Mirror of :pserver:anonymous@cvs.schmorp.de/schmorpforge libev http://software.schmorp.de/pkg/libev.html
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

5742 lines
220 KiB

  1. =encoding utf-8
  2. =head1 NAME
  3. libev - a high performance full-featured event loop written in C
  4. =head1 SYNOPSIS
  5. #include <ev.h>
  7. // a single header file is required
  8. #include <ev.h>
  9. #include <stdio.h> // for puts
  10. // every watcher type has its own typedef'd struct
  11. // with the name ev_TYPE
  12. ev_io stdin_watcher;
  13. ev_timer timeout_watcher;
  14. // all watcher callbacks have a similar signature
  15. // this callback is called when data is readable on stdin
  16. static void
  17. stdin_cb (EV_P_ ev_io *w, int revents)
  18. {
  19. puts ("stdin ready");
  20. // for one-shot events, one must manually stop the watcher
  21. // with its corresponding stop function.
  22. ev_io_stop (EV_A_ w);
  23. // this causes all nested ev_run's to stop iterating
  24. ev_break (EV_A_ EVBREAK_ALL);
  25. }
  26. // another callback, this time for a time-out
  27. static void
  28. timeout_cb (EV_P_ ev_timer *w, int revents)
  29. {
  30. puts ("timeout");
  31. // this causes the innermost ev_run to stop iterating
  32. ev_break (EV_A_ EVBREAK_ONE);
  33. }
  34. int
  35. main (void)
  36. {
  37. // use the default event loop unless you have special needs
  38. struct ev_loop *loop = EV_DEFAULT;
  39. // initialise an io watcher, then start it
  40. // this one will watch for stdin to become readable
  41. ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
  42. ev_io_start (loop, &stdin_watcher);
  43. // initialise a timer watcher, then start it
  44. // simple non-repeating 5.5 second timeout
  45. ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
  46. ev_timer_start (loop, &timeout_watcher);
  47. // now wait for events to arrive
  48. ev_run (loop, 0);
  49. // break was called, so exit
  50. return 0;
  51. }
  53. This document documents the libev software package.
  54. The newest version of this document is also available as an html-formatted
  55. web page you might find easier to navigate when reading it for the first
  56. time: L<http://pod.tst.eu/http://cvs.schmorp.de/libev/ev.pod>.
  57. While this document tries to be as complete as possible in documenting
  58. libev, its usage and the rationale behind its design, it is not a tutorial
  59. on event-based programming, nor will it introduce event-based programming
  60. with libev.
  61. Familiarity with event based programming techniques in general is assumed
  62. throughout this document.
  64. This manual tries to be very detailed, but unfortunately, this also makes
  65. it very long. If you just want to know the basics of libev, I suggest
  66. reading L</ANATOMY OF A WATCHER>, then the L</EXAMPLE PROGRAM> above and
  67. look up the missing functions in L</GLOBAL FUNCTIONS> and the C<ev_io> and
  68. C<ev_timer> sections in L</WATCHER TYPES>.
  69. =head1 ABOUT LIBEV
  70. Libev is an event loop: you register interest in certain events (such as a
  71. file descriptor being readable or a timeout occurring), and it will manage
  72. these event sources and provide your program with events.
  73. To do this, it must take more or less complete control over your process
  74. (or thread) by executing the I<event loop> handler, and will then
  75. communicate events via a callback mechanism.
  76. You register interest in certain events by registering so-called I<event
  77. watchers>, which are relatively small C structures you initialise with the
  78. details of the event, and then hand it over to libev by I<starting> the
  79. watcher.
  80. =head2 FEATURES
  81. Libev supports C<select>, C<poll>, the Linux-specific aio and C<epoll>
  82. interfaces, the BSD-specific C<kqueue> and the Solaris-specific event port
  83. mechanisms for file descriptor events (C<ev_io>), the Linux C<inotify>
  84. interface (for C<ev_stat>), Linux eventfd/signalfd (for faster and cleaner
  85. inter-thread wakeup (C<ev_async>)/signal handling (C<ev_signal>)) relative
  86. timers (C<ev_timer>), absolute timers with customised rescheduling
  87. (C<ev_periodic>), synchronous signals (C<ev_signal>), process status
  88. change events (C<ev_child>), and event watchers dealing with the event
  89. loop mechanism itself (C<ev_idle>, C<ev_embed>, C<ev_prepare> and
  90. C<ev_check> watchers) as well as file watchers (C<ev_stat>) and even
  91. limited support for fork events (C<ev_fork>).
  92. It also is quite fast (see this
  93. L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
  94. for example).
  95. =head2 CONVENTIONS
  96. Libev is very configurable. In this manual the default (and most common)
  97. configuration will be described, which supports multiple event loops. For
  98. more info about various configuration options please have a look at
  99. B<EMBED> section in this manual. If libev was configured without support
  100. for multiple event loops, then all functions taking an initial argument of
  101. name C<loop> (which is always of type C<struct ev_loop *>) will not have
  102. this argument.
  104. Libev represents time as a single floating point number, representing
  105. the (fractional) number of seconds since the (POSIX) epoch (in practice
  106. somewhere near the beginning of 1970, details are complicated, don't
  107. ask). This type is called C<ev_tstamp>, which is what you should use
  108. too. It usually aliases to the C<double> type in C. When you need to do
  109. any calculations on it, you should treat it as some floating point value.
  110. Unlike the name component C<stamp> might indicate, it is also used for
  111. time differences (e.g. delays) throughout libev.
  112. =head1 ERROR HANDLING
  113. Libev knows three classes of errors: operating system errors, usage errors
  114. and internal errors (bugs).
  115. When libev catches an operating system error it cannot handle (for example
  116. a system call indicating a condition libev cannot fix), it calls the callback
  117. set via C<ev_set_syserr_cb>, which is supposed to fix the problem or
  118. abort. The default is to print a diagnostic message and to call C<abort
  119. ()>.
  120. When libev detects a usage error such as a negative timer interval, then
  121. it will print a diagnostic message and abort (via the C<assert> mechanism,
  122. so C<NDEBUG> will disable this checking): these are programming errors in
  123. the libev caller and need to be fixed there.
  124. Via the C<EV_FREQUENT> macro you can compile in and/or enable extensive
  125. consistency checking code inside libev that can be used to check for
  126. internal inconsistencies, suually caused by application bugs.
  127. Libev also has a few internal error-checking C<assert>ions. These do not
  128. trigger under normal circumstances, as they indicate either a bug in libev
  129. or worse.
  130. =head1 GLOBAL FUNCTIONS
  131. These functions can be called anytime, even before initialising the
  132. library in any way.
  133. =over 4
  134. =item ev_tstamp ev_time ()
  135. Returns the current time as libev would use it. Please note that the
  136. C<ev_now> function is usually faster and also often returns the timestamp
  137. you actually want to know. Also interesting is the combination of
  138. C<ev_now_update> and C<ev_now>.
  139. =item ev_sleep (ev_tstamp interval)
  140. Sleep for the given interval: The current thread will be blocked
  141. until either it is interrupted or the given time interval has
  142. passed (approximately - it might return a bit earlier even if not
  143. interrupted). Returns immediately if C<< interval <= 0 >>.
  144. Basically this is a sub-second-resolution C<sleep ()>.
  145. The range of the C<interval> is limited - libev only guarantees to work
  146. with sleep times of up to one day (C<< interval <= 86400 >>).
  147. =item int ev_version_major ()
  148. =item int ev_version_minor ()
  149. You can find out the major and minor ABI version numbers of the library
  150. you linked against by calling the functions C<ev_version_major> and
  151. C<ev_version_minor>. If you want, you can compare against the global
  152. symbols C<EV_VERSION_MAJOR> and C<EV_VERSION_MINOR>, which specify the
  153. version of the library your program was compiled against.
  154. These version numbers refer to the ABI version of the library, not the
  155. release version.
  156. Usually, it's a good idea to terminate if the major versions mismatch,
  157. as this indicates an incompatible change. Minor versions are usually
  158. compatible to older versions, so a larger minor version alone is usually
  159. not a problem.
  160. Example: Make sure we haven't accidentally been linked against the wrong
  161. version (note, however, that this will not detect other ABI mismatches,
  162. such as LFS or reentrancy).
  163. assert (("libev version mismatch",
  164. ev_version_major () == EV_VERSION_MAJOR
  165. && ev_version_minor () >= EV_VERSION_MINOR));
  166. =item unsigned int ev_supported_backends ()
  167. Return the set of all backends (i.e. their corresponding C<EV_BACKEND_*>
  168. value) compiled into this binary of libev (independent of their
  169. availability on the system you are running on). See C<ev_default_loop> for
  170. a description of the set values.
  171. Example: make sure we have the epoll method, because yeah this is cool and
  172. a must have and can we have a torrent of it please!!!11
  173. assert (("sorry, no epoll, no sex",
  174. ev_supported_backends () & EVBACKEND_EPOLL));
  175. =item unsigned int ev_recommended_backends ()
  176. Return the set of all backends compiled into this binary of libev and
  177. also recommended for this platform, meaning it will work for most file
  178. descriptor types. This set is often smaller than the one returned by
  179. C<ev_supported_backends>, as for example kqueue is broken on most BSDs
  180. and will not be auto-detected unless you explicitly request it (assuming
  181. you know what you are doing). This is the set of backends that libev will
  182. probe for if you specify no backends explicitly.
  183. =item unsigned int ev_embeddable_backends ()
  184. Returns the set of backends that are embeddable in other event loops. This
  185. value is platform-specific but can include backends not available on the
  186. current system. To find which embeddable backends might be supported on
  187. the current system, you would need to look at C<ev_embeddable_backends ()
  188. & ev_supported_backends ()>, likewise for recommended ones.
  189. See the description of C<ev_embed> watchers for more info.
  190. =item ev_set_allocator (void *(*cb)(void *ptr, long size) throw ())
  191. Sets the allocation function to use (the prototype is similar - the
  192. semantics are identical to the C<realloc> C89/SuS/POSIX function). It is
  193. used to allocate and free memory (no surprises here). If it returns zero
  194. when memory needs to be allocated (C<size != 0>), the library might abort
  195. or take some potentially destructive action.
  196. Since some systems (at least OpenBSD and Darwin) fail to implement
  197. correct C<realloc> semantics, libev will use a wrapper around the system
  198. C<realloc> and C<free> functions by default.
  199. You could override this function in high-availability programs to, say,
  200. free some memory if it cannot allocate memory, to use a special allocator,
  201. or even to sleep a while and retry until some memory is available.
  202. Example: The following is the C<realloc> function that libev itself uses
  203. which should work with C<realloc> and C<free> functions of all kinds and
  204. is probably a good basis for your own implementation.
  205. static void *
  206. ev_realloc_emul (void *ptr, long size) EV_NOEXCEPT
  207. {
  208. if (size)
  209. return realloc (ptr, size);
  210. free (ptr);
  211. return 0;
  212. }
  213. Example: Replace the libev allocator with one that waits a bit and then
  214. retries.
  215. static void *
  216. persistent_realloc (void *ptr, size_t size)
  217. {
  218. if (!size)
  219. {
  220. free (ptr);
  221. return 0;
  222. }
  223. for (;;)
  224. {
  225. void *newptr = realloc (ptr, size);
  226. if (newptr)
  227. return newptr;
  228. sleep (60);
  229. }
  230. }
  231. ...
  232. ev_set_allocator (persistent_realloc);
  233. =item ev_set_syserr_cb (void (*cb)(const char *msg) throw ())
  234. Set the callback function to call on a retryable system call error (such
  235. as failed select, poll, epoll_wait). The message is a printable string
  236. indicating the system call or subsystem causing the problem. If this
  237. callback is set, then libev will expect it to remedy the situation, no
  238. matter what, when it returns. That is, libev will generally retry the
  239. requested operation, or, if the condition doesn't go away, do bad stuff
  240. (such as abort).
  241. Example: This is basically the same thing that libev does internally, too.
  242. static void
  243. fatal_error (const char *msg)
  244. {
  245. perror (msg);
  246. abort ();
  247. }
  248. ...
  249. ev_set_syserr_cb (fatal_error);
  250. =item ev_feed_signal (int signum)
  251. This function can be used to "simulate" a signal receive. It is completely
  252. safe to call this function at any time, from any context, including signal
  253. handlers or random threads.
  254. Its main use is to customise signal handling in your process, especially
  255. in the presence of threads. For example, you could block signals
  256. by default in all threads (and specifying C<EVFLAG_NOSIGMASK> when
  257. creating any loops), and in one thread, use C<sigwait> or any other
  258. mechanism to wait for signals, then "deliver" them to libev by calling
  259. C<ev_feed_signal>.
  260. =back
  262. An event loop is described by a C<struct ev_loop *> (the C<struct> is
  263. I<not> optional in this case unless libev 3 compatibility is disabled, as
  264. libev 3 had an C<ev_loop> function colliding with the struct name).
  265. The library knows two types of such loops, the I<default> loop, which
  266. supports child process events, and dynamically created event loops which
  267. do not.
  268. =over 4
  269. =item struct ev_loop *ev_default_loop (unsigned int flags)
  270. This returns the "default" event loop object, which is what you should
  271. normally use when you just need "the event loop". Event loop objects and
  272. the C<flags> parameter are described in more detail in the entry for
  273. C<ev_loop_new>.
  274. If the default loop is already initialised then this function simply
  275. returns it (and ignores the flags. If that is troubling you, check
  276. C<ev_backend ()> afterwards). Otherwise it will create it with the given
  277. flags, which should almost always be C<0>, unless the caller is also the
  278. one calling C<ev_run> or otherwise qualifies as "the main program".
  279. If you don't know what event loop to use, use the one returned from this
  280. function (or via the C<EV_DEFAULT> macro).
  281. Note that this function is I<not> thread-safe, so if you want to use it
  282. from multiple threads, you have to employ some kind of mutex (note also
  283. that this case is unlikely, as loops cannot be shared easily between
  284. threads anyway).
  285. The default loop is the only loop that can handle C<ev_child> watchers,
  286. and to do this, it always registers a handler for C<SIGCHLD>. If this is
  287. a problem for your application you can either create a dynamic loop with
  288. C<ev_loop_new> which doesn't do that, or you can simply overwrite the
  289. C<SIGCHLD> signal handler I<after> calling C<ev_default_init>.
  290. Example: This is the most typical usage.
  291. if (!ev_default_loop (0))
  292. fatal ("could not initialise libev, bad $LIBEV_FLAGS in environment?");
  293. Example: Restrict libev to the select and poll backends, and do not allow
  294. environment settings to be taken into account:
  296. =item struct ev_loop *ev_loop_new (unsigned int flags)
  297. This will create and initialise a new event loop object. If the loop
  298. could not be initialised, returns false.
  299. This function is thread-safe, and one common way to use libev with
  300. threads is indeed to create one loop per thread, and using the default
  301. loop in the "main" or "initial" thread.
  302. The flags argument can be used to specify special behaviour or specific
  303. backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
  304. The following flags are supported:
  305. =over 4
  306. =item C<EVFLAG_AUTO>
  307. The default flags value. Use this if you have no clue (it's the right
  308. thing, believe me).
  309. =item C<EVFLAG_NOENV>
  310. If this flag bit is or'ed into the flag value (or the program runs setuid
  311. or setgid) then libev will I<not> look at the environment variable
  312. C<LIBEV_FLAGS>. Otherwise (the default), this environment variable will
  313. override the flags completely if it is found in the environment. This is
  314. useful to try out specific backends to test their performance, to work
  315. around bugs, or to make libev threadsafe (accessing environment variables
  316. cannot be done in a threadsafe way, but usually it works if no other
  317. thread modifies them).
  318. =item C<EVFLAG_FORKCHECK>
  319. Instead of calling C<ev_loop_fork> manually after a fork, you can also
  320. make libev check for a fork in each iteration by enabling this flag.
  321. This works by calling C<getpid ()> on every iteration of the loop,
  322. and thus this might slow down your event loop if you do a lot of loop
  323. iterations and little real work, but is usually not noticeable (on my
  324. GNU/Linux system for example, C<getpid> is actually a simple 5-insn
  325. sequence without a system call and thus I<very> fast, but my GNU/Linux
  326. system also has C<pthread_atfork> which is even faster). (Update: glibc
  327. versions 2.25 apparently removed the C<getpid> optimisation again).
  328. The big advantage of this flag is that you can forget about fork (and
  329. forget about forgetting to tell libev about forking, although you still
  330. have to ignore C<SIGPIPE>) when you use this flag.
  331. This flag setting cannot be overridden or specified in the C<LIBEV_FLAGS>
  332. environment variable.
  333. =item C<EVFLAG_NOINOTIFY>
  334. When this flag is specified, then libev will not attempt to use the
  335. I<inotify> API for its C<ev_stat> watchers. Apart from debugging and
  336. testing, this flag can be useful to conserve inotify file descriptors, as
  337. otherwise each loop using C<ev_stat> watchers consumes one inotify handle.
  338. =item C<EVFLAG_SIGNALFD>
  339. When this flag is specified, then libev will attempt to use the
  340. I<signalfd> API for its C<ev_signal> (and C<ev_child>) watchers. This API
  341. delivers signals synchronously, which makes it both faster and might make
  342. it possible to get the queued signal data. It can also simplify signal
  343. handling with threads, as long as you properly block signals in your
  344. threads that are not interested in handling them.
  345. Signalfd will not be used by default as this changes your signal mask, and
  346. there are a lot of shoddy libraries and programs (glib's threadpool for
  347. example) that can't properly initialise their signal masks.
  348. =item C<EVFLAG_NOSIGMASK>
  349. When this flag is specified, then libev will avoid to modify the signal
  350. mask. Specifically, this means you have to make sure signals are unblocked
  351. when you want to receive them.
  352. This behaviour is useful when you want to do your own signal handling, or
  353. want to handle signals only in specific threads and want to avoid libev
  354. unblocking the signals.
  355. It's also required by POSIX in a threaded program, as libev calls
  356. C<sigprocmask>, whose behaviour is officially unspecified.
  357. =item C<EVFLAG_NOTIMERFD>
  358. When this flag is specified, the libev will avoid using a C<timerfd> to
  359. detect time jumps. It will still be able to detect time jumps, but takes
  360. longer and has a lower accuracy in doing so, but saves a file descriptor
  361. per loop.
  362. The current implementation only tries to use a C<timerfd> when the first
  363. C<ev_periodic> watcher is started and falls back on other methods if it
  364. cannot be created, but this behaviour might change in the future.
  365. =item C<EVBACKEND_SELECT> (value 1, portable select backend)
  366. This is your standard select(2) backend. Not I<completely> standard, as
  367. libev tries to roll its own fd_set with no limits on the number of fds,
  368. but if that fails, expect a fairly low limit on the number of fds when
  369. using this backend. It doesn't scale too well (O(highest_fd)), but its
  370. usually the fastest backend for a low number of (low-numbered :) fds.
  371. To get good performance out of this backend you need a high amount of
  372. parallelism (most of the file descriptors should be busy). If you are
  373. writing a server, you should C<accept ()> in a loop to accept as many
  374. connections as possible during one iteration. You might also want to have
  375. a look at C<ev_set_io_collect_interval ()> to increase the amount of
  376. readiness notifications you get per iteration.
  377. This backend maps C<EV_READ> to the C<readfds> set and C<EV_WRITE> to the
  378. C<writefds> set (and to work around Microsoft Windows bugs, also onto the
  379. C<exceptfds> set on that platform).
  380. =item C<EVBACKEND_POLL> (value 2, poll backend, available everywhere except on windows)
  381. And this is your standard poll(2) backend. It's more complicated
  382. than select, but handles sparse fds better and has no artificial
  383. limit on the number of fds you can use (except it will slow down
  384. considerably with a lot of inactive fds). It scales similarly to select,
  385. i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
  386. performance tips.
  387. This backend maps C<EV_READ> to C<POLLIN | POLLERR | POLLHUP>, and
  389. =item C<EVBACKEND_EPOLL> (value 4, Linux)
  390. Use the Linux-specific epoll(7) interface (for both pre- and post-2.6.9
  391. kernels).
  392. For few fds, this backend is a bit little slower than poll and select, but
  393. it scales phenomenally better. While poll and select usually scale like
  394. O(total_fds) where total_fds is the total number of fds (or the highest
  395. fd), epoll scales either O(1) or O(active_fds).
  396. The epoll mechanism deserves honorable mention as the most misdesigned
  397. of the more advanced event mechanisms: mere annoyances include silently
  398. dropping file descriptors, requiring a system call per change per file
  399. descriptor (and unnecessary guessing of parameters), problems with dup,
  400. returning before the timeout value, resulting in additional iterations
  401. (and only giving 5ms accuracy while select on the same platform gives
  402. 0.1ms) and so on. The biggest issue is fork races, however - if a program
  403. forks then I<both> parent and child process have to recreate the epoll
  404. set, which can take considerable time (one syscall per file descriptor)
  405. and is of course hard to detect.
  406. Epoll is also notoriously buggy - embedding epoll fds I<should> work,
  407. but of course I<doesn't>, and epoll just loves to report events for
  408. totally I<different> file descriptors (even already closed ones, so
  409. one cannot even remove them from the set) than registered in the set
  410. (especially on SMP systems). Libev tries to counter these spurious
  411. notifications by employing an additional generation counter and comparing
  412. that against the events to filter out spurious ones, recreating the set
  413. when required. Epoll also erroneously rounds down timeouts, but gives you
  414. no way to know when and by how much, so sometimes you have to busy-wait
  415. because epoll returns immediately despite a nonzero timeout. And last
  416. not least, it also refuses to work with some file descriptors which work
  417. perfectly fine with C<select> (files, many character devices...).
  418. Epoll is truly the train wreck among event poll mechanisms, a frankenpoll,
  419. cobbled together in a hurry, no thought to design or interaction with
  420. others. Oh, the pain, will it ever stop...
  421. While stopping, setting and starting an I/O watcher in the same iteration
  422. will result in some caching, there is still a system call per such
  423. incident (because the same I<file descriptor> could point to a different
  424. I<file description> now), so its best to avoid that. Also, C<dup ()>'ed
  425. file descriptors might not work very well if you register events for both
  426. file descriptors.
  427. Best performance from this backend is achieved by not unregistering all
  428. watchers for a file descriptor until it has been closed, if possible,
  429. i.e. keep at least one watcher active per fd at all times. Stopping and
  430. starting a watcher (without re-setting it) also usually doesn't cause
  431. extra overhead. A fork can both result in spurious notifications as well
  432. as in libev having to destroy and recreate the epoll object, which can
  433. take considerable time and thus should be avoided.
  434. All this means that, in practice, C<EVBACKEND_SELECT> can be as fast or
  435. faster than epoll for maybe up to a hundred file descriptors, depending on
  436. the usage. So sad.
  437. While nominally embeddable in other event loops, this feature is broken in
  438. a lot of kernel revisions, but probably(!) works in current versions.
  439. This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
  441. =item C<EVBACKEND_LINUXAIO> (value 64, Linux)
  442. Use the Linux-specific Linux AIO (I<not> C<< aio(7) >> but C<<
  443. io_submit(2) >>) event interface available in post-4.18 kernels (but libev
  444. only tries to use it in 4.19+).
  445. This is another Linux train wreck of an event interface.
  446. If this backend works for you (as of this writing, it was very
  447. experimental), it is the best event interface available on Linux and might
  448. be well worth enabling it - if it isn't available in your kernel this will
  449. be detected and this backend will be skipped.
  450. This backend can batch oneshot requests and supports a user-space ring
  451. buffer to receive events. It also doesn't suffer from most of the design
  452. problems of epoll (such as not being able to remove event sources from
  453. the epoll set), and generally sounds too good to be true. Because, this
  454. being the Linux kernel, of course it suffers from a whole new set of
  455. limitations, forcing you to fall back to epoll, inheriting all its design
  456. issues.
  457. For one, it is not easily embeddable (but probably could be done using
  458. an event fd at some extra overhead). It also is subject to a system wide
  459. limit that can be configured in F</proc/sys/fs/aio-max-nr>. If no AIO
  460. requests are left, this backend will be skipped during initialisation, and
  461. will switch to epoll when the loop is active.
  462. Most problematic in practice, however, is that not all file descriptors
  463. work with it. For example, in Linux 5.1, TCP sockets, pipes, event fds,
  464. files, F</dev/null> and many others are supported, but ttys do not work
  465. properly (a known bug that the kernel developers don't care about, see
  466. L<https://lore.kernel.org/patchwork/patch/1047453/>), so this is not
  467. (yet?) a generic event polling interface.
  468. Overall, it seems the Linux developers just don't want it to have a
  469. generic event handling mechanism other than C<select> or C<poll>.
  470. To work around all these problem, the current version of libev uses its
  471. epoll backend as a fallback for file descriptor types that do not work. Or
  472. falls back completely to epoll if the kernel acts up.
  473. This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
  475. =item C<EVBACKEND_KQUEUE> (value 8, most BSD clones)
  476. Kqueue deserves special mention, as at the time this backend was
  477. implemented, it was broken on all BSDs except NetBSD (usually it doesn't
  478. work reliably with anything but sockets and pipes, except on Darwin,
  479. where of course it's completely useless). Unlike epoll, however, whose
  480. brokenness is by design, these kqueue bugs can be (and mostly have been)
  481. fixed without API changes to existing programs. For this reason it's not
  482. being "auto-detected" on all platforms unless you explicitly specify it
  483. in the flags (i.e. using C<EVBACKEND_KQUEUE>) or libev was compiled on a
  484. known-to-be-good (-enough) system like NetBSD.
  485. You still can embed kqueue into a normal poll or select backend and use it
  486. only for sockets (after having made sure that sockets work with kqueue on
  487. the target platform). See C<ev_embed> watchers for more info.
  488. It scales in the same way as the epoll backend, but the interface to the
  489. kernel is more efficient (which says nothing about its actual speed, of
  490. course). While stopping, setting and starting an I/O watcher does never
  491. cause an extra system call as with C<EVBACKEND_EPOLL>, it still adds up to
  492. two event changes per incident. Support for C<fork ()> is very bad (you
  493. might have to leak fds on fork, but it's more sane than epoll) and it
  494. drops fds silently in similarly hard-to-detect cases.
  495. This backend usually performs well under most conditions.
  496. While nominally embeddable in other event loops, this doesn't work
  497. everywhere, so you might need to test for this. And since it is broken
  498. almost everywhere, you should only use it when you have a lot of sockets
  499. (for which it usually works), by embedding it into another event loop
  500. (e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> (but C<poll> is of course
  501. also broken on OS X)) and, did I mention it, using it only for sockets.
  502. This backend maps C<EV_READ> into an C<EVFILT_READ> kevent with
  503. C<NOTE_EOF>, and C<EV_WRITE> into an C<EVFILT_WRITE> kevent with
  504. C<NOTE_EOF>.
  505. =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
  506. This is not implemented yet (and might never be, unless you send me an
  507. implementation). According to reports, C</dev/poll> only supports sockets
  508. and is not embeddable, which would limit the usefulness of this backend
  509. immensely.
  510. =item C<EVBACKEND_PORT> (value 32, Solaris 10)
  511. This uses the Solaris 10 event port mechanism. As with everything on Solaris,
  512. it's really slow, but it still scales very well (O(active_fds)).
  513. While this backend scales well, it requires one system call per active
  514. file descriptor per loop iteration. For small and medium numbers of file
  515. descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
  516. might perform better.
  517. On the positive side, this backend actually performed fully to
  518. specification in all tests and is fully embeddable, which is a rare feat
  519. among the OS-specific backends (I vastly prefer correctness over speed
  520. hacks).
  521. On the negative side, the interface is I<bizarre> - so bizarre that
  522. even sun itself gets it wrong in their code examples: The event polling
  523. function sometimes returns events to the caller even though an error
  524. occurred, but with no indication whether it has done so or not (yes, it's
  525. even documented that way) - deadly for edge-triggered interfaces where you
  526. absolutely have to know whether an event occurred or not because you have
  527. to re-arm the watcher.
  528. Fortunately libev seems to be able to work around these idiocies.
  529. This backend maps C<EV_READ> and C<EV_WRITE> in the same way as
  531. =item C<EVBACKEND_ALL>
  532. Try all backends (even potentially broken ones that wouldn't be tried
  533. with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
  535. It is definitely not recommended to use this flag, use whatever
  536. C<ev_recommended_backends ()> returns, or simply do not specify a backend
  537. at all.
