ENABLE, EV_CHECK_ENABLE, EV_FORK_ENABLE, EV_SIGNAL_ENABLE,
4515       EV_ASYNC_ENABLE, EV_CHILD_ENABLE.
4516           If undefined or defined to be 1 (and the platform supports it),
4517           then the respective watcher type is supported. If defined to be 0,
4518           then it is not. Disabling watcher types mainly saves code size.
4519
4520       EV_FEATURES
4521           If you need to shave off some kilobytes of code at the expense of
4522           some speed (but with the full API), you can define this symbol to
4523           request certain subsets of functionality. The default is to enable
4524           all features that can be enabled on the platform.
4525
4526           A typical way to use this symbol is to define it to 0 (or to a
4527           bitset with some broad features you want) and then selectively re-
4528           enable additional parts you want, for example if you want
4529           everything minimal, but multiple event loop support, async and
4530           child watchers and the poll backend, use this:
4531
4532              #define EV_FEATURES 0
4533              #define EV_MULTIPLICITY 1
4534              #define EV_USE_POLL 1
4535              #define EV_CHILD_ENABLE 1
4536              #define EV_ASYNC_ENABLE 1
4537
4538           The actual value is a bitset, it can be a combination of the
4539           following values (by default, all of these are enabled):
4540
4541           1 - faster/larger code
4542               Use larger code to speed up some operations.
4543
4544               Currently this is used to override some inlining decisions
4545               (enlarging the code size by roughly 30% on amd64).
4546
4547               When optimising for size, use of compiler flags such as "-Os"
4548               with gcc is recommended, as well as "-DNDEBUG", as libev
4549               contains a number of assertions.
4550
4551               The default is off when "__OPTIMIZE_SIZE__" is defined by your
4552               compiler (e.g. gcc with "-Os").
4553
4554           2 - faster/larger data structures
4555               Replaces the small 2-heap for timer management by a faster
4556               4-heap, larger hash table sizes and so on. This will usually
4557               further increase code size and can additionally have an effect
4558               on the size of data structures at runtime.
4559
4560               The default is off when "__OPTIMIZE_SIZE__" is defined by your
4561               compiler (e.g. gcc with "-Os").
4562
4563           4 - full API configuration
4564               This enables priorities (sets "EV_MAXPRI"=2 and
4565               "EV_MINPRI"=-2), and enables multiplicity
4566               ("EV_MULTIPLICITY"=1).
4567
4568           8 - full API
4569               This enables a lot of the "lesser used" API functions. See
4570               "ev.h" for details on which parts of the API are still
4571               available without this feature, and do not complain if this
4572               subset changes over time.
4573
4574           16 - enable all optional watcher types
4575               Enables all optional watcher types.  If you want to selectively
4576               enable only some watcher types other than I/O and timers (e.g.
4577               prepare, embed, async, child...) you can enable them manually
4578               by defining "EV_watchertype_ENABLE" to 1 instead.
4579
4580           32 - enable all backends
4581               This enables all backends - without this feature, you need to
4582               enable at least one backend manually ("EV_USE_SELECT" is a good
4583               choice).
4584
4585           64 - enable OS-specific "helper" APIs
4586               Enable inotify, eventfd, signalfd and similar OS-specific
4587               helper APIs by default.
4588
4589           Compiling with "gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1
4590           -DEV_FEATURES=0" reduces the compiled size of libev from 24.7Kb
4591           code/2.8Kb data to 6.5Kb code/0.3Kb data on my GNU/Linux amd64
4592           system, while still giving you I/O watchers, timers and monotonic
4593           clock support.
4594
4595           With an intelligent-enough linker (gcc+binutils are intelligent
4596           enough when you use "-Wl,--gc-sections -ffunction-sections")
4597           functions unused by your program might be left out as well - a
4598           binary starting a timer and an I/O watcher then might come out at
4599           only 5Kb.
4600
4601       EV_API_STATIC
4602           If this symbol is defined (by default it is not), then all
4603           identifiers will have static linkage. This means that libev will
4604           not export any identifiers, and you cannot link against libev
4605           anymore. This can be useful when you embed libev, only want to use
4606           libev functions in a single file, and do not want its identifiers
4607           to be visible.
4608
4609           To use this, define "EV_API_STATIC" and include ev.c in the file
4610           that wants to use libev.
