--- libev/ev.pod	2007/12/20 07:12:57	1.91
+++ libev/ev.pod	2008/03/13 13:06:16	1.136
@@ -6,56 +6,70 @@
 
   #include <ev.h>
 
-=head1 EXAMPLE PROGRAM
+=head2 EXAMPLE PROGRAM
 
+  // a single header file is required
   #include <ev.h>
 
+  // every watcher type has its own typedef'd struct
+  // with the name ev_<type>
   ev_io stdin_watcher;
   ev_timer timeout_watcher;
 
-  /* called when data readable on stdin */
+  // all watcher callbacks have a similar signature
+  // this callback is called when data is readable on stdin
   static void
   stdin_cb (EV_P_ struct ev_io *w, int revents)
   {
-    /* puts ("stdin ready"); */
-    ev_io_stop (EV_A_ w); /* just a syntax example */
-    ev_unloop (EV_A_ EVUNLOOP_ALL); /* leave all loop calls */
+    puts ("stdin ready");
+    // for one-shot events, one must manually stop the watcher
+    // with its corresponding stop function.
+    ev_io_stop (EV_A_ w);
+
+    // this causes all nested ev_loop's to stop iterating
+    ev_unloop (EV_A_ EVUNLOOP_ALL);
   }
 
+  // another callback, this time for a time-out
   static void
   timeout_cb (EV_P_ struct ev_timer *w, int revents)
   {
-    /* puts ("timeout"); */
-    ev_unloop (EV_A_ EVUNLOOP_ONE); /* leave one loop call */
+    puts ("timeout");
+    // this causes the innermost ev_loop to stop iterating
+    ev_unloop (EV_A_ EVUNLOOP_ONE);
   }
 
   int
   main (void)
   {
+    // use the default event loop unless you have special needs
     struct ev_loop *loop = ev_default_loop (0);
 
-    /* initialise an io watcher, then start it */
+    // initialise an io watcher, then start it
+    // this one will watch for stdin to become readable
     ev_io_init (&stdin_watcher, stdin_cb, /*STDIN_FILENO*/ 0, EV_READ);
     ev_io_start (loop, &stdin_watcher);
 
-    /* simple non-repeating 5.5 second timeout */
+    // initialise a timer watcher, then start it
+    // simple non-repeating 5.5 second timeout
     ev_timer_init (&timeout_watcher, timeout_cb, 5.5, 0.);
     ev_timer_start (loop, &timeout_watcher);
 
-    /* loop till timeout or data ready */
+    // now wait for events to arrive
     ev_loop (loop, 0);
 
+    // unloop was called, so exit
     return 0;
   }
 
 =head1 DESCRIPTION
 
-The newest version of this document is also available as a html-formatted
+The newest version of this document is also available as an html-formatted
 web page you might find easier to navigate when reading it for the first
 time: L<http://cvs.schmorp.de/libev/ev.html>.
 
 Libev is an event loop: you register interest in certain events (such as a
-file descriptor being readable or a timeout occuring), and it will manage
+file descriptor being readable or a timeout occurring), and it will manage
 these event sources and provide your program with events.
 
 To do this, it must take more or less complete control over your process
@@ -67,7 +81,7 @@
 details of the event, and then hand it over to libev by I<starting> the
 watcher.
 
-=head1 FEATURES
+=head2 FEATURES
 
 Libev supports C<select>, C<poll>, the Linux-specific C<epoll>, the
 BSD-specific C<kqueue> and the Solaris-specific event port mechanisms
@@ -84,16 +98,17 @@
 L<benchmark|http://libev.schmorp.de/bench.html> comparing it to libevent
 for example).
 
-=head1 CONVENTIONS
+=head2 CONVENTIONS
 
-Libev is very configurable. In this manual the default configuration will
-be described, which supports multiple event loops. For more info about
-various configuration options please have a look at B<EMBED> section in
-this manual. If libev was configured without support for multiple event
-loops, then all functions taking an initial argument of name C<loop>
-(which is always of type C<struct ev_loop *>) will not have this argument.
+Libev is very configurable. In this manual the default (and most common)
+configuration will be described, which supports multiple event loops. For
+more info about various configuration options please have a look at
+B<EMBED> section in this manual. If libev was configured without support
+for multiple event loops, then all functions taking an initial argument of
+name C<loop> (which is always of type C<struct ev_loop *>) will not have
+this argument.
 
-=head1 TIME REPRESENTATION
+=head2 TIME REPRESENTATION
 
 Libev represents time as a single floating point number, representing the
 (fractional) number of seconds since the (POSIX) epoch (somewhere near
@@ -117,6 +132,12 @@
 C<ev_now> function is usually faster and also often returns the timestamp
 you actually want to know.
 
+=item ev_sleep (ev_tstamp interval)
+
+Sleep for the given interval: The current thread will be blocked until
+either it is interrupted or the given time interval has passed. Basically
+this is a subsecond-resolution C<sleep ()>.
+
 =item int ev_version_major ()
 
 =item int ev_version_minor ()
@@ -256,6 +277,13 @@
 If you don't know what event loop to use, use the one returned from this
 function.
 
+The default loop is the only loop that can handle C<ev_signal> and
+C<ev_child> watchers, and to do this, it always registers a handler
+for C<SIGCHLD>. If this is a problem for your app you can either
+create a dynamic loop with C<ev_loop_new> that doesn't do that, or you
+can simply overwrite the C<SIGCHLD> signal handler I<after> calling
+C<ev_default_init>.
+
 The flags argument can be used to specify special behaviour or specific
 backends to use, and is usually specified as C<0> (or C<EVFLAG_AUTO>).
 
@@ -286,8 +314,8 @@
 This works by calling C<getpid ()> on every iteration of the loop,
 and thus this might slow down your event loop if you do a lot of loop
 iterations and little real work, but is usually not noticeable (on my
-Linux system for example, C<getpid> is actually a simple 5-insn sequence
-without a syscall and thus I<very> fast, but my Linux system also has
+GNU/Linux system for example, C<getpid> is actually a simple 5-insn sequence
+without a syscall and thus I<very> fast, but my GNU/Linux system also has
 C<pthread_atfork> which is even faster).
 