  538. =item C<EVBACKEND_MASK>
  539. Not a backend at all, but a mask to select all backend bits from a
  540. C<flags> value, in case you want to mask out any backends from a flags
  541. value (e.g. when modifying the C<LIBEV_FLAGS> environment variable).
  542. =back
  543. If one or more of the backend flags are or'ed into the flags value,
  544. then only these backends will be tried (in the reverse order as listed
  545. here). If none are specified, all backends in C<ev_recommended_backends
  546. ()> will be tried.
  547. Example: Try to create a event loop that uses epoll and nothing else.
  548. struct ev_loop *epoller = ev_loop_new (EVBACKEND_EPOLL | EVFLAG_NOENV);
  549. if (!epoller)
  550. fatal ("no epoll found here, maybe it hides under your chair");
  551. Example: Use whatever libev has to offer, but make sure that kqueue is
  552. used if available.
  553. struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_KQUEUE);
  554. Example: Similarly, on linux, you mgiht want to take advantage of the
  555. linux aio backend if possible, but fall back to something else if that
  556. isn't available.
  557. struct ev_loop *loop = ev_loop_new (ev_recommended_backends () | EVBACKEND_LINUXAIO);
  558. =item ev_loop_destroy (loop)
  559. Destroys an event loop object (frees all memory and kernel state
  560. etc.). None of the active event watchers will be stopped in the normal
  561. sense, so e.g. C<ev_is_active> might still return true. It is your
  562. responsibility to either stop all watchers cleanly yourself I<before>
  563. calling this function, or cope with the fact afterwards (which is usually
  564. the easiest thing, you can just ignore the watchers and/or C<free ()> them
  565. for example).
  566. Note that certain global state, such as signal state (and installed signal
  567. handlers), will not be freed by this function, and related watchers (such
  568. as signal and child watchers) would need to be stopped manually.
  569. This function is normally used on loop objects allocated by
  570. C<ev_loop_new>, but it can also be used on the default loop returned by
  571. C<ev_default_loop>, in which case it is not thread-safe.
  572. Note that it is not advisable to call this function on the default loop
  573. except in the rare occasion where you really need to free its resources.
  574. If you need dynamically allocated loops it is better to use C<ev_loop_new>
  575. and C<ev_loop_destroy>.
  576. =item ev_loop_fork (loop)
  577. This function sets a flag that causes subsequent C<ev_run> iterations
  578. to reinitialise the kernel state for backends that have one. Despite
  579. the name, you can call it anytime you are allowed to start or stop
  580. watchers (except inside an C<ev_prepare> callback), but it makes most
  581. sense after forking, in the child process. You I<must> call it (or use
  582. C<EVFLAG_FORKCHECK>) in the child before resuming or calling C<ev_run>.
  583. In addition, if you want to reuse a loop (via this function or
  584. C<EVFLAG_FORKCHECK>), you I<also> have to ignore C<SIGPIPE>.
  585. Again, you I<have> to call it on I<any> loop that you want to re-use after
  586. a fork, I<even if you do not plan to use the loop in the parent>. This is
  587. because some kernel interfaces *cough* I<kqueue> *cough* do funny things
  588. during fork.
  589. On the other hand, you only need to call this function in the child
  590. process if and only if you want to use the event loop in the child. If
  591. you just fork+exec or create a new loop in the child, you don't have to
  592. call it at all (in fact, C<epoll> is so badly broken that it makes a
  593. difference, but libev will usually detect this case on its own and do a
  594. costly reset of the backend).
  595. The function itself is quite fast and it's usually not a problem to call
  596. it just in case after a fork.
  597. Example: Automate calling C<ev_loop_fork> on the default loop when
  598. using pthreads.
  599. static void
  600. post_fork_child (void)
  601. {
  602. ev_loop_fork (EV_DEFAULT);
  603. }
  604. ...
  605. pthread_atfork (0, 0, post_fork_child);
  606. =item int ev_is_default_loop (loop)
  607. Returns true when the given loop is, in fact, the default loop, and false
  608. otherwise.
  609. =item unsigned int ev_iteration (loop)
  610. Returns the current iteration count for the event loop, which is identical
  611. to the number of times libev did poll for new events. It starts at C<0>
  612. and happily wraps around with enough iterations.
  613. This value can sometimes be useful as a generation counter of sorts (it
  614. "ticks" the number of loop iterations), as it roughly corresponds with
  615. C<ev_prepare> and C<ev_check> calls - and is incremented between the
  616. prepare and check phases.
  617. =item unsigned int ev_depth (loop)
  618. Returns the number of times C<ev_run> was entered minus the number of
  619. times C<ev_run> was exited normally, in other words, the recursion depth.
  620. Outside C<ev_run>, this number is zero. In a callback, this number is
  621. C<1>, unless C<ev_run> was invoked recursively (or from another thread),
  622. in which case it is higher.
  623. Leaving C<ev_run> abnormally (setjmp/longjmp, cancelling the thread,
  624. throwing an exception etc.), doesn't count as "exit" - consider this
  625. as a hint to avoid such ungentleman-like behaviour unless it's really
  626. convenient, in which case it is fully supported.
  627. =item unsigned int ev_backend (loop)
  628. Returns one of the C<EVBACKEND_*> flags indicating the event backend in
  629. use.
  630. =item ev_tstamp ev_now (loop)
  631. Returns the current "event loop time", which is the time the event loop
  632. received events and started processing them. This timestamp does not
  633. change as long as callbacks are being processed, and this is also the base
  634. time used for relative timers. You can treat it as the timestamp of the
  635. event occurring (or more correctly, libev finding out about it).
  636. =item ev_now_update (loop)
  637. Establishes the current time by querying the kernel, updating the time
  638. returned by C<ev_now ()> in the progress. This is a costly operation and
  639. is usually done automatically within C<ev_run ()>.
  640. This function is rarely useful, but when some event callback runs for a
  641. very long time without entering the event loop, updating libev's idea of
  642. the current time is a good idea.
  643. See also L</The special problem of time updates> in the C<ev_timer> section.
  644. =item ev_suspend (loop)
  645. =item ev_resume (loop)
  646. These two functions suspend and resume an event loop, for use when the
  647. loop is not used for a while and timeouts should not be processed.
  648. A typical use case would be an interactive program such as a game: When
  649. the user presses C<^Z> to suspend the game and resumes it an hour later it
  650. would be best to handle timeouts as if no time had actually passed while
  651. the program was suspended. This can be achieved by calling C<ev_suspend>
  652. in your C<SIGTSTP> handler, sending yourself a C<SIGSTOP> and calling
  653. C<ev_resume> directly afterwards to resume timer processing.
  654. Effectively, all C<ev_timer> watchers will be delayed by the time spend
  655. between C<ev_suspend> and C<ev_resume>, and all C<ev_periodic> watchers
  656. will be rescheduled (that is, they will lose any events that would have
  657. occurred while suspended).
  658. After calling C<ev_suspend> you B<must not> call I<any> function on the
  659. given loop other than C<ev_resume>, and you B<must not> call C<ev_resume>
  660. without a previous call to C<ev_suspend>.
  661. Calling C<ev_suspend>/C<ev_resume> has the side effect of updating the
  662. event loop time (see C<ev_now_update>).
  663. =item bool ev_run (loop, int flags)
  664. Finally, this is it, the event handler. This function usually is called
  665. after you have initialised all your watchers and you want to start
  666. handling events. It will ask the operating system for any new events, call
  667. the watcher callbacks, and then repeat the whole process indefinitely: This
  668. is why event loops are called I<loops>.
  669. If the flags argument is specified as C<0>, it will keep handling events
  670. until either no event watchers are active anymore or C<ev_break> was
  671. called.
  672. The return value is false if there are no more active watchers (which
  673. usually means "all jobs done" or "deadlock"), and true in all other cases
  674. (which usually means " you should call C<ev_run> again").
  675. Please note that an explicit C<ev_break> is usually better than
  676. relying on all watchers to be stopped when deciding when a program has
  677. finished (especially in interactive programs), but having a program
  678. that automatically loops as long as it has to and no longer by virtue
  679. of relying on its watchers stopping correctly, that is truly a thing of
  680. beauty.
  681. This function is I<mostly> exception-safe - you can break out of a
  682. C<ev_run> call by calling C<longjmp> in a callback, throwing a C++
  683. exception and so on. This does not decrement the C<ev_depth> value, nor
  684. will it clear any outstanding C<EVBREAK_ONE> breaks.
  685. A flags value of C<EVRUN_NOWAIT> will look for new events, will handle
  686. those events and any already outstanding ones, but will not wait and
  687. block your process in case there are no events and will return after one
  688. iteration of the loop. This is sometimes useful to poll and handle new
  689. events while doing lengthy calculations, to keep the program responsive.
  690. A flags value of C<EVRUN_ONCE> will look for new events (waiting if
  691. necessary) and will handle those and any already outstanding ones. It
  692. will block your process until at least one new event arrives (which could
  693. be an event internal to libev itself, so there is no guarantee that a
  694. user-registered callback will be called), and will return after one
  695. iteration of the loop.
  696. This is useful if you are waiting for some external event in conjunction
  697. with something not expressible using other libev watchers (i.e. "roll your
  698. own C<ev_run>"). However, a pair of C<ev_prepare>/C<ev_check> watchers is
  699. usually a better approach for this kind of thing.
  700. Here are the gory details of what C<ev_run> does (this is for your
  701. understanding, not a guarantee that things will work exactly like this in
  702. future versions):
  703. - Increment loop depth.
  704. - Reset the ev_break status.
  705. - Before the first iteration, call any pending watchers.
  706. LOOP:
  707. - If EVFLAG_FORKCHECK was used, check for a fork.
  708. - If a fork was detected (by any means), queue and call all fork watchers.
  709. - Queue and call all prepare watchers.
  710. - If ev_break was called, goto FINISH.
  711. - If we have been forked, detach and recreate the kernel state
  712. as to not disturb the other process.
  713. - Update the kernel state with all outstanding changes.
  714. - Update the "event loop time" (ev_now ()).
  715. - Calculate for how long to sleep or block, if at all
  716. (active idle watchers, EVRUN_NOWAIT or not having
  717. any active watchers at all will result in not sleeping).
  718. - Sleep if the I/O and timer collect interval say so.
  719. - Increment loop iteration counter.
  720. - Block the process, waiting for any events.
  721. - Queue all outstanding I/O (fd) events.
  722. - Update the "event loop time" (ev_now ()), and do time jump adjustments.
  723. - Queue all expired timers.
  724. - Queue all expired periodics.
  725. - Queue all idle watchers with priority higher than that of pending events.
  726. - Queue all check watchers.
  727. - Call all queued watchers in reverse order (i.e. check watchers first).
  728. Signals and child watchers are implemented as I/O watchers, and will
  729. be handled here by queueing them when their watcher gets executed.
  730. - If ev_break has been called, or EVRUN_ONCE or EVRUN_NOWAIT
  731. were used, or there are no active watchers, goto FINISH, otherwise
  732. continue with step LOOP.
  733. FINISH:
  734. - Reset the ev_break status iff it was EVBREAK_ONE.
  735. - Decrement the loop depth.
  736. - Return.
  737. Example: Queue some jobs and then loop until no events are outstanding
  738. anymore.
  739. ... queue jobs here, make sure they register event watchers as long
  740. ... as they still have work to do (even an idle watcher will do..)
  741. ev_run (my_loop, 0);
  742. ... jobs done or somebody called break. yeah!
  743. =item ev_break (loop, how)
  744. Can be used to make a call to C<ev_run> return early (but only after it
  745. has processed all outstanding events). The C<how> argument must be either
  746. C<EVBREAK_ONE>, which will make the innermost C<ev_run> call return, or
  747. C<EVBREAK_ALL>, which will make all nested C<ev_run> calls return.
  748. This "break state" will be cleared on the next call to C<ev_run>.
  749. It is safe to call C<ev_break> from outside any C<ev_run> calls, too, in
  750. which case it will have no effect.
  751. =item ev_ref (loop)
  752. =item ev_unref (loop)
  753. Ref/unref can be used to add or remove a reference count on the event
  754. loop: Every watcher keeps one reference, and as long as the reference
  755. count is nonzero, C<ev_run> will not return on its own.
  756. This is useful when you have a watcher that you never intend to
  757. unregister, but that nevertheless should not keep C<ev_run> from
  758. returning. In such a case, call C<ev_unref> after starting, and C<ev_ref>
  759. before stopping it.
  760. As an example, libev itself uses this for its internal signal pipe: It
  761. is not visible to the libev user and should not keep C<ev_run> from
  762. exiting if no event watchers registered by it are active. It is also an
  763. excellent way to do this for generic recurring timers or from within
  764. third-party libraries. Just remember to I<unref after start> and I<ref
  765. before stop> (but only if the watcher wasn't active before, or was active
  766. before, respectively. Note also that libev might stop watchers itself
  767. (e.g. non-repeating timers) in which case you have to C<ev_ref>
  768. in the callback).
  769. Example: Create a signal watcher, but keep it from keeping C<ev_run>
  770. running when nothing else is active.
  771. ev_signal exitsig;
  772. ev_signal_init (&exitsig, sig_cb, SIGINT);
  773. ev_signal_start (loop, &exitsig);
  774. ev_unref (loop);
  775. Example: For some weird reason, unregister the above signal handler again.
  776. ev_ref (loop);
  777. ev_signal_stop (loop, &exitsig);
  778. =item ev_set_io_collect_interval (loop, ev_tstamp interval)
  779. =item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
  780. These advanced functions influence the time that libev will spend waiting
  781. for events. Both time intervals are by default C<0>, meaning that libev
  782. will try to invoke timer/periodic callbacks and I/O callbacks with minimum
  783. latency.
  784. Setting these to a higher value (the C<interval> I<must> be >= C<0>)
  785. allows libev to delay invocation of I/O and timer/periodic callbacks
  786. to increase efficiency of loop iterations (or to increase power-saving
  787. opportunities).
  788. The idea is that sometimes your program runs just fast enough to handle
  789. one (or very few) event(s) per loop iteration. While this makes the
  790. program responsive, it also wastes a lot of CPU time to poll for new
  791. events, especially with backends like C<select ()> which have a high
  792. overhead for the actual polling but can deliver many events at once.
  793. By setting a higher I<io collect interval> you allow libev to spend more
  794. time collecting I/O events, so you can handle more events per iteration,
  795. at the cost of increasing latency. Timeouts (both C<ev_periodic> and
  796. C<ev_timer>) will not be affected. Setting this to a non-null value will
  797. introduce an additional C<ev_sleep ()> call into most loop iterations. The
  798. sleep time ensures that libev will not poll for I/O events more often then
  799. once per this interval, on average (as long as the host time resolution is
  800. good enough).
  801. Likewise, by setting a higher I<timeout collect interval> you allow libev
  802. to spend more time collecting timeouts, at the expense of increased
  803. latency/jitter/inexactness (the watcher callback will be called
  804. later). C<ev_io> watchers will not be affected. Setting this to a non-null
  805. value will not introduce any overhead in libev.
  806. Many (busy) programs can usually benefit by setting the I/O collect
  807. interval to a value near C<0.1> or so, which is often enough for
  808. interactive servers (of course not for games), likewise for timeouts. It
  809. usually doesn't make much sense to set it to a lower value than C<0.01>,
  810. as this approaches the timing granularity of most systems. Note that if
  811. you do transactions with the outside world and you can't increase the
  812. parallelity, then this setting will limit your transaction rate (if you
  813. need to poll once per transaction and the I/O collect interval is 0.01,
  814. then you can't do more than 100 transactions per second).
  815. Setting the I<timeout collect interval> can improve the opportunity for
  816. saving power, as the program will "bundle" timer callback invocations that
  817. are "near" in time together, by delaying some, thus reducing the number of
  818. times the process sleeps and wakes up again. Another useful technique to
  819. reduce iterations/wake-ups is to use C<ev_periodic> watchers and make sure
  820. they fire on, say, one-second boundaries only.
  821. Example: we only need 0.1s timeout granularity, and we wish not to poll
  822. more often than 100 times per second:
  823. ev_set_timeout_collect_interval (EV_DEFAULT_UC_ 0.1);
  824. ev_set_io_collect_interval (EV_DEFAULT_UC_ 0.01);
  825. =item ev_invoke_pending (loop)
  826. This call will simply invoke all pending watchers while resetting their
  827. pending state. Normally, C<ev_run> does this automatically when required,
  828. but when overriding the invoke callback this call comes handy. This
  829. function can be invoked from a watcher - this can be useful for example
  830. when you want to do some lengthy calculation and want to pass further
  831. event handling to another thread (you still have to make sure only one
  832. thread executes within C<ev_invoke_pending> or C<ev_run> of course).
  833. =item int ev_pending_count (loop)
  834. Returns the number of pending watchers - zero indicates that no watchers
  835. are pending.
  836. =item ev_set_invoke_pending_cb (loop, void (*invoke_pending_cb)(EV_P))
  837. This overrides the invoke pending functionality of the loop: Instead of
  838. invoking all pending watchers when there are any, C<ev_run> will call
  839. this callback instead. This is useful, for example, when you want to
  840. invoke the actual watchers inside another context (another thread etc.).
  841. If you want to reset the callback, use C<ev_invoke_pending> as new
  842. callback.
  843. =item ev_set_loop_release_cb (loop, void (*release)(EV_P) throw (), void (*acquire)(EV_P) throw ())
  844. Sometimes you want to share the same loop between multiple threads. This
  845. can be done relatively simply by putting mutex_lock/unlock calls around
  846. each call to a libev function.
  847. However, C<ev_run> can run an indefinite time, so it is not feasible
  848. to wait for it to return. One way around this is to wake up the event
  849. loop via C<ev_break> and C<ev_async_send>, another way is to set these
  850. I<release> and I<acquire> callbacks on the loop.
  851. When set, then C<release> will be called just before the thread is
  852. suspended waiting for new events, and C<acquire> is called just
  853. afterwards.
  854. Ideally, C<release> will just call your mutex_unlock function, and
  855. C<acquire> will just call the mutex_lock function again.
  856. While event loop modifications are allowed between invocations of
  857. C<release> and C<acquire> (that's their only purpose after all), no
  858. modifications done will affect the event loop, i.e. adding watchers will
  859. have no effect on the set of file descriptors being watched, or the time
  860. waited. Use an C<ev_async> watcher to wake up C<ev_run> when you want it
  861. to take note of any changes you made.
  862. In theory, threads executing C<ev_run> will be async-cancel safe between
  863. invocations of C<release> and C<acquire>.
  864. See also the locking example in the C<THREADS> section later in this
  865. document.
  866. =item ev_set_userdata (loop, void *data)
  867. =item void *ev_userdata (loop)
  868. Set and retrieve a single C<void *> associated with a loop. When
  869. C<ev_set_userdata> has never been called, then C<ev_userdata> returns
  870. C<0>.
  871. These two functions can be used to associate arbitrary data with a loop,
  872. and are intended solely for the C<invoke_pending_cb>, C<release> and
  873. C<acquire> callbacks described above, but of course can be (ab-)used for
  874. any other purpose as well.
  875. =item ev_verify (loop)
  876. This function only does something when C<EV_VERIFY> support has been
  877. compiled in, which is the default for non-minimal builds. It tries to go
  878. through all internal structures and checks them for validity. If anything
  879. is found to be inconsistent, it will print an error message to standard
  880. error and call C<abort ()>.
  881. This can be used to catch bugs inside libev itself: under normal
  882. circumstances, this function will never abort as of course libev keeps its
  883. data structures consistent.
  884. =back
  885. =head1 ANATOMY OF A WATCHER
  886. In the following description, uppercase C<TYPE> in names stands for the
  887. watcher type, e.g. C<ev_TYPE_start> can mean C<ev_timer_start> for timer
  888. watchers and C<ev_io_start> for I/O watchers.
  889. A watcher is an opaque structure that you allocate and register to record
  890. your interest in some event. To make a concrete example, imagine you want
  891. to wait for STDIN to become readable, you would create an C<ev_io> watcher
  892. for that:
  893. static void my_cb (struct ev_loop *loop, ev_io *w, int revents)
  894. {
  895. ev_io_stop (w);
  896. ev_break (loop, EVBREAK_ALL);
  897. }
  898. struct ev_loop *loop = ev_default_loop (0);
  899. ev_io stdin_watcher;
  900. ev_init (&stdin_watcher, my_cb);
  901. ev_io_set (&stdin_watcher, STDIN_FILENO, EV_READ);
  902. ev_io_start (loop, &stdin_watcher);
  903. ev_run (loop, 0);
  904. As you can see, you are responsible for allocating the memory for your
  905. watcher structures (and it is I<usually> a bad idea to do this on the
  906. stack).
  907. Each watcher has an associated watcher structure (called C<struct ev_TYPE>
  908. or simply C<ev_TYPE>, as typedefs are provided for all watcher structs).
  909. Each watcher structure must be initialised by a call to C<ev_init (watcher
  910. *, callback)>, which expects a callback to be provided. This callback is
  911. invoked each time the event occurs (or, in the case of I/O watchers, each
  912. time the event loop detects that the file descriptor given is readable
  913. and/or writable).
  914. Each watcher type further has its own C<< ev_TYPE_set (watcher *, ...) >>
  915. macro to configure it, with arguments specific to the watcher type. There
  916. is also a macro to combine initialisation and setting in one call: C<<
  917. ev_TYPE_init (watcher *, callback, ...) >>.
  918. To make the watcher actually watch out for events, you have to start it
  919. with a watcher-specific start function (C<< ev_TYPE_start (loop, watcher
  920. *) >>), and you can stop watching for events at any time by calling the
  921. corresponding stop function (C<< ev_TYPE_stop (loop, watcher *) >>.
  922. As long as your watcher is active (has been started but not stopped) you
  923. must not touch the values stored in it except when explicitly documented
  924. otherwise. Most specifically you must never reinitialise it or call its
  925. C<ev_TYPE_set> macro.
  926. Each and every callback receives the event loop pointer as first, the
  927. registered watcher structure as second, and a bitset of received events as
  928. third argument.
  929. The received events usually include a single bit per event type received
  930. (you can receive multiple events at the same time). The possible bit masks
  931. are:
  932. =over 4
  933. =item C<EV_READ>
  934. =item C<EV_WRITE>
  935. The file descriptor in the C<ev_io> watcher has become readable and/or
  936. writable.
  937. =item C<EV_TIMER>
  938. The C<ev_timer> watcher has timed out.
  939. =item C<EV_PERIODIC>
  940. The C<ev_periodic> watcher has timed out.
  941. =item C<EV_SIGNAL>
  942. The signal specified in the C<ev_signal> watcher has been received by a thread.
  943. =item C<EV_CHILD>
  944. The pid specified in the C<ev_child> watcher has received a status change.
  945. =item C<EV_STAT>
  946. The path specified in the C<ev_stat> watcher changed its attributes somehow.
  947. =item C<EV_IDLE>
  948. The C<ev_idle> watcher has determined that you have nothing better to do.
  949. =item C<EV_PREPARE>
  950. =item C<EV_CHECK>
  951. All C<ev_prepare> watchers are invoked just I<before> C<ev_run> starts to
  952. gather new events, and all C<ev_check> watchers are queued (not invoked)
  953. just after C<ev_run> has gathered them, but before it queues any callbacks
  954. for any received events. That means C<ev_prepare> watchers are the last
  955. watchers invoked before the event loop sleeps or polls for new events, and
  956. C<ev_check> watchers will be invoked before any other watchers of the same
  957. or lower priority within an event loop iteration.
  958. Callbacks of both watcher types can start and stop as many watchers as
  959. they want, and all of them will be taken into account (for example, a
  960. C<ev_prepare> watcher might start an idle watcher to keep C<ev_run> from
  961. blocking).
  962. =item C<EV_EMBED>
  963. The embedded event loop specified in the C<ev_embed> watcher needs attention.
  964. =item C<EV_FORK>
  965. The event loop has been resumed in the child process after fork (see
  966. C<ev_fork>).
  967. =item C<EV_CLEANUP>
  968. The event loop is about to be destroyed (see C<ev_cleanup>).
  969. =item C<EV_ASYNC>
  970. The given async watcher has been asynchronously notified (see C<ev_async>).
  971. =item C<EV_CUSTOM>
  972. Not ever sent (or otherwise used) by libev itself, but can be freely used
  973. by libev users to signal watchers (e.g. via C<ev_feed_event>).
  974. =item C<EV_ERROR>
  975. An unspecified error has occurred, the watcher has been stopped. This might
  976. happen because the watcher could not be properly started because libev
  977. ran out of memory, a file descriptor was found to be closed or any other
  978. problem. Libev considers these application bugs.
  979. You best act on it by reporting the problem and somehow coping with the
  980. watcher being stopped. Note that well-written programs should not receive
  981. an error ever, so when your watcher receives it, this usually indicates a
  982. bug in your program.
  983. Libev will usually signal a few "dummy" events together with an error, for
  984. example it might indicate that a fd is readable or writable, and if your
  985. callbacks is well-written it can just attempt the operation and cope with
  986. the error from read() or write(). This will not work in multi-threaded
  987. programs, though, as the fd could already be closed and reused for another
  988. thing, so beware.
  989. =back
  991. =over 4
  992. =item C<ev_init> (ev_TYPE *watcher, callback)
  993. This macro initialises the generic portion of a watcher. The contents
  994. of the watcher object can be arbitrary (so C<malloc> will do). Only
  995. the generic parts of the watcher are initialised, you I<need> to call
  996. the type-specific C<ev_TYPE_set> macro afterwards to initialise the
  997. type-specific parts. For each type there is also a C<ev_TYPE_init> macro
  998. which rolls both calls into one.
  999. You can reinitialise a watcher at any time as long as it has been stopped
  1000. (or never started) and there are no pending events outstanding.
  1001. The callback is always of type C<void (*)(struct ev_loop *loop, ev_TYPE *watcher,
  1002. int revents)>.
  1003. Example: Initialise an C<ev_io> watcher in two steps.
  1004. ev_io w;
  1005. ev_init (&w, my_cb);
  1006. ev_io_set (&w, STDIN_FILENO, EV_READ);
  1007. =item C<ev_TYPE_set> (ev_TYPE *watcher, [args])
  1008. This macro initialises the type-specific parts of a watcher. You need to
  1009. call C<ev_init> at least once before you call this macro, but you can
  1010. call C<ev_TYPE_set> any number of times. You must not, however, call this
  1011. macro on a watcher that is active (it can be pending, however, which is a
  1012. difference to the C<ev_init> macro).
  1013. Although some watcher types do not have type-specific arguments
  1014. (e.g. C<ev_prepare>) you still need to call its C<set> macro.
  1015. See C<ev_init>, above, for an example.
  1016. =item C<ev_TYPE_init> (ev_TYPE *watcher, callback, [args])
  1017. This convenience macro rolls both C<ev_init> and C<ev_TYPE_set> macro
  1018. calls into a single call. This is the most convenient method to initialise
  1019. a watcher. The same limitations apply, of course.
  1020. Example: Initialise and set an C<ev_io> watcher in one step.
  1021. ev_io_init (&w, my_cb, STDIN_FILENO, EV_READ);
  1022. =item C<ev_TYPE_start> (loop, ev_TYPE *watcher)
  1023. Starts (activates) the given watcher. Only active watchers will receive
  1024. events. If the watcher is already active nothing will happen.
  1025. Example: Start the C<ev_io> watcher that is being abused as example in this
  1026. whole section.
  1027. ev_io_start (EV_DEFAULT_UC, &w);
  1028. =item C<ev_TYPE_stop> (loop, ev_TYPE *watcher)
  1029. Stops the given watcher if active, and clears the pending status (whether
  1030. the watcher was active or not).
  1031. It is possible that stopped watchers are pending - for example,
  1032. non-repeating timers are being stopped when they become pending - but
  1033. calling C<ev_TYPE_stop> ensures that the watcher is neither active nor
  1034. pending. If you want to free or reuse the memory used by the watcher it is
  1035. therefore a good idea to always call its C<ev_TYPE_stop> function.
  1036. =item bool ev_is_active (ev_TYPE *watcher)
  1037. Returns a true value iff the watcher is active (i.e. it has been started
  1038. and not yet been stopped). As long as a watcher is active you must not modify
  1039. it.
  1040. =item bool ev_is_pending (ev_TYPE *watcher)
  1041. Returns a true value iff the watcher is pending, (i.e. it has outstanding
  1042. events but its callback has not yet been invoked). As long as a watcher
  1043. is pending (but not active) you must not call an init function on it (but
  1044. C<ev_TYPE_set> is safe), you must not change its priority, and you must
  1045. make sure the watcher is available to libev (e.g. you cannot C<free ()>
  1046. it).