4611
4612           This option only works when libev is compiled with a C compiler, as
4613           C++ doesn't support the required declaration syntax.
4614
4615       EV_AVOID_STDIO
4616           If this is set to 1 at compiletime, then libev will avoid using
4617           stdio functions (printf, scanf, perror etc.). This will increase
4618           the code size somewhat, but if your program doesn't otherwise
4619           depend on stdio and your libc allows it, this avoids linking in the
4620           stdio library which is quite big.
4621
4622           Note that error messages might become less precise when this option
4623           is enabled.
4624
4625       EV_NSIG
4626           The highest supported signal number, +1 (or, the number of
4627           signals): Normally, libev tries to deduce the maximum number of
4628           signals automatically, but sometimes this fails, in which case it
4629           can be specified. Also, using a lower number than detected (32
4630           should be good for about any system in existence) can save some
4631           memory, as libev statically allocates some 12-24 bytes per signal
4632           number.
4633
4634       EV_PID_HASHSIZE
4635           "ev_child" watchers use a small hash table to distribute workload
4636           by pid. The default size is 16 (or 1 with "EV_FEATURES" disabled),
4637           usually more than enough. If you need to manage thousands of
4638           children you might want to increase this value (must be a power of
4639           two).
4640
4641       EV_INOTIFY_HASHSIZE
4642           "ev_stat" watchers use a small hash table to distribute workload by
4643           inotify watch id. The default size is 16 (or 1 with "EV_FEATURES"
4644           disabled), usually more than enough. If you need to manage
4645           thousands of "ev_stat" watchers you might want to increase this
4646           value (must be a power of two).
4647
4648       EV_USE_4HEAP
4649           Heaps are not very cache-efficient. To improve the cache-efficiency
4650           of the timer and periodics heaps, libev uses a 4-heap when this
4651           symbol is defined to 1. The 4-heap uses more complicated (longer)
4652           code but has noticeably faster performance with many (thousands) of
4653           watchers.
4654
4655           The default is 1, unless "EV_FEATURES" overrides it, in which case
4656           it will be 0.
4657
4658       EV_HEAP_CACHE_AT
4659           Heaps are not very cache-efficient. To improve the cache-efficiency
4660           of the timer and periodics heaps, libev can cache the timestamp
4661           (at) within the heap structure (selected by defining
4662           "EV_HEAP_CACHE_AT" to 1), which uses 8-12 bytes more per watcher
4663           and a few hundred bytes more code, but avoids random read accesses
4664           on heap changes. This improves performance noticeably with many
4665           (hundreds) of watchers.
4666
4667           The default is 1, unless "EV_FEATURES" overrides it, in which case
4668           it will be 0.
4669
4670       EV_VERIFY
4671           Controls how much internal verification (see "ev_verify ()") will
4672           be done: If set to 0, no internal verification code will be
4673           compiled in. If set to 1, then verification code will be compiled
4674           in, but not called. If set to 2, then the internal verification
4675           code will be called once per loop, which can slow down libev. If
4676           set to 3, then the verification code will be called very
4677           frequently, which will slow down libev considerably.
4678
4679           The default is 1, unless "EV_FEATURES" overrides it, in which case
4680           it will be 0.
4681
4682       EV_COMMON
4683           By default, all watchers have a "void *data" member. By redefining
4684           this macro to something else you can include more and other types
4685           of members. You have to define it each time you include one of the
4686           files, though, and it must be identical each time.
4687
4688           For example, the perl EV module uses something like this:
4689
4690              #define EV_COMMON                       \
4691                SV *self; /* contains this struct */  \
4692                SV *cb_sv, *fh /* note no trailing ";" */
4693
4694       EV_CB_DECLARE (type)
4695       EV_CB_INVOKE (watcher, revents)
4696       ev_set_cb (ev, cb)
4697           Can be used to change the callback member declaration in each
4698           watcher, and the way callbacks are invoked and set. Must expand to
4699           a struct member definition and a statement, respectively. See the
4700           ev.h header file for their default definitions. One possible use
4701           for overriding these is to avoid the "struct ev_loop *" as first
4702           argument in all cases, or to use method calls instead of plain
4703           function calls in C++.