 The big advantage of this flag is that you can forget about fork (and
@@ -302,73 +330,120 @@
 This is your standard select(2) backend. Not I<completely> standard, as
 libev tries to roll its own fd_set with no limits on the number of fds,
 but if that fails, expect a fairly low limit on the number of fds when
-using this backend. It doesn't scale too well (O(highest_fd)), but its usually
-the fastest backend for a low number of fds.
+using this backend. It doesn't scale too well (O(highest_fd)), but its
+usually the fastest backend for a low number of (low-numbered :) fds.
+
+To get good performance out of this backend you need a high amount of
+parallelity (most of the file descriptors should be busy). If you are
+writing a server, you should C<accept ()> in a loop to accept as many
+connections as possible during one iteration. You might also want to have
+a look at C<ev_set_io_collect_interval ()> to increase the amount of
+readyness notifications you get per iteration.
 
 =item C<EVBACKEND_POLL>    (value 2, poll backend, available everywhere except on windows)
 
-And this is your standard poll(2) backend. It's more complicated than
-select, but handles sparse fds better and has no artificial limit on the
-number of fds you can use (except it will slow down considerably with a
-lot of inactive fds). It scales similarly to select, i.e. O(total_fds).
+And this is your standard poll(2) backend. It's more complicated
+than select, but handles sparse fds better and has no artificial
+limit on the number of fds you can use (except it will slow down
+considerably with a lot of inactive fds). It scales similarly to select,
+i.e. O(total_fds). See the entry for C<EVBACKEND_SELECT>, above, for
+performance tips.
 
 =item C<EVBACKEND_EPOLL>   (value 4, Linux)
 
 For few fds, this backend is a bit little slower than poll and select,
-but it scales phenomenally better. While poll and select usually scale like
-O(total_fds) where n is the total number of fds (or the highest fd), epoll scales
-either O(1) or O(active_fds).
+but it scales phenomenally better. While poll and select usually scale
+like O(total_fds) where n is the total number of fds (or the highest fd),
+epoll scales either O(1) or O(active_fds). The epoll design has a number
+of shortcomings, such as silently dropping events in some hard-to-detect
+cases and rewiring a syscall per fd change, no fork support and bad
+support for dup.
 
-While stopping and starting an I/O watcher in the same iteration will
-result in some caching, there is still a syscall per such incident
+While stopping, setting and starting an I/O watcher in the same iteration
+will result in some caching, there is still a syscall per such incident
 (because the fd could point to a different file description now), so its
-best to avoid that. Also, dup()ed file descriptors might not work very
-well if you register events for both fds.
+best to avoid that. Also, C<dup ()>'ed file descriptors might not work
+very well if you register events for both fds.
 
 Please note that epoll sometimes generates spurious notifications, so you
 need to use non-blocking I/O or other means to avoid blocking when no data
 (or space) is available.
 
+Best performance from this backend is achieved by not unregistering all
+watchers for a file descriptor until it has been closed, if possible, i.e.
+keep at least one watcher active per fd at all times.
+
+While nominally embeddeble in other event loops, this feature is broken in
+all kernel versions tested so far.
+
 =item C<EVBACKEND_KQUEUE>  (value 8, most BSD clones)
 
 Kqueue deserves special mention, as at the time of this writing, it
-was broken on all BSDs except NetBSD (usually it doesn't work with
-anything but sockets and pipes, except on Darwin, where of course it's
-completely useless). For this reason it's not being "autodetected"
+was broken on all BSDs except NetBSD (usually it doesn't work reliably
+with anything but sockets and pipes, except on Darwin, where of course
+it's completely useless). For this reason it's not being "autodetected"
 unless you explicitly specify it explicitly in the flags (i.e. using
-C<EVBACKEND_KQUEUE>).
+C<EVBACKEND_KQUEUE>) or libev was compiled on a known-to-be-good (-enough)
+system like NetBSD.
+
+You still can embed kqueue into a normal poll or select backend and use it
+only for sockets (after having made sure that sockets work with kqueue on
+the target platform). See C<ev_embed> watchers for more info.
 
 It scales in the same way as the epoll backend, but the interface to the
 kernel is more efficient (which says nothing about its actual speed, of
-course). While starting and stopping an I/O watcher does not cause an
-extra syscall as with epoll, it still adds up to four event changes per
-incident, so its best to avoid that.
+course). While stopping, setting and starting an I/O watcher does never
+cause an extra syscall as with C<EVBACKEND_EPOLL>, it still adds up to
+two event changes per incident, support for C<fork ()> is very bad and it
+drops fds silently in similarly hard-to-detect cases.
+
+This backend usually performs well under most conditions.
+
+While nominally embeddable in other event loops, this doesn't work
+everywhere, so you might need to test for this. And since it is broken
+almost everywhere, you should only use it when you have a lot of sockets
+(for which it usually works), by embedding it into another event loop
+(e.g. C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>) and using it only for
+sockets.
 
 =item C<EVBACKEND_DEVPOLL> (value 16, Solaris 8)
 
-This is not implemented yet (and might never be).
+This is not implemented yet (and might never be, unless you send me an
+implementation). According to reports, C</dev/poll> only supports sockets
+and is not embeddable, which would limit the usefulness of this backend
+immensely.
 
 =item C<EVBACKEND_PORT>    (value 32, Solaris 10)
 
-This uses the Solaris 10 port mechanism. As with everything on Solaris,
+This uses the Solaris 10 event port mechanism. As with everything on Solaris,
 it's really slow, but it still scales very well (O(active_fds)).
 
-Please note that solaris ports can result in a lot of spurious
+Please note that solaris event ports can deliver a lot of spurious
 notifications, so you need to use non-blocking I/O or other means to avoid
 blocking when no data (or space) is available.
 
+While this backend scales well, it requires one system call per active
+file descriptor per loop iteration. For small and medium numbers of file
+descriptors a "slow" C<EVBACKEND_SELECT> or C<EVBACKEND_POLL> backend
+might perform better.
+
+On the positive side, ignoring the spurious readyness notifications, this
+backend actually performed to specification in all tests and is fully
+embeddable, which is a rare feat among the OS-specific backends.
+
 =item C<EVBACKEND_ALL>
 
 Try all backends (even potentially broken ones that wouldn't be tried
 with C<EVFLAG_AUTO>). Since this is a mask, you can do stuff such as
 C<EVBACKEND_ALL & ~EVBACKEND_KQUEUE>.
 
+It is definitely not recommended to use this flag.
+
 =back
 
 If one or more of these are ored into the flags value, then only these
-backends will be tried (in the reverse order as given here). If none are
-specified, most compiled-in backend will be tried, usually in reverse
-order of their flag values :)
+backends will be tried (in the reverse order as listed here). If none are
+specified, all backends in C<ev_recommended_backends ()> will be tried.
 