  1047. =item callback ev_cb (ev_TYPE *watcher)
  1048. Returns the callback currently set on the watcher.
  1049. =item ev_set_cb (ev_TYPE *watcher, callback)
  1050. Change the callback. You can change the callback at virtually any time
  1051. (modulo threads).
  1052. =item ev_set_priority (ev_TYPE *watcher, int priority)
  1053. =item int ev_priority (ev_TYPE *watcher)
  1054. Set and query the priority of the watcher. The priority is a small
  1055. integer between C<EV_MAXPRI> (default: C<2>) and C<EV_MINPRI>
  1056. (default: C<-2>). Pending watchers with higher priority will be invoked
  1057. before watchers with lower priority, but priority will not keep watchers
  1058. from being executed (except for C<ev_idle> watchers).
  1059. If you need to suppress invocation when higher priority events are pending
  1060. you need to look at C<ev_idle> watchers, which provide this functionality.
  1061. You I<must not> change the priority of a watcher as long as it is active or
  1062. pending.
  1063. Setting a priority outside the range of C<EV_MINPRI> to C<EV_MAXPRI> is
  1064. fine, as long as you do not mind that the priority value you query might
  1065. or might not have been clamped to the valid range.
  1066. The default priority used by watchers when no priority has been set is
  1067. always C<0>, which is supposed to not be too high and not be too low :).
  1068. See L</WATCHER PRIORITY MODELS>, below, for a more thorough treatment of
  1069. priorities.
  1070. =item ev_invoke (loop, ev_TYPE *watcher, int revents)
  1071. Invoke the C<watcher> with the given C<loop> and C<revents>. Neither
  1072. C<loop> nor C<revents> need to be valid as long as the watcher callback
  1073. can deal with that fact, as both are simply passed through to the
  1074. callback.
  1075. =item int ev_clear_pending (loop, ev_TYPE *watcher)
  1076. If the watcher is pending, this function clears its pending status and
  1077. returns its C<revents> bitset (as if its callback was invoked). If the
  1078. watcher isn't pending it does nothing and returns C<0>.
  1079. Sometimes it can be useful to "poll" a watcher instead of waiting for its
  1080. callback to be invoked, which can be accomplished with this function.
  1081. =item ev_feed_event (loop, ev_TYPE *watcher, int revents)
  1082. Feeds the given event set into the event loop, as if the specified event
  1083. had happened for the specified watcher (which must be a pointer to an
  1084. initialised but not necessarily started event watcher). Obviously you must
  1085. not free the watcher as long as it has pending events.
  1086. Stopping the watcher, letting libev invoke it, or calling
  1087. C<ev_clear_pending> will clear the pending event, even if the watcher was
  1088. not started in the first place.
  1089. See also C<ev_feed_fd_event> and C<ev_feed_signal_event> for related
  1090. functions that do not need a watcher.
  1091. =back
  1094. =head2 WATCHER STATES
  1095. There are various watcher states mentioned throughout this manual -
  1096. active, pending and so on. In this section these states and the rules to
  1097. transition between them will be described in more detail - and while these
  1098. rules might look complicated, they usually do "the right thing".
  1099. =over 4
  1100. =item initialised
  1101. Before a watcher can be registered with the event loop it has to be
  1102. initialised. This can be done with a call to C<ev_TYPE_init>, or calls to
  1103. C<ev_init> followed by the watcher-specific C<ev_TYPE_set> function.
  1104. In this state it is simply some block of memory that is suitable for
  1105. use in an event loop. It can be moved around, freed, reused etc. at
  1106. will - as long as you either keep the memory contents intact, or call
  1107. C<ev_TYPE_init> again.
  1108. =item started/running/active
  1109. Once a watcher has been started with a call to C<ev_TYPE_start> it becomes
  1110. property of the event loop, and is actively waiting for events. While in
  1111. this state it cannot be accessed (except in a few documented ways), moved,
  1112. freed or anything else - the only legal thing is to keep a pointer to it,
  1113. and call libev functions on it that are documented to work on active watchers.
  1114. =item pending
  1115. If a watcher is active and libev determines that an event it is interested
  1116. in has occurred (such as a timer expiring), it will become pending. It will
  1117. stay in this pending state until either it is stopped or its callback is
  1118. about to be invoked, so it is not normally pending inside the watcher
  1119. callback.
  1120. The watcher might or might not be active while it is pending (for example,
  1121. an expired non-repeating timer can be pending but no longer active). If it
  1122. is stopped, it can be freely accessed (e.g. by calling C<ev_TYPE_set>),
  1123. but it is still property of the event loop at this time, so cannot be
  1124. moved, freed or reused. And if it is active the rules described in the
  1125. previous item still apply.
  1126. It is also possible to feed an event on a watcher that is not active (e.g.
  1127. via C<ev_feed_event>), in which case it becomes pending without being
  1128. active.
  1129. =item stopped
  1130. A watcher can be stopped implicitly by libev (in which case it might still
  1131. be pending), or explicitly by calling its C<ev_TYPE_stop> function. The
  1132. latter will clear any pending state the watcher might be in, regardless
  1133. of whether it was active or not, so stopping a watcher explicitly before
  1134. freeing it is often a good idea.
  1135. While stopped (and not pending) the watcher is essentially in the
  1136. initialised state, that is, it can be reused, moved, modified in any way
  1137. you wish (but when you trash the memory block, you need to C<ev_TYPE_init>
  1138. it again).
  1139. =back
  1141. Many event loops support I<watcher priorities>, which are usually small
  1142. integers that influence the ordering of event callback invocation
  1143. between watchers in some way, all else being equal.
  1144. In libev, watcher priorities can be set using C<ev_set_priority>. See its
  1145. description for the more technical details such as the actual priority
  1146. range.
  1147. There are two common ways how these these priorities are being interpreted
  1148. by event loops:
  1149. In the more common lock-out model, higher priorities "lock out" invocation
  1150. of lower priority watchers, which means as long as higher priority
  1151. watchers receive events, lower priority watchers are not being invoked.
  1152. The less common only-for-ordering model uses priorities solely to order
  1153. callback invocation within a single event loop iteration: Higher priority
  1154. watchers are invoked before lower priority ones, but they all get invoked
  1155. before polling for new events.
  1156. Libev uses the second (only-for-ordering) model for all its watchers
  1157. except for idle watchers (which use the lock-out model).
  1158. The rationale behind this is that implementing the lock-out model for
  1159. watchers is not well supported by most kernel interfaces, and most event
  1160. libraries will just poll for the same events again and again as long as
  1161. their callbacks have not been executed, which is very inefficient in the
  1162. common case of one high-priority watcher locking out a mass of lower
  1163. priority ones.
  1164. Static (ordering) priorities are most useful when you have two or more
  1165. watchers handling the same resource: a typical usage example is having an
  1166. C<ev_io> watcher to receive data, and an associated C<ev_timer> to handle
  1167. timeouts. Under load, data might be received while the program handles
  1168. other jobs, but since timers normally get invoked first, the timeout
  1169. handler will be executed before checking for data. In that case, giving
  1170. the timer a lower priority than the I/O watcher ensures that I/O will be
  1171. handled first even under adverse conditions (which is usually, but not
  1172. always, what you want).
  1173. Since idle watchers use the "lock-out" model, meaning that idle watchers
  1174. will only be executed when no same or higher priority watchers have
  1175. received events, they can be used to implement the "lock-out" model when
  1176. required.
  1177. For example, to emulate how many other event libraries handle priorities,
  1178. you can associate an C<ev_idle> watcher to each such watcher, and in
  1179. the normal watcher callback, you just start the idle watcher. The real
  1180. processing is done in the idle watcher callback. This causes libev to
  1181. continuously poll and process kernel event data for the watcher, but when
  1182. the lock-out case is known to be rare (which in turn is rare :), this is
  1183. workable.
  1184. Usually, however, the lock-out model implemented that way will perform
  1185. miserably under the type of load it was designed to handle. In that case,
  1186. it might be preferable to stop the real watcher before starting the
  1187. idle watcher, so the kernel will not have to process the event in case
  1188. the actual processing will be delayed for considerable time.
  1189. Here is an example of an I/O watcher that should run at a strictly lower
  1190. priority than the default, and which should only process data when no
  1191. other events are pending:
  1192. ev_idle idle; // actual processing watcher
  1193. ev_io io; // actual event watcher
  1194. static void
  1195. io_cb (EV_P_ ev_io *w, int revents)
  1196. {
  1197. // stop the I/O watcher, we received the event, but
  1198. // are not yet ready to handle it.
  1199. ev_io_stop (EV_A_ w);
  1200. // start the idle watcher to handle the actual event.
  1201. // it will not be executed as long as other watchers
  1202. // with the default priority are receiving events.
  1203. ev_idle_start (EV_A_ &idle);
  1204. }
  1205. static void
  1206. idle_cb (EV_P_ ev_idle *w, int revents)
  1207. {
  1208. // actual processing
  1209. read (STDIN_FILENO, ...);
  1210. // have to start the I/O watcher again, as
  1211. // we have handled the event
  1212. ev_io_start (EV_P_ &io);
  1213. }
  1214. // initialisation
  1215. ev_idle_init (&idle, idle_cb);
  1216. ev_io_init (&io, io_cb, STDIN_FILENO, EV_READ);
  1217. ev_io_start (EV_DEFAULT_ &io);
  1218. In the "real" world, it might also be beneficial to start a timer, so that
  1219. low-priority connections can not be locked out forever under load. This
  1220. enables your program to keep a lower latency for important connections
  1221. during short periods of high load, while not completely locking out less
  1222. important ones.
  1223. =head1 WATCHER TYPES
  1224. This section describes each watcher in detail, but will not repeat
  1225. information given in the last section. Any initialisation/set macros,
  1226. functions and members specific to the watcher type are explained.
  1227. Most members are additionally marked with either I<[read-only]>, meaning
  1228. that, while the watcher is active, you can look at the member and expect
  1229. some sensible content, but you must not modify it (you can modify it while
  1230. the watcher is stopped to your hearts content), or I<[read-write]>, which
  1231. means you can expect it to have some sensible content while the watcher is
  1232. active, but you can also modify it (within the same thread as the event
  1233. loop, i.e. without creating data races). Modifying it may not do something
  1234. sensible or take immediate effect (or do anything at all), but libev will
  1235. not crash or malfunction in any way.
  1236. In any case, the documentation for each member will explain what the
  1237. effects are, and if there are any additional access restrictions.
  1238. =head2 C<ev_io> - is this file descriptor readable or writable?
  1239. I/O watchers check whether a file descriptor is readable or writable
  1240. in each iteration of the event loop, or, more precisely, when reading
  1241. would not block the process and writing would at least be able to write
  1242. some data. This behaviour is called level-triggering because you keep
  1243. receiving events as long as the condition persists. Remember you can stop
  1244. the watcher if you don't want to act on the event and neither want to
  1245. receive future events.
  1246. In general you can register as many read and/or write event watchers per
  1247. fd as you want (as long as you don't confuse yourself). Setting all file
  1248. descriptors to non-blocking mode is also usually a good idea (but not
  1249. required if you know what you are doing).
  1250. Another thing you have to watch out for is that it is quite easy to
  1251. receive "spurious" readiness notifications, that is, your callback might
  1252. be called with C<EV_READ> but a subsequent C<read>(2) will actually block
  1253. because there is no data. It is very easy to get into this situation even
  1254. with a relatively standard program structure. Thus it is best to always
  1255. use non-blocking I/O: An extra C<read>(2) returning C<EAGAIN> is far
  1256. preferable to a program hanging until some data arrives.
  1257. If you cannot run the fd in non-blocking mode (for example you should
  1258. not play around with an Xlib connection), then you have to separately
  1259. re-test whether a file descriptor is really ready with a known-to-be good
  1260. interface such as poll (fortunately in the case of Xlib, it already does
  1261. this on its own, so its quite safe to use). Some people additionally
  1262. use C<SIGALRM> and an interval timer, just to be sure you won't block
  1263. indefinitely.
  1264. But really, best use non-blocking mode.
  1265. =head3 The special problem of disappearing file descriptors
  1266. Some backends (e.g. kqueue, epoll, linuxaio) need to be told about closing
  1267. a file descriptor (either due to calling C<close> explicitly or any other
  1268. means, such as C<dup2>). The reason is that you register interest in some
  1269. file descriptor, but when it goes away, the operating system will silently
  1270. drop this interest. If another file descriptor with the same number then
  1271. is registered with libev, there is no efficient way to see that this is,
  1272. in fact, a different file descriptor.
  1273. To avoid having to explicitly tell libev about such cases, libev follows
  1274. the following policy: Each time C<ev_io_set> is being called, libev
  1275. will assume that this is potentially a new file descriptor, otherwise
  1276. it is assumed that the file descriptor stays the same. That means that
  1277. you I<have> to call C<ev_io_set> (or C<ev_io_init>) when you change the
  1278. descriptor even if the file descriptor number itself did not change.
  1279. This is how one would do it normally anyway, the important point is that
  1280. the libev application should not optimise around libev but should leave
  1281. optimisations to libev.
  1282. =head3 The special problem of dup'ed file descriptors
  1283. Some backends (e.g. epoll), cannot register events for file descriptors,
  1284. but only events for the underlying file descriptions. That means when you
  1285. have C<dup ()>'ed file descriptors or weirder constellations, and register
  1286. events for them, only one file descriptor might actually receive events.
  1287. There is no workaround possible except not registering events
  1288. for potentially C<dup ()>'ed file descriptors, or to resort to
  1290. =head3 The special problem of files
  1291. Many people try to use C<select> (or libev) on file descriptors
  1292. representing files, and expect it to become ready when their program
  1293. doesn't block on disk accesses (which can take a long time on their own).
  1294. However, this cannot ever work in the "expected" way - you get a readiness
  1295. notification as soon as the kernel knows whether and how much data is
  1296. there, and in the case of open files, that's always the case, so you
  1297. always get a readiness notification instantly, and your read (or possibly
  1298. write) will still block on the disk I/O.
  1299. Another way to view it is that in the case of sockets, pipes, character
  1300. devices and so on, there is another party (the sender) that delivers data
  1301. on its own, but in the case of files, there is no such thing: the disk
  1302. will not send data on its own, simply because it doesn't know what you
  1303. wish to read - you would first have to request some data.
  1304. Since files are typically not-so-well supported by advanced notification
  1305. mechanism, libev tries hard to emulate POSIX behaviour with respect
  1306. to files, even though you should not use it. The reason for this is
  1307. convenience: sometimes you want to watch STDIN or STDOUT, which is
  1308. usually a tty, often a pipe, but also sometimes files or special devices
  1309. (for example, C<epoll> on Linux works with F</dev/random> but not with
  1310. F</dev/urandom>), and even though the file might better be served with
  1311. asynchronous I/O instead of with non-blocking I/O, it is still useful when
  1312. it "just works" instead of freezing.
  1313. So avoid file descriptors pointing to files when you know it (e.g. use
  1314. libeio), but use them when it is convenient, e.g. for STDIN/STDOUT, or
  1315. when you rarely read from a file instead of from a socket, and want to
  1316. reuse the same code path.
  1317. =head3 The special problem of fork
  1318. Some backends (epoll, kqueue, linuxaio, iouring) do not support C<fork ()>
  1319. at all or exhibit useless behaviour. Libev fully supports fork, but needs
  1320. to be told about it in the child if you want to continue to use it in the
  1321. child.
  1322. To support fork in your child processes, you have to call C<ev_loop_fork
  1323. ()> after a fork in the child, enable C<EVFLAG_FORKCHECK>, or resort to
  1325. =head3 The special problem of SIGPIPE
  1326. While not really specific to libev, it is easy to forget about C<SIGPIPE>:
  1327. when writing to a pipe whose other end has been closed, your program gets
  1328. sent a SIGPIPE, which, by default, aborts your program. For most programs
  1329. this is sensible behaviour, for daemons, this is usually undesirable.
  1330. So when you encounter spurious, unexplained daemon exits, make sure you
  1331. ignore SIGPIPE (and maybe make sure you log the exit status of your daemon
  1332. somewhere, as that would have given you a big clue).
  1333. =head3 The special problem of accept()ing when you can't
  1334. Many implementations of the POSIX C<accept> function (for example,
  1335. found in post-2004 Linux) have the peculiar behaviour of not removing a
  1336. connection from the pending queue in all error cases.
  1337. For example, larger servers often run out of file descriptors (because
  1338. of resource limits), causing C<accept> to fail with C<ENFILE> but not
  1339. rejecting the connection, leading to libev signalling readiness on
  1340. the next iteration again (the connection still exists after all), and
  1341. typically causing the program to loop at 100% CPU usage.
  1342. Unfortunately, the set of errors that cause this issue differs between
  1343. operating systems, there is usually little the app can do to remedy the
  1344. situation, and no known thread-safe method of removing the connection to
  1345. cope with overload is known (to me).
  1346. One of the easiest ways to handle this situation is to just ignore it
  1347. - when the program encounters an overload, it will just loop until the
  1348. situation is over. While this is a form of busy waiting, no OS offers an
  1349. event-based way to handle this situation, so it's the best one can do.
  1350. A better way to handle the situation is to log any errors other than
  1351. C<EAGAIN> and C<EWOULDBLOCK>, making sure not to flood the log with such
  1352. messages, and continue as usual, which at least gives the user an idea of
  1353. what could be wrong ("raise the ulimit!"). For extra points one could stop
  1354. the C<ev_io> watcher on the listening fd "for a while", which reduces CPU
  1355. usage.
  1356. If your program is single-threaded, then you could also keep a dummy file
  1357. descriptor for overload situations (e.g. by opening F</dev/null>), and
  1358. when you run into C<ENFILE> or C<EMFILE>, close it, run C<accept>,
  1359. close that fd, and create a new dummy fd. This will gracefully refuse
  1360. clients under typical overload conditions.
  1361. The last way to handle it is to simply log the error and C<exit>, as
  1362. is often done with C<malloc> failures, but this results in an easy
  1363. opportunity for a DoS attack.
  1364. =head3 Watcher-Specific Functions
  1365. =over 4
  1366. =item ev_io_init (ev_io *, callback, int fd, int events)
  1367. =item ev_io_set (ev_io *, int fd, int events)
  1368. Configures an C<ev_io> watcher. The C<fd> is the file descriptor to
  1369. receive events for and C<events> is either C<EV_READ>, C<EV_WRITE>, both
  1370. C<EV_READ | EV_WRITE> or C<0>, to express the desire to receive the given
  1371. events.
  1372. Note that setting the C<events> to C<0> and starting the watcher is
  1373. supported, but not specially optimized - if your program sometimes happens
  1374. to generate this combination this is fine, but if it is easy to avoid
  1375. starting an io watcher watching for no events you should do so.
  1376. =item ev_io_modify (ev_io *, int events)
  1377. Similar to C<ev_io_set>, but only changes the event mask. Using this might
  1378. be faster with some backends, as libev can assume that the C<fd> still
  1379. refers to the same underlying file description, something it cannot do
  1380. when using C<ev_io_set>.
  1381. =item int fd [no-modify]
  1382. The file descriptor being watched. While it can be read at any time, you
  1383. must not modify this member even when the watcher is stopped - always use
  1384. C<ev_io_set> for that.
  1385. =item int events [no-modify]
  1386. The set of events the fd is being watched for, among other flags. Remember
  1387. that this is a bit set - to test for C<EV_READ>, use C<< w->events &
  1388. EV_READ >>, and similarly for C<EV_WRITE>.
  1389. As with C<fd>, you must not modify this member even when the watcher is
  1390. stopped, always use C<ev_io_set> or C<ev_io_modify> for that.
  1391. =back
  1392. =head3 Examples
  1393. Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
  1394. readable, but only once. Since it is likely line-buffered, you could
  1395. attempt to read a whole line in the callback.
  1396. static void
  1397. stdin_readable_cb (struct ev_loop *loop, ev_io *w, int revents)
  1398. {
  1399. ev_io_stop (loop, w);
  1400. .. read from stdin here (or from w->fd) and handle any I/O errors
  1401. }
  1402. ...
  1403. struct ev_loop *loop = ev_default_init (0);
  1404. ev_io stdin_readable;
  1405. ev_io_init (&stdin_readable, stdin_readable_cb, STDIN_FILENO, EV_READ);
  1406. ev_io_start (loop, &stdin_readable);
  1407. ev_run (loop, 0);
  1408. =head2 C<ev_timer> - relative and optionally repeating timeouts
  1409. Timer watchers are simple relative timers that generate an event after a
  1410. given time, and optionally repeating in regular intervals after that.
  1411. The timers are based on real time, that is, if you register an event that
  1412. times out after an hour and you reset your system clock to January last
  1413. year, it will still time out after (roughly) one hour. "Roughly" because
  1414. detecting time jumps is hard, and some inaccuracies are unavoidable (the
  1415. monotonic clock option helps a lot here).
  1416. The callback is guaranteed to be invoked only I<after> its timeout has
  1417. passed (not I<at>, so on systems with very low-resolution clocks this
  1418. might introduce a small delay, see "the special problem of being too
  1419. early", below). If multiple timers become ready during the same loop
  1420. iteration then the ones with earlier time-out values are invoked before
  1421. ones of the same priority with later time-out values (but this is no
  1422. longer true when a callback calls C<ev_run> recursively).
  1423. =head3 Be smart about timeouts
  1424. Many real-world problems involve some kind of timeout, usually for error
  1425. recovery. A typical example is an HTTP request - if the other side hangs,
  1426. you want to raise some error after a while.
  1427. What follows are some ways to handle this problem, from obvious and
  1428. inefficient to smart and efficient.
  1429. In the following, a 60 second activity timeout is assumed - a timeout that
  1430. gets reset to 60 seconds each time there is activity (e.g. each time some
  1431. data or other life sign was received).
  1432. =over 4
  1433. =item 1. Use a timer and stop, reinitialise and start it on activity.
  1434. This is the most obvious, but not the most simple way: In the beginning,
  1435. start the watcher:
  1436. ev_timer_init (timer, callback, 60., 0.);
  1437. ev_timer_start (loop, timer);
  1438. Then, each time there is some activity, C<ev_timer_stop> it, initialise it
  1439. and start it again:
  1440. ev_timer_stop (loop, timer);
  1441. ev_timer_set (timer, 60., 0.);
  1442. ev_timer_start (loop, timer);
  1443. This is relatively simple to implement, but means that each time there is
  1444. some activity, libev will first have to remove the timer from its internal
  1445. data structure and then add it again. Libev tries to be fast, but it's
  1446. still not a constant-time operation.
  1447. =item 2. Use a timer and re-start it with C<ev_timer_again> inactivity.
  1448. This is the easiest way, and involves using C<ev_timer_again> instead of
  1449. C<ev_timer_start>.
  1450. To implement this, configure an C<ev_timer> with a C<repeat> value
  1451. of C<60> and then call C<ev_timer_again> at start and each time you
  1452. successfully read or write some data. If you go into an idle state where
  1453. you do not expect data to travel on the socket, you can C<ev_timer_stop>
  1454. the timer, and C<ev_timer_again> will automatically restart it if need be.
  1455. That means you can ignore both the C<ev_timer_start> function and the
  1456. C<after> argument to C<ev_timer_set>, and only ever use the C<repeat>
  1457. member and C<ev_timer_again>.
  1458. At start:
  1459. ev_init (timer, callback);
  1460. timer->repeat = 60.;
  1461. ev_timer_again (loop, timer);
  1462. Each time there is some activity:
  1463. ev_timer_again (loop, timer);
  1464. It is even possible to change the time-out on the fly, regardless of
  1465. whether the watcher is active or not:
  1466. timer->repeat = 30.;
  1467. ev_timer_again (loop, timer);
  1468. This is slightly more efficient then stopping/starting the timer each time
  1469. you want to modify its timeout value, as libev does not have to completely
  1470. remove and re-insert the timer from/into its internal data structure.
  1471. It is, however, even simpler than the "obvious" way to do it.
  1472. =item 3. Let the timer time out, but then re-arm it as required.
  1473. This method is more tricky, but usually most efficient: Most timeouts are
  1474. relatively long compared to the intervals between other activity - in
  1475. our example, within 60 seconds, there are usually many I/O events with
  1476. associated activity resets.
  1477. In this case, it would be more efficient to leave the C<ev_timer> alone,
  1478. but remember the time of last activity, and check for a real timeout only
  1479. within the callback:
  1480. ev_tstamp timeout = 60.;
  1481. ev_tstamp last_activity; // time of last activity
  1482. ev_timer timer;
  1483. static void
  1484. callback (EV_P_ ev_timer *w, int revents)
  1485. {
  1486. // calculate when the timeout would happen
  1487. ev_tstamp after = last_activity - ev_now (EV_A) + timeout;
  1488. // if negative, it means we the timeout already occurred
  1489. if (after < 0.)
  1490. {
  1491. // timeout occurred, take action
  1492. }
  1493. else
  1494. {
  1495. // callback was invoked, but there was some recent
  1496. // activity. simply restart the timer to time out
  1497. // after "after" seconds, which is the earliest time
  1498. // the timeout can occur.
  1499. ev_timer_set (w, after, 0.);
  1500. ev_timer_start (EV_A_ w);
  1501. }
  1502. }
  1503. To summarise the callback: first calculate in how many seconds the
  1504. timeout will occur (by calculating the absolute time when it would occur,
  1505. C<last_activity + timeout>, and subtracting the current time, C<ev_now
  1506. (EV_A)> from that).
  1507. If this value is negative, then we are already past the timeout, i.e. we
  1508. timed out, and need to do whatever is needed in this case.
  1509. Otherwise, we now the earliest time at which the timeout would trigger,
  1510. and simply start the timer with this timeout value.
  1511. In other words, each time the callback is invoked it will check whether
  1512. the timeout occurred. If not, it will simply reschedule itself to check
  1513. again at the earliest time it could time out. Rinse. Repeat.
  1514. This scheme causes more callback invocations (about one every 60 seconds
  1515. minus half the average time between activity), but virtually no calls to
  1516. libev to change the timeout.
  1517. To start the machinery, simply initialise the watcher and set
  1518. C<last_activity> to the current time (meaning there was some activity just
  1519. now), then call the callback, which will "do the right thing" and start
  1520. the timer:
  1521. last_activity = ev_now (EV_A);
  1522. ev_init (&timer, callback);
  1523. callback (EV_A_ &timer, 0);
  1524. When there is some activity, simply store the current time in
  1525. C<last_activity>, no libev calls at all:
  1526. if (activity detected)
  1527. last_activity = ev_now (EV_A);
  1528. When your timeout value changes, then the timeout can be changed by simply
  1529. providing a new value, stopping the timer and calling the callback, which
  1530. will again do the right thing (for example, time out immediately :).
  1531. timeout = new_value;
  1532. ev_timer_stop (EV_A_ &timer);
  1533. callback (EV_A_ &timer, 0);
  1534. This technique is slightly more complex, but in most cases where the
  1535. time-out is unlikely to be triggered, much more efficient.
  1536. =item 4. Wee, just use a double-linked list for your timeouts.
  1537. If there is not one request, but many thousands (millions...), all
  1538. employing some kind of timeout with the same timeout value, then one can
  1539. do even better:
  1540. When starting the timeout, calculate the timeout value and put the timeout
  1541. at the I<end> of the list.
  1542. Then use an C<ev_timer> to fire when the timeout at the I<beginning> of
  1543. the list is expected to fire (for example, using the technique #3).
  1544. When there is some activity, remove the timer from the list, recalculate
  1545. the timeout, append it to the end of the list again, and make sure to
  1546. update the C<ev_timer> if it was taken from the beginning of the list.
  1547. This way, one can manage an unlimited number of timeouts in O(1) time for
  1548. starting, stopping and updating the timers, at the expense of a major
  1549. complication, and having to use a constant timeout. The constant timeout
  1550. ensures that the list stays sorted.
  1551. =back
  1552. So which method the best?
  1553. Method #2 is a simple no-brain-required solution that is adequate in most
  1554. situations. Method #3 requires a bit more thinking, but handles many cases
  1555. better, and isn't very complicated either. In most case, choosing either
  1556. one is fine, with #3 being better in typical situations.
  1557. Method #1 is almost always a bad idea, and buys you nothing. Method #4 is
  1558. rather complicated, but extremely efficient, something that really pays
  1559. off after the first million or so of active timers, i.e. it's usually
  1560. overkill :)
  1561. =head3 The special problem of being too early
  1562. If you ask a timer to call your callback after three seconds, then
  1563. you expect it to be invoked after three seconds - but of course, this
  1564. cannot be guaranteed to infinite precision. Less obviously, it cannot be
  1565. guaranteed to any precision by libev - imagine somebody suspending the
  1566. process with a STOP signal for a few hours for example.