4704
4705   EXPORTED API SYMBOLS
4706       If you need to re-export the API (e.g. via a DLL) and you need a list
4707       of exported symbols, you can use the provided Symbol.* files which list
4708       all public symbols, one per line:
4709
4710          Symbols.ev      for libev proper
4711          Symbols.event   for the libevent emulation
4712
4713       This can also be used to rename all public symbols to avoid clashes
4714       with multiple versions of libev linked together (which is obviously bad
4715       in itself, but sometimes it is inconvenient to avoid this).
4716
4717       A sed command like this will create wrapper "#define"'s that you need
4718       to include before including ev.h:
4719
4720          <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
4721
4722       This would create a file wrap.h which essentially looks like this:
4723
4724          #define ev_backend     myprefix_ev_backend
4725          #define ev_check_start myprefix_ev_check_start
4726          #define ev_check_stop  myprefix_ev_check_stop
4727          ...
4728
4729   EXAMPLES
4730       For a real-world example of a program the includes libev verbatim, you
4731       can have a look at the EV perl module
4732       (<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
4733       the libev/ subdirectory and includes them in the EV/EVAPI.h (public
4734       interface) and EV.xs (implementation) files. Only the EV.xs file will
4735       be compiled. It is pretty complex because it provides its own header
4736       file.
4737
4738       The usage in rxvt-unicode is simpler. It has a ev_cpp.h header file
4739       that everybody includes and which overrides some configure choices:
4740
4741          #define EV_FEATURES 8
4742          #define EV_USE_SELECT 1
4743          #define EV_PREPARE_ENABLE 1
4744          #define EV_IDLE_ENABLE 1
4745          #define EV_SIGNAL_ENABLE 1
4746          #define EV_CHILD_ENABLE 1
4747          #define EV_USE_STDEXCEPT 0
4748          #define EV_CONFIG_H <config.h>
4749
4750          #include "ev++.h"
4751
4752       And a ev_cpp.C implementation file that contains libev proper and is
4753       compiled:
4754
4755          #include "ev_cpp.h"
4756          #include "ev.c"
4757

INTERACTION WITH OTHER PROGRAMS, LIBRARIES OR THE ENVIRONMENT

4759   THREADS AND COROUTINES
4760       THREADS
4761
4762       All libev functions are reentrant and thread-safe unless explicitly
4763       documented otherwise, but libev implements no locking itself. This
4764       means that you can use as many loops as you want in parallel, as long
4765       as there are no concurrent calls into any libev function with the same
4766       loop parameter ("ev_default_*" calls have an implicit default loop
4767       parameter, of course): libev guarantees that different event loops
4768       share no data structures that need any locking.
4769
4770       Or to put it differently: calls with different loop parameters can be
4771       done concurrently from multiple threads, calls with the same loop
4772       parameter must be done serially (but can be done from different
4773       threads, as long as only one thread ever is inside a call at any point
4774       in time, e.g. by using a mutex per loop).
4775
4776       Specifically to support threads (and signal handlers), libev implements
4777       so-called "ev_async" watchers, which allow some limited form of
4778       concurrency on the same event loop, namely waking it up "from the
4779       outside".
4780
4781       If you want to know which design (one loop, locking, or multiple loops
4782       without or something else still) is best for your problem, then I
4783       cannot help you, but here is some generic advice:
4784
4785       ·   most applications have a main thread: use the default libev loop in
4786           that thread, or create a separate thread running only the default
4787           loop.
4788
4789           This helps integrating other libraries or software modules that use
4790           libev themselves and don't care/know about threading.
4791
4792       ·   one loop per thread is usually a good model.
4793
4794           Doing this is almost never wrong, sometimes a better-performance
4795           model exists, but it is always a good start.
4796
4797       ·   other models exist, such as the leader/follower pattern, where one
4798           loop is handed through multiple threads in a kind of round-robin
4799           fashion.
4800
4801           Choosing a model is hard - look around, learn, know that usually
4802           you can do better than you currently do :-)
4803
4804       ·   often you need to talk to some other thread which blocks in the
4805           event loop.
4806
4807           "ev_async" watchers can be used to wake them up from other threads
4808           safely (or from signal contexts...).
4809
4810           An example use would be to communicate signals or other events that
4811           only work in the default loop by registering the signal watcher
4812           with the default loop and triggering an "ev_async" watcher from the
4813           default loop watcher callback into the event loop interested in the
4814           signal.
