--- cvsroot/EV/README 2007/11/03 16:25:49 1.7 +++ cvsroot/EV/README 2007/12/08 14:44:21 1.17 @@ -11,20 +11,20 @@ warn "is called after 2s"; }; - my $w = EV::timer 2, 1, sub { - warn "is called roughly every 2s (repeat = 1)"; + my $w = EV::timer 2, 2, sub { + warn "is called roughly every 2s (repeat = 2)"; }; undef $w; # destroy event watcher again - my $w = EV::periodic 0, 60, sub { + my $w = EV::periodic 0, 60, 0, sub { warn "is called every minute, on the minute, exactly"; }; # IO my $w = EV::io *STDIN, EV::READ, sub { - my ($w, $revents) = @_; # all callbacks get the watcher object and event mask + my ($w, $revents) = @_; # all callbacks receive the watcher and event mask warn "stdin is readable, you entered: ", ; }; @@ -34,26 +34,32 @@ warn "sigquit received\n"; }; - my $w = EV::signal 3, sub { - warn "sigquit received (this is GNU/Linux, right?)\n"; - }; - # CHILD/PID STATUS CHANGES my $w = EV::child 666, sub { my ($w, $revents) = @_; - # my $pid = $w->rpid; my $status = $w->rstatus; }; + + # STAT CHANGES + my $w = EV::stat "/etc/passwd", 10, sub { + my ($w, $revents) = @_; + warn $w->path, " has changed somehow.\n"; + }; # MAINLOOP - EV::loop; # loop until EV::loop_done is called + EV::loop; # loop until EV::unloop is called or all watchers stop EV::loop EV::LOOP_ONESHOT; # block until at least one event could be handled EV::loop EV::LOOP_NONBLOCK; # try to handle same events, but do not block DESCRIPTION This module provides an interface to libev - (). + (). While the documentation + below is comprehensive, one might also consult the documentation of + libev itself () for more subtle + details on watcher semantics or some discussion on the available + backends, or how to force a specific backend with "LIBEV_FLAGS", or just + about in any case because it has much more detailed information. BASIC INTERFACE $EV::DIED @@ -71,13 +77,13 @@ This is the time that (relative) timers are based on, and refering to it is usually faster then calling EV::time. - $method = EV::ev_method + $method = EV::method Returns an integer describing the backend used by libev (EV::METHOD_SELECT or EV::METHOD_EPOLL). EV::loop [$flags] Begin checking for events and calling callbacks. It returns when a - callback calls EV::loop_done. + callback calls EV::unloop. The $flags argument can be one of the following: @@ -85,14 +91,50 @@ EV::LOOP_ONESHOT block at most once (wait, but do not loop) EV::LOOP_NONBLOCK do not block at all (fetch/handle events but do not wait) - EV::loop_done [$how] - When called with no arguments or an argument of 1, makes the - innermost call to EV::loop return. - - When called with an agrument of 2, all calls to EV::loop will return - as fast as possible. + EV::unloop [$how] + When called with no arguments or an argument of EV::UNLOOP_ONE, + makes the innermost call to EV::loop return. + + When called with an argument of EV::UNLOOP_ALL, all calls to + EV::loop will return as fast as possible. + + $count = EV::loop_count + Return the number of times the event loop has polled for new events. + Sometiems useful as a generation counter. + + EV::once $fh_or_undef, $events, $timeout, $cb->($revents) + This function rolls together an I/O and a timer watcher for a single + one-shot event without the need for managing a watcher object. + + If $fh_or_undef is a filehandle or file descriptor, then $events + must be a bitset containing either "EV::READ", "EV::WRITE" or + "EV::READ | EV::WRITE", indicating the type of I/O event you want to + wait for. If you do not want to wait for some I/O event, specify + "undef" for $fh_or_undef and 0 for $events). + + If timeout is "undef" or negative, then there will be no timeout. + Otherwise a EV::timer with this value will be started. + + When an error occurs or either the timeout or I/O watcher triggers, + then the callback will be called with the received event set (in + general you can expect it to be a combination of "EV:ERROR", + "EV::READ", "EV::WRITE" and "EV::TIMEOUT"). + + EV::once doesn't return anything: the watchers stay active till + either of them triggers, then they will be stopped and freed, and + the callback invoked. + + EV::feed_fd_event ($fd, $revents) + Feed an event on a file descriptor into EV. EV will react to this + call as if the readyness notifications specified by $revents (a + combination of "EV::READ" and "EV::WRITE") happened on the file + descriptor $fd. + + EV::feed_signal_event ($signal) + Feed a signal event into EV. EV will