  1567. So, libev tries to invoke your callback as soon as possible I<after> the
  1568. delay has occurred, but cannot guarantee this.
  1569. A less obvious failure mode is calling your callback too early: many event
  1570. loops compare timestamps with a "elapsed delay >= requested delay", but
  1571. this can cause your callback to be invoked much earlier than you would
  1572. expect.
  1573. To see why, imagine a system with a clock that only offers full second
  1574. resolution (think windows if you can't come up with a broken enough OS
  1575. yourself). If you schedule a one-second timer at the time 500.9, then the
  1576. event loop will schedule your timeout to elapse at a system time of 500
  1577. (500.9 truncated to the resolution) + 1, or 501.
  1578. If an event library looks at the timeout 0.1s later, it will see "501 >=
  1579. 501" and invoke the callback 0.1s after it was started, even though a
  1580. one-second delay was requested - this is being "too early", despite best
  1581. intentions.
  1582. This is the reason why libev will never invoke the callback if the elapsed
  1583. delay equals the requested delay, but only when the elapsed delay is
  1584. larger than the requested delay. In the example above, libev would only invoke
  1585. the callback at system time 502, or 1.1s after the timer was started.
  1586. So, while libev cannot guarantee that your callback will be invoked
  1587. exactly when requested, it I<can> and I<does> guarantee that the requested
  1588. delay has actually elapsed, or in other words, it always errs on the "too
  1589. late" side of things.
  1590. =head3 The special problem of time updates
  1591. Establishing the current time is a costly operation (it usually takes
  1592. at least one system call): EV therefore updates its idea of the current
  1593. time only before and after C<ev_run> collects new events, which causes a
  1594. growing difference between C<ev_now ()> and C<ev_time ()> when handling
  1595. lots of events in one iteration.
  1596. The relative timeouts are calculated relative to the C<ev_now ()>
  1597. time. This is usually the right thing as this timestamp refers to the time
  1598. of the event triggering whatever timeout you are modifying/starting. If
  1599. you suspect event processing to be delayed and you I<need> to base the
  1600. timeout on the current time, use something like the following to adjust
  1601. for it:
  1602. ev_timer_set (&timer, after + (ev_time () - ev_now ()), 0.);
  1603. If the event loop is suspended for a long time, you can also force an
  1604. update of the time returned by C<ev_now ()> by calling C<ev_now_update
  1605. ()>, although that will push the event time of all outstanding events
  1606. further into the future.
  1607. =head3 The special problem of unsynchronised clocks
  1608. Modern systems have a variety of clocks - libev itself uses the normal
  1609. "wall clock" clock and, if available, the monotonic clock (to avoid time
  1610. jumps).
  1611. Neither of these clocks is synchronised with each other or any other clock
  1612. on the system, so C<ev_time ()> might return a considerably different time
  1613. than C<gettimeofday ()> or C<time ()>. On a GNU/Linux system, for example,
  1614. a call to C<gettimeofday> might return a second count that is one higher
  1615. than a directly following call to C<time>.
  1616. The moral of this is to only compare libev-related timestamps with
  1617. C<ev_time ()> and C<ev_now ()>, at least if you want better precision than
  1618. a second or so.
  1619. One more problem arises due to this lack of synchronisation: if libev uses
  1620. the system monotonic clock and you compare timestamps from C<ev_time>
  1621. or C<ev_now> from when you started your timer and when your callback is
  1622. invoked, you will find that sometimes the callback is a bit "early".
  1623. This is because C<ev_timer>s work in real time, not wall clock time, so
  1624. libev makes sure your callback is not invoked before the delay happened,
  1625. I<measured according to the real time>, not the system clock.
  1626. If your timeouts are based on a physical timescale (e.g. "time out this
  1627. connection after 100 seconds") then this shouldn't bother you as it is
  1628. exactly the right behaviour.
  1629. If you want to compare wall clock/system timestamps to your timers, then
  1630. you need to use C<ev_periodic>s, as these are based on the wall clock
  1631. time, where your comparisons will always generate correct results.
  1632. =head3 The special problems of suspended animation
  1633. When you leave the server world it is quite customary to hit machines that
  1634. can suspend/hibernate - what happens to the clocks during such a suspend?
  1635. Some quick tests made with a Linux 2.6.28 indicate that a suspend freezes
  1636. all processes, while the clocks (C<times>, C<CLOCK_MONOTONIC>) continue
  1637. to run until the system is suspended, but they will not advance while the
  1638. system is suspended. That means, on resume, it will be as if the program
  1639. was frozen for a few seconds, but the suspend time will not be counted
  1640. towards C<ev_timer> when a monotonic clock source is used. The real time
  1641. clock advanced as expected, but if it is used as sole clocksource, then a
  1642. long suspend would be detected as a time jump by libev, and timers would
  1643. be adjusted accordingly.
  1644. I would not be surprised to see different behaviour in different between
  1645. operating systems, OS versions or even different hardware.
  1646. The other form of suspend (job control, or sending a SIGSTOP) will see a
  1647. time jump in the monotonic clocks and the realtime clock. If the program
  1648. is suspended for a very long time, and monotonic clock sources are in use,
  1649. then you can expect C<ev_timer>s to expire as the full suspension time
  1650. will be counted towards the timers. When no monotonic clock source is in
  1651. use, then libev will again assume a timejump and adjust accordingly.
  1652. It might be beneficial for this latter case to call C<ev_suspend>
  1653. and C<ev_resume> in code that handles C<SIGTSTP>, to at least get
  1654. deterministic behaviour in this case (you can do nothing against
  1655. C<SIGSTOP>).
  1656. =head3 Watcher-Specific Functions and Data Members
  1657. =over 4
  1658. =item ev_timer_init (ev_timer *, callback, ev_tstamp after, ev_tstamp repeat)
  1659. =item ev_timer_set (ev_timer *, ev_tstamp after, ev_tstamp repeat)
  1660. Configure the timer to trigger after C<after> seconds (fractional and
  1661. negative values are supported). If C<repeat> is C<0.>, then it will
  1662. automatically be stopped once the timeout is reached. If it is positive,
  1663. then the timer will automatically be configured to trigger again C<repeat>
  1664. seconds later, again, and again, until stopped manually.
  1665. The timer itself will do a best-effort at avoiding drift, that is, if
  1666. you configure a timer to trigger every 10 seconds, then it will normally
  1667. trigger at exactly 10 second intervals. If, however, your program cannot
  1668. keep up with the timer (because it takes longer than those 10 seconds to
  1669. do stuff) the timer will not fire more than once per event loop iteration.
  1670. =item ev_timer_again (loop, ev_timer *)
  1671. This will act as if the timer timed out, and restarts it again if it is
  1672. repeating. It basically works like calling C<ev_timer_stop>, updating the
  1673. timeout to the C<repeat> value and calling C<ev_timer_start>.
  1674. The exact semantics are as in the following rules, all of which will be
  1675. applied to the watcher:
  1676. =over 4
  1677. =item If the timer is pending, the pending status is always cleared.
  1678. =item If the timer is started but non-repeating, stop it (as if it timed
  1679. out, without invoking it).
  1680. =item If the timer is repeating, make the C<repeat> value the new timeout
  1681. and start the timer, if necessary.
  1682. =back
  1683. This sounds a bit complicated, see L</Be smart about timeouts>, above, for a
  1684. usage example.
  1685. =item ev_tstamp ev_timer_remaining (loop, ev_timer *)
  1686. Returns the remaining time until a timer fires. If the timer is active,
  1687. then this time is relative to the current event loop time, otherwise it's
  1688. the timeout value currently configured.
  1689. That is, after an C<ev_timer_set (w, 5, 7)>, C<ev_timer_remaining> returns
  1690. C<5>. When the timer is started and one second passes, C<ev_timer_remaining>
  1691. will return C<4>. When the timer expires and is restarted, it will return
  1692. roughly C<7> (likely slightly less as callback invocation takes some time,
  1693. too), and so on.
  1694. =item ev_tstamp repeat [read-write]
  1695. The current C<repeat> value. Will be used each time the watcher times out
  1696. or C<ev_timer_again> is called, and determines the next timeout (if any),
  1697. which is also when any modifications are taken into account.
  1698. =back
  1699. =head3 Examples
  1700. Example: Create a timer that fires after 60 seconds.
  1701. static void
  1702. one_minute_cb (struct ev_loop *loop, ev_timer *w, int revents)
  1703. {
  1704. .. one minute over, w is actually stopped right here
  1705. }
  1706. ev_timer mytimer;
  1707. ev_timer_init (&mytimer, one_minute_cb, 60., 0.);
  1708. ev_timer_start (loop, &mytimer);
  1709. Example: Create a timeout timer that times out after 10 seconds of
  1710. inactivity.
  1711. static void
  1712. timeout_cb (struct ev_loop *loop, ev_timer *w, int revents)
  1713. {
  1714. .. ten seconds without any activity
  1715. }
  1716. ev_timer mytimer;
  1717. ev_timer_init (&mytimer, timeout_cb, 0., 10.); /* note, only repeat used */
  1718. ev_timer_again (&mytimer); /* start timer */
  1719. ev_run (loop, 0);
  1720. // and in some piece of code that gets executed on any "activity":
  1721. // reset the timeout to start ticking again at 10 seconds
  1722. ev_timer_again (&mytimer);
  1723. =head2 C<ev_periodic> - to cron or not to cron?
  1724. Periodic watchers are also timers of a kind, but they are very versatile
  1725. (and unfortunately a bit complex).
  1726. Unlike C<ev_timer>, periodic watchers are not based on real time (or
  1727. relative time, the physical time that passes) but on wall clock time
  1728. (absolute time, the thing you can read on your calendar or clock). The
  1729. difference is that wall clock time can run faster or slower than real
  1730. time, and time jumps are not uncommon (e.g. when you adjust your
  1731. wrist-watch).
  1732. You can tell a periodic watcher to trigger after some specific point
  1733. in time: for example, if you tell a periodic watcher to trigger "in 10
  1734. seconds" (by specifying e.g. C<ev_now () + 10.>, that is, an absolute time
  1735. not a delay) and then reset your system clock to January of the previous
  1736. year, then it will take a year or more to trigger the event (unlike an
  1737. C<ev_timer>, which would still trigger roughly 10 seconds after starting
  1738. it, as it uses a relative timeout).
  1739. C<ev_periodic> watchers can also be used to implement vastly more complex
  1740. timers, such as triggering an event on each "midnight, local time", or
  1741. other complicated rules. This cannot easily be done with C<ev_timer>
  1742. watchers, as those cannot react to time jumps.
  1743. As with timers, the callback is guaranteed to be invoked only when the
  1744. point in time where it is supposed to trigger has passed. If multiple
  1745. timers become ready during the same loop iteration then the ones with
  1746. earlier time-out values are invoked before ones with later time-out values
  1747. (but this is no longer true when a callback calls C<ev_run> recursively).
  1748. =head3 Watcher-Specific Functions and Data Members
  1749. =over 4
  1750. =item ev_periodic_init (ev_periodic *, callback, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
  1751. =item ev_periodic_set (ev_periodic *, ev_tstamp offset, ev_tstamp interval, reschedule_cb)
  1752. Lots of arguments, let's sort it out... There are basically three modes of
  1753. operation, and we will explain them from simplest to most complex:
  1754. =over 4
  1755. =item * absolute timer (offset = absolute time, interval = 0, reschedule_cb = 0)
  1756. In this configuration the watcher triggers an event after the wall clock
  1757. time C<offset> has passed. It will not repeat and will not adjust when a
  1758. time jump occurs, that is, if it is to be run at January 1st 2011 then it
  1759. will be stopped and invoked when the system clock reaches or surpasses
  1760. this point in time.
  1761. =item * repeating interval timer (offset = offset within interval, interval > 0, reschedule_cb = 0)
  1762. In this mode the watcher will always be scheduled to time out at the next
  1763. C<offset + N * interval> time (for some integer N, which can also be
  1764. negative) and then repeat, regardless of any time jumps. The C<offset>
  1765. argument is merely an offset into the C<interval> periods.
  1766. This can be used to create timers that do not drift with respect to the
  1767. system clock, for example, here is an C<ev_periodic> that triggers each
  1768. hour, on the hour (with respect to UTC):
  1769. ev_periodic_set (&periodic, 0., 3600., 0);
  1770. This doesn't mean there will always be 3600 seconds in between triggers,
  1771. but only that the callback will be called when the system time shows a
  1772. full hour (UTC), or more correctly, when the system time is evenly divisible
  1773. by 3600.
  1774. Another way to think about it (for the mathematically inclined) is that
  1775. C<ev_periodic> will try to run the callback in this mode at the next possible
  1776. time where C<time = offset (mod interval)>, regardless of any time jumps.
  1777. The C<interval> I<MUST> be positive, and for numerical stability, the
  1778. interval value should be higher than C<1/8192> (which is around 100
  1779. microseconds) and C<offset> should be higher than C<0> and should have
  1780. at most a similar magnitude as the current time (say, within a factor of
  1781. ten). Typical values for offset are, in fact, C<0> or something between
  1782. C<0> and C<interval>, which is also the recommended range.
  1783. Note also that there is an upper limit to how often a timer can fire (CPU
  1784. speed for example), so if C<interval> is very small then timing stability
  1785. will of course deteriorate. Libev itself tries to be exact to be about one
  1786. millisecond (if the OS supports it and the machine is fast enough).
  1787. =item * manual reschedule mode (offset ignored, interval ignored, reschedule_cb = callback)
  1788. In this mode the values for C<interval> and C<offset> are both being
  1789. ignored. Instead, each time the periodic watcher gets scheduled, the
  1790. reschedule callback will be called with the watcher as first, and the
  1791. current time as second argument.
  1792. NOTE: I<This callback MUST NOT stop or destroy any periodic watcher, ever,
  1793. or make ANY other event loop modifications whatsoever, unless explicitly
  1794. allowed by documentation here>.
  1795. If you need to stop it, return C<now + 1e30> (or so, fudge fudge) and stop
  1796. it afterwards (e.g. by starting an C<ev_prepare> watcher, which is the
  1797. only event loop modification you are allowed to do).
  1798. The callback prototype is C<ev_tstamp (*reschedule_cb)(ev_periodic
  1799. *w, ev_tstamp now)>, e.g.:
  1800. static ev_tstamp
  1801. my_rescheduler (ev_periodic *w, ev_tstamp now)
  1802. {
  1803. return now + 60.;
  1804. }
  1805. It must return the next time to trigger, based on the passed time value
  1806. (that is, the lowest time value larger than to the second argument). It
  1807. will usually be called just before the callback will be triggered, but
  1808. might be called at other times, too.
  1809. NOTE: I<< This callback must always return a time that is higher than or
  1810. equal to the passed C<now> value >>.
  1811. This can be used to create very complex timers, such as a timer that
  1812. triggers on "next midnight, local time". To do this, you would calculate
  1813. the next midnight after C<now> and return the timestamp value for
  1814. this. Here is a (completely untested, no error checking) example on how to
  1815. do this:
  1816. #include <time.h>
  1817. static ev_tstamp
  1818. my_rescheduler (ev_periodic *w, ev_tstamp now)
  1819. {
  1820. time_t tnow = (time_t)now;
  1821. struct tm tm;
  1822. localtime_r (&tnow, &tm);
  1823. tm.tm_sec = tm.tm_min = tm.tm_hour = 0; // midnight current day
  1824. ++tm.tm_mday; // midnight next day
  1825. return mktime (&tm);
  1826. }
  1827. Note: this code might run into trouble on days that have more then two
  1828. midnights (beginning and end).
  1829. =back
  1830. =item ev_periodic_again (loop, ev_periodic *)
  1831. Simply stops and restarts the periodic watcher again. This is only useful
  1832. when you changed some parameters or the reschedule callback would return
  1833. a different time than the last time it was called (e.g. in a crond like
  1834. program when the crontabs have changed).
  1835. =item ev_tstamp ev_periodic_at (ev_periodic *)
  1836. When active, returns the absolute time that the watcher is supposed
  1837. to trigger next. This is not the same as the C<offset> argument to
  1838. C<ev_periodic_set>, but indeed works even in interval and manual
  1839. rescheduling modes.
  1840. =item ev_tstamp offset [read-write]
  1841. When repeating, this contains the offset value, otherwise this is the
  1842. absolute point in time (the C<offset> value passed to C<ev_periodic_set>,
  1843. although libev might modify this value for better numerical stability).
  1844. Can be modified any time, but changes only take effect when the periodic
  1845. timer fires or C<ev_periodic_again> is being called.
  1846. =item ev_tstamp interval [read-write]
  1847. The current interval value. Can be modified any time, but changes only
  1848. take effect when the periodic timer fires or C<ev_periodic_again> is being
  1849. called.
  1850. =item ev_tstamp (*reschedule_cb)(ev_periodic *w, ev_tstamp now) [read-write]
  1851. The current reschedule callback, or C<0>, if this functionality is
  1852. switched off. Can be changed any time, but changes only take effect when
  1853. the periodic timer fires or C<ev_periodic_again> is being called.
  1854. =back
  1855. =head3 Examples
  1856. Example: Call a callback every hour, or, more precisely, whenever the
  1857. system time is divisible by 3600. The callback invocation times have
  1858. potentially a lot of jitter, but good long-term stability.
  1859. static void
  1860. clock_cb (struct ev_loop *loop, ev_periodic *w, int revents)
  1861. {
  1862. ... its now a full hour (UTC, or TAI or whatever your clock follows)
  1863. }
  1864. ev_periodic hourly_tick;
  1865. ev_periodic_init (&hourly_tick, clock_cb, 0., 3600., 0);
  1866. ev_periodic_start (loop, &hourly_tick);
  1867. Example: The same as above, but use a reschedule callback to do it:
  1868. #include <math.h>
  1869. static ev_tstamp
  1870. my_scheduler_cb (ev_periodic *w, ev_tstamp now)
  1871. {
  1872. return now + (3600. - fmod (now, 3600.));
  1873. }
  1874. ev_periodic_init (&hourly_tick, clock_cb, 0., 0., my_scheduler_cb);
  1875. Example: Call a callback every hour, starting now:
  1876. ev_periodic hourly_tick;
  1877. ev_periodic_init (&hourly_tick, clock_cb,
  1878. fmod (ev_now (loop), 3600.), 3600., 0);
  1879. ev_periodic_start (loop, &hourly_tick);
  1880. =head2 C<ev_signal> - signal me when a signal gets signalled!
  1881. Signal watchers will trigger an event when the process receives a specific
  1882. signal one or more times. Even though signals are very asynchronous, libev
  1883. will try its best to deliver signals synchronously, i.e. as part of the
  1884. normal event processing, like any other event.
  1885. If you want signals to be delivered truly asynchronously, just use
  1886. C<sigaction> as you would do without libev and forget about sharing
  1887. the signal. You can even use C<ev_async> from a signal handler to
  1888. synchronously wake up an event loop.
  1889. You can configure as many watchers as you like for the same signal, but
  1890. only within the same loop, i.e. you can watch for C<SIGINT> in your
  1891. default loop and for C<SIGIO> in another loop, but you cannot watch for
  1892. C<SIGINT> in both the default loop and another loop at the same time. At
  1893. the moment, C<SIGCHLD> is permanently tied to the default loop.
  1894. Only after the first watcher for a signal is started will libev actually
  1895. register something with the kernel. It thus coexists with your own signal
  1896. handlers as long as you don't register any with libev for the same signal.
  1897. If possible and supported, libev will install its handlers with
  1898. C<SA_RESTART> (or equivalent) behaviour enabled, so system calls should
  1899. not be unduly interrupted. If you have a problem with system calls getting
  1900. interrupted by signals you can block all signals in an C<ev_check> watcher
  1901. and unblock them in an C<ev_prepare> watcher.
  1902. =head3 The special problem of inheritance over fork/execve/pthread_create
  1903. Both the signal mask (C<sigprocmask>) and the signal disposition
  1904. (C<sigaction>) are unspecified after starting a signal watcher (and after
  1905. stopping it again), that is, libev might or might not block the signal,
  1906. and might or might not set or restore the installed signal handler (but
  1907. see C<EVFLAG_NOSIGMASK>).
  1908. While this does not matter for the signal disposition (libev never
  1909. sets signals to C<SIG_IGN>, so handlers will be reset to C<SIG_DFL> on
  1910. C<execve>), this matters for the signal mask: many programs do not expect
  1911. certain signals to be blocked.
  1912. This means that before calling C<exec> (from the child) you should reset
  1913. the signal mask to whatever "default" you expect (all clear is a good
  1914. choice usually).
  1915. The simplest way to ensure that the signal mask is reset in the child is
  1916. to install a fork handler with C<pthread_atfork> that resets it. That will
  1917. catch fork calls done by libraries (such as the libc) as well.
  1918. In current versions of libev, the signal will not be blocked indefinitely
  1919. unless you use the C<signalfd> API (C<EV_SIGNALFD>). While this reduces
  1920. the window of opportunity for problems, it will not go away, as libev
  1921. I<has> to modify the signal mask, at least temporarily.
  1922. So I can't stress this enough: I<If you do not reset your signal mask when
  1923. you expect it to be empty, you have a race condition in your code>. This
  1924. is not a libev-specific thing, this is true for most event libraries.
  1925. =head3 The special problem of threads signal handling
  1926. POSIX threads has problematic signal handling semantics, specifically,
  1927. a lot of functionality (sigfd, sigwait etc.) only really works if all
  1928. threads in a process block signals, which is hard to achieve.
  1929. When you want to use sigwait (or mix libev signal handling with your own
  1930. for the same signals), you can tackle this problem by globally blocking
  1931. all signals before creating any threads (or creating them with a fully set
  1932. sigprocmask) and also specifying the C<EVFLAG_NOSIGMASK> when creating
  1933. loops. Then designate one thread as "signal receiver thread" which handles
  1934. these signals. You can pass on any signals that libev might be interested
  1935. in by calling C<ev_feed_signal>.
  1936. =head3 Watcher-Specific Functions and Data Members
  1937. =over 4
  1938. =item ev_signal_init (ev_signal *, callback, int signum)
  1939. =item ev_signal_set (ev_signal *, int signum)
  1940. Configures the watcher to trigger on the given signal number (usually one
  1941. of the C<SIGxxx> constants).
  1942. =item int signum [read-only]
  1943. The signal the watcher watches out for.
  1944. =back
  1945. =head3 Examples
  1946. Example: Try to exit cleanly on SIGINT.
  1947. static void
  1948. sigint_cb (struct ev_loop *loop, ev_signal *w, int revents)
  1949. {
  1950. ev_break (loop, EVBREAK_ALL);
  1951. }
  1952. ev_signal signal_watcher;
  1953. ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
  1954. ev_signal_start (loop, &signal_watcher);
  1955. =head2 C<ev_child> - watch out for process status changes
  1956. Child watchers trigger when your process receives a SIGCHLD in response to
  1957. some child status changes (most typically when a child of yours dies or
  1958. exits). It is permissible to install a child watcher I<after> the child
  1959. has been forked (which implies it might have already exited), as long
  1960. as the event loop isn't entered (or is continued from a watcher), i.e.,
  1961. forking and then immediately registering a watcher for the child is fine,
  1962. but forking and registering a watcher a few event loop iterations later or
  1963. in the next callback invocation is not.
  1964. Only the default event loop is capable of handling signals, and therefore
  1965. you can only register child watchers in the default event loop.
  1966. Due to some design glitches inside libev, child watchers will always be
  1967. handled at maximum priority (their priority is set to C<EV_MAXPRI> by
  1968. libev)
  1969. =head3 Process Interaction
  1970. Libev grabs C<SIGCHLD> as soon as the default event loop is
  1971. initialised. This is necessary to guarantee proper behaviour even if the
  1972. first child watcher is started after the child exits. The occurrence
  1973. of C<SIGCHLD> is recorded asynchronously, but child reaping is done
  1974. synchronously as part of the event loop processing. Libev always reaps all
  1975. children, even ones not watched.
  1976. =head3 Overriding the Built-In Processing
  1977. Libev offers no special support for overriding the built-in child
  1978. processing, but if your application collides with libev's default child
  1979. handler, you can override it easily by installing your own handler for
  1980. C<SIGCHLD> after initialising the default loop, and making sure the
  1981. default loop never gets destroyed. You are encouraged, however, to use an
  1982. event-based approach to child reaping and thus use libev's support for
  1983. that, so other libev users can use C<ev_child> watchers freely.
  1984. =head3 Stopping the Child Watcher
  1985. Currently, the child watcher never gets stopped, even when the
  1986. child terminates, so normally one needs to stop the watcher in the
  1987. callback. Future versions of libev might stop the watcher automatically
  1988. when a child exit is detected (calling C<ev_child_stop> twice is not a
  1989. problem).
  1990. =head3 Watcher-Specific Functions and Data Members
  1991. =over 4
  1992. =item ev_child_init (ev_child *, callback, int pid, int trace)
  1993. =item ev_child_set (ev_child *, int pid, int trace)
  1994. Configures the watcher to wait for status changes of process C<pid> (or
  1995. I<any> process if C<pid> is specified as C<0>). The callback can look
  1996. at the C<rstatus> member of the C<ev_child> watcher structure to see
  1997. the status word (use the macros from C<sys/wait.h> and see your systems
  1998. C<waitpid> documentation). The C<rpid> member contains the pid of the
  1999. process causing the status change. C<trace> must be either C<0> (only
  2000. activate the watcher when the process terminates) or C<1> (additionally
  2001. activate the watcher when the process is stopped or continued).
  2002. =item int pid [read-only]
  2003. The process id this watcher watches out for, or C<0>, meaning any process id.
  2004. =item int rpid [read-write]
  2005. The process id that detected a status change.
  2006. =item int rstatus [read-write]
  2007. The process exit/trace status caused by C<rpid> (see your systems
  2008. C<waitpid> and C<sys/wait.h> documentation for details).
  2009. =back
  2010. =head3 Examples
  2011. Example: C<fork()> a new process and install a child handler to wait for
  2012. its completion.
  2013. ev_child cw;
  2014. static void
  2015. child_cb (EV_P_ ev_child *w, int revents)
  2016. {
  2017. ev_child_stop (EV_A_ w);
  2018. printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
  2019. }
  2020. pid_t pid = fork ();
  2021. if (pid < 0)
  2022. // error
  2023. else if (pid == 0)
  2024. {
  2025. // the forked child executes here
  2026. exit (1);
  2027. }
  2028. else
  2029. {
  2030. ev_child_init (&cw, child_cb, pid, 0);
  2031. ev_child_start (EV_DEFAULT_ &cw);
  2032. }
  2033. =head2 C<ev_stat> - did the file attributes just change?
  2034. This watches a file system path for attribute changes. That is, it calls
  2035. C<stat> on that path in regular intervals (or when the OS says it changed)
  2036. and sees if it changed compared to the last time, invoking the callback
  2037. if it did. Starting the watcher C<stat>'s the file, so only changes that
  2038. happen after the watcher has been started will be reported.
  2039. The path does not need to exist: changing from "path exists" to "path does
  2040. not exist" is a status change like any other. The condition "path does not
  2041. exist" (or more correctly "path cannot be stat'ed") is signified by the
  2042. C<st_nlink> field being zero (which is otherwise always forced to be at
  2043. least one) and all the other fields of the stat buffer having unspecified
  2044. contents.
  2045. The path I<must not> end in a slash or contain special components such as
  2046. C<.> or C<..>. The path I<should> be absolute: If it is relative and
  2047. your working directory changes, then the behaviour is undefined.
  2048. Since there is no portable change notification interface available, the
  2049. portable implementation simply calls C<stat(2)> regularly on the path
  2050. to see if it changed somehow. You can specify a recommended polling
  2051. interval for this case. If you specify a polling interval of C<0> (highly
  2052. recommended!) then a I<suitable, unspecified default> value will be used
  2053. (which you can expect to be around five seconds, although this might
  2054. change dynamically). Libev will also impose a minimum interval which is
  2055. currently around C<0.1>, but that's usually overkill.
  2056. This watcher type is not meant for massive numbers of stat watchers,
  2057. as even with OS-supported change notifications, this can be
  2058. resource-intensive.
  2059. At the time of this writing, the only OS-specific interface implemented
  2060. is the Linux inotify interface (implementing kqueue support is left as an
  2061. exercise for the reader. Note, however, that the author sees no way of
  2062. implementing C<ev_stat> semantics with kqueue, except as a hint).