4815
4816       See also "THREAD LOCKING EXAMPLE".
4817
4818       COROUTINES
4819
4820       Libev is very accommodating to coroutines ("cooperative threads"):
4821       libev fully supports nesting calls to its functions from different
4822       coroutines (e.g. you can call "ev_run" on the same loop from two
4823       different coroutines, and switch freely between both coroutines running
4824       the loop, as long as you don't confuse yourself). The only exception is
4825       that you must not do this from "ev_periodic" reschedule callbacks.
4826
4827       Care has been taken to ensure that libev does not keep local state
4828       inside "ev_run", and other calls do not usually allow for coroutine
4829       switches as they do not call any callbacks.
4830
4831   COMPILER WARNINGS
4832       Depending on your compiler and compiler settings, you might get no or a
4833       lot of warnings when compiling libev code. Some people are apparently
4834       scared by this.
4835
4836       However, these are unavoidable for many reasons. For one, each compiler
4837       has different warnings, and each user has different tastes regarding
4838       warning options. "Warn-free" code therefore cannot be a goal except
4839       when targeting a specific compiler and compiler-version.
4840
4841       Another reason is that some compiler warnings require elaborate
4842       workarounds, or other changes to the code that make it less clear and
4843       less maintainable.
4844
4845       And of course, some compiler warnings are just plain stupid, or simply
4846       wrong (because they don't actually warn about the condition their
4847       message seems to warn about). For example, certain older gcc versions
4848       had some warnings that resulted in an extreme number of false
4849       positives. These have been fixed, but some people still insist on
4850       making code warn-free with such buggy versions.
4851
4852       While libev is written to generate as few warnings as possible, "warn-
4853       free" code is not a goal, and it is recommended not to build libev with
4854       any compiler warnings enabled unless you are prepared to cope with them
4855       (e.g. by ignoring them). Remember that warnings are just that:
4856       warnings, not errors, or proof of bugs.
4857
4858   VALGRIND
4859       Valgrind has a special section here because it is a popular tool that
4860       is highly useful. Unfortunately, valgrind reports are very hard to
4861       interpret.
4862
4863       If you think you found a bug (memory leak, uninitialised data access
4864       etc.)  in libev, then check twice: If valgrind reports something like:
4865
4866          ==2274==    definitely lost: 0 bytes in 0 blocks.
4867          ==2274==      possibly lost: 0 bytes in 0 blocks.
4868          ==2274==    still reachable: 256 bytes in 1 blocks.
4869
4870       Then there is no memory leak, just as memory accounted to global
4871       variables is not a memleak - the memory is still being referenced, and
4872       didn't leak.
4873
4874       Similarly, under some circumstances, valgrind might report kernel bugs
4875       as if it were a bug in libev (e.g. in realloc or in the poll backend,
4876       although an acceptable workaround has been found here), or it might be
4877       confused.
4878
4879       Keep in mind that valgrind is a very good tool, but only a tool. Don't
4880       make it into some kind of religion.
4881
4882       If you are unsure about something, feel free to contact the mailing
4883       list with the full valgrind report and an explanation on why you think
4884       this is a bug in libev (best check the archives, too :). However, don't
4885       be annoyed when you get a brisk "this is no bug" answer and take the
4886       chance of learning how to interpret valgrind properly.
4887
4888       If you need, for some reason, empty reports from valgrind for your
4889       project I suggest using suppression lists.
4890

PORTABILITY NOTES

4892   GNU/LINUX 32 BIT LIMITATIONS
4893       GNU/Linux is the only common platform that supports 64 bit file/large
4894       file interfaces but disables them by default.
4895
4896       That means that libev compiled in the default environment doesn't
4897       support files larger than 2GiB or so, which mainly affects "ev_stat"
4898       watchers.
4899
4900       Unfortunately, many programs try to work around this GNU/Linux issue by
4901       enabling the large file API, which makes them incompatible with the
4902       standard libev compiled for their system.
4903
4904       Likewise, libev cannot enable the large file API itself as this would
4905       suddenly make it incompatible to the default compile time environment,
4906       i.e. all programs not using special compile switches.
4907
4908   OS/X AND DARWIN BUGS
4909       The whole thing is a bug if you ask me - basically any system interface
4910       you touch is broken, whether it is locales, poll, kqueue or even the
4911       OpenGL drivers.
4912
4913       "kqueue" is buggy
4914
4915       The kqueue syscall is broken in all known versions - most versions
4916       support only sockets, many support pipes.