react to this call as if the + signal specified by $signal had occured. - WATCHER + WATCHER OBJECTS A watcher is an object that gets created to record your interest in some event. For instance, if you want to wait for STDIN to become readable, you would create an EV::io watcher for that: @@ -110,7 +152,7 @@ Each watcher type has its associated bit in revents, so you can use the same callback for multiple watchers. The event mask is named after the type, i..e. EV::child sets EV::CHILD, EV::prepare sets EV::PREPARE, - EV::periodic sets EV::PERIODIC and so on, with the exception of IO + EV::periodic sets EV::PERIODIC and so on, with the exception of I/O events (which can set both EV::READ and EV::WRITE bits), and EV::timer (which uses EV::TIMEOUT). @@ -126,13 +168,8 @@ ->fh and so on) automatically stop and start it again if it is active, which means pending events get lost. - WATCHER TYPES - Now lets move to the existing watcher types and asociated methods. - - The following methods are available for all watchers. Then followes a - description of each watcher constructor (EV::io, EV::timer, - EV::periodic, EV::signal, EV::child, EV::idle, EV::prepare and - EV::check), followed by any type-specific methods (if any). + COMMON WATCHER METHODS + This section lists methods common to all watchers. $w->start Starts a watcher if it isn't active already. Does nothing to an @@ -143,12 +180,22 @@ $w->stop Stop a watcher if it is active. Also clear any pending events (events that have been received but that didn't yet result in a - callback invocation), regardless of wether the watcher was active or - not. + callback invocation), regardless of whether the watcher was active + or not. $bool = $w->is_active Returns true if the watcher is active, false otherwise. + $current_data = $w->data + $old_data = $w->data ($new_data) + Queries a freely usable data scalar on the watcher and optionally + changes it. This is a way to associate custom data with a watcher: + + my $w = EV::timer 60, 0, sub { + warn $_[0]->data; + }; + $w->data ("print me!"); + $current_cb = $w->cb $old_cb = $w->cb ($new_cb) Queries the callback on the watcher and optionally changes it. You @@ -162,15 +209,62 @@ EV::MINPRI (default -2). If the priority is outside this range it will automatically be normalised to the nearest valid priority. - The default priority of any newly-created weatcher is 0. + The default priority of any newly-created watcher is 0. + + Note that the priority semantics have not yet been fleshed out and + are subject to almost certain change. - $w->trigger ($revents) + $w->invoke ($revents) Call the callback *now* with the given event mask. + $w->feed_event ($revents) + Feed some events on this watcher into EV. EV will react to this call + as if the watcher had received the given $revents mask. + + $revents = $w->clear_pending + If the watcher is pending, this function returns clears its pending + status and returns its $revents bitset (as if its callback was + invoked). If the watcher isn't pending it does nothing and returns + 0. + + $previous_state = $w->keepalive ($bool) + Normally, "EV::loop" will return when there are no active watchers + (which is a "deadlock" because no progress can be made anymore). + This is convinient because it allows you to start your watchers (and + your jobs), call "EV::loop" once and when it returns you know that + all your jobs are finished (or they forgot to register some watchers + for their task :). + + Sometimes, however, this gets in your way, for example when you the + module that calls "EV::loop" (usually the main program) is not the + same module as a long-living watcher (for example a DNS client + module written by somebody else even). Then you might want any + outstanding requests to be handled, but you would not want to keep + "EV::loop" from returning just because you happen to have this + long-running UDP port watcher. + + In this case you can clear the keepalive status, which means that + even though your watcher is active, it won't keep "EV::loop" from + returning. + + The initial value for keepalive is true (enabled), and you cna + change it any time. + + Example: Register an I/O watcher for some UDP socket but do not keep + the event loop from running just because of that watcher. + + my $udp_socket = ... + my $udp_watcher = EV::io $udp_socket, EV::READ, sub { ... }; + $udp_watcher->keepalive (0); + + WATCHER TYPES + Each of the following subsections describes a single