  2063. =head3 ABI Issues (Largefile Support)
  2064. Libev by default (unless the user overrides this) uses the default
  2065. compilation environment, which means that on systems with large file
  2066. support disabled by default, you get the 32 bit version of the stat
  2067. structure. When using the library from programs that change the ABI to
  2068. use 64 bit file offsets the programs will fail. In that case you have to
  2069. compile libev with the same flags to get binary compatibility. This is
  2070. obviously the case with any flags that change the ABI, but the problem is
  2071. most noticeably displayed with ev_stat and large file support.
  2072. The solution for this is to lobby your distribution maker to make large
  2073. file interfaces available by default (as e.g. FreeBSD does) and not
  2074. optional. Libev cannot simply switch on large file support because it has
  2075. to exchange stat structures with application programs compiled using the
  2076. default compilation environment.
  2077. =head3 Inotify and Kqueue
  2078. When C<inotify (7)> support has been compiled into libev and present at
  2079. runtime, it will be used to speed up change detection where possible. The
  2080. inotify descriptor will be created lazily when the first C<ev_stat>
  2081. watcher is being started.
  2082. Inotify presence does not change the semantics of C<ev_stat> watchers
  2083. except that changes might be detected earlier, and in some cases, to avoid
  2084. making regular C<stat> calls. Even in the presence of inotify support
  2085. there are many cases where libev has to resort to regular C<stat> polling,
  2086. but as long as kernel 2.6.25 or newer is used (2.6.24 and older have too
  2087. many bugs), the path exists (i.e. stat succeeds), and the path resides on
  2088. a local filesystem (libev currently assumes only ext2/3, jfs, reiserfs and
  2089. xfs are fully working) libev usually gets away without polling.
  2090. There is no support for kqueue, as apparently it cannot be used to
  2091. implement this functionality, due to the requirement of having a file
  2092. descriptor open on the object at all times, and detecting renames, unlinks
  2093. etc. is difficult.
  2094. =head3 C<stat ()> is a synchronous operation
  2095. Libev doesn't normally do any kind of I/O itself, and so is not blocking
  2096. the process. The exception are C<ev_stat> watchers - those call C<stat
  2097. ()>, which is a synchronous operation.
  2098. For local paths, this usually doesn't matter: unless the system is very
  2099. busy or the intervals between stat's are large, a stat call will be fast,
  2100. as the path data is usually in memory already (except when starting the
  2101. watcher).
  2102. For networked file systems, calling C<stat ()> can block an indefinite
  2103. time due to network issues, and even under good conditions, a stat call
  2104. often takes multiple milliseconds.
  2105. Therefore, it is best to avoid using C<ev_stat> watchers on networked
  2106. paths, although this is fully supported by libev.
  2107. =head3 The special problem of stat time resolution
  2108. The C<stat ()> system call only supports full-second resolution portably,
  2109. and even on systems where the resolution is higher, most file systems
  2110. still only support whole seconds.
  2111. That means that, if the time is the only thing that changes, you can
  2112. easily miss updates: on the first update, C<ev_stat> detects a change and
  2113. calls your callback, which does something. When there is another update
  2114. within the same second, C<ev_stat> will be unable to detect unless the
  2115. stat data does change in other ways (e.g. file size).
  2116. The solution to this is to delay acting on a change for slightly more
  2117. than a second (or till slightly after the next full second boundary), using
  2118. a roughly one-second-delay C<ev_timer> (e.g. C<ev_timer_set (w, 0., 1.02);
  2119. ev_timer_again (loop, w)>).
  2120. The C<.02> offset is added to work around small timing inconsistencies
  2121. of some operating systems (where the second counter of the current time
  2122. might be be delayed. One such system is the Linux kernel, where a call to
  2123. C<gettimeofday> might return a timestamp with a full second later than
  2124. a subsequent C<time> call - if the equivalent of C<time ()> is used to
  2125. update file times then there will be a small window where the kernel uses
  2126. the previous second to update file times but libev might already execute
  2127. the timer callback).
  2128. =head3 Watcher-Specific Functions and Data Members
  2129. =over 4
  2130. =item ev_stat_init (ev_stat *, callback, const char *path, ev_tstamp interval)
  2131. =item ev_stat_set (ev_stat *, const char *path, ev_tstamp interval)
  2132. Configures the watcher to wait for status changes of the given
  2133. C<path>. The C<interval> is a hint on how quickly a change is expected to
  2134. be detected and should normally be specified as C<0> to let libev choose
  2135. a suitable value. The memory pointed to by C<path> must point to the same
  2136. path for as long as the watcher is active.
  2137. The callback will receive an C<EV_STAT> event when a change was detected,
  2138. relative to the attributes at the time the watcher was started (or the
  2139. last change was detected).
  2140. =item ev_stat_stat (loop, ev_stat *)
  2141. Updates the stat buffer immediately with new values. If you change the
  2142. watched path in your callback, you could call this function to avoid
  2143. detecting this change (while introducing a race condition if you are not
  2144. the only one changing the path). Can also be useful simply to find out the
  2145. new values.
  2146. =item ev_statdata attr [read-only]
  2147. The most-recently detected attributes of the file. Although the type is
  2148. C<ev_statdata>, this is usually the (or one of the) C<struct stat> types
  2149. suitable for your system, but you can only rely on the POSIX-standardised
  2150. members to be present. If the C<st_nlink> member is C<0>, then there was
  2151. some error while C<stat>ing the file.
  2152. =item ev_statdata prev [read-only]
  2153. The previous attributes of the file. The callback gets invoked whenever
  2154. C<prev> != C<attr>, or, more precisely, one or more of these members
  2155. differ: C<st_dev>, C<st_ino>, C<st_mode>, C<st_nlink>, C<st_uid>,
  2156. C<st_gid>, C<st_rdev>, C<st_size>, C<st_atime>, C<st_mtime>, C<st_ctime>.
  2157. =item ev_tstamp interval [read-only]
  2158. The specified interval.
  2159. =item const char *path [read-only]
  2160. The file system path that is being watched.
  2161. =back
  2162. =head3 Examples
  2163. Example: Watch C</etc/passwd> for attribute changes.
  2164. static void
  2165. passwd_cb (struct ev_loop *loop, ev_stat *w, int revents)
  2166. {
  2167. /* /etc/passwd changed in some way */
  2168. if (w->attr.st_nlink)
  2169. {
  2170. printf ("passwd current size %ld\n", (long)w->attr.st_size);
  2171. printf ("passwd current atime %ld\n", (long)w->attr.st_mtime);
  2172. printf ("passwd current mtime %ld\n", (long)w->attr.st_mtime);
  2173. }
  2174. else
  2175. /* you shalt not abuse printf for puts */
  2176. puts ("wow, /etc/passwd is not there, expect problems. "
  2177. "if this is windows, they already arrived\n");
  2178. }
  2179. ...
  2180. ev_stat passwd;
  2181. ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
  2182. ev_stat_start (loop, &passwd);
  2183. Example: Like above, but additionally use a one-second delay so we do not
  2184. miss updates (however, frequent updates will delay processing, too, so
  2185. one might do the work both on C<ev_stat> callback invocation I<and> on
  2186. C<ev_timer> callback invocation).
  2187. static ev_stat passwd;
  2188. static ev_timer timer;
  2189. static void
  2190. timer_cb (EV_P_ ev_timer *w, int revents)
  2191. {
  2192. ev_timer_stop (EV_A_ w);
  2193. /* now it's one second after the most recent passwd change */
  2194. }
  2195. static void
  2196. stat_cb (EV_P_ ev_stat *w, int revents)
  2197. {
  2198. /* reset the one-second timer */
  2199. ev_timer_again (EV_A_ &timer);
  2200. }
  2201. ...
  2202. ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
  2203. ev_stat_start (loop, &passwd);
  2204. ev_timer_init (&timer, timer_cb, 0., 1.02);
  2205. =head2 C<ev_idle> - when you've got nothing better to do...
  2206. Idle watchers trigger events when no other events of the same or higher
  2207. priority are pending (prepare, check and other idle watchers do not count
  2208. as receiving "events").
  2209. That is, as long as your process is busy handling sockets or timeouts
  2210. (or even signals, imagine) of the same or higher priority it will not be
  2211. triggered. But when your process is idle (or only lower-priority watchers
  2212. are pending), the idle watchers are being called once per event loop
  2213. iteration - until stopped, that is, or your process receives more events
  2214. and becomes busy again with higher priority stuff.
  2215. The most noteworthy effect is that as long as any idle watchers are
  2216. active, the process will not block when waiting for new events.
  2217. Apart from keeping your process non-blocking (which is a useful
  2218. effect on its own sometimes), idle watchers are a good place to do
  2219. "pseudo-background processing", or delay processing stuff to after the
  2220. event loop has handled all outstanding events.
  2221. =head3 Abusing an C<ev_idle> watcher for its side-effect
  2222. As long as there is at least one active idle watcher, libev will never
  2223. sleep unnecessarily. Or in other words, it will loop as fast as possible.
  2224. For this to work, the idle watcher doesn't need to be invoked at all - the
  2225. lowest priority will do.
  2226. This mode of operation can be useful together with an C<ev_check> watcher,
  2227. to do something on each event loop iteration - for example to balance load
  2228. between different connections.
  2229. See L</Abusing an ev_check watcher for its side-effect> for a longer
  2230. example.
  2231. =head3 Watcher-Specific Functions and Data Members
  2232. =over 4
  2233. =item ev_idle_init (ev_idle *, callback)
  2234. Initialises and configures the idle watcher - it has no parameters of any
  2235. kind. There is a C<ev_idle_set> macro, but using it is utterly pointless,
  2236. believe me.
  2237. =back
  2238. =head3 Examples
  2239. Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
  2240. callback, free it. Also, use no error checking, as usual.
  2241. static void
  2242. idle_cb (struct ev_loop *loop, ev_idle *w, int revents)
  2243. {
  2244. // stop the watcher
  2245. ev_idle_stop (loop, w);
  2246. // now we can free it
  2247. free (w);
  2248. // now do something you wanted to do when the program has
  2249. // no longer anything immediate to do.
  2250. }
  2251. ev_idle *idle_watcher = malloc (sizeof (ev_idle));
  2252. ev_idle_init (idle_watcher, idle_cb);
  2253. ev_idle_start (loop, idle_watcher);
  2254. =head2 C<ev_prepare> and C<ev_check> - customise your event loop!
  2255. Prepare and check watchers are often (but not always) used in pairs:
  2256. prepare watchers get invoked before the process blocks and check watchers
  2257. afterwards.
  2258. You I<must not> call C<ev_run> (or similar functions that enter the
  2259. current event loop) or C<ev_loop_fork> from either C<ev_prepare> or
  2260. C<ev_check> watchers. Other loops than the current one are fine,
  2261. however. The rationale behind this is that you do not need to check
  2262. for recursion in those watchers, i.e. the sequence will always be
  2263. C<ev_prepare>, blocking, C<ev_check> so if you have one watcher of each
  2264. kind they will always be called in pairs bracketing the blocking call.
  2265. Their main purpose is to integrate other event mechanisms into libev and
  2266. their use is somewhat advanced. They could be used, for example, to track
  2267. variable changes, implement your own watchers, integrate net-snmp or a
  2268. coroutine library and lots more. They are also occasionally useful if
  2269. you cache some data and want to flush it before blocking (for example,
  2270. in X programs you might want to do an C<XFlush ()> in an C<ev_prepare>
  2271. watcher).
  2272. This is done by examining in each prepare call which file descriptors
  2273. need to be watched by the other library, registering C<ev_io> watchers
  2274. for them and starting an C<ev_timer> watcher for any timeouts (many
  2275. libraries provide exactly this functionality). Then, in the check watcher,
  2276. you check for any events that occurred (by checking the pending status
  2277. of all watchers and stopping them) and call back into the library. The
  2278. I/O and timer callbacks will never actually be called (but must be valid
  2279. nevertheless, because you never know, you know?).
  2280. As another example, the Perl Coro module uses these hooks to integrate
  2281. coroutines into libev programs, by yielding to other active coroutines
  2282. during each prepare and only letting the process block if no coroutines
  2283. are ready to run (it's actually more complicated: it only runs coroutines
  2284. with priority higher than or equal to the event loop and one coroutine
  2285. of lower priority, but only once, using idle watchers to keep the event
  2286. loop from blocking if lower-priority coroutines are active, thus mapping
  2287. low-priority coroutines to idle/background tasks).
  2288. When used for this purpose, it is recommended to give C<ev_check> watchers
  2289. highest (C<EV_MAXPRI>) priority, to ensure that they are being run before
  2290. any other watchers after the poll (this doesn't matter for C<ev_prepare>
  2291. watchers).
  2292. Also, C<ev_check> watchers (and C<ev_prepare> watchers, too) should not
  2293. activate ("feed") events into libev. While libev fully supports this, they
  2294. might get executed before other C<ev_check> watchers did their job. As
  2295. C<ev_check> watchers are often used to embed other (non-libev) event
  2296. loops those other event loops might be in an unusable state until their
  2297. C<ev_check> watcher ran (always remind yourself to coexist peacefully with
  2298. others).
  2299. =head3 Abusing an C<ev_check> watcher for its side-effect
  2300. C<ev_check> (and less often also C<ev_prepare>) watchers can also be
  2301. useful because they are called once per event loop iteration. For
  2302. example, if you want to handle a large number of connections fairly, you
  2303. normally only do a bit of work for each active connection, and if there
  2304. is more work to do, you wait for the next event loop iteration, so other
  2305. connections have a chance of making progress.
  2306. Using an C<ev_check> watcher is almost enough: it will be called on the
  2307. next event loop iteration. However, that isn't as soon as possible -
  2308. without external events, your C<ev_check> watcher will not be invoked.
  2309. This is where C<ev_idle> watchers come in handy - all you need is a
  2310. single global idle watcher that is active as long as you have one active
  2311. C<ev_check> watcher. The C<ev_idle> watcher makes sure the event loop
  2312. will not sleep, and the C<ev_check> watcher makes sure a callback gets
  2313. invoked. Neither watcher alone can do that.
  2314. =head3 Watcher-Specific Functions and Data Members
  2315. =over 4
  2316. =item ev_prepare_init (ev_prepare *, callback)
  2317. =item ev_check_init (ev_check *, callback)
  2318. Initialises and configures the prepare or check watcher - they have no
  2319. parameters of any kind. There are C<ev_prepare_set> and C<ev_check_set>
  2320. macros, but using them is utterly, utterly, utterly and completely
  2321. pointless.
  2322. =back
  2323. =head3 Examples
  2324. There are a number of principal ways to embed other event loops or modules
  2325. into libev. Here are some ideas on how to include libadns into libev
  2326. (there is a Perl module named C<EV::ADNS> that does this, which you could
  2327. use as a working example. Another Perl module named C<EV::Glib> embeds a
  2328. Glib main context into libev, and finally, C<Glib::EV> embeds EV into the
  2329. Glib event loop).
  2330. Method 1: Add IO watchers and a timeout watcher in a prepare handler,
  2331. and in a check watcher, destroy them and call into libadns. What follows
  2332. is pseudo-code only of course. This requires you to either use a low
  2333. priority for the check watcher or use C<ev_clear_pending> explicitly, as
  2334. the callbacks for the IO/timeout watchers might not have been called yet.
  2335. static ev_io iow [nfd];
  2336. static ev_timer tw;
  2337. static void
  2338. io_cb (struct ev_loop *loop, ev_io *w, int revents)
  2339. {
  2340. }
  2341. // create io watchers for each fd and a timer before blocking
  2342. static void
  2343. adns_prepare_cb (struct ev_loop *loop, ev_prepare *w, int revents)
  2344. {
  2345. int timeout = 3600000;
  2346. struct pollfd fds [nfd];
  2347. // actual code will need to loop here and realloc etc.
  2348. adns_beforepoll (ads, fds, &nfd, &timeout, timeval_from (ev_time ()));
  2349. /* the callback is illegal, but won't be called as we stop during check */
  2350. ev_timer_init (&tw, 0, timeout * 1e-3, 0.);
  2351. ev_timer_start (loop, &tw);
  2352. // create one ev_io per pollfd
  2353. for (int i = 0; i < nfd; ++i)
  2354. {
  2355. ev_io_init (iow + i, io_cb, fds [i].fd,
  2356. ((fds [i].events & POLLIN ? EV_READ : 0)
  2357. | (fds [i].events & POLLOUT ? EV_WRITE : 0)));
  2358. fds [i].revents = 0;
  2359. ev_io_start (loop, iow + i);
  2360. }
  2361. }
  2362. // stop all watchers after blocking
  2363. static void
  2364. adns_check_cb (struct ev_loop *loop, ev_check *w, int revents)
  2365. {
  2366. ev_timer_stop (loop, &tw);
  2367. for (int i = 0; i < nfd; ++i)
  2368. {
  2369. // set the relevant poll flags
  2370. // could also call adns_processreadable etc. here
  2371. struct pollfd *fd = fds + i;
  2372. int revents = ev_clear_pending (iow + i);
  2373. if (revents & EV_READ ) fd->revents |= fd->events & POLLIN;
  2374. if (revents & EV_WRITE) fd->revents |= fd->events & POLLOUT;
  2375. // now stop the watcher
  2376. ev_io_stop (loop, iow + i);
  2377. }
  2378. adns_afterpoll (adns, fds, nfd, timeval_from (ev_now (loop));
  2379. }
  2380. Method 2: This would be just like method 1, but you run C<adns_afterpoll>
  2381. in the prepare watcher and would dispose of the check watcher.
  2382. Method 3: If the module to be embedded supports explicit event
  2383. notification (libadns does), you can also make use of the actual watcher
  2384. callbacks, and only destroy/create the watchers in the prepare watcher.
  2385. static void
  2386. timer_cb (EV_P_ ev_timer *w, int revents)
  2387. {
  2388. adns_state ads = (adns_state)w->data;
  2389. update_now (EV_A);
  2390. adns_processtimeouts (ads, &tv_now);
  2391. }
  2392. static void
  2393. io_cb (EV_P_ ev_io *w, int revents)
  2394. {
  2395. adns_state ads = (adns_state)w->data;
  2396. update_now (EV_A);
  2397. if (revents & EV_READ ) adns_processreadable (ads, w->fd, &tv_now);
  2398. if (revents & EV_WRITE) adns_processwriteable (ads, w->fd, &tv_now);
  2399. }
  2400. // do not ever call adns_afterpoll
  2401. Method 4: Do not use a prepare or check watcher because the module you
  2402. want to embed is not flexible enough to support it. Instead, you can
  2403. override their poll function. The drawback with this solution is that the
  2404. main loop is now no longer controllable by EV. The C<Glib::EV> module uses
  2405. this approach, effectively embedding EV as a client into the horrible
  2406. libglib event loop.
  2407. static gint
  2408. event_poll_func (GPollFD *fds, guint nfds, gint timeout)
  2409. {
  2410. int got_events = 0;
  2411. for (n = 0; n < nfds; ++n)
  2412. // create/start io watcher that sets the relevant bits in fds[n] and increment got_events
  2413. if (timeout >= 0)
  2414. // create/start timer
  2415. // poll
  2416. ev_run (EV_A_ 0);
  2417. // stop timer again
  2418. if (timeout >= 0)
  2419. ev_timer_stop (EV_A_ &to);
  2420. // stop io watchers again - their callbacks should have set
  2421. for (n = 0; n < nfds; ++n)
  2422. ev_io_stop (EV_A_ iow [n]);
  2423. return got_events;
  2424. }
  2425. =head2 C<ev_embed> - when one backend isn't enough...
  2426. This is a rather advanced watcher type that lets you embed one event loop
  2427. into another (currently only C<ev_io> events are supported in the embedded
  2428. loop, other types of watchers might be handled in a delayed or incorrect
  2429. fashion and must not be used).
  2430. There are primarily two reasons you would want that: work around bugs and
  2431. prioritise I/O.
  2432. As an example for a bug workaround, the kqueue backend might only support
  2433. sockets on some platform, so it is unusable as generic backend, but you
  2434. still want to make use of it because you have many sockets and it scales
  2435. so nicely. In this case, you would create a kqueue-based loop and embed
  2436. it into your default loop (which might use e.g. poll). Overall operation
  2437. will be a bit slower because first libev has to call C<poll> and then
  2438. C<kevent>, but at least you can use both mechanisms for what they are
  2439. best: C<kqueue> for scalable sockets and C<poll> if you want it to work :)
  2440. As for prioritising I/O: under rare circumstances you have the case where
  2441. some fds have to be watched and handled very quickly (with low latency),
  2442. and even priorities and idle watchers might have too much overhead. In
  2443. this case you would put all the high priority stuff in one loop and all
  2444. the rest in a second one, and embed the second one in the first.
  2445. As long as the watcher is active, the callback will be invoked every
  2446. time there might be events pending in the embedded loop. The callback
  2447. must then call C<ev_embed_sweep (mainloop, watcher)> to make a single
  2448. sweep and invoke their callbacks (the callback doesn't need to invoke the
  2449. C<ev_embed_sweep> function directly, it could also start an idle watcher
  2450. to give the embedded loop strictly lower priority for example).
  2451. You can also set the callback to C<0>, in which case the embed watcher
  2452. will automatically execute the embedded loop sweep whenever necessary.
  2453. Fork detection will be handled transparently while the C<ev_embed> watcher
  2454. is active, i.e., the embedded loop will automatically be forked when the
  2455. embedding loop forks. In other cases, the user is responsible for calling
  2456. C<ev_loop_fork> on the embedded loop.
  2457. Unfortunately, not all backends are embeddable: only the ones returned by
  2458. C<ev_embeddable_backends> are, which, unfortunately, does not include any
  2459. portable one.
  2460. So when you want to use this feature you will always have to be prepared
  2461. that you cannot get an embeddable loop. The recommended way to get around
  2462. this is to have a separate variables for your embeddable loop, try to
  2463. create it, and if that fails, use the normal loop for everything.
  2464. =head3 C<ev_embed> and fork
  2465. While the C<ev_embed> watcher is running, forks in the embedding loop will
  2466. automatically be applied to the embedded loop as well, so no special
  2467. fork handling is required in that case. When the watcher is not running,
  2468. however, it is still the task of the libev user to call C<ev_loop_fork ()>
  2469. as applicable.
  2470. =head3 Watcher-Specific Functions and Data Members
  2471. =over 4
  2472. =item ev_embed_init (ev_embed *, callback, struct ev_loop *embedded_loop)
  2473. =item ev_embed_set (ev_embed *, struct ev_loop *embedded_loop)
  2474. Configures the watcher to embed the given loop, which must be
  2475. embeddable. If the callback is C<0>, then C<ev_embed_sweep> will be
  2476. invoked automatically, otherwise it is the responsibility of the callback
  2477. to invoke it (it will continue to be called until the sweep has been done,
  2478. if you do not want that, you need to temporarily stop the embed watcher).
  2479. =item ev_embed_sweep (loop, ev_embed *)
  2480. Make a single, non-blocking sweep over the embedded loop. This works
  2481. similarly to C<ev_run (embedded_loop, EVRUN_NOWAIT)>, but in the most
  2482. appropriate way for embedded loops.
  2483. =item struct ev_loop *other [read-only]
  2484. The embedded event loop.
  2485. =back
  2486. =head3 Examples
  2487. Example: Try to get an embeddable event loop and embed it into the default
  2488. event loop. If that is not possible, use the default loop. The default
  2489. loop is stored in C<loop_hi>, while the embeddable loop is stored in
  2490. C<loop_lo> (which is C<loop_hi> in the case no embeddable loop can be
  2491. used).
  2492. struct ev_loop *loop_hi = ev_default_init (0);
  2493. struct ev_loop *loop_lo = 0;
  2494. ev_embed embed;
  2495. // see if there is a chance of getting one that works
  2496. // (remember that a flags value of 0 means autodetection)
  2497. loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
  2498. ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
  2499. : 0;
  2500. // if we got one, then embed it, otherwise default to loop_hi
  2501. if (loop_lo)
  2502. {
  2503. ev_embed_init (&embed, 0, loop_lo);
  2504. ev_embed_start (loop_hi, &embed);
  2505. }
  2506. else
  2507. loop_lo = loop_hi;
  2508. Example: Check if kqueue is available but not recommended and create
  2509. a kqueue backend for use with sockets (which usually work with any
  2510. kqueue implementation). Store the kqueue/socket-only event loop in
  2511. C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
  2512. struct ev_loop *loop = ev_default_init (0);
  2513. struct ev_loop *loop_socket = 0;
  2514. ev_embed embed;
  2515. if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
  2516. if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
  2517. {
  2518. ev_embed_init (&embed, 0, loop_socket);
  2519. ev_embed_start (loop, &embed);
  2520. }
  2521. if (!loop_socket)
  2522. loop_socket = loop;
  2523. // now use loop_socket for all sockets, and loop for everything else
  2524. =head2 C<ev_fork> - the audacity to resume the event loop after a fork
  2525. Fork watchers are called when a C<fork ()> was detected (usually because
  2526. whoever is a good citizen cared to tell libev about it by calling
  2527. C<ev_loop_fork>). The invocation is done before the event loop blocks next
  2528. and before C<ev_check> watchers are being called, and only in the child
  2529. after the fork. If whoever good citizen calling C<ev_default_fork> cheats
  2530. and calls it in the wrong process, the fork handlers will be invoked, too,
  2531. of course.
  2532. =head3 The special problem of life after fork - how is it possible?
  2533. Most uses of C<fork ()> consist of forking, then some simple calls to set
  2534. up/change the process environment, followed by a call to C<exec()>. This
  2535. sequence should be handled by libev without any problems.
  2536. This changes when the application actually wants to do event handling
  2537. in the child, or both parent in child, in effect "continuing" after the
  2538. fork.
  2539. The default mode of operation (for libev, with application help to detect
  2540. forks) is to duplicate all the state in the child, as would be expected
  2541. when I<either> the parent I<or> the child process continues.
  2542. When both processes want to continue using libev, then this is usually the
  2543. wrong result. In that case, usually one process (typically the parent) is
  2544. supposed to continue with all watchers in place as before, while the other
  2545. process typically wants to start fresh, i.e. without any active watchers.
  2546. The cleanest and most efficient way to achieve that with libev is to
  2547. simply create a new event loop, which of course will be "empty", and
  2548. use that for new watchers. This has the advantage of not touching more
  2549. memory than necessary, and thus avoiding the copy-on-write, and the
  2550. disadvantage of having to use multiple event loops (which do not support
  2551. signal watchers).
  2552. When this is not possible, or you want to use the default loop for
  2553. other reasons, then in the process that wants to start "fresh", call
  2554. C<ev_loop_destroy (EV_DEFAULT)> followed by C<ev_default_loop (...)>.
  2555. Destroying the default loop will "orphan" (not stop) all registered
  2556. watchers, so you have to be careful not to execute code that modifies
  2557. those watchers. Note also that in that case, you have to re-register any
  2558. signal watchers.
  2559. =head3 Watcher-Specific Functions and Data Members
  2560. =over 4
  2561. =item ev_fork_init (ev_fork *, callback)
  2562. Initialises and configures the fork watcher - it has no parameters of any
  2563. kind. There is a C<ev_fork_set> macro, but using it is utterly pointless,
  2564. really.
  2565. =back
  2566. =head2 C<ev_cleanup> - even the best things end
  2567. Cleanup watchers are called just before the event loop is being destroyed
  2568. by a call to C<ev_loop_destroy>.
  2569. While there is no guarantee that the event loop gets destroyed, cleanup
  2570. watchers provide a convenient method to install cleanup hooks for your
  2571. program, worker threads and so on - you just to make sure to destroy the
  2572. loop when you want them to be invoked.
  2573. Cleanup watchers are invoked in the same way as any other watcher. Unlike
  2574. all other watchers, they do not keep a reference to the event loop (which
  2575. makes a lot of sense if you think about it). Like all other watchers, you
  2576. can call libev functions in the callback, except C<ev_cleanup_start>.
  2577. =head3 Watcher-Specific Functions and Data Members
  2578. =over 4
  2579. =item ev_cleanup_init (ev_cleanup *, callback)
  2580. Initialises and configures the cleanup watcher - it has no parameters of
  2581. any kind. There is a C<ev_cleanup_set> macro, but using it is utterly
  2582. pointless, I assure you.
  2583. =back
  2584. Example: Register an atexit handler to destroy the default loop, so any
  2585. cleanup functions are called.