4917
4918       Libev tries to work around this by not using "kqueue" by default on
4919       this rotten platform, but of course you can still ask for it when
4920       creating a loop - embedding a socket-only kqueue loop into a select-
4921       based one is probably going to work well.
4922
4923       "poll" is buggy
4924
4925       Instead of fixing "kqueue", Apple replaced their (working) "poll"
4926       implementation by something calling "kqueue" internally around the
4927       10.5.6 release, so now "kqueue" and "poll" are broken.
4928
4929       Libev tries to work around this by not using "poll" by default on this
4930       rotten platform, but of course you can still ask for it when creating a
4931       loop.
4932
4933       "select" is buggy
4934
4935       All that's left is "select", and of course Apple found a way to fuck
4936       this one up as well: On OS/X, "select" actively limits the number of
4937       file descriptors you can pass in to 1024 - your program suddenly
4938       crashes when you use more.
4939
4940       There is an undocumented "workaround" for this - defining
4941       "_DARWIN_UNLIMITED_SELECT", which libev tries to use, so select should
4942       work on OS/X.
4943
4944   SOLARIS PROBLEMS AND WORKAROUNDS
4945       "errno" reentrancy
4946
4947       The default compile environment on Solaris is unfortunately so thread-
4948       unsafe that you can't even use components/libraries compiled without
4949       "-D_REENTRANT" in a threaded program, which, of course, isn't defined
4950       by default. A valid, if stupid, implementation choice.
4951
4952       If you want to use libev in threaded environments you have to make sure
4953       it's compiled with "_REENTRANT" defined.
4954
4955       Event port backend
4956
4957       The scalable event interface for Solaris is called "event ports".
4958       Unfortunately, this mechanism is very buggy in all major releases. If
4959       you run into high CPU usage, your program freezes or you get a large
4960       number of spurious wakeups, make sure you have all the relevant and
4961       latest kernel patches applied. No, I don't know which ones, but there
4962       are multiple ones to apply, and afterwards, event ports actually work
4963       great.
4964
4965       If you can't get it to work, you can try running the program by setting
4966       the environment variable "LIBEV_FLAGS=3" to only allow "poll" and
4967       "select" backends.
4968
4969   AIX POLL BUG
4970       AIX unfortunately has a broken "poll.h" header. Libev works around this
4971       by trying to avoid the poll backend altogether (i.e. it's not even
4972       compiled in), which normally isn't a big problem as "select" works fine
4973       with large bitsets on AIX, and AIX is dead anyway.
4974
4975   WIN32 PLATFORM LIMITATIONS AND WORKAROUNDS
4976       General issues
4977
4978       Win32 doesn't support any of the standards (e.g. POSIX) that libev
4979       requires, and its I/O model is fundamentally incompatible with the
4980       POSIX model. Libev still offers limited functionality on this platform
4981       in the form of the "EVBACKEND_SELECT" backend, and only supports socket
4982       descriptors. This only applies when using Win32 natively, not when
4983       using e.g. cygwin. Actually, it only applies to the microsofts own
4984       compilers, as every compiler comes with a slightly differently
4985       broken/incompatible environment.
4986
4987       Lifting these limitations would basically require the full re-
4988       implementation of the I/O system. If you are into this kind of thing,
4989       then note that glib does exactly that for you in a very portable way
4990       (note also that glib is the slowest event library known to man).
4991
4992       There is no supported compilation method available on windows except
4993       embedding it into other applications.
4994
4995       Sensible signal handling is officially unsupported by Microsoft - libev
4996       tries its best, but under most conditions, signals will simply not
4997       work.
4998
4999       Not a libev limitation but worth mentioning: windows apparently doesn't
5000       accept large writes: instead of resulting in a partial write, windows
5001       will either accept everything or return "ENOBUFS" if the buffer is too
5002       large, so make sure you only write small amounts into your sockets
5003       (less than a megabyte seems safe, but this apparently depends on the
5004       amount of memory available).
5005
5006       Due to the many, low, and arbitrary limits on the win32 platform and
5007       the abysmal performance of winsockets, using a large number of sockets
5008       is not recommended (and not reasonable). If your program needs to use
5009       more than a hundred or so sockets, then likely it needs to use a
5010       totally different implementation for windows, as libev offers the POSIX
5011       readiness notification model, which cannot be implemented efficiently
5012       on windows (due to Microsoft monopoly games).
