1 |
=head1 NAME |
2 |
|
3 |
Coro - the only real threads in perl |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
7 |
use Coro; |
8 |
|
9 |
async { |
10 |
# some asynchronous thread of execution |
11 |
print "2\n"; |
12 |
cede; # yield back to main |
13 |
print "4\n"; |
14 |
}; |
15 |
print "1\n"; |
16 |
cede; # yield to coro |
17 |
print "3\n"; |
18 |
cede; # and again |
19 |
|
20 |
# use locking |
21 |
my $lock = new Coro::Semaphore; |
22 |
my $locked; |
23 |
|
24 |
$lock->down; |
25 |
$locked = 1; |
26 |
$lock->up; |
27 |
|
28 |
=head1 DESCRIPTION |
29 |
|
30 |
For a tutorial-style introduction, please read the L<Coro::Intro> |
31 |
manpage. This manpage mainly contains reference information. |
32 |
|
33 |
This module collection manages continuations in general, most often in |
34 |
the form of cooperative threads (also called coros, or simply "coro" |
35 |
in the documentation). They are similar to kernel threads but don't (in |
36 |
general) run in parallel at the same time even on SMP machines. The |
37 |
specific flavor of thread offered by this module also guarantees you that |
38 |
it will not switch between threads unless necessary, at easily-identified |
39 |
points in your program, so locking and parallel access are rarely an |
40 |
issue, making thread programming much safer and easier than using other |
41 |
thread models. |
42 |
|
43 |
Unlike the so-called "Perl threads" (which are not actually real threads |
44 |
but only the windows process emulation (see section of same name for |
45 |
more details) ported to UNIX, and as such act as processes), Coro |
46 |
provides a full shared address space, which makes communication between |
47 |
threads very easy. And coro threads are fast, too: disabling the Windows |
48 |
process emulation code in your perl and using Coro can easily result in |
49 |
a two to four times speed increase for your programs. A parallel matrix |
50 |
multiplication benchmark (very communication-intensive) runs over 300 |
51 |
times faster on a single core than perls pseudo-threads on a quad core |
52 |
using all four cores. |
53 |
|
54 |
Coro achieves that by supporting multiple running interpreters that share |
55 |
data, which is especially useful to code pseudo-parallel processes and |
56 |
for event-based programming, such as multiple HTTP-GET requests running |
57 |
concurrently. See L<Coro::AnyEvent> to learn more on how to integrate Coro |
58 |
into an event-based environment. |
59 |
|
60 |
In this module, a thread is defined as "callchain + lexical variables + |
61 |
some package variables + C stack), that is, a thread has its own callchain, |
62 |
its own set of lexicals and its own set of perls most important global |
63 |
variables (see L<Coro::State> for more configuration and background info). |
64 |
|
65 |
See also the C<SEE ALSO> section at the end of this document - the Coro |
66 |
module family is quite large. |
67 |
|
68 |
=head1 CORO THREAD LIFE CYCLE |
69 |
|
70 |
During the long and exciting (or not) life of a coro thread, it goes |
71 |
through a number of states: |
72 |
|
73 |
=over 4 |
74 |
|
75 |
=item 1. Creation |
76 |
|
77 |
The first thing in the life of a coro thread is it's creation - |
78 |
obviously. The typical way to create a thread is to call the C<async |
79 |
BLOCK> function: |
80 |
|
81 |
async { |
82 |
# thread code goes here |
83 |
}; |
84 |
|
85 |
You can also pass arguments, which are put in C<@_>: |
86 |
|
87 |
async { |
88 |
print $_[1]; # prints 2 |
89 |
} 1, 2, 3; |
90 |
|
91 |
This creates a new coro thread and puts it into the ready queue, meaning |
92 |
it will run as soon as the CPU is free for it. |
93 |
|
94 |
C<async> will return a Coro object - you can store this for future |
95 |
reference or ignore it - a thread that is running, ready to run or waiting |
96 |
for some event is alive on it's own. |
97 |
|
98 |
Another way to create a thread is to call the C<new> constructor with a |
99 |
code-reference: |
100 |
|
101 |
new Coro sub { |
102 |
# thread code goes here |
103 |
}, @optional_arguments; |
104 |
|
105 |
This is quite similar to calling C<async>, but the important difference is |
106 |
that the new thread is not put into the ready queue, so the thread will |
107 |
not run until somebody puts it there. C<async> is, therefore, identical to |
108 |
this sequence: |
109 |
|
110 |