watcher type. + + I/O WATCHERS - is this file descriptor readable or writable? $w = EV::io $fileno_or_fh, $eventmask, $callback $w = EV::io_ns $fileno_or_fh, $eventmask, $callback As long as the returned watcher object is alive, call the $callback - when the events specified in $eventmask. + when at least one of events specified in $eventmask occurs. The $eventmask can be one or more of these constants ORed together: @@ -192,18 +286,21 @@ $old_eventmask = $w->events ($new_eventmask) Returns the previously set event mask and optionally set a new one. + TIMER WATCHERS - relative and optionally repeating timeouts $w = EV::timer $after, $repeat, $callback $w = EV::timer_ns $after, $repeat, $callback - Calls the callback after $after seconds. If $repeat is non-zero, the - timer will be restarted (with the $repeat value as $after) after the - callback returns. + Calls the callback after $after seconds (which may be fractional). + If $repeat is non-zero, the timer will be restarted (with the + $repeat value as $after) after the callback returns. This means that the callback would be called roughly after $after - seconds, and then every $repeat seconds. "Roughly" because the time - of callback processing is not taken into account, so the timer will - slowly drift. If that isn't acceptable, look at EV::periodic. + seconds, and then every $repeat seconds. The timer does his best not + to drift, but it will not invoke the timer more often then once per + event loop iteration, and might drift in other cases. If that isn't + acceptable, look at EV::periodic, which can provide long-term stable + timers. - The timer is based on a monotonic clock, that is if somebody is + The timer is based on a monotonic clock, that is, if somebody is sitting in front of the machine while the timer is running and changes the system clock, the timer will nevertheless run (roughly) the same time. @@ -213,18 +310,19 @@ $w->set ($after, $repeat) Reconfigures the watcher, see the constructor above for details. Can - be at any time. + be called at any time. $w->again Similar to the "start" method, but has special semantics for repeating timers: + If the timer is active and non-repeating, it will be stopped. + If the timer is active and repeating, reset the timeout to occur $repeat seconds after now. - If the timer is active and non-repeating, it will be stopped. - - If the timer is in active and repeating, start it. + If the timer is inactive and repeating, start it using the repeat + value. Otherwise do nothing. @@ -233,43 +331,91 @@ and $repeat, and then, in the read/write watcher, run the "again" method on the timeout. - $w = EV::periodic $at, $interval, $callback - $w = EV::periodic_ns $at, $interval, $callback - Similar to EV::timer, but the time is given as an absolute point in - time ($at), plus an optional $interval. - - If the $interval is zero, then the callback will be called at the - time $at if that is in the future, or as soon as possible if it is - in the past. It will not automatically repeat. - - If the $interval is nonzero, then the watcher will always be - scheduled to time out at the next "$at + N * $interval" time. - - This can be used to schedule a callback to run at very regular - intervals, as long as the processing time is less then the interval - (otherwise obviously events will be skipped). - - Another way to think about it (for the mathematically inclined) is - that EV::periodic will try to run the callback at the next possible - time where "$time = $at (mod $interval)", regardless of any time - jumps. - - This periodic timer is based on "wallclock time", that is, if the - clock changes ("ntp", "date -s" etc.), then the timer will - nevertheless run at the specified time. This means it will never - drift (it might jitter, but it will not drift). + PERIODIC WATCHERS - to cron or not to cron? + $w = EV::periodic $at, $interval, $reschedule_cb, $callback + $w = EV::periodic_ns $at, $interval, $reschedule_cb, $callback + Similar to EV::timer, but is not based on relative timeouts but on + absolute times. Apart from creating "simple" timers that trigger + "at" the specified time, it can also be used for non-drifting + absolute timers and more complex, cron-like, setups that are not + adversely affected by time jumps (i.e. when the system clock is + changed by explicit date -s or other means such as ntpd). It is also + the most complex watcher type in EV. + + It has three distinct "modes": + + * absolute