5013
5014       A typical way to use libev under windows is to embed it (see the
5015       embedding section for details) and use the following evwrap.h header
5016       file instead of ev.h:
5017
5018          #define EV_STANDALONE              /* keeps ev from requiring config.h */
5019          #define EV_SELECT_IS_WINSOCKET 1   /* configure libev for windows select */
5020
5021          #include "ev.h"
5022
5023       And compile the following evwrap.c file into your project (make sure
5024       you do not compile the ev.c or any other embedded source files!):
5025
5026          #include "evwrap.h"
5027          #include "ev.c"
5028
5029       The winsocket "select" function
5030
5031       The winsocket "select" function doesn't follow POSIX in that it
5032       requires socket handles and not socket file descriptors (it is also
5033       extremely buggy). This makes select very inefficient, and also requires
5034       a mapping from file descriptors to socket handles (the Microsoft C
5035       runtime provides the function "_open_osfhandle" for this). See the
5036       discussion of the "EV_SELECT_USE_FD_SET", "EV_SELECT_IS_WINSOCKET" and
5037       "EV_FD_TO_WIN32_HANDLE" preprocessor symbols for more info.
5038
5039       The configuration for a "naked" win32 using the Microsoft runtime
5040       libraries and raw winsocket select is:
5041
5042          #define EV_USE_SELECT 1
5043          #define EV_SELECT_IS_WINSOCKET 1   /* forces EV_SELECT_USE_FD_SET, too */
5044
5045       Note that winsockets handling of fd sets is O(n), so you can easily get
5046       a complexity in the O(nX) range when using win32.
5047
5048       Limited number of file descriptors
5049
5050       Windows has numerous arbitrary (and low) limits on things.
5051
5052       Early versions of winsocket's select only supported waiting for a
5053       maximum of 64 handles (probably owning to the fact that all windows
5054       kernels can only wait for 64 things at the same time internally;
5055       Microsoft recommends spawning a chain of threads and wait for 63
5056       handles and the previous thread in each. Sounds great!).
5057
5058       Newer versions support more handles, but you need to define
5059       "FD_SETSIZE" to some high number (e.g. 2048) before compiling the
5060       winsocket select call (which might be in libev or elsewhere, for
5061       example, perl and many other interpreters do their own select emulation
5062       on windows).
5063
5064       Another limit is the number of file descriptors in the Microsoft
5065       runtime libraries, which by default is 64 (there must be a hidden 64
5066       fetish or something like this inside Microsoft). You can increase this
5067       by calling "_setmaxstdio", which can increase this limit to 2048
5068       (another arbitrary limit), but is broken in many versions of the
5069       Microsoft runtime libraries. This might get you to about 512 or 2048
5070       sockets (depending on windows version and/or the phase of the moon). To
5071       get more, you need to wrap all I/O functions and provide your own fd
5072       management, but the cost of calling select (O(nX)) will likely make
5073       this unworkable.
5074
5075   PORTABILITY REQUIREMENTS
5076       In addition to a working ISO-C implementation and of course the
5077       backend-specific APIs, libev relies on a few additional extensions:
5078
5079       "void (*)(ev_watcher_type *, int revents)" must have compatible calling
5080       conventions regardless of "ev_watcher_type *".
5081           Libev assumes not only that all watcher pointers have the same
5082           internal structure (guaranteed by POSIX but not by ISO C for
5083           example), but it also assumes that the same (machine) code can be
5084           used to call any watcher callback: The watcher callbacks have
5085           different type signatures, but libev calls them using an
5086           "ev_watcher *" internally.
5087
5088       null pointers and integer zero are represented by 0 bytes
5089           Libev uses "memset" to initialise structs and arrays to 0 bytes,
5090           and relies on this setting pointers and integers to null.
5091
5092       pointer accesses must be thread-atomic
5093           Accessing a pointer value must be atomic, it must both be readable
5094           and writable in one piece - this is the case on all current
5095           architectures.
5096
5097       "sig_atomic_t volatile" must be thread-atomic as well
5098           The type "sig_atomic_t volatile" (or whatever is defined as
5099           "EV_ATOMIC_T") must be atomic with respect to accesses from
5100           different threads. This is not part of the specification for
5101           "sig_atomic_t", but is believed to be sufficiently portable.