my $coro = new Coro sub { |
111 |
# thread code goes here |
112 |
}; |
113 |
$coro->ready; |
114 |
return $coro; |
115 |
|
116 |
=item 2. Startup |
117 |
|
118 |
When a new coro thread is created, only a copy of the code reference |
119 |
and the arguments are stored, no extra memory for stacks and so on is |
120 |
allocated, keeping the coro thread in a low-memory state. |
121 |
|
122 |
Only when it actually starts executing will all the resources be finally |
123 |
allocated. |
124 |
|
125 |
The optional arguments specified at coro creation are available in C<@_>, |
126 |
similar to function calls. |
127 |
|
128 |
=item 3. Running / Blocking |
129 |
|
130 |
A lot can happen after the coro thread has started running. Quite usually, |
131 |
it will not run to the end in one go (because you could use a function |
132 |
instead), but it will give up the CPU regularly because it waits for |
133 |
external events. |
134 |
|
135 |
As long as a coro thread runs, its Coro object is available in the global |
136 |
variable C<$Coro::current>. |
137 |
|
138 |
The low-level way to give up the CPU is to call the scheduler, which |
139 |
selects a new coro thread to run: |
140 |
|
141 |
Coro::schedule; |
142 |
|
143 |
Since running threads are not in the ready queue, calling the scheduler |
144 |
without doing anything else will block the coro thread forever - you need |
145 |
to arrange either for the coro to put woken up (readied) by some other |
146 |
event or some other thread, or you can put it into the ready queue before |
147 |
scheduling: |
148 |
|
149 |
# this is exactly what Coro::cede does |
150 |
$Coro::current->ready; |
151 |
Coro::schedule; |
152 |
|
153 |
All the higher-level synchronisation methods (Coro::Semaphore, |
154 |
Coro::rouse_*...) are actually implemented via C<< ->ready >> and C<< |
155 |
Coro::schedule >>. |
156 |
|
157 |
While the coro thread is running it also might get assigned a C-level |
158 |
thread, or the C-level thread might be unassigned from it, as the Coro |
159 |
runtime wishes. A C-level thread needs to be assigned when your perl |
160 |
thread calls into some C-level function and that function in turn calls |
161 |
perl and perl then wants to switch coroutines. This happens most often |
162 |
when you run an event loop and block in the callback, or when perl |
163 |
itself calls some function such as C<AUTOLOAD> or methods via the C<tie> |
164 |
mechanism. |
165 |
|
166 |
=item 4. Termination |
167 |
|
168 |
Many threads actually terminate after some time. There are a number of |
169 |
ways to terminate a coro thread, the simplest is returning from the |
170 |
top-level code reference: |
171 |
|
172 |
async { |
173 |
# after returning from here, the coro thread is terminated |
174 |
}; |
175 |
|
176 |
async { |
177 |
return if 0.5 < rand; # terminate a little earlier, maybe |
178 |
print "got a chance to print this\n"; |
179 |
# or here |
180 |
}; |
181 |
|
182 |
Any values returned from the coroutine can be recovered using C<< ->join |
183 |
>>: |
184 |
|
185 |
my $coro = async { |
186 |
"hello, world\n" # return a string |
187 |
}; |
188 |
|
189 |
my $hello_world = $coro->join; |
190 |
|
191 |
print $hello_world; |
192 |
|
193 |
Another way to terminate is to call C<< Coro::terminate >>, which at any |
194 |
subroutine call nesting level: |
195 |
|
196 |
async { |
197 |
Coro::terminate "return value 1", "return value 2"; |
198 |
}; |
199 |
|
200 |
And yet another way is to C<< ->cancel >> (or C<< ->safe_cancel >>) the |
201 |
coro thread from another thread: |
202 |
|
203 |
my $coro = async { |
204 |
exit 1; |
205 |
}; |
206 |
|
207 |
$coro->cancel; # also accepts values for ->join to retrieve |
208 |
|
209 |
Cancellation I<can> be dangerous - it's a bit like calling C<exit> without |
210 |
actually exiting, and might leave C libraries and XS modules in a weird |
211 |
state. Unlike other thread implementations, however, Coro is exceptionally |
212 |
safe with regards to cancellation, as perl will always be in a consistent |
213 |
state, and for those cases where you want to do truly marvellous things |
214 |
with your coro while it is being cancelled - that is, make sure all |
215 |
cleanup code is executed from the thread being cancelled - there is even a |
216 |
C<< ->safe_cancel >> method. |
217 |
|
218 |
So, cancelling a thread that runs in an XS event loop might not be the |