 The most typical usage is like this:
 
@@ -425,14 +500,16 @@
 
 =item ev_default_fork ()
 
-This function reinitialises the kernel state for backends that have
-one. Despite the name, you can call it anytime, but it makes most sense
-after forking, in either the parent or child process (or both, but that
-again makes little sense).
-
-You I<must> call this function in the child process after forking if and
-only if you want to use the event library in both processes. If you just
-fork+exec, you don't have to call it.
+This function sets a flag that causes subsequent C<ev_loop> iterations
+to reinitialise the kernel state for backends that have one. Despite the
+name, you can call it anytime, but it makes most sense after forking, in
+the child process (or both child and parent, but that again makes little
+sense). You I<must> call it in the child before using any of the libev
+functions, and it will only take effect at the next C<ev_loop> iteration.
+
+On the other hand, you only need to call this function in the child
+process if and only if you want to use the event library in the child. If
+you just fork+exec, you don't have to call it at all.
 
 The function itself is quite fast and it's usually not a problem to call
 it just in case after a fork. To make this easy, the function will fit in
@@ -440,16 +517,16 @@
 
     pthread_atfork (0, 0, ev_default_fork);
 
-At the moment, C<EVBACKEND_SELECT> and C<EVBACKEND_POLL> are safe to use
-without calling this function, so if you force one of those backends you
-do not need to care.
-
 =item ev_loop_fork (loop)
 
 Like C<ev_default_fork>, but acts on an event loop created by
 C<ev_loop_new>. Yes, you have to call this on every allocated event loop
 after fork, and how you do this is entirely your own problem.
 
+=item int ev_is_default_loop (loop)
+
+Returns true when the given loop actually is the default loop, false otherwise.
+
 =item unsigned int ev_loop_count (loop)
 
 Returns the count of loop iterations for the loop, which is identical to
@@ -471,7 +548,7 @@
 received events and started processing them. This timestamp does not
 change as long as callbacks are being processed, and this is also the base
 time used for relative timers. You can treat it as the timestamp of the
-event occuring (or more correctly, libev finding out about it).
+event occurring (or more correctly, libev finding out about it).
 
 =item ev_loop (loop, int flags)
 
@@ -503,12 +580,16 @@
 Here are the gory details of what C<ev_loop> does:
 
    - Before the first iteration, call any pending watchers.
-   * If there are no active watchers (reference count is zero), return.
-   - Queue all prepare watchers and then call all outstanding watchers.
+   * If EVFLAG_FORKCHECK was used, check for a fork.
+   - If a fork was detected, queue and call all fork watchers.
+   - Queue and call all prepare watchers.
    - If we have been forked, recreate the kernel state.
    - Update the kernel state with all outstanding changes.
    - Update the "event loop time".
-   - Calculate for how long to block.
+   - Calculate for how long to sleep or block, if at all
+     (active idle watchers, EVLOOP_NONBLOCK or not having
+     any active watchers at all will result in not sleeping).
+   - Sleep if the I/O and timer collect interval say so.
    - Block the process, waiting for any events.
    - Queue all outstanding I/O (fd) events.
    - Update the "event loop time" and do time jump handling.
@@ -519,10 +600,11 @@
    - Call all queued watchers in reverse order (i.e. check watchers first).
      Signals and child watchers are implemented as I/O watchers, and will
      be handled here by queueing them when their watcher gets executed.
-   - If ev_unloop has been called or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
-     were used, return, otherwise continue with step *.
+   - If ev_unloop has been called, or EVLOOP_ONESHOT or EVLOOP_NONBLOCK
+     were used, or there are no active watchers, return, otherwise
+     continue with step *.
 
-Example: Queue some jobs and then loop until no events are outsanding
+Example: Queue some jobs and then loop until no events are outstanding
 anymore.
 
    ... queue jobs here, make sure they register event watchers as long
@@ -537,6 +619,8 @@
 C<EVUNLOOP_ONE>, which will make the innermost C<ev_loop> call return, or
 C<EVUNLOOP_ALL>, which will make all nested C<ev_loop> calls return.
 
+This "unloop state" will be cleared when entering C<ev_loop> again.
+
 =item ev_ref (loop)
 
 =item ev_unref (loop)
@@ -550,7 +634,9 @@
 visible to the libev user and should not keep C<ev_loop> from exiting if
 no event watchers registered by it are active. It is also an excellent
 way to do this for generic recurring timers or from within third-party
-libraries. Just remember to I<unref after start> and I<ref before stop>.
+libraries. Just remember to I<unref after start> and I<ref before stop>
+(but only if the watcher wasn't active before, or was active before,
+respectively).
 
 Example: Create a signal watcher, but keep it from keeping C<ev_loop>
 running when nothing else is active.
@@ -565,6 +651,42 @@
   ev_ref (loop);
   ev_signal_stop (loop, &exitsig);
 
+=item ev_set_io_collect_interval (loop, ev_tstamp interval)
+
+=item ev_set_timeout_collect_interval (loop, ev_tstamp interval)
+
+These advanced functions influence the time that libev will spend waiting
+for events. Both are by default C<0>, meaning that libev will try to
+invoke timer/periodic callbacks and I/O callbacks with minimum latency.
+
+Setting these to a higher value (the C<interval> I<must> be >= C<0>)
+allows libev to delay invocation of I/O and timer/periodic callbacks to
+increase efficiency of loop iterations.
+
+The background is that sometimes your program runs just fast enough to
+handle one (or very few) event(s) per loop iteration. While this makes
+the program responsive, it also wastes a lot of CPU time to poll for new
+events, especially with backends like C<select ()> which have a high
+overhead for the actual polling but can deliver many events at once.
+
+By setting a higher I<io collect interval> you allow libev to spend more
+time collecting I/O events, so you can handle more events per iteration,
+at the cost of increasing latency. Timeouts (both C<ev_periodic> and
+C<ev_timer>) will be not affected. Setting this to a non-null value will
+introduce an additional C<ev_sleep ()> call into most loop iterations.
+
+Likewise, by setting a higher I<timeout collect interval> you allow libev
+to spend more time collecting timeouts, at the expense of increased
+latency (the watcher callback will be called later). C<ev_io> watchers
+will not be affected. Setting this to a non-null value will not introduce
+any overhead in libev.
+
+Many (busy) programs can usually benefit by setting the io collect
+interval to a value near C<0.1> or so, which is often enough for
+interactive servers (of course not for games), likewise for timeouts. It
+usually doesn't make much sense to set it to a lower value than C<0.01>,
+as this approsaches the timing granularity of most systems.
+
 =back
 
 
@@ -673,6 +795,10 @@
 The event loop has been resumed in the child process after fork (see
 C<ev_fork>).
 