timer ($interval = $reschedule_cb = 0) + This time simply fires at the wallclock time $at and doesn't + repeat. It will not adjust when a time jump occurs, that is, if + it is to be run at January 1st 2011 then it will run when the + system time reaches or surpasses this time. + + * non-repeating interval timer ($interval > 0, $reschedule_cb = 0) + In this mode the watcher will always be scheduled to time out at + the next "$at + N * $interval" time (for some integer N) and + then repeat, regardless of any time jumps. + + This can be used to create timers that do not drift with respect + to system time: + + my $hourly = EV::periodic 0, 3600, 0, sub { print "once/hour\n" }; + + That doesn't mean there will always be 3600 seconds in between + triggers, but only that the the clalback will be called when the + system time shows a full hour (UTC). + + Another way to think about it (for the mathematically inclined) + is that EV::periodic will try to run the callback in this mode + at the next possible time where "$time = $at (mod $interval)", + regardless of any time jumps. + + * manual reschedule mode ($reschedule_cb = coderef) + In this mode $interval and $at are both being ignored. Instead, + each time the periodic watcher gets scheduled, the reschedule + callback ($reschedule_cb) will be called with the watcher as + first, and the current time as second argument. + + *This callback MUST NOT stop or destroy this or any other + periodic watcher, ever*. If you need to stop it, return 1e30 and + stop it afterwards. + + It must return the next time to trigger, based on the passed + time value (that is, the lowest time value larger than to the + second argument). It will usually be called just before the + callback will be triggered, but might be called at other times, + too. + + This can be used to create very complex timers, such as a timer + that triggers on each midnight, local time (actually 24 hours + after the last midnight, to keep the example simple. If you know + a way to do it correctly in about the same space (without + requiring elaborate modules), drop me a note :): + + my $daily = EV::periodic 0, 0, sub { + my ($w, $now) = @_; + + use Time::Local (); + my (undef, undef, undef, $d, $m, $y) = localtime $now; + 86400 + Time::Local::timelocal 0, 0, 0, $d, $m, $y + }, sub { + print "it's midnight or likely shortly after, now\n"; + }; The "periodic_ns" variant doesn't start (activate) the newly created watcher. - $w->set ($at, $interval) + $w->set ($at, $interval, $reschedule_cb) Reconfigures the watcher, see the constructor above for details. Can - be at any time. + be called at any time. + + $w->again + Simply stops and starts the watcher again. + SIGNAL WATCHERS - signal me when a signal gets signalled! $w = EV::signal $signal, $callback $w = EV::signal_ns $signal, $callback Call the callback when $signal is received (the signal can be - specified by number or by name, just as with kill or %SIG). + specified by number or by name, just as with "kill" or %SIG). EV will grab the signal for the process (the kernel only allows one component to receive a signal at a time) when you start a signal @@ -283,31 +429,39 @@ $w->set ($signal) Reconfigures the watcher, see the constructor above for details. Can - be at any time. + be called at any time. $current_signum = $w->signal $old_signum = $w->signal ($new_signal) Returns the previously set signal (always as a number not name) and optionally set a new one. + CHILD WATCHERS - watch out for process status changes $w = EV::child $pid, $callback $w = EV::child_ns $pid, $callback Call the callback when a status change for pid $pid (or any pid if $pid is 0) has been received. More precisely: when the process - receives a SIGCHLD, EV will fetch the outstanding exit/wait status + receives a "SIGCHLD", EV will fetch the outstanding exit/wait status for all changed/zombie children and call the callback. - You can access both status and pid by using the "rstatus" and "rpid" - methods on the watcher object. + It is valid (and fully supported) to install a child watcher after a + child has exited but before the event loop has started its next + iteration (for example, first you "fork", then the new child process + might exit, and only then do you install a child watcher in the + parent for the new pid). + + You can access both exit (or tracing) status and pid by using the + "rstatus" and "rpid" methods on the watcher object. - You can have as many pid watchers