219 |
best idea, but any other combination that deals with perl only (cancelling |
220 |
when a thread is in a C<tie> method or an C<AUTOLOAD> for example) is |
221 |
safe. |
222 |
|
223 |
Lastly, a coro thread object that isn't referenced is C<< ->cancel >>'ed |
224 |
automatically - just like other objects in Perl. This is not such a common |
225 |
case, however - a running thread is referencedy b C<$Coro::current>, a |
226 |
thread ready to run is referenced by the ready queue, a thread waiting |
227 |
on a lock or semaphore is referenced by being in some wait list and so |
228 |
on. But a thread that isn't in any of those queues gets cancelled: |
229 |
|
230 |
async { |
231 |
schedule; # cede to other coros, don't go into the ready queue |
232 |
}; |
233 |
|
234 |
cede; |
235 |
# now the async above is destroyed, as it is not referenced by anything. |
236 |
|
237 |
=item 5. Cleanup |
238 |
|
239 |
Threads will allocate various resources. Most but not all will be returned |
240 |
when a thread terminates, during clean-up. |
241 |
|
242 |
Cleanup is quite similar to throwing an uncaught exception: perl will |
243 |
work it's way up through all subroutine calls and blocks. On it's way, it |
244 |
will release all C<my> variables, undo all C<local>'s and free any other |
245 |
resources truly local to the thread. |
246 |
|
247 |
So, a common way to free resources is to keep them referenced only by my |
248 |
variables: |
249 |
|
250 |
async { |
251 |
my $big_cache = new Cache ...; |
252 |
}; |
253 |
|
254 |
If there are no other references, then the C<$big_cache> object will be |
255 |
freed when the thread terminates, regardless of how it does so. |
256 |
|
257 |
What it does C<NOT> do is unlock any Coro::Semaphores or similar |
258 |
resources, but that's where the C<guard> methods come in handy: |
259 |
|
260 |
my $sem = new Coro::Semaphore; |
261 |
|
262 |
async { |
263 |
my $lock_guard = $sem->guard; |
264 |
# if we reutrn, or die or get cancelled, here, |
265 |
# then the semaphore will be "up"ed. |
266 |
}; |
267 |
|
268 |
The C<Guard::guard> function comes in handy for any custom cleanup you |
269 |
might want to do (but you cannot switch to other coroutines form those |
270 |
code blocks): |
271 |
|
272 |
async { |
273 |
my $window = new Gtk2::Window "toplevel"; |
274 |
# The window will not be cleaned up automatically, even when $window |
275 |
# gets freed, so use a guard to ensure it's destruction |
276 |
# in case of an error: |
277 |
my $window_guard = Guard::guard { $window->destroy }; |
278 |
|
279 |
# we are safe here |
280 |
}; |
281 |
|
282 |
Last not least, C<local> can often be handy, too, e.g. when temporarily |
283 |
replacing the coro thread description: |
284 |
|
285 |
sub myfunction { |
286 |
local $Coro::current->{desc} = "inside myfunction(@_)"; |
287 |
|
288 |
# if we return or die here, the description will be restored |
289 |
} |
290 |
|
291 |
=item 6. Viva La Zombie Muerte |
292 |
|
293 |
Even after a thread has terminated and cleaned up its resources, the Coro |
294 |
object still is there and stores the return values of the thread. |
295 |
|
296 |
The means the Coro object gets freed automatically when the thread has |
297 |
terminated and cleaned up and there arenot other references. |
298 |
|
299 |
If there are, the Coro object will stay around, and you can call C<< |
300 |
->join >> as many times as you wish to retrieve the result values: |
301 |
|
302 |
async { |
303 |
print "hi\n"; |
304 |
1 |
305 |
}; |
306 |
|
307 |
# run the async above, and free everything before returning |
308 |
# from Coro::cede: |
309 |
Coro::cede; |
310 |
|
311 |
{ |
312 |
my $coro = async { |
313 |
print "hi\n"; |
314 |
1 |
315 |
}; |
316 |
|
317 |
# run the async above, and clean up, but do not free the coro |
318 |
# object: |
319 |
Coro::cede; |
320 |
|
321 |
# optionally retrieve the result values |
322 |
my @results = $coro->join; |
323 |
|
324 |
# now $coro goes out of scope, and presumably gets freed |
325 |
}; |
326 |
|
327 |
=back |
328 |
|
329 |
=cut |
330 |
|
331 |
package Coro; |
332 |
|
333 |
use common::sense; |
334 |
|
335 |
use Carp (); |
336 |
|
337 |
use Guard (); |
338 |
|
339 |
use Coro::State; |
340 |
|
341 |
use base qw(Coro::State Exporter); |
342 |
|
343 |
our $idle; # idle handler |
344 |
our $main; # main coro |