+=item C<EV_ASYNC>
+
+The given async watcher has been asynchronously notified (see C<ev_async>).
+
 =item C<EV_ERROR>
 
 An unspecified error has occured, the watcher has been stopped. This might
@@ -899,12 +1025,6 @@
 descriptors to non-blocking mode is also usually a good idea (but not
 required if you know what you are doing).
 
-You have to be careful with dup'ed file descriptors, though. Some backends
-(the linux epoll backend is a notable example) cannot handle dup'ed file
-descriptors correctly if you register interest in two or more fds pointing
-to the same underlying file/socket/etc. description (that is, they share
-the same underlying "file open").
-
 If you must do this, then force the use of a known-to-be-good backend
 (at the time of this writing, this includes only C<EVBACKEND_SELECT> and
 C<EVBACKEND_POLL>).
@@ -926,7 +1046,7 @@
 
 =head3 The special problem of disappearing file descriptors
 
-Some backends (e.g kqueue, epoll) need to be told about closing a file
+Some backends (e.g. kqueue, epoll) need to be told about closing a file
 descriptor (either by calling C<close> explicitly or by any other means,
 such as C<dup>). The reason is that you register interest in some file
 descriptor, but when it goes away, the operating system will silently drop
@@ -945,6 +1065,28 @@
 the libev application should not optimise around libev but should leave
 optimisations to libev.
 
+=head3 The special problem of dup'ed file descriptors
+
+Some backends (e.g. epoll), cannot register events for file descriptors,
+but only events for the underlying file descriptions. That means when you
+have C<dup ()>'ed file descriptors or weirder constellations, and register
+events for them, only one file descriptor might actually receive events.
+
+There is no workaround possible except not registering events
+for potentially C<dup ()>'ed file descriptors, or to resort to
+C<EVBACKEND_SELECT> or C<EVBACKEND_POLL>.
+
+=head3 The special problem of fork
+
+Some backends (epoll, kqueue) do not support C<fork ()> at all or exhibit
+useless behaviour. Libev fully supports fork, but needs to be told about
+it in the child.
+
+To support fork in your programs, you either have to call
+C<ev_default_fork ()> or C<ev_loop_fork ()> after a fork in the child,
+enable C<EVFLAG_FORKCHECK>, or resort to C<EVBACKEND_SELECT> or
+C<EVBACKEND_POLL>.
+
 
 =head3 Watcher-Specific Functions
 
@@ -968,6 +1110,8 @@
 
 =back
 
+=head3 Examples
+
 Example: Call C<stdin_readable_cb> when STDIN_FILENO has become, well
 readable, but only once. Since it is likely line-buffered, you could
 attempt to read a whole line in the callback.
@@ -1029,7 +1173,7 @@
 the timer (because it takes longer than those 10 seconds to do stuff) the
 timer will not fire more than once per event loop iteration.
 
-=item ev_timer_again (loop)
+=item ev_timer_again (loop, ev_timer *)
 
 This will act as if the timer timed out and restart it again if it is
 repeating. The exact semantics are:
@@ -1074,6 +1218,8 @@
 
 =back
 
+=head3 Examples
+
 Example: Create a timer that fires after 60 seconds.
 
   static void
@@ -1146,7 +1292,7 @@
 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.
 
-=item * non-repeating interval timer (at = offset, interval > 0, reschedule_cb = 0)
+=item * repeating interval timer (at = offset, interval > 0, reschedule_cb = 0)
 
 In this mode the watcher will always be scheduled to time out at the next
 C<at + N * interval> time (for some integer N, which can also be negative)
@@ -1240,6 +1386,8 @@
 
 =back
 
+=head3 Examples
+
 Example: Call a callback every hour, or, more precisely, whenever the
 system clock is divisible by 3600. The callback invocation times have
 potentially a lot of jittering, but good long-term stability.
@@ -1288,6 +1436,12 @@
 watcher for a signal is stopped libev will reset the signal handler to
 SIG_DFL (regardless of what it was set to before).
 
+If possible and supported, libev will install its handlers with
+C<SA_RESTART> behaviour enabled, so syscalls should not be unduly
+interrupted. If you have a problem with syscalls getting interrupted by
+signals you can block all signals in an C<ev_check> watcher and unblock
+them in an C<ev_prepare> watcher.
+
 =head3 Watcher-Specific Functions and Data Members
 
 =over 4
@@ -1305,26 +1459,67 @@
 
 =back
 
+=head3 Examples
+
+Example: Try to exit cleanly on SIGINT and SIGTERM.
+
+  static void
+  sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents)
+  {
+    ev_unloop (loop, EVUNLOOP_ALL);
+  }
+
+  struct ev_signal signal_watcher;
+  ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
+  ev_signal_start (loop, &sigint_cb);
+
 
 =head2 C<ev_child> - watch out for process status changes
 
 Child watchers trigger when your process receives a SIGCHLD in response to
-some child status changes (most typically when a child of yours dies).
+some child status changes (most typically when a child of yours dies). It
+is permissible to install a child watcher I<after> the child has been
+forked (which implies it might have already exited), as long as the event
+loop isn't entered (or is continued from a watcher).
+
+Only the default event loop is capable of handling signals, and therefore
+you can only rgeister child watchers in the default event loop.
+
+=head3 Process Interaction
+
+Libev grabs C<SIGCHLD> as soon as the default event loop is
+initialised. This is necessary to guarantee proper behaviour even if
+the first child watcher is started after the child exits. The occurance
+of C<SIGCHLD> is recorded asynchronously, but child reaping is done
+synchronously as part of the event loop processing. Libev always reaps all
+children, even ones not watched.
+
+=head3 Overriding the Built-In Processing
+
+Libev offers no special support for overriding the built-in child
+processing, but if your application collides with libev's default child
+handler, you can override it easily by installing your own handler for
+C<SIGCHLD> after initialising the default loop, and making sure the
+default loop never gets destroyed. You are encouraged, however, to use an
+event-based approach to child reaping and thus use libev's support for
+that, so other libev users can use C<ev_child> watchers freely.
 
 =head3 Watcher-Specific Functions and Data Members
 
 =over 4
 
-=item ev_child_init (ev_child *, callback, int pid)
+=item ev_child_init (ev_child *, callback, int pid, int trace)
 
-=item ev_child_set (ev_child *, int pid)
+=item ev_child_set (ev_child *, int pid, int trace)
 
 Configures the watcher to wait for status changes of process C<pid> (or
 I<any> process if C<pid> is specified as C<0>). The callback can look
 at the C<rstatus> member of the C<ev_child> watcher structure to see
 the status word (use the macros from C<sys/wait.h> and see your systems
 C<waitpid> documentation). The C<rpid> member contains the pid of the
-process causing the status change.
+process causing the status change. C<trace> must be either C<0> (only
+activate the watcher when the process terminates) or C<1> (additionally
+activate the watcher when the process is stopped or continued).
 