per pid as you want. + You can have as many pid watchers per pid as you want, they will all + be called. The "child_ns" variant doesn't start (activate) the newly created watcher. $w->set ($pid) Reconfigures the watcher, see the constructor above for details. Can - be at any time. + be called at any time. $current_pid = $w->pid $old_pid = $w->pid ($new_pid) @@ -321,17 +475,101 @@ Return the pid of the awaited child (useful when you have installed a watcher for all pids). + STAT WATCHERS - did the file attributes just change? + $w = EV::stat $path, $interval, $callback + $w = EV::stat_ns $path, $interval, $callback + Call the callback when a file status change has been detected on + $path. The $path does not need to exist, changing from "path exists" + to "path does not exist" is a status change like any other. + + The $interval is a recommended polling interval for systems where + OS-supported change notifications don't exist or are not supported. + If you use 0 then an unspecified default is used (which is highly + recommended!), which is to be expected to be around five seconds + usually. + + This watcher type is not meant for massive numbers of stat watchers, + as even with OS-supported change notifications, this can be + resource-intensive. + + The "stat_ns" variant doesn't start (activate) the newly created + watcher. + + ... = $w->stat + This call is very similar to the perl "stat" built-in: It stats + (using "lstat") the path specified in the watcher and sets perls + stat cache (as well as EV's idea of the current stat values) to the + values found. + + In scalar context, a boolean is return indicating success or failure + of the stat. In list context, the same 13-value list as with stat is + returned (except that the blksize and blocks fields are not + reliable). + + In the case of an error, errno is set to "ENOENT" (regardless of the + actual error value) and the "nlink" value is forced to zero (if the + stat was successful then nlink is guaranteed to be non-zero). + + See also the next two entries for more info. + + ... = $w->attr + Just like "$w->stat", but without the initial stat'ing: this returns + the values most recently detected by EV. See the next entry for more + info. + + ... = $w->prev + Just like "$w->stat", but without the initial stat'ing: this returns + the previous set of values, before the change. + + That is, when the watcher callback is invoked, "$w->prev" will be + set to the values found *before* a change was detected, while + "$w->attr" returns the values found leading to the change detection. + The difference (if any) between "prev" and "attr" is what triggered + the callback. + + If you did something to the filesystem object and do not want to + trigger yet another change, you can call "stat" to update EV's idea + of what the current attributes are. + + $w->set ($path, $interval) + Reconfigures the watcher, see the constructor above for details. Can + be called at any time. + + $current_path = $w->path + $old_path = $w->path ($new_path) + Returns the previously set path and optionally set a new one. + + $current_interval = $w->interval + $old_interval = $w->interval ($new_interval) + Returns the previously set interval and optionally set a new one. + Can be used to query the actual interval used. + + IDLE WATCHERS - when you've got nothing better to do... $w = EV::idle $callback $w = EV::idle_ns $callback - Call the callback when there are no pending io, timer/periodic, - signal or child events, i.e. when the process is idle. + Call the callback when there are no other pending watchers of the + same or higher priority (excluding check, prepare and other idle + watchers of the same or lower priority, of course). They are called + idle watchers because when the watcher is the highest priority + pending event in the process, the process is considered to be idle + at that priority. + + If you want a watcher that is only ever called when *no* other + events are outstanding you have to set the priority to "EV::MINPRI". The process will not block as long as any idle watchers are active, and they will be called repeatedly until stopped. + For example, if you have idle watchers at priority 0 and 1, and an + I/O watcher at priority 0, then the idle watcher at priority 1 and + the I/O watcher will always run when ready. Only when the idle + watcher at priority 1 is stopped and the I/O watcher at priority 0 + is not pending with the 0-priority idle watcher be