5102
5103       "sigprocmask" must work in a threaded environment
5104           Libev uses "sigprocmask" to temporarily block signals. This is not
5105           allowed in a threaded program ("pthread_sigmask" has to be used).
5106           Typical pthread implementations will either allow "sigprocmask" in
5107           the "main thread" or will block signals process-wide, both
5108           behaviours would be compatible with libev. Interaction between
5109           "sigprocmask" and "pthread_sigmask" could complicate things,
5110           however.
5111
5112           The most portable way to handle signals is to block signals in all
5113           threads except the initial one, and run the signal handling loop in
5114           the initial thread as well.
5115
5116       "long" must be large enough for common memory allocation sizes
5117           To improve portability and simplify its API, libev uses "long"
5118           internally instead of "size_t" when allocating its data structures.
5119           On non-POSIX systems (Microsoft...) this might be unexpectedly low,
5120           but is still at least 31 bits everywhere, which is enough for
5121           hundreds of millions of watchers.
5122
5123       "double" must hold a time value in seconds with enough accuracy
5124           The type "double" is used to represent timestamps. It is required
5125           to have at least 51 bits of mantissa (and 9 bits of exponent),
5126           which is good enough for at least into the year 4000 with
5127           millisecond accuracy (the design goal for libev). This requirement
5128           is overfulfilled by implementations using IEEE 754, which is
5129           basically all existing ones.
5130
5131           With IEEE 754 doubles, you get microsecond accuracy until at least
5132           the year 2255 (and millisecond accuracy till the year 287396 - by
5133           then, libev is either obsolete or somebody patched it to use "long
5134           double" or something like that, just kidding).
5135
5136       If you know of other additional requirements drop me a note.
5137

ALGORITHMIC COMPLEXITIES

5139       In this section the complexities of (many of) the algorithms used
5140       inside libev will be documented. For complexity discussions about
5141       backends see the documentation for "ev_default_init".
5142
5143       All of the following are about amortised time: If an array needs to be
5144       extended, libev needs to realloc and move the whole array, but this
5145       happens asymptotically rarer with higher number of elements, so O(1)
5146       might mean that libev does a lengthy realloc operation in rare cases,
5147       but on average it is much faster and asymptotically approaches constant
5148       time.
5149
5150       Starting and stopping timer/periodic watchers: O(log
5151       skipped_other_timers)
5152           This means that, when you have a watcher that triggers in one hour
5153           and there are 100 watchers that would trigger before that, then
5154           inserting will have to skip roughly seven ("ld 100") of these
5155           watchers.
5156
5157       Changing timer/periodic watchers (by autorepeat or calling again):
5158       O(log skipped_other_timers)
5159           That means that changing a timer costs less than removing/adding
5160           them, as only the relative motion in the event queue has to be paid
5161           for.
5162
5163       Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
5164           These just add the watcher into an array or at the head of a list.
5165
5166       Stopping check/prepare/idle/fork/async watchers: O(1)
5167       Stopping an io/signal/child watcher:
5168       O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
5169           These watchers are stored in lists, so they need to be walked to
5170           find the correct watcher to remove. The lists are usually short
5171           (you don't usually have many watchers waiting for the same fd or
5172           signal: one is typical, two is rare).
5173
5174       Finding the next timer in each loop iteration: O(1)
5175           By virtue of using a binary or 4-heap, the next timer is always
5176           found at a fixed position in the storage array.
5177
5178       Each change on a file descriptor per loop iteration:
5179       O(number_of_watchers_for_this_fd)
5180           A change means an I/O watcher gets started or stopped, which
5181           requires libev to recalculate its status (and possibly tell the
5182           kernel, depending on backend and whether "ev_io_set" was used).
5183
5184       Activating one watcher (putting it into the pending state): O(1)
5185       Priority handling: O(number_of_priorities)
5186           Priorities are implemented by allocating some space for each
5187           priority. When doing priority-based operations, libev usually has
5188           to linearly search all the priorities, but starting/stopping and
5189           activating watchers becomes O(1) with respect to priority handling.
5190
5191       Sending an ev_async: O(1)
5192       Processing ev_async_send: O(number_of_async_watchers)
5193       Processing signals: O(max_signal_number)
5194           Sending involves a system call iff there were no other
5195           "ev_async_send" calls in the current loop iteration and the loop is
5196           currently blocked. Checking for async and signal events involves
5197           iterating over all running async watchers or all signal numbers.