 =item int pid [read-only]
 
@@ -1341,17 +1536,34 @@
 
 =back
 
-Example: Try to exit cleanly on SIGINT and SIGTERM.
+=head3 Examples
+
+Example: C<fork()> a new process and install a child handler to wait for
+its completion.
+
+  ev_child cw;
 
   static void
-  sigint_cb (struct ev_loop *loop, struct ev_signal *w, int revents)
+  child_cb (EV_P_ struct ev_child *w, int revents)
   {
-    ev_unloop (loop, EVUNLOOP_ALL);
+    ev_child_stop (EV_A_ w);
+    printf ("process %d exited with status %x\n", w->rpid, w->rstatus);
   }
 
-  struct ev_signal signal_watcher;
-  ev_signal_init (&signal_watcher, sigint_cb, SIGINT);
-  ev_signal_start (loop, &sigint_cb);
+  pid_t pid = fork ();
+
+  if (pid < 0)
+    // error
+  else if (pid == 0)
+    {
+      // the forked child executes here
+      exit (1);
+    }
+  else
+    {
+      ev_child_init (&cw, child_cb, pid, 0);
+      ev_child_start (EV_DEFAULT_ &cw);
+    }
 
 
 =head2 C<ev_stat> - did the file attributes just change?
@@ -1390,6 +1602,39 @@
 usually detected immediately, and if the file exists there will be no
 polling.
 
+=head3 Inotify
+
+When C<inotify (7)> support has been compiled into libev (generally only
+available on Linux) and present at runtime, it will be used to speed up
+change detection where possible. The inotify descriptor will be created lazily
+when the first C<ev_stat> watcher is being started.
+
+Inotify presense does not change the semantics of C<ev_stat> watchers
+except that changes might be detected earlier, and in some cases, to avoid
+making regular C<stat> calls. Even in the presense of inotify support
+there are many cases where libev has to resort to regular C<stat> polling.
+
+(There is no support for kqueue, as apparently it cannot be used to
+implement this functionality, due to the requirement of having a file
+descriptor open on the object at all times).
+
+=head3 The special problem of stat time resolution
+
+The C<stat ()> syscall only supports full-second resolution portably, and
+even on systems where the resolution is higher, many filesystems still
+only support whole seconds.
+
+That means that, if the time is the only thing that changes, you might
+miss updates: on the first update, C<ev_stat> detects a change and calls
+your callback, which does something. When there is another update within
+the same second, C<ev_stat> will be unable to detect it.
+
+The solution to this is to delay acting on a change for a second (or till
+the next second boundary), using a roughly one-second delay C<ev_timer>
+(C<ev_timer_set (w, 0., 1.01); ev_timer_again (loop, w)>). The C<.01>
+is added to work around small timing inconsistencies of some operating
+systems.
+
 =head3 Watcher-Specific Functions and Data Members
 
 =over 4
@@ -1408,7 +1653,7 @@
 relative to the attributes at the time the watcher was started (or the
 last change was detected).
 
-=item ev_stat_stat (ev_stat *)
+=item ev_stat_stat (loop, ev_stat *)
 
 Updates the stat buffer immediately with new values. If you change the
 watched path in your callback, you could call this fucntion to avoid
@@ -1437,6 +1682,8 @@
 
 =back
 
+=head3 Examples
+
 Example: Watch C</etc/passwd> for attribute changes.
 
   static void
@@ -1458,8 +1705,36 @@
   ...
   ev_stat passwd;
 
-  ev_stat_init (&passwd, passwd_cb, "/etc/passwd");
+  ev_stat_init (&passwd, passwd_cb, "/etc/passwd", 0.);
+  ev_stat_start (loop, &passwd);
+
+Example: Like above, but additionally use a one-second delay so we do not
+miss updates (however, frequent updates will delay processing, too, so
+one might do the work both on C<ev_stat> callback invocation I<and> on
+C<ev_timer> callback invocation).
+
+  static ev_stat passwd;
+  static ev_timer timer;
+
+  static void
+  timer_cb (EV_P_ ev_timer *w, int revents)
+  {
+    ev_timer_stop (EV_A_ w);
+
+    /* now it's one second after the most recent passwd change */
+  }
+
+  static void
+  stat_cb (EV_P_ ev_stat *w, int revents)
+  {
+    /* reset the one-second timer */
+    ev_timer_again (EV_A_ &timer);
+  }
+
+  ...
+  ev_stat_init (&passwd, stat_cb, "/etc/passwd", 0.);
   ev_stat_start (loop, &passwd);
+  ev_timer_init (&timer, timer_cb, 0., 1.01);
 
 
 =head2 C<ev_idle> - when you've got nothing better to do...
@@ -1495,6 +1770,8 @@
 
 =back
 
+=head3 Examples
+
 Example: Dynamically allocate an C<ev_idle> watcher, start it, and in the
 callback, free it. Also, use no error checking, as usual.
 
@@ -1503,7 +1780,7 @@
   {
     free (w);
     // now do something you wanted to do when the program has
-    // no longer asnything immediate to do.
+    // no longer anything immediate to do.
   }
 
   struct ev_idle *idle_watcher = malloc (sizeof (struct ev_idle));
@@ -1555,11 +1832,11 @@
 priority, to ensure that they are being run before any other watchers
 after the poll. Also, C<ev_check> watchers (and C<ev_prepare> watchers,
 too) should not activate ("feed") events into libev. While libev fully
-supports this, they will be called before other C<ev_check> watchers did
-their job. As C<ev_check> watchers are often used to embed other event
-loops those other event loops might be in an unusable state until their
-C<ev_check> watcher ran (always remind yourself to coexist peacefully with
-others).
+supports this, they will be called before other C<ev_check> watchers
+did their job. As C<ev_check> watchers are often used to embed other
+(non-libev) event loops those other event loops might be in an unusable
+state until their C<ev_check> watcher ran (always remind yourself to
+coexist peacefully with others).
 