invoked. + The "idle_ns" variant doesn't start (activate) the newly created watcher. + PREPARE WATCHERS - customise your event loop! $w = EV::prepare $callback $w = EV::prepare_ns $callback Call the callback just before the process would block. You can still @@ -342,6 +580,7 @@ The "prepare_ns" variant doesn't start (activate) the newly created watcher. + CHECK WATCHERS - customise your event loop even more! $w = EV::check $callback $w = EV::check_ns $callback Call the callback just after the process wakes up again (after it @@ -362,28 +601,30 @@ or return; # make the dispatcher handle any outstanding stuff + ... not shown - # create an IO watcher for each and every socket + # create an I/O watcher for each and every socket @snmp_watcher = ( (map { EV::io $_, EV::READ, sub { } } keys %{ $dispatcher->{_descriptors} }), - ); - # if there are any timeouts, also create a timer - push @snmp_watcher, EV::timer $event->[Net::SNMP::Dispatcher::_TIME] - EV::now, 0, sub { } - if $event->[Net::SNMP::Dispatcher::_ACTIVE]; + EV::timer +($event->[Net::SNMP::Dispatcher::_ACTIVE] + ? $event->[Net::SNMP::Dispatcher::_TIME] - EV::now : 0), + 0, sub { }, + ); }; - The callbacks are irrelevant, the only purpose of those watchers is - to wake up the process as soon as one of those events occurs (socket - readable, or timer timed out). The corresponding EV::check watcher - will then clean up: + The callbacks are irrelevant (and are not even being called), the + only purpose of those watchers is to wake up the process as soon as + one of those events occurs (socket readable, or timer timed out). + The corresponding EV::check watcher will then clean up: our $snmp_check = EV::check sub { # destroy all watchers @snmp_watcher = (); # make the dispatcher handle any new stuff + ... not shown }; The callbacks of the created watchers will not be called as the @@ -393,12 +634,62 @@ The "check_ns" variant doesn't start (activate) the newly created watcher. + FORK WATCHERS - the audacity to resume the event loop after a fork + Fork watchers are called when a "fork ()" was detected. The invocation + is done before the event loop blocks next and before "check" watchers + are being called, and only in the child after the fork. + + $w = EV::fork $callback + $w = EV::fork_ns $callback + Call the callback before the event loop is resumed in the child + process after a fork. + + The "fork_ns" variant doesn't start (activate) the newly created + watcher. + +PERL SIGNALS + While Perl signal handling (%SIG) is not affected by EV, the behaviour + with EV is as the same as any other C library: Perl-signals will only be + handled when Perl runs, which means your signal handler might be invoked + only the next time an event callback is invoked. + + The solution is to use EV signal watchers (see "EV::signal"), which will + ensure proper operations with regards to other event watchers. + + If you cannot do this for whatever reason, you can also force a watcher + to be called on every event loop iteration by installing a "EV::check" + watcher: + + my $async_check = EV::check sub { }; + + This ensures that perl shortly gets into control for a short time, and + also ensures slower overall operation. + THREADS - Threads are not supported by this in any way. Perl pseudo-threads is - evil stuff and must die. + Threads are not supported by this module in any way. Perl pseudo-threads + is evil stuff and must die. As soon as Perl gains real threads I will + work on thread support for it. + +FORK + Most of the "improved" event delivering mechanisms of modern operating + systems have quite a few problems with fork(2) (to put it bluntly: it is + not supported and usually destructive). Libev makes it possible to work + around this by having a function that recreates the kernel state after + fork in the child. + + On non-win32 platforms, this module requires the pthread_atfork + functionality to do this automatically for you. This function is quite + buggy on most BSDs, though, so YMMV. The overhead for this is quite + negligible, because everything the function currently does is set a flag + that is checked only when the event loop gets used the next time, so + when you do fork but not use EV, the overhead is minimal. + + On win32, there is no notion of fork so all this doesn't apply, of + course. SEE ALSO - L, L. + EV::ADNS (asynchronous dns), Glib::EV (makes Glib/Gtk2 use EV as event + loop), Coro::EV (efficient coroutines with EV). AUTHOR Marc Lehmann