  2586. static void
  2587. program_exits (void)
  2588. {
  2589. ev_loop_destroy (EV_DEFAULT_UC);
  2590. }
  2591. ...
  2592. atexit (program_exits);
  2593. =head2 C<ev_async> - how to wake up an event loop
  2594. In general, you cannot use an C<ev_loop> from multiple threads or other
  2595. asynchronous sources such as signal handlers (as opposed to multiple event
  2596. loops - those are of course safe to use in different threads).
  2597. Sometimes, however, you need to wake up an event loop you do not control,
  2598. for example because it belongs to another thread. This is what C<ev_async>
  2599. watchers do: as long as the C<ev_async> watcher is active, you can signal
  2600. it by calling C<ev_async_send>, which is thread- and signal safe.
  2601. This functionality is very similar to C<ev_signal> watchers, as signals,
  2602. too, are asynchronous in nature, and signals, too, will be compressed
  2603. (i.e. the number of callback invocations may be less than the number of
  2604. C<ev_async_send> calls). In fact, you could use signal watchers as a kind
  2605. of "global async watchers" by using a watcher on an otherwise unused
  2606. signal, and C<ev_feed_signal> to signal this watcher from another thread,
  2607. even without knowing which loop owns the signal.
  2608. =head3 Queueing
  2609. C<ev_async> does not support queueing of data in any way. The reason
  2610. is that the author does not know of a simple (or any) algorithm for a
  2611. multiple-writer-single-reader queue that works in all cases and doesn't
  2612. need elaborate support such as pthreads or unportable memory access
  2613. semantics.
  2614. That means that if you want to queue data, you have to provide your own
  2615. queue. But at least I can tell you how to implement locking around your
  2616. queue:
  2617. =over 4
  2618. =item queueing from a signal handler context
  2619. To implement race-free queueing, you simply add to the queue in the signal
  2620. handler but you block the signal handler in the watcher callback. Here is
  2621. an example that does that for some fictitious SIGUSR1 handler:
  2622. static ev_async mysig;
  2623. static void
  2624. sigusr1_handler (void)
  2625. {
  2626. sometype data;
  2627. // no locking etc.
  2628. queue_put (data);
  2629. ev_async_send (EV_DEFAULT_ &mysig);
  2630. }
  2631. static void
  2632. mysig_cb (EV_P_ ev_async *w, int revents)
  2633. {
  2634. sometype data;
  2635. sigset_t block, prev;
  2636. sigemptyset (&block);
  2637. sigaddset (&block, SIGUSR1);
  2638. sigprocmask (SIG_BLOCK, &block, &prev);
  2639. while (queue_get (&data))
  2640. process (data);
  2641. if (sigismember (&prev, SIGUSR1)
  2642. sigprocmask (SIG_UNBLOCK, &block, 0);
  2643. }
  2644. (Note: pthreads in theory requires you to use C<pthread_setmask>
  2645. instead of C<sigprocmask> when you use threads, but libev doesn't do it
  2646. either...).
  2647. =item queueing from a thread context
  2648. The strategy for threads is different, as you cannot (easily) block
  2649. threads but you can easily preempt them, so to queue safely you need to
  2650. employ a traditional mutex lock, such as in this pthread example:
  2651. static ev_async mysig;
  2652. static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
  2653. static void
  2654. otherthread (void)
  2655. {
  2656. // only need to lock the actual queueing operation
  2657. pthread_mutex_lock (&mymutex);
  2658. queue_put (data);
  2659. pthread_mutex_unlock (&mymutex);
  2660. ev_async_send (EV_DEFAULT_ &mysig);
  2661. }
  2662. static void
  2663. mysig_cb (EV_P_ ev_async *w, int revents)
  2664. {
  2665. pthread_mutex_lock (&mymutex);
  2666. while (queue_get (&data))
  2667. process (data);
  2668. pthread_mutex_unlock (&mymutex);
  2669. }
  2670. =back
  2671. =head3 Watcher-Specific Functions and Data Members
  2672. =over 4
  2673. =item ev_async_init (ev_async *, callback)
  2674. Initialises and configures the async watcher - it has no parameters of any
  2675. kind. There is a C<ev_async_set> macro, but using it is utterly pointless,
  2676. trust me.
  2677. =item ev_async_send (loop, ev_async *)
  2678. Sends/signals/activates the given C<ev_async> watcher, that is, feeds
  2679. an C<EV_ASYNC> event on the watcher into the event loop, and instantly
  2680. returns.
  2681. Unlike C<ev_feed_event>, this call is safe to do from other threads,
  2682. signal or similar contexts (see the discussion of C<EV_ATOMIC_T> in the
  2683. embedding section below on what exactly this means).
  2684. Note that, as with other watchers in libev, multiple events might get
  2685. compressed into a single callback invocation (another way to look at
  2686. this is that C<ev_async> watchers are level-triggered: they are set on
  2687. C<ev_async_send>, reset when the event loop detects that).
  2688. This call incurs the overhead of at most one extra system call per event
  2689. loop iteration, if the event loop is blocked, and no syscall at all if
  2690. the event loop (or your program) is processing events. That means that
  2691. repeated calls are basically free (there is no need to avoid calls for
  2692. performance reasons) and that the overhead becomes smaller (typically
  2693. zero) under load.
  2694. =item bool = ev_async_pending (ev_async *)
  2695. Returns a non-zero value when C<ev_async_send> has been called on the
  2696. watcher but the event has not yet been processed (or even noted) by the
  2697. event loop.
  2698. C<ev_async_send> sets a flag in the watcher and wakes up the loop. When
  2699. the loop iterates next and checks for the watcher to have become active,
  2700. it will reset the flag again. C<ev_async_pending> can be used to very
  2701. quickly check whether invoking the loop might be a good idea.
  2702. Not that this does I<not> check whether the watcher itself is pending,
  2703. only whether it has been requested to make this watcher pending: there
  2704. is a time window between the event loop checking and resetting the async
  2705. notification, and the callback being invoked.
  2706. =back
  2707. =head1 OTHER FUNCTIONS
  2708. There are some other functions of possible interest. Described. Here. Now.
  2709. =over 4
  2710. =item ev_once (loop, int fd, int events, ev_tstamp timeout, callback, arg)
  2711. This function combines a simple timer and an I/O watcher, calls your
  2712. callback on whichever event happens first and automatically stops both
  2713. watchers. This is useful if you want to wait for a single event on an fd
  2714. or timeout without having to allocate/configure/start/stop/free one or
  2715. more watchers yourself.
  2716. If C<fd> is less than 0, then no I/O watcher will be started and the
  2717. C<events> argument is being ignored. Otherwise, an C<ev_io> watcher for
  2718. the given C<fd> and C<events> set will be created and started.
  2719. If C<timeout> is less than 0, then no timeout watcher will be
  2720. started. Otherwise an C<ev_timer> watcher with after = C<timeout> (and
  2721. repeat = 0) will be started. C<0> is a valid timeout.
  2722. The callback has the type C<void (*cb)(int revents, void *arg)> and is
  2723. passed an C<revents> set like normal event callbacks (a combination of
  2724. C<EV_ERROR>, C<EV_READ>, C<EV_WRITE> or C<EV_TIMER>) and the C<arg>
  2725. value passed to C<ev_once>. Note that it is possible to receive I<both>
  2726. a timeout and an io event at the same time - you probably should give io
  2727. events precedence.
  2728. Example: wait up to ten seconds for data to appear on STDIN_FILENO.
  2729. static void stdin_ready (int revents, void *arg)
  2730. {
  2731. if (revents & EV_READ)
  2732. /* stdin might have data for us, joy! */;
  2733. else if (revents & EV_TIMER)
  2734. /* doh, nothing entered */;
  2735. }
  2736. ev_once (STDIN_FILENO, EV_READ, 10., stdin_ready, 0);
  2737. =item ev_feed_fd_event (loop, int fd, int revents)
  2738. Feed an event on the given fd, as if a file descriptor backend detected
  2739. the given events.
  2740. =item ev_feed_signal_event (loop, int signum)
  2741. Feed an event as if the given signal occurred. See also C<ev_feed_signal>,
  2742. which is async-safe.
  2743. =back
  2745. This section explains some common idioms that are not immediately
  2746. obvious. Note that examples are sprinkled over the whole manual, and this
  2747. section only contains stuff that wouldn't fit anywhere else.
  2749. Each watcher has, by default, a C<void *data> member that you can read
  2750. or modify at any time: libev will completely ignore it. This can be used
  2751. to associate arbitrary data with your watcher. If you need more data and
  2752. don't want to allocate memory separately and store a pointer to it in that
  2753. data member, you can also "subclass" the watcher type and provide your own
  2754. data:
  2755. struct my_io
  2756. {
  2757. ev_io io;
  2758. int otherfd;
  2759. void *somedata;
  2760. struct whatever *mostinteresting;
  2761. };
  2762. ...
  2763. struct my_io w;
  2764. ev_io_init (&w.io, my_cb, fd, EV_READ);
  2765. And since your callback will be called with a pointer to the watcher, you
  2766. can cast it back to your own type:
  2767. static void my_cb (struct ev_loop *loop, ev_io *w_, int revents)
  2768. {
  2769. struct my_io *w = (struct my_io *)w_;
  2770. ...
  2771. }
  2772. More interesting and less C-conformant ways of casting your callback
  2773. function type instead have been omitted.
  2775. Another common scenario is to use some data structure with multiple
  2776. embedded watchers, in effect creating your own watcher that combines
  2777. multiple libev event sources into one "super-watcher":
  2778. struct my_biggy
  2779. {
  2780. int some_data;
  2781. ev_timer t1;
  2782. ev_timer t2;
  2783. }
  2784. In this case getting the pointer to C<my_biggy> is a bit more
  2785. complicated: Either you store the address of your C<my_biggy> struct in
  2786. the C<data> member of the watcher (for woozies or C++ coders), or you need
  2787. to use some pointer arithmetic using C<offsetof> inside your watchers (for
  2788. real programmers):
  2789. #include <stddef.h>
  2790. static void
  2791. t1_cb (EV_P_ ev_timer *w, int revents)
  2792. {
  2793. struct my_biggy big = (struct my_biggy *)
  2794. (((char *)w) - offsetof (struct my_biggy, t1));
  2795. }
  2796. static void
  2797. t2_cb (EV_P_ ev_timer *w, int revents)
  2798. {
  2799. struct my_biggy big = (struct my_biggy *)
  2800. (((char *)w) - offsetof (struct my_biggy, t2));
  2801. }
  2803. Often you have structures like this in event-based programs:
  2804. callback ()
  2805. {
  2806. free (request);
  2807. }
  2808. request = start_new_request (..., callback);
  2809. The intent is to start some "lengthy" operation. The C<request> could be
  2810. used to cancel the operation, or do other things with it.
  2811. It's not uncommon to have code paths in C<start_new_request> that
  2812. immediately invoke the callback, for example, to report errors. Or you add
  2813. some caching layer that finds that it can skip the lengthy aspects of the
  2814. operation and simply invoke the callback with the result.
  2815. The problem here is that this will happen I<before> C<start_new_request>
  2816. has returned, so C<request> is not set.
  2817. Even if you pass the request by some safer means to the callback, you
  2818. might want to do something to the request after starting it, such as
  2819. canceling it, which probably isn't working so well when the callback has
  2820. already been invoked.
  2821. A common way around all these issues is to make sure that
  2822. C<start_new_request> I<always> returns before the callback is invoked. If
  2823. C<start_new_request> immediately knows the result, it can artificially
  2824. delay invoking the callback by using a C<prepare> or C<idle> watcher for
  2825. example, or more sneakily, by reusing an existing (stopped) watcher and
  2826. pushing it into the pending queue:
  2827. ev_set_cb (watcher, callback);
  2828. ev_feed_event (EV_A_ watcher, 0);
  2829. This way, C<start_new_request> can safely return before the callback is
  2830. invoked, while not delaying callback invocation too much.
  2832. Often (especially in GUI toolkits) there are places where you have
  2833. I<modal> interaction, which is most easily implemented by recursively
  2834. invoking C<ev_run>.
  2835. This brings the problem of exiting - a callback might want to finish the
  2836. main C<ev_run> call, but not the nested one (e.g. user clicked "Quit", but
  2837. a modal "Are you sure?" dialog is still waiting), or just the nested one
  2838. and not the main one (e.g. user clocked "Ok" in a modal dialog), or some
  2839. other combination: In these cases, a simple C<ev_break> will not work.
  2840. The solution is to maintain "break this loop" variable for each C<ev_run>
  2841. invocation, and use a loop around C<ev_run> until the condition is
  2842. triggered, using C<EVRUN_ONCE>:
  2843. // main loop
  2844. int exit_main_loop = 0;
  2845. while (!exit_main_loop)
  2846. ev_run (EV_DEFAULT_ EVRUN_ONCE);
  2847. // in a modal watcher
  2848. int exit_nested_loop = 0;
  2849. while (!exit_nested_loop)
  2850. ev_run (EV_A_ EVRUN_ONCE);
  2851. To exit from any of these loops, just set the corresponding exit variable:
  2852. // exit modal loop
  2853. exit_nested_loop = 1;
  2854. // exit main program, after modal loop is finished
  2855. exit_main_loop = 1;
  2856. // exit both
  2857. exit_main_loop = exit_nested_loop = 1;
  2859. Here is a fictitious example of how to run an event loop in a different
  2860. thread from where callbacks are being invoked and watchers are
  2861. created/added/removed.
  2862. For a real-world example, see the C<EV::Loop::Async> perl module,
  2863. which uses exactly this technique (which is suited for many high-level
  2864. languages).
  2865. The example uses a pthread mutex to protect the loop data, a condition
  2866. variable to wait for callback invocations, an async watcher to notify the
  2867. event loop thread and an unspecified mechanism to wake up the main thread.
  2868. First, you need to associate some data with the event loop:
  2869. typedef struct {
  2870. mutex_t lock; /* global loop lock */
  2871. ev_async async_w;
  2872. thread_t tid;
  2873. cond_t invoke_cv;
  2874. } userdata;
  2875. void prepare_loop (EV_P)
  2876. {
  2877. // for simplicity, we use a static userdata struct.
  2878. static userdata u;
  2879. ev_async_init (&u->async_w, async_cb);
  2880. ev_async_start (EV_A_ &u->async_w);
  2881. pthread_mutex_init (&u->lock, 0);
  2882. pthread_cond_init (&u->invoke_cv, 0);
  2883. // now associate this with the loop
  2884. ev_set_userdata (EV_A_ u);
  2885. ev_set_invoke_pending_cb (EV_A_ l_invoke);
  2886. ev_set_loop_release_cb (EV_A_ l_release, l_acquire);
  2887. // then create the thread running ev_run
  2888. pthread_create (&u->tid, 0, l_run, EV_A);
  2889. }
  2890. The callback for the C<ev_async> watcher does nothing: the watcher is used
  2891. solely to wake up the event loop so it takes notice of any new watchers
  2892. that might have been added:
  2893. static void
  2894. async_cb (EV_P_ ev_async *w, int revents)
  2895. {
  2896. // just used for the side effects
  2897. }
  2898. The C<l_release> and C<l_acquire> callbacks simply unlock/lock the mutex
  2899. protecting the loop data, respectively.
  2900. static void
  2901. l_release (EV_P)
  2902. {
  2903. userdata *u = ev_userdata (EV_A);
  2904. pthread_mutex_unlock (&u->lock);
  2905. }
  2906. static void
  2907. l_acquire (EV_P)
  2908. {
  2909. userdata *u = ev_userdata (EV_A);
  2910. pthread_mutex_lock (&u->lock);
  2911. }
  2912. The event loop thread first acquires the mutex, and then jumps straight
  2913. into C<ev_run>:
  2914. void *
  2915. l_run (void *thr_arg)
  2916. {
  2917. struct ev_loop *loop = (struct ev_loop *)thr_arg;
  2918. l_acquire (EV_A);
  2919. pthread_setcanceltype (PTHREAD_CANCEL_ASYNCHRONOUS, 0);
  2920. ev_run (EV_A_ 0);
  2921. l_release (EV_A);
  2922. return 0;
  2923. }
  2924. Instead of invoking all pending watchers, the C<l_invoke> callback will
  2925. signal the main thread via some unspecified mechanism (signals? pipe
  2926. writes? C<Async::Interrupt>?) and then waits until all pending watchers
  2927. have been called (in a while loop because a) spurious wakeups are possible
  2928. and b) skipping inter-thread-communication when there are no pending
  2929. watchers is very beneficial):
  2930. static void
  2931. l_invoke (EV_P)
  2932. {
  2933. userdata *u = ev_userdata (EV_A);
  2934. while (ev_pending_count (EV_A))
  2935. {
  2936. wake_up_other_thread_in_some_magic_or_not_so_magic_way ();
  2937. pthread_cond_wait (&u->invoke_cv, &u->lock);
  2938. }
  2939. }
  2940. Now, whenever the main thread gets told to invoke pending watchers, it
  2941. will grab the lock, call C<ev_invoke_pending> and then signal the loop
  2942. thread to continue:
  2943. static void
  2944. real_invoke_pending (EV_P)
  2945. {
  2946. userdata *u = ev_userdata (EV_A);
  2947. pthread_mutex_lock (&u->lock);
  2948. ev_invoke_pending (EV_A);
  2949. pthread_cond_signal (&u->invoke_cv);
  2950. pthread_mutex_unlock (&u->lock);
  2951. }
  2952. Whenever you want to start/stop a watcher or do other modifications to an
  2953. event loop, you will now have to lock:
  2954. ev_timer timeout_watcher;
  2955. userdata *u = ev_userdata (EV_A);
  2956. ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
  2957. pthread_mutex_lock (&u->lock);
  2958. ev_timer_start (EV_A_ &timeout_watcher);
  2959. ev_async_send (EV_A_ &u->async_w);
  2960. pthread_mutex_unlock (&u->lock);
  2961. Note that sending the C<ev_async> watcher is required because otherwise
  2962. an event loop currently blocking in the kernel will have no knowledge
  2963. about the newly added timer. By waking up the loop it will pick up any new
  2964. watchers in the next event loop iteration.
  2966. While the overhead of a callback that e.g. schedules a thread is small, it
  2967. is still an overhead. If you embed libev, and your main usage is with some
  2968. kind of threads or coroutines, you might want to customise libev so that
  2969. doesn't need callbacks anymore.
  2970. Imagine you have coroutines that you can switch to using a function
  2971. C<switch_to (coro)>, that libev runs in a coroutine called C<libev_coro>
  2972. and that due to some magic, the currently active coroutine is stored in a
  2973. global called C<current_coro>. Then you can build your own "wait for libev
  2974. event" primitive by changing C<EV_CB_DECLARE> and C<EV_CB_INVOKE> (note
  2975. the differing C<;> conventions):
  2976. #define EV_CB_DECLARE(type) struct my_coro *cb;
  2977. #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
  2978. That means instead of having a C callback function, you store the
  2979. coroutine to switch to in each watcher, and instead of having libev call
  2980. your callback, you instead have it switch to that coroutine.
  2981. A coroutine might now wait for an event with a function called
  2982. C<wait_for_event>. (the watcher needs to be started, as always, but it doesn't
  2983. matter when, or whether the watcher is active or not when this function is
  2984. called):
  2985. void
  2986. wait_for_event (ev_watcher *w)
  2987. {
  2988. ev_set_cb (w, current_coro);
  2989. switch_to (libev_coro);
  2990. }
  2991. That basically suspends the coroutine inside C<wait_for_event> and
  2992. continues the libev coroutine, which, when appropriate, switches back to
  2993. this or any other coroutine.
  2994. You can do similar tricks if you have, say, threads with an event queue -
  2995. instead of storing a coroutine, you store the queue object and instead of
  2996. switching to a coroutine, you push the watcher onto the queue and notify
  2997. any waiters.
  2998. To embed libev, see L</EMBEDDING>, but in short, it's easiest to create two
  2999. files, F<my_ev.h> and F<my_ev.c> that include the respective libev files:
  3000. // my_ev.h
  3001. #define EV_CB_DECLARE(type) struct my_coro *cb;
  3002. #define EV_CB_INVOKE(watcher) switch_to ((watcher)->cb)
  3003. #include "../libev/ev.h"
  3004. // my_ev.c
  3005. #define EV_H "my_ev.h"
  3006. #include "../libev/ev.c"
  3007. And then use F<my_ev.h> when you would normally use F<ev.h>, and compile
  3008. F<my_ev.c> into your project. When properly specifying include paths, you
  3009. can even use F<ev.h> as header file name directly.
  3011. Libev offers a compatibility emulation layer for libevent. It cannot
  3012. emulate the internals of libevent, so here are some usage hints:
  3013. =over 4
  3014. =item * Only the libevent-1.4.1-beta API is being emulated.
  3015. This was the newest libevent version available when libev was implemented,
  3016. and is still mostly unchanged in 2010.
  3017. =item * Use it by including <event.h>, as usual.
  3018. =item * The following members are fully supported: ev_base, ev_callback,
  3019. ev_arg, ev_fd, ev_res, ev_events.
  3020. =item * Avoid using ev_flags and the EVLIST_*-macros, while it is
  3021. maintained by libev, it does not work exactly the same way as in libevent (consider
  3022. it a private API).
  3023. =item * Priorities are not currently supported. Initialising priorities
  3024. will fail and all watchers will have the same priority, even though there
  3025. is an ev_pri field.
  3026. =item * In libevent, the last base created gets the signals, in libev, the
  3027. base that registered the signal gets the signals.
  3028. =item * Other members are not supported.
  3029. =item * The libev emulation is I<not> ABI compatible to libevent, you need
  3030. to use the libev header file and library.
  3031. =back
  3032. =head1 C++ SUPPORT
  3033. =head2 C API
  3034. The normal C API should work fine when used from C++: both ev.h and the
  3035. libev sources can be compiled as C++. Therefore, code that uses the C API
  3036. will work fine.
  3037. Proper exception specifications might have to be added to callbacks passed
  3038. to libev: exceptions may be thrown only from watcher callbacks, all other
  3039. callbacks (allocator, syserr, loop acquire/release and periodic reschedule
  3040. callbacks) must not throw exceptions, and might need a C<noexcept>
  3041. specification. If you have code that needs to be compiled as both C and
  3042. C++ you can use the C<EV_NOEXCEPT> macro for this:
  3043. static void
  3044. fatal_error (const char *msg) EV_NOEXCEPT
  3045. {
  3046. perror (msg);
  3047. abort ();
  3048. }
  3049. ...
  3050. ev_set_syserr_cb (fatal_error);
  3051. The only API functions that can currently throw exceptions are C<ev_run>,
  3052. C<ev_invoke>, C<ev_invoke_pending> and C<ev_loop_destroy> (the latter
  3053. because it runs cleanup watchers).
  3054. Throwing exceptions in watcher callbacks is only supported if libev itself
  3055. is compiled with a C++ compiler or your C and C++ environments allow
  3056. throwing exceptions through C libraries (most do).
  3057. =head2 C++ API
  3058. Libev comes with some simplistic wrapper classes for C++ that mainly allow
  3059. you to use some convenience methods to start/stop watchers and also change
  3060. the callback model to a model using method callbacks on objects.
  3061. To use it,
  3062. #include <ev++.h>
  3063. This automatically includes F<ev.h> and puts all of its definitions (many
  3064. of them macros) into the global namespace. All C++ specific things are
  3065. put into the C<ev> namespace. It should support all the same embedding
  3066. options as F<ev.h>, most notably C<EV_MULTIPLICITY>.
  3067. Care has been taken to keep the overhead low. The only data member the C++
  3068. classes add (compared to plain C-style watchers) is the event loop pointer
  3069. that the watcher is associated with (or no additional members at all if
  3070. you disable C<EV_MULTIPLICITY> when embedding libev).
  3071. Currently, functions, static and non-static member functions and classes
  3072. with C<operator ()> can be used as callbacks. Other types should be easy
  3073. to add as long as they only need one additional pointer for context. If
  3074. you need support for other types of functors please contact the author
  3075. (preferably after implementing it).
  3076. For all this to work, your C++ compiler either has to use the same calling
  3077. conventions as your C compiler (for static member functions), or you have
  3078. to embed libev and compile libev itself as C++.
  3079. Here is a list of things available in the C<ev> namespace:
  3080. =over 4
  3081. =item C<ev::READ>, C<ev::WRITE> etc.
  3082. These are just enum values with the same values as the C<EV_READ> etc.
  3083. macros from F<ev.h>.
  3084. =item C<ev::tstamp>, C<ev::now>
  3085. Aliases to the same types/functions as with the C<ev_> prefix.
  3086. =item C<ev::io>, C<ev::timer>, C<ev::periodic>, C<ev::idle>, C<ev::sig> etc.
  3087. For each C<ev_TYPE> watcher in F<ev.h> there is a corresponding class of
  3088. the same name in the C<ev> namespace, with the exception of C<ev_signal>
  3089. which is called C<ev::sig> to avoid clashes with the C<signal> macro
  3090. defined by many implementations.
  3091. All of those classes have these methods:
  3092. =over 4
  3093. =item ev::TYPE::TYPE ()
  3094. =item ev::TYPE::TYPE (loop)
  3095. =item ev::TYPE::~TYPE
  3096. The constructor (optionally) takes an event loop to associate the watcher
  3097. with. If it is omitted, it will use C<EV_DEFAULT>.
  3098. The constructor calls C<ev_init> for you, which means you have to call the
  3099. C<set> method before starting it.
  3100. It will not set a callback, however: You have to call the templated C<set>
  3101. method to set a callback before you can start the watcher.
  3102. (The reason why you have to use a method is a limitation in C++ which does
  3103. not allow explicit template arguments for constructors).
  3104. The destructor automatically stops the watcher if it is active.
  3105. =item w->set<class, &class::method> (object *)
  3106. This method sets the callback method to call. The method has to have a
  3107. signature of C<void (*)(ev_TYPE &, int)>, it receives the watcher as
  3108. first argument and the C<revents> as second. The object must be given as
  3109. parameter and is stored in the C<data> member of the watcher.
  3110. This method synthesizes efficient thunking code to call your method from
  3111. the C callback that libev requires. If your compiler can inline your
  3112. callback (i.e. it is visible to it at the place of the C<set> call and
  3113. your compiler is good :), then the method will be fully inlined into the
  3114. thunking function, making it as fast as a direct C callback.
  3115. Example: simple class declaration and watcher initialisation
  3116. struct myclass
  3117. {
  3118. void io_cb (ev::io &w, int revents) { }
  3119. }
  3120. myclass obj;
  3121. ev::io iow;
  3122. iow.set <myclass, &myclass::io_cb> (&obj);
  3123. =item w->set (object *)
  3124. This is a variation of a method callback - leaving out the method to call
  3125. will default the method to C<operator ()>, which makes it possible to use
  3126. functor objects without having to manually specify the C<operator ()> all
  3127. the time. Incidentally, you can then also leave out the template argument
  3128. list.
  3129. The C<operator ()> method prototype must be C<void operator ()(watcher &w,
  3130. int revents)>.
  3131. See the method-C<set> above for more details.
  3132. Example: use a functor object as callback.
  3133. struct myfunctor
  3134. {
  3135. void operator() (ev::io &w, int revents)
  3136. {
  3137. ...
  3138. }
  3139. }
  3140. myfunctor f;
  3141. ev::io w;
  3142. w.set (&f);
  3143. =item w->set<function> (void *data = 0)
  3144. Also sets a callback, but uses a static method or plain function as
  3145. callback. The optional C<data> argument will be stored in the watcher's
  3146. C<data> member and is free for you to use.
  3147. The prototype of the C<function> must be C<void (*)(ev::TYPE &w, int)>.
  3148. See the method-C<set> above for more details.
  3149. Example: Use a plain function as callback.
  3150. static void io_cb (ev::io &w, int revents) { }
  3151. iow.set <io_cb> ();
  3152. =item w->set (loop)
  3153. Associates a different C<struct ev_loop> with this watcher. You can only
  3154. do this when the watcher is inactive (and not pending either).
  3155. =item w->set ([arguments])
  3156. Basically the same as C<ev_TYPE_set> (except for C<ev::embed> watchers>),
  3157. with the same arguments. Either this method or a suitable start method
  3158. must be called at least once. Unlike the C counterpart, an active watcher
  3159. gets automatically stopped and restarted when reconfiguring it with this
  3160. method.
  3161. For C<ev::embed> watchers this method is called C<set_embed>, to avoid
  3162. clashing with the C<set (loop)> method.