 =head3 Watcher-Specific Functions and Data Members
 
@@ -1575,6 +1852,8 @@
 
 =back
 
+=head3 Examples
+
 There are a number of principal ways to embed other event loops or modules
 into libev. Here are some ideas on how to include libadns into libev
 (there is a Perl module named C<EV::ADNS> that does this, which you could
@@ -1752,26 +2031,7 @@
 So when you want to use this feature you will always have to be prepared
 that you cannot get an embeddable loop. The recommended way to get around
 this is to have a separate variables for your embeddable loop, try to
-create it, and if that fails, use the normal loop for everything:
-
-  struct ev_loop *loop_hi = ev_default_init (0);
-  struct ev_loop *loop_lo = 0;
-  struct ev_embed embed;
-  
-  // see if there is a chance of getting one that works
-  // (remember that a flags value of 0 means autodetection)
-  loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
-    ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
-    : 0;
-
-  // if we got one, then embed it, otherwise default to loop_hi
-  if (loop_lo)
-    {
-      ev_embed_init (&embed, 0, loop_lo);
-      ev_embed_start (loop_hi, &embed);
-    }
-  else
-    loop_lo = loop_hi;
+create it, and if that fails, use the normal loop for everything.
 
 =head3 Watcher-Specific Functions and Data Members
 
@@ -1799,6 +2059,54 @@
 
 =back
 
+=head3 Examples
+
+Example: Try to get an embeddable event loop and embed it into the default
+event loop. If that is not possible, use the default loop. The default
+loop is stored in C<loop_hi>, while the mebeddable loop is stored in
+C<loop_lo> (which is C<loop_hi> in the acse no embeddable loop can be
+used).
+
+  struct ev_loop *loop_hi = ev_default_init (0);
+  struct ev_loop *loop_lo = 0;
+  struct ev_embed embed;
+  
+  // see if there is a chance of getting one that works
+  // (remember that a flags value of 0 means autodetection)
+  loop_lo = ev_embeddable_backends () & ev_recommended_backends ()
+    ? ev_loop_new (ev_embeddable_backends () & ev_recommended_backends ())
+    : 0;
+
+  // if we got one, then embed it, otherwise default to loop_hi
+  if (loop_lo)
+    {
+      ev_embed_init (&embed, 0, loop_lo);
+      ev_embed_start (loop_hi, &embed);
+    }
+  else
+    loop_lo = loop_hi;
+
+Example: Check if kqueue is available but not recommended and create
+a kqueue backend for use with sockets (which usually work with any
+kqueue implementation). Store the kqueue/socket-only event loop in
+C<loop_socket>. (One might optionally use C<EVFLAG_NOENV>, too).
+
+  struct ev_loop *loop = ev_default_init (0);
+  struct ev_loop *loop_socket = 0;
+  struct ev_embed embed;
+  
+  if (ev_supported_backends () & ~ev_recommended_backends () & EVBACKEND_KQUEUE)
+    if ((loop_socket = ev_loop_new (EVBACKEND_KQUEUE))
+      {
+        ev_embed_init (&embed, 0, loop_socket);
+        ev_embed_start (loop, &embed);
+      }
+
+  if (!loop_socket)
+    loop_socket = loop;
+
+  // now use loop_socket for all sockets, and loop for everything else
+
 
 =head2 C<ev_fork> - the audacity to resume the event loop after a fork
 
@@ -1823,6 +2131,137 @@
 =back
 
 
+=head2 C<ev_async> - how to wake up another event loop
+
+In general, you cannot use an C<ev_loop> from multiple threads or other
+asynchronous sources such as signal handlers (as opposed to multiple event
+loops - those are of course safe to use in different threads).
+
+Sometimes, however, you need to wake up another event loop you do not
+control, for example because it belongs to another thread. This is what
+C<ev_async> watchers do: as long as the C<ev_async> watcher is active, you
+can signal it by calling C<ev_async_send>, which is thread- and signal
+safe.
+
+This functionality is very similar to C<ev_signal> watchers, as signals,
+too, are asynchronous in nature, and signals, too, will be compressed
+(i.e. the number of callback invocations may be less than the number of
+C<ev_async_sent> calls).
+
+Unlike C<ev_signal> watchers, C<ev_async> works with any event loop, not
+just the default loop.
+
+=head3 Queueing
+
+C<ev_async> does not support queueing of data in any way. The reason
+is that the author does not know of a simple (or any) algorithm for a
+multiple-writer-single-reader queue that works in all cases and doesn't
+need elaborate support such as pthreads.
+
+That means that if you want to queue data, you have to provide your own
+queue. But at least I can tell you would implement locking around your
+queue:
+
+=over 4
+
+=item queueing from a signal handler context
+
+To implement race-free queueing, you simply add to the queue in the signal
+handler but you block the signal handler in the watcher callback. Here is an example that does that for
+some fictitiuous SIGUSR1 handler:
+
+   static ev_async mysig;
+
+   static void
+   sigusr1_handler (void)
+   {
+     sometype data;
+
+     // no locking etc.
+     queue_put (data);
+     ev_async_send (EV_DEFAULT_ &mysig);
+   }
+
+   static void
+   mysig_cb (EV_P_ ev_async *w, int revents)
+   {
+     sometype data;
+     sigset_t block, prev;
+
+     sigemptyset (&block);
+     sigaddset (&block, SIGUSR1);
+     sigprocmask (SIG_BLOCK, &block, &prev);
+
+     while (queue_get (&data))
+       process (data);
+
+     if (sigismember (&prev, SIGUSR1)
+       sigprocmask (SIG_UNBLOCK, &block, 0);
+   }
+
+(Note: pthreads in theory requires you to use C<pthread_setmask>
+instead of C<sigprocmask> when you use threads, but libev doesn't do it
+either...).
+
+=item queueing from a thread context
+
+The strategy for threads is different, as you cannot (easily) block
+threads but you can easily preempt them, so to queue safely you need to
+employ a traditional mutex lock, such as in this pthread example:
+
+   static ev_async mysig;
+   static pthread_mutex_t mymutex = PTHREAD_MUTEX_INITIALIZER;
+
+   static void
+   otherthread (void)
+   {
+     // only need to lock the actual queueing operation
+     pthread_mutex_lock (&mymutex);
+     queue_put (data);
+     pthread_mutex_unlock (&mymutex);
+
+     ev_async_send (EV_DEFAULT_ &mysig);
+   }
+
+   static void
+   mysig_cb (EV_P_ ev_async *w, int revents)
+   {
+     pthread_mutex_lock (&mymutex);
+
+     while (queue_get (&data))
+       process (data);
+
+     pthread_mutex_unlock (&mymutex);
+   }
+
+=back
+
+
+=head3 Watcher-Specific Functions and Data Members
+
+=over 4
+
+=item ev_async_init (ev_async *, callback)
+
+Initialises and configures the async watcher - it has no parameters of any
+kind. There is a C<ev_asynd_set> macro, but using it is utterly pointless,
+believe me.
+
+=item ev_async_send (loop, ev_async *)
+
+Sends/signals/activates the given C<ev_async> watcher, that is, feeds
+an C<EV_ASYNC> event on the watcher into the event loop. Unlike
+C<ev_feed_event>, this call is safe to do in other threads, signal or
+similar contexts (see the dicusssion of C<EV_ATOMIC_T> in the embedding
+section below on what exactly this means).
+
+This call incurs the overhead of a syscall only once per loop iteration,
+so while the overhead might be noticable, it doesn't apply to repeated
+calls to C<ev_async_send>.
+
+=back
+
+
 =head1 OTHER FUNCTIONS
 