5198

PORTING FROM LIBEV 3.X TO 4.X

5200       The major version 4 introduced some incompatible changes to the API.
5201
5202       At the moment, the "ev.h" header file provides compatibility
5203       definitions for all changes, so most programs should still compile. The
5204       compatibility layer might be removed in later versions of libev, so
5205       better update to the new API early than late.
5206
5207       "EV_COMPAT3" backwards compatibility mechanism
5208           The backward compatibility mechanism can be controlled by
5209           "EV_COMPAT3". See "PREPROCESSOR SYMBOLS/MACROS" in the "EMBEDDING"
5210           section.
5211
5212       "ev_default_destroy" and "ev_default_fork" have been removed
5213           These calls can be replaced easily by their "ev_loop_xxx"
5214           counterparts:
5215
5216              ev_loop_destroy (EV_DEFAULT_UC);
5217              ev_loop_fork (EV_DEFAULT);
5218
5219       function/symbol renames
5220           A number of functions and symbols have been renamed:
5221
5222             ev_loop         => ev_run
5223             EVLOOP_NONBLOCK => EVRUN_NOWAIT
5224             EVLOOP_ONESHOT  => EVRUN_ONCE
5225
5226             ev_unloop       => ev_break
5227             EVUNLOOP_CANCEL => EVBREAK_CANCEL
5228             EVUNLOOP_ONE    => EVBREAK_ONE
5229             EVUNLOOP_ALL    => EVBREAK_ALL
5230
5231             EV_TIMEOUT      => EV_TIMER
5232
5233             ev_loop_count   => ev_iteration
5234             ev_loop_depth   => ev_depth
5235             ev_loop_verify  => ev_verify
5236
5237           Most functions working on "struct ev_loop" objects don't have an
5238           "ev_loop_" prefix, so it was removed; "ev_loop", "ev_unloop" and
5239           associated constants have been renamed to not collide with the
5240           "struct ev_loop" anymore and "EV_TIMER" now follows the same naming
5241           scheme as all other watcher types. Note that "ev_loop_fork" is
5242           still called "ev_loop_fork" because it would otherwise clash with
5243           the "ev_fork" typedef.
5244
5245       "EV_MINIMAL" mechanism replaced by "EV_FEATURES"
5246           The preprocessor symbol "EV_MINIMAL" has been replaced by a
5247           different mechanism, "EV_FEATURES". Programs using "EV_MINIMAL"
5248           usually compile and work, but the library code will of course be
5249           larger.
5250

GLOSSARY

5252       active
5253           A watcher is active as long as it has been started and not yet
5254           stopped.  See "WATCHER STATES" for details.
5255
5256       application
5257           In this document, an application is whatever is using libev.
5258
5259       backend
5260           The part of the code dealing with the operating system interfaces.
5261
5262       callback
5263           The address of a function that is called when some event has been
5264           detected. Callbacks are being passed the event loop, the watcher
5265           that received the event, and the actual event bitset.
5266
5267       callback/watcher invocation
5268           The act of calling the callback associated with a watcher.
5269
5270       event
5271           A change of state of some external event, such as data now being
5272           available for reading on a file descriptor, time having passed or
5273           simply not having any other events happening anymore.
5274
5275           In libev, events are represented as single bits (such as "EV_READ"
5276           or "EV_TIMER").
5277
5278       event library
5279           A software package implementing an event model and loop.
5280
5281       event loop
5282           An entity that handles and processes external events and converts
5283           them into callback invocations.
5284
5285       event model
5286           The model used to describe how an event loop handles and processes
5287           watchers and events.
5288
5289       pending
5290           A watcher is pending as soon as the corresponding event has been
5291           detected. See "WATCHER STATES" for details.
5292
5293       real time
5294           The physical time that is observed. It is apparently strictly
5295           monotonic :)
5296
5297       wall-clock time
5298           The time and date as shown on clocks. Unlike real time, it can
5299           actually be wrong and jump forwards and backwards, e.g. when you
5300           adjust your clock.
5301
5302       watcher
5303           A data structure that describes interest in certain events.
5304           Watchers need to be started (attached to an event loop) before they
5305           can receive events.
5306

AUTHOR

5308       Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
5309       Magnusson and Emanuele Giaquinta, and minor corrections by many others.
5310
5311
5312
5313libev-4.25                        2018-12-21                          LIBEV(3)
Impressum