  3163. For C<ev::io> watchers there is an additional C<set> method that acepts a
  3164. new event mask only, and internally calls C<ev_io_modfify>.
  3165. =item w->start ()
  3166. Starts the watcher. Note that there is no C<loop> argument, as the
  3167. constructor already stores the event loop.
  3168. =item w->start ([arguments])
  3169. Instead of calling C<set> and C<start> methods separately, it is often
  3170. convenient to wrap them in one call. Uses the same type of arguments as
  3171. the configure C<set> method of the watcher.
  3172. =item w->stop ()
  3173. Stops the watcher if it is active. Again, no C<loop> argument.
  3174. =item w->again () (C<ev::timer>, C<ev::periodic> only)
  3175. For C<ev::timer> and C<ev::periodic>, this invokes the corresponding
  3176. C<ev_TYPE_again> function.
  3177. =item w->sweep () (C<ev::embed> only)
  3178. Invokes C<ev_embed_sweep>.
  3179. =item w->update () (C<ev::stat> only)
  3180. Invokes C<ev_stat_stat>.
  3181. =back
  3182. =back
  3183. Example: Define a class with two I/O and idle watchers, start the I/O
  3184. watchers in the constructor.
  3185. class myclass
  3186. {
  3187. ev::io io ; void io_cb (ev::io &w, int revents);
  3188. ev::io io2 ; void io2_cb (ev::io &w, int revents);
  3189. ev::idle idle; void idle_cb (ev::idle &w, int revents);
  3190. myclass (int fd)
  3191. {
  3192. io .set <myclass, &myclass::io_cb > (this);
  3193. io2 .set <myclass, &myclass::io2_cb > (this);
  3194. idle.set <myclass, &myclass::idle_cb> (this);
  3195. io.set (fd, ev::WRITE); // configure the watcher
  3196. io.start (); // start it whenever convenient
  3197. io2.start (fd, ev::READ); // set + start in one call
  3198. }
  3199. };
  3201. Libev does not offer other language bindings itself, but bindings for a
  3202. number of languages exist in the form of third-party packages. If you know
  3203. any interesting language binding in addition to the ones listed here, drop
  3204. me a note.
  3205. =over 4
  3206. =item Perl
  3207. The EV module implements the full libev API and is actually used to test
  3208. libev. EV is developed together with libev. Apart from the EV core module,
  3209. there are additional modules that implement libev-compatible interfaces
  3210. to C<libadns> (C<EV::ADNS>, but C<AnyEvent::DNS> is preferred nowadays),
  3211. C<Net::SNMP> (C<Net::SNMP::EV>) and the C<libglib> event core (C<Glib::EV>
  3212. and C<EV::Glib>).
  3213. It can be found and installed via CPAN, its homepage is at
  3214. L<http://software.schmorp.de/pkg/EV>.
  3215. =item Python
  3216. Python bindings can be found at L<http://code.google.com/p/pyev/>. It
  3217. seems to be quite complete and well-documented.
  3218. =item Ruby
  3219. Tony Arcieri has written a ruby extension that offers access to a subset
  3220. of the libev API and adds file handle abstractions, asynchronous DNS and
  3221. more on top of it. It can be found via gem servers. Its homepage is at
  3222. L<http://rev.rubyforge.org/>.
  3223. Roger Pack reports that using the link order C<-lws2_32 -lmsvcrt-ruby-190>
  3224. makes rev work even on mingw.
  3225. =item Haskell
  3226. A haskell binding to libev is available at
  3227. L<http://hackage.haskell.org/cgi-bin/hackage-scripts/package/hlibev>.
  3228. =item D
  3229. Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
  3230. be found at L<http://www.llucax.com.ar/proj/ev.d/index.html>.
  3231. =item Ocaml
  3232. Erkki Seppala has written Ocaml bindings for libev, to be found at
  3233. L<http://modeemi.cs.tut.fi/~flux/software/ocaml-ev/>.
  3234. =item Lua
  3235. Brian Maher has written a partial interface to libev for lua (at the
  3236. time of this writing, only C<ev_io> and C<ev_timer>), to be found at
  3237. L<http://github.com/brimworks/lua-ev>.
  3238. =item Javascript
  3239. Node.js (L<http://nodejs.org>) uses libev as the underlying event library.
  3240. =item Others
  3241. There are others, and I stopped counting.
  3242. =back
  3243. =head1 MACRO MAGIC
  3244. Libev can be compiled with a variety of options, the most fundamental
  3245. of which is C<EV_MULTIPLICITY>. This option determines whether (most)
  3246. functions and callbacks have an initial C<struct ev_loop *> argument.
  3247. To make it easier to write programs that cope with either variant, the
  3248. following macros are defined:
  3249. =over 4
  3250. =item C<EV_A>, C<EV_A_>
  3251. This provides the loop I<argument> for functions, if one is required ("ev
  3252. loop argument"). The C<EV_A> form is used when this is the sole argument,
  3253. C<EV_A_> is used when other arguments are following. Example:
  3254. ev_unref (EV_A);
  3255. ev_timer_add (EV_A_ watcher);
  3256. ev_run (EV_A_ 0);
  3257. It assumes the variable C<loop> of type C<struct ev_loop *> is in scope,
  3258. which is often provided by the following macro.
  3259. =item C<EV_P>, C<EV_P_>
  3260. This provides the loop I<parameter> for functions, if one is required ("ev
  3261. loop parameter"). The C<EV_P> form is used when this is the sole parameter,
  3262. C<EV_P_> is used when other parameters are following. Example:
  3263. // this is how ev_unref is being declared
  3264. static void ev_unref (EV_P);
  3265. // this is how you can declare your typical callback
  3266. static void cb (EV_P_ ev_timer *w, int revents)
  3267. It declares a parameter C<loop> of type C<struct ev_loop *>, quite
  3268. suitable for use with C<EV_A>.
  3269. =item C<EV_DEFAULT>, C<EV_DEFAULT_>
  3270. Similar to the other two macros, this gives you the value of the default
  3271. loop, if multiple loops are supported ("ev loop default"). The default loop
  3272. will be initialised if it isn't already initialised.
  3273. For non-multiplicity builds, these macros do nothing, so you always have
  3274. to initialise the loop somewhere.
  3275. =item C<EV_DEFAULT_UC>, C<EV_DEFAULT_UC_>
  3276. Usage identical to C<EV_DEFAULT> and C<EV_DEFAULT_>, but requires that the
  3277. default loop has been initialised (C<UC> == unchecked). Their behaviour
  3278. is undefined when the default loop has not been initialised by a previous
  3279. execution of C<EV_DEFAULT>, C<EV_DEFAULT_> or C<ev_default_init (...)>.
  3280. It is often prudent to use C<EV_DEFAULT> when initialising the first
  3281. watcher in a function but use C<EV_DEFAULT_UC> afterwards.
  3282. =back
  3283. Example: Declare and initialise a check watcher, utilising the above
  3284. macros so it will work regardless of whether multiple loops are supported
  3285. or not.
  3286. static void
  3287. check_cb (EV_P_ ev_timer *w, int revents)
  3288. {
  3289. ev_check_stop (EV_A_ w);
  3290. }
  3291. ev_check check;
  3292. ev_check_init (&check, check_cb);
  3293. ev_check_start (EV_DEFAULT_ &check);
  3294. ev_run (EV_DEFAULT_ 0);
  3295. =head1 EMBEDDING
  3296. Libev can (and often is) directly embedded into host
  3297. applications. Examples of applications that embed it include the Deliantra
  3298. Game Server, the EV perl module, the GNU Virtual Private Ethernet (gvpe)
  3299. and rxvt-unicode.
  3300. The goal is to enable you to just copy the necessary files into your
  3301. source directory without having to change even a single line in them, so
  3302. you can easily upgrade by simply copying (or having a checked-out copy of
  3303. libev somewhere in your source tree).
  3304. =head2 FILESETS
  3305. Depending on what features you need you need to include one or more sets of files
  3306. in your application.
  3307. =head3 CORE EVENT LOOP
  3308. To include only the libev core (all the C<ev_*> functions), with manual
  3309. configuration (no autoconf):
  3310. #define EV_STANDALONE 1
  3311. #include "ev.c"
  3312. This will automatically include F<ev.h>, too, and should be done in a
  3313. single C source file only to provide the function implementations. To use
  3314. it, do the same for F<ev.h> in all files wishing to use this API (best
  3315. done by writing a wrapper around F<ev.h> that you can include instead and
  3316. where you can put other configuration options):
  3317. #define EV_STANDALONE 1
  3318. #include "ev.h"
  3319. Both header files and implementation files can be compiled with a C++
  3320. compiler (at least, that's a stated goal, and breakage will be treated
  3321. as a bug).
  3322. You need the following files in your source tree, or in a directory
  3323. in your include path (e.g. in libev/ when using -Ilibev):
  3324. ev.h
  3325. ev.c
  3326. ev_vars.h
  3327. ev_wrap.h
  3328. ev_win32.c required on win32 platforms only
  3329. ev_select.c only when select backend is enabled
  3330. ev_poll.c only when poll backend is enabled
  3331. ev_epoll.c only when the epoll backend is enabled
  3332. ev_linuxaio.c only when the linux aio backend is enabled
  3333. ev_iouring.c only when the linux io_uring backend is enabled
  3334. ev_kqueue.c only when the kqueue backend is enabled
  3335. ev_port.c only when the solaris port backend is enabled
  3336. F<ev.c> includes the backend files directly when enabled, so you only need
  3337. to compile this single file.
  3339. To include the libevent compatibility API, also include:
  3340. #include "event.c"
  3341. in the file including F<ev.c>, and:
  3342. #include "event.h"
  3343. in the files that want to use the libevent API. This also includes F<ev.h>.
  3344. You need the following additional files for this:
  3345. event.h
  3346. event.c
  3347. =head3 AUTOCONF SUPPORT
  3348. Instead of using C<EV_STANDALONE=1> and providing your configuration in
  3349. whatever way you want, you can also C<m4_include([libev.m4])> in your
  3350. F<configure.ac> and leave C<EV_STANDALONE> undefined. F<ev.c> will then
  3351. include F<config.h> and configure itself accordingly.
  3352. For this of course you need the m4 file:
  3353. libev.m4
  3355. Libev can be configured via a variety of preprocessor symbols you have to
  3356. define before including (or compiling) any of its files. The default in
  3357. the absence of autoconf is documented for every option.
  3358. Symbols marked with "(h)" do not change the ABI, and can have different
  3359. values when compiling libev vs. including F<ev.h>, so it is permissible
  3360. to redefine them before including F<ev.h> without breaking compatibility
  3361. to a compiled library. All other symbols change the ABI, which means all
  3362. users of libev and the libev code itself must be compiled with compatible
  3363. settings.
  3364. =over 4
  3365. =item EV_COMPAT3 (h)
  3366. Backwards compatibility is a major concern for libev. This is why this
  3367. release of libev comes with wrappers for the functions and symbols that
  3368. have been renamed between libev version 3 and 4.
  3369. You can disable these wrappers (to test compatibility with future
  3370. versions) by defining C<EV_COMPAT3> to C<0> when compiling your
  3371. sources. This has the additional advantage that you can drop the C<struct>
  3372. from C<struct ev_loop> declarations, as libev will provide an C<ev_loop>
  3373. typedef in that case.
  3374. In some future version, the default for C<EV_COMPAT3> will become C<0>,
  3375. and in some even more future version the compatibility code will be
  3376. removed completely.
  3377. =item EV_STANDALONE (h)
  3378. Must always be C<1> if you do not use autoconf configuration, which
  3379. keeps libev from including F<config.h>, and it also defines dummy
  3380. implementations for some libevent functions (such as logging, which is not
  3381. supported). It will also not define any of the structs usually found in
  3382. F<event.h> that are not directly supported by the libev core alone.
  3383. In standalone mode, libev will still try to automatically deduce the
  3384. configuration, but has to be more conservative.
  3385. =item EV_USE_FLOOR
  3386. If defined to be C<1>, libev will use the C<floor ()> function for its
  3387. periodic reschedule calculations, otherwise libev will fall back on a
  3388. portable (slower) implementation. If you enable this, you usually have to
  3389. link against libm or something equivalent. Enabling this when the C<floor>
  3390. function is not available will fail, so the safe default is to not enable
  3391. this.
  3392. =item EV_USE_MONOTONIC
  3393. If defined to be C<1>, libev will try to detect the availability of the
  3394. monotonic clock option at both compile time and runtime. Otherwise no
  3395. use of the monotonic clock option will be attempted. If you enable this,
  3396. you usually have to link against librt or something similar. Enabling it
  3397. when the functionality isn't available is safe, though, although you have
  3398. to make sure you link against any libraries where the C<clock_gettime>
  3399. function is hiding in (often F<-lrt>). See also C<EV_USE_CLOCK_SYSCALL>.
  3400. =item EV_USE_REALTIME
  3401. If defined to be C<1>, libev will try to detect the availability of the
  3402. real-time clock option at compile time (and assume its availability
  3403. at runtime if successful). Otherwise no use of the real-time clock
  3404. option will be attempted. This effectively replaces C<gettimeofday>
  3405. by C<clock_get (CLOCK_REALTIME, ...)> and will not normally affect
  3406. correctness. See the note about libraries in the description of
  3407. C<EV_USE_MONOTONIC>, though. Defaults to the opposite value of
  3409. =item EV_USE_CLOCK_SYSCALL
  3410. If defined to be C<1>, libev will try to use a direct syscall instead
  3411. of calling the system-provided C<clock_gettime> function. This option
  3412. exists because on GNU/Linux, C<clock_gettime> is in C<librt>, but C<librt>
  3413. unconditionally pulls in C<libpthread>, slowing down single-threaded
  3414. programs needlessly. Using a direct syscall is slightly slower (in
  3415. theory), because no optimised vdso implementation can be used, but avoids
  3416. the pthread dependency. Defaults to C<1> on GNU/Linux with glibc 2.x or
  3417. higher, as it simplifies linking (no need for C<-lrt>).
  3418. =item EV_USE_NANOSLEEP
  3419. If defined to be C<1>, libev will assume that C<nanosleep ()> is available
  3420. and will use it for delays. Otherwise it will use C<select ()>.
  3421. =item EV_USE_EVENTFD
  3422. If defined to be C<1>, then libev will assume that C<eventfd ()> is
  3423. available and will probe for kernel support at runtime. This will improve
  3424. C<ev_signal> and C<ev_async> performance and reduce resource consumption.
  3425. If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
  3426. 2.7 or newer, otherwise disabled.
  3427. =item EV_USE_SIGNALFD
  3428. If defined to be C<1>, then libev will assume that C<signalfd ()> is
  3429. available and will probe for kernel support at runtime. This enables
  3430. the use of EVFLAG_SIGNALFD for faster and simpler signal handling. If
  3431. undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
  3432. 2.7 or newer, otherwise disabled.
  3433. =item EV_USE_TIMERFD
  3434. If defined to be C<1>, then libev will assume that C<timerfd ()> is
  3435. available and will probe for kernel support at runtime. This allows
  3436. libev to detect time jumps accurately. If undefined, it will be enabled
  3437. if the headers indicate GNU/Linux + Glibc 2.8 or newer and define
  3438. C<TFD_TIMER_CANCEL_ON_SET>, otherwise disabled.
  3439. =item EV_USE_EVENTFD
  3440. If defined to be C<1>, then libev will assume that C<eventfd ()> is
  3441. available and will probe for kernel support at runtime. This will improve
  3442. C<ev_signal> and C<ev_async> performance and reduce resource consumption.
  3443. If undefined, it will be enabled if the headers indicate GNU/Linux + Glibc
  3444. 2.7 or newer, otherwise disabled.
  3445. =item EV_USE_SELECT
  3446. If undefined or defined to be C<1>, libev will compile in support for the
  3447. C<select>(2) backend. No attempt at auto-detection will be done: if no
  3448. other method takes over, select will be it. Otherwise the select backend
  3449. will not be compiled in.
  3450. =item EV_SELECT_USE_FD_SET
  3451. If defined to C<1>, then the select backend will use the system C<fd_set>
  3452. structure. This is useful if libev doesn't compile due to a missing
  3453. C<NFDBITS> or C<fd_mask> definition or it mis-guesses the bitset layout
  3454. on exotic systems. This usually limits the range of file descriptors to
  3455. some low limit such as 1024 or might have other limitations (winsocket
  3456. only allows 64 sockets). The C<FD_SETSIZE> macro, set before compilation,
  3457. configures the maximum size of the C<fd_set>.
  3459. When defined to C<1>, the select backend will assume that
  3460. select/socket/connect etc. don't understand file descriptors but
  3461. wants osf handles on win32 (this is the case when the select to
  3462. be used is the winsock select). This means that it will call
  3463. C<_get_osfhandle> on the fd to convert it to an OS handle. Otherwise,
  3464. it is assumed that all these functions actually work on fds, even
  3465. on win32. Should not be defined on non-win32 platforms.
  3466. =item EV_FD_TO_WIN32_HANDLE(fd)
  3467. If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
  3468. file descriptors to socket handles. When not defining this symbol (the
  3469. default), then libev will call C<_get_osfhandle>, which is usually
  3470. correct. In some cases, programs use their own file descriptor management,
  3471. in which case they can provide this function to map fds to socket handles.
  3472. =item EV_WIN32_HANDLE_TO_FD(handle)
  3473. If C<EV_SELECT_IS_WINSOCKET> then libev maps handles to file descriptors
  3474. using the standard C<_open_osfhandle> function. For programs implementing
  3475. their own fd to handle mapping, overwriting this function makes it easier
  3476. to do so. This can be done by defining this macro to an appropriate value.
  3477. =item EV_WIN32_CLOSE_FD(fd)
  3478. If programs implement their own fd to handle mapping on win32, then this
  3479. macro can be used to override the C<close> function, useful to unregister
  3480. file descriptors again. Note that the replacement function has to close
  3481. the underlying OS handle.
  3482. =item EV_USE_WSASOCKET
  3483. If defined to be C<1>, libev will use C<WSASocket> to create its internal
  3484. communication socket, which works better in some environments. Otherwise,
  3485. the normal C<socket> function will be used, which works better in other
  3486. environments.
  3487. =item EV_USE_POLL
  3488. If defined to be C<1>, libev will compile in support for the C<poll>(2)
  3489. backend. Otherwise it will be enabled on non-win32 platforms. It
  3490. takes precedence over select.
  3491. =item EV_USE_EPOLL
  3492. If defined to be C<1>, libev will compile in support for the Linux
  3493. C<epoll>(7) backend. Its availability will be detected at runtime,
  3494. otherwise another method will be used as fallback. This is the preferred
  3495. backend for GNU/Linux systems. If undefined, it will be enabled if the
  3496. headers indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
  3497. =item EV_USE_LINUXAIO
  3498. If defined to be C<1>, libev will compile in support for the Linux aio
  3499. backend (C<EV_USE_EPOLL> must also be enabled). If undefined, it will be
  3500. enabled on linux, otherwise disabled.
  3501. =item EV_USE_IOURING
  3502. If defined to be C<1>, libev will compile in support for the Linux
  3503. io_uring backend (C<EV_USE_EPOLL> must also be enabled). Due to it's
  3504. current limitations it has to be requested explicitly. If undefined, it
  3505. will be enabled on linux, otherwise disabled.
  3506. =item EV_USE_KQUEUE
  3507. If defined to be C<1>, libev will compile in support for the BSD style
  3508. C<kqueue>(2) backend. Its actual availability will be detected at runtime,
  3509. otherwise another method will be used as fallback. This is the preferred
  3510. backend for BSD and BSD-like systems, although on most BSDs kqueue only
  3511. supports some types of fds correctly (the only platform we found that
  3512. supports ptys for example was NetBSD), so kqueue might be compiled in, but
  3513. not be used unless explicitly requested. The best way to use it is to find
  3514. out whether kqueue supports your type of fd properly and use an embedded
  3515. kqueue loop.
  3516. =item EV_USE_PORT
  3517. If defined to be C<1>, libev will compile in support for the Solaris
  3518. 10 port style backend. Its availability will be detected at runtime,
  3519. otherwise another method will be used as fallback. This is the preferred
  3520. backend for Solaris 10 systems.
  3521. =item EV_USE_DEVPOLL
  3522. Reserved for future expansion, works like the USE symbols above.
  3523. =item EV_USE_INOTIFY
  3524. If defined to be C<1>, libev will compile in support for the Linux inotify
  3525. interface to speed up C<ev_stat> watchers. Its actual availability will
  3526. be detected at runtime. If undefined, it will be enabled if the headers
  3527. indicate GNU/Linux + Glibc 2.4 or newer, otherwise disabled.
  3528. =item EV_NO_SMP
  3529. If defined to be C<1>, libev will assume that memory is always coherent
  3530. between threads, that is, threads can be used, but threads never run on
  3531. different cpus (or different cpu cores). This reduces dependencies
  3532. and makes libev faster.
  3533. =item EV_NO_THREADS
  3534. If defined to be C<1>, libev will assume that it will never be called from
  3535. different threads (that includes signal handlers), which is a stronger
  3536. assumption than C<EV_NO_SMP>, above. This reduces dependencies and makes
  3537. libev faster.
  3538. =item EV_ATOMIC_T
  3539. Libev requires an integer type (suitable for storing C<0> or C<1>) whose
  3540. access is atomic with respect to other threads or signal contexts. No
  3541. such type is easily found in the C language, so you can provide your own
  3542. type that you know is safe for your purposes. It is used both for signal
  3543. handler "locking" as well as for signal and thread safety in C<ev_async>
  3544. watchers.
  3545. In the absence of this define, libev will use C<sig_atomic_t volatile>
  3546. (from F<signal.h>), which is usually good enough on most platforms.
  3547. =item EV_H (h)
  3548. The name of the F<ev.h> header file used to include it. The default if
  3549. undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
  3550. used to virtually rename the F<ev.h> header file in case of conflicts.
  3551. =item EV_CONFIG_H (h)
  3552. If C<EV_STANDALONE> isn't C<1>, this variable can be used to override
  3553. F<ev.c>'s idea of where to find the F<config.h> file, similarly to
  3554. C<EV_H>, above.
  3555. =item EV_EVENT_H (h)
  3556. Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
  3557. of how the F<event.h> header can be found, the default is C<"event.h">.
  3558. =item EV_PROTOTYPES (h)
  3559. If defined to be C<0>, then F<ev.h> will not define any function
  3560. prototypes, but still define all the structs and other symbols. This is
  3561. occasionally useful if you want to provide your own wrapper functions
  3562. around libev functions.
  3563. =item EV_MULTIPLICITY
  3564. If undefined or defined to C<1>, then all event-loop-specific functions
  3565. will have the C<struct ev_loop *> as first argument, and you can create
  3566. additional independent event loops. Otherwise there will be no support
  3567. for multiple event loops and there is no first event loop pointer
  3568. argument. Instead, all functions act on the single default loop.
  3569. Note that C<EV_DEFAULT> and C<EV_DEFAULT_> will no longer provide a
  3570. default loop when multiplicity is switched off - you always have to
  3571. initialise the loop manually in this case.
  3572. =item EV_MINPRI
  3573. =item EV_MAXPRI
  3574. The range of allowed priorities. C<EV_MINPRI> must be smaller or equal to
  3575. C<EV_MAXPRI>, but otherwise there are no non-obvious limitations. You can
  3576. provide for more priorities by overriding those symbols (usually defined
  3577. to be C<-2> and C<2>, respectively).
  3578. When doing priority-based operations, libev usually has to linearly search
  3579. all the priorities, so having many of them (hundreds) uses a lot of space
  3580. and time, so using the defaults of five priorities (-2 .. +2) is usually
  3581. fine.
  3582. If your embedding application does not need any priorities, defining these
  3583. both to C<0> will save some memory and CPU.
  3587. If undefined or defined to be C<1> (and the platform supports it), then
  3588. the respective watcher type is supported. If defined to be C<0>, then it
  3589. is not. Disabling watcher types mainly saves code size.
  3590. =item EV_FEATURES
  3591. If you need to shave off some kilobytes of code at the expense of some
  3592. speed (but with the full API), you can define this symbol to request
  3593. certain subsets of functionality. The default is to enable all features
  3594. that can be enabled on the platform.
  3595. A typical way to use this symbol is to define it to C<0> (or to a bitset
  3596. with some broad features you want) and then selectively re-enable
  3597. additional parts you want, for example if you want everything minimal,
  3598. but multiple event loop support, async and child watchers and the poll
  3599. backend, use this:
  3600. #define EV_FEATURES 0
  3601. #define EV_MULTIPLICITY 1
  3602. #define EV_USE_POLL 1
  3603. #define EV_CHILD_ENABLE 1
  3604. #define EV_ASYNC_ENABLE 1
  3605. The actual value is a bitset, it can be a combination of the following
  3606. values (by default, all of these are enabled):
  3607. =over 4
  3608. =item C<1> - faster/larger code
  3609. Use larger code to speed up some operations.
  3610. Currently this is used to override some inlining decisions (enlarging the
  3611. code size by roughly 30% on amd64).
  3612. When optimising for size, use of compiler flags such as C<-Os> with
  3613. gcc is recommended, as well as C<-DNDEBUG>, as libev contains a number of
  3614. assertions.
  3615. The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
  3616. (e.g. gcc with C<-Os>).
  3617. =item C<2> - faster/larger data structures
  3618. Replaces the small 2-heap for timer management by a faster 4-heap, larger
  3619. hash table sizes and so on. This will usually further increase code size
  3620. and can additionally have an effect on the size of data structures at
  3621. runtime.
  3622. The default is off when C<__OPTIMIZE_SIZE__> is defined by your compiler
  3623. (e.g. gcc with C<-Os>).
  3624. =item C<4> - full API configuration
  3625. This enables priorities (sets C<EV_MAXPRI>=2 and C<EV_MINPRI>=-2), and
  3626. enables multiplicity (C<EV_MULTIPLICITY>=1).
  3627. =item C<8> - full API
  3628. This enables a lot of the "lesser used" API functions. See C<ev.h> for
  3629. details on which parts of the API are still available without this
  3630. feature, and do not complain if this subset changes over time.
  3631. =item C<16> - enable all optional watcher types
  3632. Enables all optional watcher types. If you want to selectively enable
  3633. only some watcher types other than I/O and timers (e.g. prepare,
  3634. embed, async, child...) you can enable them manually by defining
  3635. C<EV_watchertype_ENABLE> to C<1> instead.
  3636. =item C<32> - enable all backends
  3637. This enables all backends - without this feature, you need to enable at
  3638. least one backend manually (C<EV_USE_SELECT> is a good choice).
  3639. =item C<64> - enable OS-specific "helper" APIs
  3640. Enable inotify, eventfd, signalfd and similar OS-specific helper APIs by
  3641. default.
  3642. =back
  3643. Compiling with C<gcc -Os -DEV_STANDALONE -DEV_USE_EPOLL=1 -DEV_FEATURES=0>
  3644. reduces the compiled size of libev from 24.7Kb code/2.8Kb data to 6.5Kb
  3645. code/0.3Kb data on my GNU/Linux amd64 system, while still giving you I/O
  3646. watchers, timers and monotonic clock support.
  3647. With an intelligent-enough linker (gcc+binutils are intelligent enough
  3648. when you use C<-Wl,--gc-sections -ffunction-sections>) functions unused by
  3649. your program might be left out as well - a binary starting a timer and an
  3650. I/O watcher then might come out at only 5Kb.
  3651. =item EV_API_STATIC
  3652. If this symbol is defined (by default it is not), then all identifiers
  3653. will have static linkage. This means that libev will not export any
  3654. identifiers, and you cannot link against libev anymore. This can be useful
  3655. when you embed libev, only want to use libev functions in a single file,
  3656. and do not want its identifiers to be visible.
  3657. To use this, define C<EV_API_STATIC> and include F<ev.c> in the file that
  3658. wants to use libev.
  3659. This option only works when libev is compiled with a C compiler, as C++
  3660. doesn't support the required declaration syntax.
  3661. =item EV_AVOID_STDIO
  3662. If this is set to C<1> at compiletime, then libev will avoid using stdio
  3663. functions (printf, scanf, perror etc.). This will increase the code size
  3664. somewhat, but if your program doesn't otherwise depend on stdio and your
  3665. libc allows it, this avoids linking in the stdio library which is quite
  3666. big.
  3667. Note that error messages might become less precise when this option is
  3668. enabled.