 There are some other functions of possible interest. Described. Here. Now.
@@ -2059,19 +2498,52 @@
 
   class myclass
   {
-    ev_io   io;   void io_cb   (ev::io   &w, int revents);
-    ev_idle idle  void idle_cb (ev::idle &w, int revents);
+    ev::io   io;  void io_cb   (ev::io   &w, int revents);
+    ev:idle idle  void idle_cb (ev::idle &w, int revents);
 
-    myclass ();
-  }
+    myclass (int fd)
+    {
+      io  .set <myclass, &myclass::io_cb  > (this);
+      idle.set <myclass, &myclass::idle_cb> (this);
 
-  myclass::myclass (int fd)
-  {
-    io  .set <myclass, &myclass::io_cb  > (this);
-    idle.set <myclass, &myclass::idle_cb> (this);
+      io.start (fd, ev::READ);
+    }
+  };
 
-    io.start (fd, ev::READ);
-  }
+
+=head1 OTHER LANGUAGE BINDINGS
+
+Libev does not offer other language bindings itself, but bindings for a
+numbe rof languages exist in the form of third-party packages. If you know
+any interesting language binding in addition to the ones listed here, drop
+me a note.
+
+=over 4
+
+=item Perl
+
+The EV module implements the full libev API and is actually used to test
+libev. EV is developed together with libev. Apart from the EV core module,
+there are additional modules that implement libev-compatible interfaces
+to C<libadns> (C<EV::ADNS>), C<Net::SNMP> (C<Net::SNMP::EV>) and the
+C<libglib> event core (C<Glib::EV> and C<EV::Glib>).
+
+It can be found and installed via CPAN, its homepage is found at
+L<http://software.schmorp.de/pkg/EV>.
+
+=item Ruby
+
+Tony Arcieri has written a ruby extension that offers access to a subset
+of the libev API and adds filehandle abstractions, asynchronous DNS and
+more on top of it. It can be found via gem servers. Its homepage is at
+L<http://rev.rubyforge.org/>.
+
+=item D
+
+Leandro Lucarella has written a D language binding (F<ev.d>) for libev, to
+be found at L<http://git.llucax.com.ar/?p=software/ev.d.git;a=summary>.
+
+=back
 
 
 =head1 MACRO MAGIC
@@ -2242,7 +2714,7 @@
 monotonic clock option at both compiletime and runtime. Otherwise no use
 of the monotonic clock option will be attempted. If you enable this, you
 usually have to link against librt or something similar. Enabling it when
-the functionality isn't available is safe, though, althoguh you have
+the functionality isn't available is safe, though, although you have
 to make sure you link against any libraries where the C<clock_gettime>
 function is hiding in (often F<-lrt>).
 
@@ -2255,6 +2727,11 @@
 (CLOCK_REALTIME, ...)> and will not normally affect correctness. See the
 note about libraries in the description of C<EV_USE_MONOTONIC>, though.
 
+=item EV_USE_NANOSLEEP
+
+If defined to be C<1>, libev will assume that C<nanosleep ()> is available
+and will use it for delays. Otherwise it will use C<select ()>.
+
 =item EV_USE_SELECT
 
 If undefined or defined to be C<1>, libev will compile in support for the
@@ -2282,6 +2759,14 @@
 it is assumed that all these functions actually work on fds, even
 on win32. Should not be defined on non-win32 platforms.
 
+=item EV_FD_TO_WIN32_HANDLE
+
+If C<EV_SELECT_IS_WINSOCKET> is enabled, then libev needs a way to map
+file descriptors to socket handles. When not defining this symbol (the
+default), then libev will call C<_get_osfhandle>, which is usually
+correct. In some cases, programs use their own file descriptor management,
+in which case they can provide this function to map fds to socket handles.
+
 =item EV_USE_POLL
 
 If defined to be C<1>, libev will compile in support for the C<poll>(2)
@@ -2324,11 +2809,22 @@
 interface to speed up C<ev_stat> watchers. Its actual availability will
 be detected at runtime.
 
+=item EV_ATOMIC_T
+
+Libev requires an integer type (suitable for storing C<0> or C<1>) whose
+access is atomic with respect to other threads or signal contexts. No such
+type is easily found in the C language, so you can provide your own type
+that you know is safe for your purposes. It is used both for signal handler "locking"
+as well as for signal and thread safety in C<ev_async> watchers.
+
+In the absense of this define, libev will use C<sig_atomic_t volatile>
+(from F<signal.h>), which is usually good enough on most platforms.
+
 =item EV_H
 
 The name of the F<ev.h> header file used to include it. The default if
-undefined is C<< <ev.h> >> in F<event.h> and C<"ev.h"> in F<ev.c>. This
-can be used to virtually rename the F<ev.h> header file in case of conflicts.
+undefined is C<"ev.h"> in F<event.h>, F<ev.c> and F<ev++.h>. This can be
+used to virtually rename the F<ev.h> header file in case of conflicts.
 
 =item EV_CONFIG_H
 
@@ -2339,7 +2835,7 @@
 =item EV_EVENT_H
 
 Similarly to C<EV_H>, this macro can be used to override F<event.c>'s idea
-of how the F<event.h> header can be found.
+of how the F<event.h> header can be found, the default is C<"event.h">.
 
 =item EV_PROTOTYPES
 
@@ -2400,6 +2896,11 @@
 If undefined or defined to be C<1>, then fork watchers are supported. If
 defined to be C<0>, then they are not.
 