  3669. =item EV_NSIG
  3670. The highest supported signal number, +1 (or, the number of
  3671. signals): Normally, libev tries to deduce the maximum number of signals
  3672. automatically, but sometimes this fails, in which case it can be
  3673. specified. Also, using a lower number than detected (C<32> should be
  3674. good for about any system in existence) can save some memory, as libev
  3675. statically allocates some 12-24 bytes per signal number.
  3676. =item EV_PID_HASHSIZE
  3677. C<ev_child> watchers use a small hash table to distribute workload by
  3678. pid. The default size is C<16> (or C<1> with C<EV_FEATURES> disabled),
  3679. usually more than enough. If you need to manage thousands of children you
  3680. might want to increase this value (I<must> be a power of two).
  3682. C<ev_stat> watchers use a small hash table to distribute workload by
  3683. inotify watch id. The default size is C<16> (or C<1> with C<EV_FEATURES>
  3684. disabled), usually more than enough. If you need to manage thousands of
  3685. C<ev_stat> watchers you might want to increase this value (I<must> be a
  3686. power of two).
  3687. =item EV_USE_4HEAP
  3688. Heaps are not very cache-efficient. To improve the cache-efficiency of the
  3689. timer and periodics heaps, libev uses a 4-heap when this symbol is defined
  3690. to C<1>. The 4-heap uses more complicated (longer) code but has noticeably
  3691. faster performance with many (thousands) of watchers.
  3692. The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
  3693. will be C<0>.
  3694. =item EV_HEAP_CACHE_AT
  3695. Heaps are not very cache-efficient. To improve the cache-efficiency of the
  3696. timer and periodics heaps, libev can cache the timestamp (I<at>) within
  3697. the heap structure (selected by defining C<EV_HEAP_CACHE_AT> to C<1>),
  3698. which uses 8-12 bytes more per watcher and a few hundred bytes more code,
  3699. but avoids random read accesses on heap changes. This improves performance
  3700. noticeably with many (hundreds) of watchers.
  3701. The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
  3702. will be C<0>.
  3703. =item EV_VERIFY
  3704. Controls how much internal verification (see C<ev_verify ()>) will
  3705. be done: If set to C<0>, no internal verification code will be compiled
  3706. in. If set to C<1>, then verification code will be compiled in, but not
  3707. called. If set to C<2>, then the internal verification code will be
  3708. called once per loop, which can slow down libev. If set to C<3>, then the
  3709. verification code will be called very frequently, which will slow down
  3710. libev considerably.
  3711. Verification errors are reported via C's C<assert> mechanism, so if you
  3712. disable that (e.g. by defining C<NDEBUG>) then no errors will be reported.
  3713. The default is C<1>, unless C<EV_FEATURES> overrides it, in which case it
  3714. will be C<0>.
  3715. =item EV_COMMON
  3716. By default, all watchers have a C<void *data> member. By redefining
  3717. this macro to something else you can include more and other types of
  3718. members. You have to define it each time you include one of the files,
  3719. though, and it must be identical each time.
  3720. For example, the perl EV module uses something like this:
  3721. #define EV_COMMON \
  3722. SV *self; /* contains this struct */ \
  3723. SV *cb_sv, *fh /* note no trailing ";" */
  3724. =item EV_CB_DECLARE (type)
  3725. =item EV_CB_INVOKE (watcher, revents)
  3726. =item ev_set_cb (ev, cb)
  3727. Can be used to change the callback member declaration in each watcher,
  3728. and the way callbacks are invoked and set. Must expand to a struct member
  3729. definition and a statement, respectively. See the F<ev.h> header file for
  3730. their default definitions. One possible use for overriding these is to
  3731. avoid the C<struct ev_loop *> as first argument in all cases, or to use
  3732. method calls instead of plain function calls in C++.
  3733. =back
  3735. If you need to re-export the API (e.g. via a DLL) and you need a list of
  3736. exported symbols, you can use the provided F<Symbol.*> files which list
  3737. all public symbols, one per line:
  3738. Symbols.ev for libev proper
  3739. Symbols.event for the libevent emulation
  3740. This can also be used to rename all public symbols to avoid clashes with
  3741. multiple versions of libev linked together (which is obviously bad in
  3742. itself, but sometimes it is inconvenient to avoid this).
  3743. A sed command like this will create wrapper C<#define>'s that you need to
  3744. include before including F<ev.h>:
  3745. <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
  3746. This would create a file F<wrap.h> which essentially looks like this:
  3747. #define ev_backend myprefix_ev_backend
  3748. #define ev_check_start myprefix_ev_check_start
  3749. #define ev_check_stop myprefix_ev_check_stop
  3750. ...
  3751. =head2 EXAMPLES
  3752. For a real-world example of a program the includes libev
  3753. verbatim, you can have a look at the EV perl module
  3754. (L<http://software.schmorp.de/pkg/EV.html>). It has the libev files in
  3755. the F<libev/> subdirectory and includes them in the F<EV/EVAPI.h> (public
  3756. interface) and F<EV.xs> (implementation) files. Only the F<EV.xs> file
  3757. will be compiled. It is pretty complex because it provides its own header
  3758. file.
  3759. The usage in rxvt-unicode is simpler. It has a F<ev_cpp.h> header file
  3760. that everybody includes and which overrides some configure choices:
  3761. #define EV_FEATURES 8
  3762. #define EV_USE_SELECT 1
  3763. #define EV_PREPARE_ENABLE 1
  3764. #define EV_IDLE_ENABLE 1
  3765. #define EV_SIGNAL_ENABLE 1
  3766. #define EV_CHILD_ENABLE 1
  3767. #define EV_USE_STDEXCEPT 0
  3768. #define EV_CONFIG_H <config.h>
  3769. #include "ev++.h"
  3770. And a F<ev_cpp.C> implementation file that contains libev proper and is compiled:
  3771. #include "ev_cpp.h"
  3772. #include "ev.c"
  3775. =head3 THREADS
  3776. All libev functions are reentrant and thread-safe unless explicitly
  3777. documented otherwise, but libev implements no locking itself. This means
  3778. that you can use as many loops as you want in parallel, as long as there
  3779. are no concurrent calls into any libev function with the same loop
  3780. parameter (C<ev_default_*> calls have an implicit default loop parameter,
  3781. of course): libev guarantees that different event loops share no data
  3782. structures that need any locking.
  3783. Or to put it differently: calls with different loop parameters can be done
  3784. concurrently from multiple threads, calls with the same loop parameter
  3785. must be done serially (but can be done from different threads, as long as
  3786. only one thread ever is inside a call at any point in time, e.g. by using
  3787. a mutex per loop).
  3788. Specifically to support threads (and signal handlers), libev implements
  3789. so-called C<ev_async> watchers, which allow some limited form of
  3790. concurrency on the same event loop, namely waking it up "from the
  3791. outside".
  3792. If you want to know which design (one loop, locking, or multiple loops
  3793. without or something else still) is best for your problem, then I cannot
  3794. help you, but here is some generic advice:
  3795. =over 4
  3796. =item * most applications have a main thread: use the default libev loop
  3797. in that thread, or create a separate thread running only the default loop.
  3798. This helps integrating other libraries or software modules that use libev
  3799. themselves and don't care/know about threading.
  3800. =item * one loop per thread is usually a good model.
  3801. Doing this is almost never wrong, sometimes a better-performance model
  3802. exists, but it is always a good start.
  3803. =item * other models exist, such as the leader/follower pattern, where one
  3804. loop is handed through multiple threads in a kind of round-robin fashion.
  3805. Choosing a model is hard - look around, learn, know that usually you can do
  3806. better than you currently do :-)
  3807. =item * often you need to talk to some other thread which blocks in the
  3808. event loop.
  3809. C<ev_async> watchers can be used to wake them up from other threads safely
  3810. (or from signal contexts...).
  3811. An example use would be to communicate signals or other events that only
  3812. work in the default loop by registering the signal watcher with the
  3813. default loop and triggering an C<ev_async> watcher from the default loop
  3814. watcher callback into the event loop interested in the signal.
  3815. =back
  3816. See also L</THREAD LOCKING EXAMPLE>.
  3817. =head3 COROUTINES
  3818. Libev is very accommodating to coroutines ("cooperative threads"):
  3819. libev fully supports nesting calls to its functions from different
  3820. coroutines (e.g. you can call C<ev_run> on the same loop from two
  3821. different coroutines, and switch freely between both coroutines running
  3822. the loop, as long as you don't confuse yourself). The only exception is
  3823. that you must not do this from C<ev_periodic> reschedule callbacks.
  3824. Care has been taken to ensure that libev does not keep local state inside
  3825. C<ev_run>, and other calls do not usually allow for coroutine switches as
  3826. they do not call any callbacks.
  3827. =head2 COMPILER WARNINGS
  3828. Depending on your compiler and compiler settings, you might get no or a
  3829. lot of warnings when compiling libev code. Some people are apparently
  3830. scared by this.
  3831. However, these are unavoidable for many reasons. For one, each compiler
  3832. has different warnings, and each user has different tastes regarding
  3833. warning options. "Warn-free" code therefore cannot be a goal except when
  3834. targeting a specific compiler and compiler-version.
  3835. Another reason is that some compiler warnings require elaborate
  3836. workarounds, or other changes to the code that make it less clear and less
  3837. maintainable.
  3838. And of course, some compiler warnings are just plain stupid, or simply
  3839. wrong (because they don't actually warn about the condition their message
  3840. seems to warn about). For example, certain older gcc versions had some
  3841. warnings that resulted in an extreme number of false positives. These have
  3842. been fixed, but some people still insist on making code warn-free with
  3843. such buggy versions.
  3844. While libev is written to generate as few warnings as possible,
  3845. "warn-free" code is not a goal, and it is recommended not to build libev
  3846. with any compiler warnings enabled unless you are prepared to cope with
  3847. them (e.g. by ignoring them). Remember that warnings are just that:
  3848. warnings, not errors, or proof of bugs.
  3849. =head2 VALGRIND
  3850. Valgrind has a special section here because it is a popular tool that is
  3851. highly useful. Unfortunately, valgrind reports are very hard to interpret.
  3852. If you think you found a bug (memory leak, uninitialised data access etc.)
  3853. in libev, then check twice: If valgrind reports something like:
  3854. ==2274== definitely lost: 0 bytes in 0 blocks.
  3855. ==2274== possibly lost: 0 bytes in 0 blocks.
  3856. ==2274== still reachable: 256 bytes in 1 blocks.
  3857. Then there is no memory leak, just as memory accounted to global variables
  3858. is not a memleak - the memory is still being referenced, and didn't leak.
  3859. Similarly, under some circumstances, valgrind might report kernel bugs
  3860. as if it were a bug in libev (e.g. in realloc or in the poll backend,
  3861. although an acceptable workaround has been found here), or it might be
  3862. confused.
  3863. Keep in mind that valgrind is a very good tool, but only a tool. Don't
  3864. make it into some kind of religion.
  3865. If you are unsure about something, feel free to contact the mailing list
  3866. with the full valgrind report and an explanation on why you think this
  3867. is a bug in libev (best check the archives, too :). However, don't be
  3868. annoyed when you get a brisk "this is no bug" answer and take the chance
  3869. of learning how to interpret valgrind properly.
  3870. If you need, for some reason, empty reports from valgrind for your project
  3871. I suggest using suppression lists.
  3872. =head1 PORTABILITY NOTES
  3874. GNU/Linux is the only common platform that supports 64 bit file/large file
  3875. interfaces but I<disables> them by default.
  3876. That means that libev compiled in the default environment doesn't support
  3877. files larger than 2GiB or so, which mainly affects C<ev_stat> watchers.
  3878. Unfortunately, many programs try to work around this GNU/Linux issue
  3879. by enabling the large file API, which makes them incompatible with the
  3880. standard libev compiled for their system.
  3881. Likewise, libev cannot enable the large file API itself as this would
  3882. suddenly make it incompatible to the default compile time environment,
  3883. i.e. all programs not using special compile switches.
  3884. =head2 OS/X AND DARWIN BUGS
  3885. The whole thing is a bug if you ask me - basically any system interface
  3886. you touch is broken, whether it is locales, poll, kqueue or even the
  3887. OpenGL drivers.
  3888. =head3 C<kqueue> is buggy
  3889. The kqueue syscall is broken in all known versions - most versions support
  3890. only sockets, many support pipes.
  3891. Libev tries to work around this by not using C<kqueue> by default on this
  3892. rotten platform, but of course you can still ask for it when creating a
  3893. loop - embedding a socket-only kqueue loop into a select-based one is
  3894. probably going to work well.
  3895. =head3 C<poll> is buggy
  3896. Instead of fixing C<kqueue>, Apple replaced their (working) C<poll>
  3897. implementation by something calling C<kqueue> internally around the 10.5.6
  3898. release, so now C<kqueue> I<and> C<poll> are broken.
  3899. Libev tries to work around this by not using C<poll> by default on
  3900. this rotten platform, but of course you can still ask for it when creating
  3901. a loop.
  3902. =head3 C<select> is buggy
  3903. All that's left is C<select>, and of course Apple found a way to fuck this
  3904. one up as well: On OS/X, C<select> actively limits the number of file
  3905. descriptors you can pass in to 1024 - your program suddenly crashes when
  3906. you use more.
  3907. There is an undocumented "workaround" for this - defining
  3908. C<_DARWIN_UNLIMITED_SELECT>, which libev tries to use, so select I<should>
  3909. work on OS/X.
  3911. =head3 C<errno> reentrancy
  3912. The default compile environment on Solaris is unfortunately so
  3913. thread-unsafe that you can't even use components/libraries compiled
  3914. without C<-D_REENTRANT> in a threaded program, which, of course, isn't
  3915. defined by default. A valid, if stupid, implementation choice.
  3916. If you want to use libev in threaded environments you have to make sure
  3917. it's compiled with C<_REENTRANT> defined.
  3918. =head3 Event port backend
  3919. The scalable event interface for Solaris is called "event
  3920. ports". Unfortunately, this mechanism is very buggy in all major
  3921. releases. If you run into high CPU usage, your program freezes or you get
  3922. a large number of spurious wakeups, make sure you have all the relevant
  3923. and latest kernel patches applied. No, I don't know which ones, but there
  3924. are multiple ones to apply, and afterwards, event ports actually work
  3925. great.
  3926. If you can't get it to work, you can try running the program by setting
  3927. the environment variable C<LIBEV_FLAGS=3> to only allow C<poll> and
  3928. C<select> backends.
  3929. =head2 AIX POLL BUG
  3930. AIX unfortunately has a broken C<poll.h> header. Libev works around
  3931. this by trying to avoid the poll backend altogether (i.e. it's not even
  3932. compiled in), which normally isn't a big problem as C<select> works fine
  3933. with large bitsets on AIX, and AIX is dead anyway.
  3935. =head3 General issues
  3936. Win32 doesn't support any of the standards (e.g. POSIX) that libev
  3937. requires, and its I/O model is fundamentally incompatible with the POSIX
  3938. model. Libev still offers limited functionality on this platform in
  3939. the form of the C<EVBACKEND_SELECT> backend, and only supports socket
  3940. descriptors. This only applies when using Win32 natively, not when using
  3941. e.g. cygwin. Actually, it only applies to the microsofts own compilers,
  3942. as every compiler comes with a slightly differently broken/incompatible
  3943. environment.
  3944. Lifting these limitations would basically require the full
  3945. re-implementation of the I/O system. If you are into this kind of thing,
  3946. then note that glib does exactly that for you in a very portable way (note
  3947. also that glib is the slowest event library known to man).
  3948. There is no supported compilation method available on windows except
  3949. embedding it into other applications.
  3950. Sensible signal handling is officially unsupported by Microsoft - libev
  3951. tries its best, but under most conditions, signals will simply not work.
  3952. Not a libev limitation but worth mentioning: windows apparently doesn't
  3953. accept large writes: instead of resulting in a partial write, windows will
  3954. either accept everything or return C<ENOBUFS> if the buffer is too large,
  3955. so make sure you only write small amounts into your sockets (less than a
  3956. megabyte seems safe, but this apparently depends on the amount of memory
  3957. available).
  3958. Due to the many, low, and arbitrary limits on the win32 platform and
  3959. the abysmal performance of winsockets, using a large number of sockets
  3960. is not recommended (and not reasonable). If your program needs to use
  3961. more than a hundred or so sockets, then likely it needs to use a totally
  3962. different implementation for windows, as libev offers the POSIX readiness
  3963. notification model, which cannot be implemented efficiently on windows
  3964. (due to Microsoft monopoly games).
  3965. A typical way to use libev under windows is to embed it (see the embedding
  3966. section for details) and use the following F<evwrap.h> header file instead
  3967. of F<ev.h>:
  3968. #define EV_STANDALONE /* keeps ev from requiring config.h */
  3969. #define EV_SELECT_IS_WINSOCKET 1 /* configure libev for windows select */
  3970. #include "ev.h"
  3971. And compile the following F<evwrap.c> file into your project (make sure
  3972. you do I<not> compile the F<ev.c> or any other embedded source files!):
  3973. #include "evwrap.h"
  3974. #include "ev.c"
  3975. =head3 The winsocket C<select> function
  3976. The winsocket C<select> function doesn't follow POSIX in that it
  3977. requires socket I<handles> and not socket I<file descriptors> (it is
  3978. also extremely buggy). This makes select very inefficient, and also
  3979. requires a mapping from file descriptors to socket handles (the Microsoft
  3980. C runtime provides the function C<_open_osfhandle> for this). See the
  3981. discussion of the C<EV_SELECT_USE_FD_SET>, C<EV_SELECT_IS_WINSOCKET> and
  3982. C<EV_FD_TO_WIN32_HANDLE> preprocessor symbols for more info.
  3983. The configuration for a "naked" win32 using the Microsoft runtime
  3984. libraries and raw winsocket select is:
  3985. #define EV_USE_SELECT 1
  3986. #define EV_SELECT_IS_WINSOCKET 1 /* forces EV_SELECT_USE_FD_SET, too */
  3987. Note that winsockets handling of fd sets is O(n), so you can easily get a
  3988. complexity in the O(n²) range when using win32.
  3989. =head3 Limited number of file descriptors
  3990. Windows has numerous arbitrary (and low) limits on things.
  3991. Early versions of winsocket's select only supported waiting for a maximum
  3992. of C<64> handles (probably owning to the fact that all windows kernels
  3993. can only wait for C<64> things at the same time internally; Microsoft
  3994. recommends spawning a chain of threads and wait for 63 handles and the
  3995. previous thread in each. Sounds great!).
  3996. Newer versions support more handles, but you need to define C<FD_SETSIZE>
  3997. to some high number (e.g. C<2048>) before compiling the winsocket select
  3998. call (which might be in libev or elsewhere, for example, perl and many
  3999. other interpreters do their own select emulation on windows).
  4000. Another limit is the number of file descriptors in the Microsoft runtime
  4001. libraries, which by default is C<64> (there must be a hidden I<64>
  4002. fetish or something like this inside Microsoft). You can increase this
  4003. by calling C<_setmaxstdio>, which can increase this limit to C<2048>
  4004. (another arbitrary limit), but is broken in many versions of the Microsoft
  4005. runtime libraries. This might get you to about C<512> or C<2048> sockets
  4006. (depending on windows version and/or the phase of the moon). To get more,
  4007. you need to wrap all I/O functions and provide your own fd management, but
  4008. the cost of calling select (O(n²)) will likely make this unworkable.
  4010. In addition to a working ISO-C implementation and of course the
  4011. backend-specific APIs, libev relies on a few additional extensions:
  4012. =over 4
  4013. =item C<void (*)(ev_watcher_type *, int revents)> must have compatible
  4014. calling conventions regardless of C<ev_watcher_type *>.
  4015. Libev assumes not only that all watcher pointers have the same internal
  4016. structure (guaranteed by POSIX but not by ISO C for example), but it also
  4017. assumes that the same (machine) code can be used to call any watcher
  4018. callback: The watcher callbacks have different type signatures, but libev
  4019. calls them using an C<ev_watcher *> internally.
  4020. =item null pointers and integer zero are represented by 0 bytes
  4021. Libev uses C<memset> to initialise structs and arrays to C<0> bytes, and
  4022. relies on this setting pointers and integers to null.
  4023. =item pointer accesses must be thread-atomic
  4024. Accessing a pointer value must be atomic, it must both be readable and
  4025. writable in one piece - this is the case on all current architectures.
  4026. =item C<sig_atomic_t volatile> must be thread-atomic as well
  4027. The type C<sig_atomic_t volatile> (or whatever is defined as
  4028. C<EV_ATOMIC_T>) must be atomic with respect to accesses from different
  4029. threads. This is not part of the specification for C<sig_atomic_t>, but is
  4030. believed to be sufficiently portable.
  4031. =item C<sigprocmask> must work in a threaded environment
  4032. Libev uses C<sigprocmask> to temporarily block signals. This is not
  4033. allowed in a threaded program (C<pthread_sigmask> has to be used). Typical
  4034. pthread implementations will either allow C<sigprocmask> in the "main
  4035. thread" or will block signals process-wide, both behaviours would
  4036. be compatible with libev. Interaction between C<sigprocmask> and
  4037. C<pthread_sigmask> could complicate things, however.
  4038. The most portable way to handle signals is to block signals in all threads
  4039. except the initial one, and run the signal handling loop in the initial
  4040. thread as well.
  4041. =item C<long> must be large enough for common memory allocation sizes
  4042. To improve portability and simplify its API, libev uses C<long> internally
  4043. instead of C<size_t> when allocating its data structures. On non-POSIX
  4044. systems (Microsoft...) this might be unexpectedly low, but is still at
  4045. least 31 bits everywhere, which is enough for hundreds of millions of
  4046. watchers.
  4047. =item C<double> must hold a time value in seconds with enough accuracy
  4048. The type C<double> is used to represent timestamps. It is required to
  4049. have at least 51 bits of mantissa (and 9 bits of exponent), which is
  4050. good enough for at least into the year 4000 with millisecond accuracy
  4051. (the design goal for libev). This requirement is overfulfilled by
  4052. implementations using IEEE 754, which is basically all existing ones.
  4053. With IEEE 754 doubles, you get microsecond accuracy until at least the
  4054. year 2255 (and millisecond accuracy till the year 287396 - by then, libev
  4055. is either obsolete or somebody patched it to use C<long double> or
  4056. something like that, just kidding).
  4057. =back
  4058. If you know of other additional requirements drop me a note.
  4060. In this section the complexities of (many of) the algorithms used inside
  4061. libev will be documented. For complexity discussions about backends see
  4062. the documentation for C<ev_default_init>.
  4063. All of the following are about amortised time: If an array needs to be
  4064. extended, libev needs to realloc and move the whole array, but this
  4065. happens asymptotically rarer with higher number of elements, so O(1) might
  4066. mean that libev does a lengthy realloc operation in rare cases, but on
  4067. average it is much faster and asymptotically approaches constant time.
  4068. =over 4
  4069. =item Starting and stopping timer/periodic watchers: O(log skipped_other_timers)
  4070. This means that, when you have a watcher that triggers in one hour and
  4071. there are 100 watchers that would trigger before that, then inserting will
  4072. have to skip roughly seven (C<ld 100>) of these watchers.
  4073. =item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
  4074. That means that changing a timer costs less than removing/adding them,
  4075. as only the relative motion in the event queue has to be paid for.
  4076. =item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
  4077. These just add the watcher into an array or at the head of a list.
  4078. =item Stopping check/prepare/idle/fork/async watchers: O(1)
  4079. =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
  4080. These watchers are stored in lists, so they need to be walked to find the
  4081. correct watcher to remove. The lists are usually short (you don't usually
  4082. have many watchers waiting for the same fd or signal: one is typical, two
  4083. is rare).
  4084. =item Finding the next timer in each loop iteration: O(1)
  4085. By virtue of using a binary or 4-heap, the next timer is always found at a
  4086. fixed position in the storage array.
  4087. =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
  4088. A change means an I/O watcher gets started or stopped, which requires
  4089. libev to recalculate its status (and possibly tell the kernel, depending
  4090. on backend and whether C<ev_io_set> was used).
  4091. =item Activating one watcher (putting it into the pending state): O(1)
  4092. =item Priority handling: O(number_of_priorities)
  4093. Priorities are implemented by allocating some space for each
  4094. priority. When doing priority-based operations, libev usually has to
  4095. linearly search all the priorities, but starting/stopping and activating
  4096. watchers becomes O(1) with respect to priority handling.
  4097. =item Sending an ev_async: O(1)
  4098. =item Processing ev_async_send: O(number_of_async_watchers)
  4099. =item Processing signals: O(max_signal_number)
  4100. Sending involves a system call I<iff> there were no other C<ev_async_send>
  4101. calls in the current loop iteration and the loop is currently
  4102. blocked. Checking for async and signal events involves iterating over all
  4103. running async watchers or all signal numbers.
  4104. =back
  4105. =head1 PORTING FROM LIBEV 3.X TO 4.X
  4106. The major version 4 introduced some incompatible changes to the API.
  4107. At the moment, the C<ev.h> header file provides compatibility definitions
  4108. for all changes, so most programs should still compile. The compatibility
  4109. layer might be removed in later versions of libev, so better update to the
  4110. new API early than late.
  4111. =over 4
  4112. =item C<EV_COMPAT3> backwards compatibility mechanism
  4113. The backward compatibility mechanism can be controlled by
  4115. section.
  4116. =item C<ev_default_destroy> and C<ev_default_fork> have been removed
  4117. These calls can be replaced easily by their C<ev_loop_xxx> counterparts:
  4118. ev_loop_destroy (EV_DEFAULT_UC);
  4119. ev_loop_fork (EV_DEFAULT);
  4120. =item function/symbol renames
  4121. A number of functions and symbols have been renamed:
  4122. ev_loop => ev_run
  4125. ev_unloop => ev_break
  4130. ev_loop_count => ev_iteration
  4131. ev_loop_depth => ev_depth
  4132. ev_loop_verify => ev_verify
  4133. Most functions working on C<struct ev_loop> objects don't have an
  4134. C<ev_loop_> prefix, so it was removed; C<ev_loop>, C<ev_unloop> and
  4135. associated constants have been renamed to not collide with the C<struct
  4136. ev_loop> anymore and C<EV_TIMER> now follows the same naming scheme
  4137. as all other watcher types. Note that C<ev_loop_fork> is still called
  4138. C<ev_loop_fork> because it would otherwise clash with the C<ev_fork>
  4139. typedef.
  4140. =item C<EV_MINIMAL> mechanism replaced by C<EV_FEATURES>
  4141. The preprocessor symbol C<EV_MINIMAL> has been replaced by a different
  4142. mechanism, C<EV_FEATURES>. Programs using C<EV_MINIMAL> usually compile
  4143. and work, but the library code will of course be larger.
  4144. =back
  4145. =head1 GLOSSARY
  4146. =over 4
  4147. =item active
  4148. A watcher is active as long as it has been started and not yet stopped.
  4149. See L</WATCHER STATES> for details.
  4150. =item application
  4151. In this document, an application is whatever is using libev.
  4152. =item backend
  4153. The part of the code dealing with the operating system interfaces.
  4154. =item callback
  4155. The address of a function that is called when some event has been
  4156. detected. Callbacks are being passed the event loop, the watcher that
  4157. received the event, and the actual event bitset.
  4158. =item callback/watcher invocation
  4159. The act of calling the callback associated with a watcher.
  4160. =item event
  4161. A change of state of some external event, such as data now being available
  4162. for reading on a file descriptor, time having passed or simply not having
  4163. any other events happening anymore.
  4164. In libev, events are represented as single bits (such as C<EV_READ> or
  4165. C<EV_TIMER>).
  4166. =item event library
  4167. A software package implementing an event model and loop.
  4168. =item event loop
  4169. An entity that handles and processes external events and converts them
  4170. into callback invocations.
  4171. =item event model
  4172. The model used to describe how an event loop handles and processes
  4173. watchers and events.
  4174. =item pending
  4175. A watcher is pending as soon as the corresponding event has been
  4176. detected. See L</WATCHER STATES> for details.
  4177. =item real time
  4178. The physical time that is observed. It is apparently strictly monotonic :)
  4179. =item wall-clock time
  4180. The time and date as shown on clocks. Unlike real time, it can actually
  4181. be wrong and jump forwards and backwards, e.g. when you adjust your
  4182. clock.
  4183. =item watcher
  4184. A data structure that describes interest in certain events. Watchers need
  4185. to be started (attached to an event loop) before they can receive events.
  4186. =back
  4187. =head1 AUTHOR
  4188. Marc Lehmann <libev@schmorp.de>, with repeated corrections by Mikael
  4189. Magnusson and Emanuele Giaquinta, and minor corrections by many others.