+=item EV_ASYNC_ENABLE
+
+If undefined or defined to be C<1>, then async watchers are supported. If
+defined to be C<0>, then they are not.
+
 =item EV_MINIMAL
 
 If you need to shave off some kilobytes of code at the expense of some
@@ -2415,7 +2916,7 @@
 
 =item EV_INOTIFY_HASHSIZE
 
-C<ev_staz> watchers use a small hash table to distribute workload by
+C<ev_stat> watchers use a small hash table to distribute workload by
 inotify watch id. The default size is C<16> (or C<1> with C<EV_MINIMAL>),
 usually more than enough. If you need to manage thousands of C<ev_stat>
 watchers you might want to increase this value (I<must> be a power of
@@ -2442,7 +2943,7 @@
 
 Can be used to change the callback member declaration in each watcher,
 and the way callbacks are invoked and set. Must expand to a struct member
-definition and a statement, respectively. See the F<ev.v> header file for
+definition and a statement, respectively. See the F<ev.h> header file for
 their default definitions. One possible use for overriding these is to
 avoid the C<struct ev_loop *> as first argument in all cases, or to use
 method calls instead of plain function calls in C++.
@@ -2460,7 +2961,7 @@
 multiple versions of libev linked together (which is obviously bad in
 itself, but sometimes it is inconvinient to avoid this).
 
-A sed comamnd like this will create wrapper C<#define>'s that you need to
+A sed command like this will create wrapper C<#define>'s that you need to
 include before including F<ev.h>:
 
    <Symbols.ev sed -e "s/.*/#define & myprefix_&/" >wrap.h
@@ -2521,17 +3022,18 @@
 
 This means that, when you have a watcher that triggers in one hour and
 there are 100 watchers that would trigger before that then inserting will
-have to skip those 100 watchers.
+have to skip roughly seven (C<ld 100>) of these watchers.
 
-=item Changing timer/periodic watchers (by autorepeat, again): O(log skipped_other_timers)
+=item Changing timer/periodic watchers (by autorepeat or calling again): O(log skipped_other_timers)
 
-That means that for changing a timer costs less than removing/adding them
+That means that changing a timer costs less than removing/adding them
 as only the relative motion in the event queue has to be paid for.
 
-=item Starting io/check/prepare/idle/signal/child watchers: O(1)
+=item Starting io/check/prepare/idle/signal/child/fork/async watchers: O(1)
 
 These just add the watcher into an array or at the head of a list.
-=item Stopping check/prepare/idle watchers: O(1)
+
+=item Stopping check/prepare/idle/fork/async watchers: O(1)
 
 =item Stopping an io/signal/child watcher: O(number_of_watchers_for_this_(fd/signal/pid % EV_PID_HASHSIZE))
 
@@ -2539,20 +3041,102 @@
 correct watcher to remove. The lists are usually short (you don't usually
 have many watchers waiting for the same fd or signal).
 
-=item Finding the next timer per loop iteration: O(1)
+=item Finding the next timer in each loop iteration: O(1)
+
+By virtue of using a binary heap, the next timer is always found at the
+beginning of the storage array.
 
 =item Each change on a file descriptor per loop iteration: O(number_of_watchers_for_this_fd)
 
 A change means an I/O watcher gets started or stopped, which requires
-libev to recalculate its status (and possibly tell the kernel).
+libev to recalculate its status (and possibly tell the kernel, depending
+on backend and wether C<ev_io_set> was used).
 
-=item Activating one watcher: O(1)
+=item Activating one watcher (putting it into the pending state): O(1)
 
 =item Priority handling: O(number_of_priorities)
 
 Priorities are implemented by allocating some space for each
 priority. When doing priority-based operations, libev usually has to
-linearly search all the priorities.
+linearly search all the priorities, but starting/stopping and activating
+watchers becomes O(1) w.r.t. priority handling.
+
+=item Sending an ev_async: O(1)
+
+=item Processing ev_async_send: O(number_of_async_watchers)
+
+=item Processing signals: O(max_signal_number)
+
+Sending involves a syscall I<iff> there were no other C<ev_async_send>
+calls in the current loop iteration. Checking for async and signal events
+involves iterating over all running async watchers or all signal numbers.
+
+=back
+
+
+=head1 Win32 platform limitations and workarounds
+
+Win32 doesn't support any of the standards (e.g. POSIX) that libev
+requires, and its I/O model is fundamentally incompatible with the POSIX
+model. Libev still offers limited functionality on this platform in
+the form of the C<EVBACKEND_SELECT> backend, and only supports socket
+descriptors. This only applies when using Win32 natively, not when using
+e.g. cygwin.
+
+There is no supported compilation method available on windows except
+embedding it into other applications.
+
+Due to the many, low, and arbitrary limits on the win32 platform and the
+abysmal performance of winsockets, using a large number of sockets is not
+recommended (and not reasonable). If your program needs to use more than
+a hundred or so sockets, then likely it needs to use a totally different
+implementation for windows, as libev offers the POSIX model, which cannot
+be implemented efficiently on windows (microsoft monopoly games).
+
+=over 4
+
+=item The winsocket select function
+
+The winsocket C<select> function doesn't follow POSIX in that it requires
+socket I<handles> and not socket I<file descriptors>. This makes select
+very inefficient, and also requires a mapping from file descriptors
+to socket handles. See the discussion of the C<EV_SELECT_USE_FD_SET>,
+C<EV_SELECT_IS_WINSOCKET> and C<EV_FD_TO_WIN32_HANDLE> preprocessor
+symbols for more info.
+
+The configuration for a "naked" win32 using the microsoft runtime
+libraries and raw winsocket select is:
+
+  #define EV_USE_SELECT 1
+  #define EV_SELECT_IS_WINSOCKET 1   /* forces EV_SELECT_USE_FD_SET, too */
+
+Note that winsockets handling of fd sets is O(n), so you can easily get a
+complexity in the O(n²) range when using win32.
+
+=item Limited number of file descriptors
+
+Windows has numerous arbitrary (and low) limits on things. Early versions
+of winsocket's select only supported waiting for a max. of C<64> handles
+(probably owning to the fact that all windows kernels can only wait for
+C<64> things at the same time internally; microsoft recommends spawning a
+chain of threads and wait for 63 handles and the previous thread in each).
+
+Newer versions support more handles, but you need to define C<FD_SETSIZE>
+to some high number (e.g. C<2048>) before compiling the winsocket select
+call (which might be in libev or elsewhere, for example, perl does its own
+select emulation on windows).
+
+Another limit is the number of file descriptors in the microsoft runtime
+libraries, which by default is C<64> (there must be a hidden I<64> fetish
+or something like this inside microsoft). You can increase this by calling
+C<_setmaxstdio>, which can increase this limit to C<2048> (another
+arbitrary limit), but is broken in many versions of the microsoft runtime
+libraries.
+
+This might get you to about C<512> or C<2048> sockets (depending on
+windows version and/or the phase of the moon). To get more, you need to
+wrap all I/O functions and provide your own fd management, but the cost of
+calling select (O(n²)) will likely make this unworkable.
 
 =back