345 |
our $current; # current coro |
346 |
|
347 |
our $VERSION = 5.372; |
348 |
|
349 |
our @EXPORT = qw(async async_pool cede schedule terminate current unblock_sub rouse_cb rouse_wait); |
350 |
our %EXPORT_TAGS = ( |
351 |
prio => [qw(PRIO_MAX PRIO_HIGH PRIO_NORMAL PRIO_LOW PRIO_IDLE PRIO_MIN)], |
352 |
); |
353 |
our @EXPORT_OK = (@{$EXPORT_TAGS{prio}}, qw(nready)); |
354 |
|
355 |
=head1 GLOBAL VARIABLES |
356 |
|
357 |
=over 4 |
358 |
|
359 |
=item $Coro::main |
360 |
|
361 |
This variable stores the Coro object that represents the main |
362 |
program. While you cna C<ready> it and do most other things you can do to |
363 |
coro, it is mainly useful to compare again C<$Coro::current>, to see |
364 |
whether you are running in the main program or not. |
365 |
|
366 |
=cut |
367 |
|
368 |
# $main is now being initialised by Coro::State |
369 |
|
370 |
=item $Coro::current |
371 |
|
372 |
The Coro object representing the current coro (the last |
373 |
coro that the Coro scheduler switched to). The initial value is |
374 |
C<$Coro::main> (of course). |
375 |
|
376 |
This variable is B<strictly> I<read-only>. You can take copies of the |
377 |
value stored in it and use it as any other Coro object, but you must |
378 |
not otherwise modify the variable itself. |
379 |
|
380 |
=cut |
381 |
|
382 |
sub current() { $current } # [DEPRECATED] |
383 |
|
384 |
=item $Coro::idle |
385 |
|
386 |
This variable is mainly useful to integrate Coro into event loops. It is |
387 |
usually better to rely on L<Coro::AnyEvent> or L<Coro::EV>, as this is |
388 |
pretty low-level functionality. |
389 |
|
390 |
This variable stores a Coro object that is put into the ready queue when |
391 |
there are no other ready threads (without invoking any ready hooks). |
392 |
|
393 |
The default implementation dies with "FATAL: deadlock detected.", followed |
394 |
by a thread listing, because the program has no other way to continue. |
395 |
|
396 |
This hook is overwritten by modules such as C<Coro::EV> and |
397 |
C<Coro::AnyEvent> to wait on an external event that hopefully wakes up a |
398 |
coro so the scheduler can run it. |
399 |
|
400 |
See L<Coro::EV> or L<Coro::AnyEvent> for examples of using this technique. |
401 |
|
402 |
=cut |
403 |
|
404 |
# ||= because other modules could have provided their own by now |
405 |
$idle ||= new Coro sub { |
406 |
require Coro::Debug; |
407 |
die "FATAL: deadlock detected.\n" |
408 |
. Coro::Debug::ps_listing (); |
409 |
}; |
410 |
|
411 |
# this coro is necessary because a coro |
412 |
# cannot destroy itself. |
413 |
our @destroy; |
414 |
our $manager; |
415 |
|
416 |
$manager = new Coro sub { |
417 |
while () { |
418 |
_destroy shift @destroy |
419 |
while @destroy; |
420 |
|
421 |
&schedule; |
422 |
} |
423 |
}; |
424 |
$manager->{desc} = "[coro manager]"; |
425 |
$manager->prio (PRIO_MAX); |
426 |
|
427 |
=back |
428 |
|
429 |
=head1 SIMPLE CORO CREATION |
430 |
|
431 |
=over 4 |
432 |
|
433 |
=item async { ... } [@args...] |
434 |
|
435 |
Create a new coro and return its Coro object (usually |
436 |
unused). The coro will be put into the ready queue, so |
437 |
it will start running automatically on the next scheduler run. |
438 |
|
439 |
The first argument is a codeblock/closure that should be executed in the |
440 |
coro. When it returns argument returns the coro is automatically |
441 |
terminated. |
442 |
|
443 |
The remaining arguments are passed as arguments to the closure. |
444 |
|
445 |
See the C<Coro::State::new> constructor for info about the coro |
446 |
environment in which coro are executed. |
447 |
|
448 |
Calling C<exit> in a coro will do the same as calling exit outside |
449 |
the coro. Likewise, when the coro dies, the program will exit, |
450 |
just as it would in the main program. |
451 |
|
452 |
If you do not want that, you can provide a default C<die> handler, or |
453 |
simply avoid dieing (by use of C<eval>). |
454 |
|
455 |
Example: Create a new coro that just prints its arguments. |
456 |
|
457 |
async { |
458 |
print "@_\n"; |
459 |
} 1,2,3,4; |
460 |
|
461 |
=item async_pool { ... } [@args...] |
462 |
|
463 |
Similar to C<async>, but uses a coro pool, so you should not call |
464 |
terminate or join on it (although you are allowed to), and you get a |
465 |
coro that might have executed other code already (which can be good |
466 |
or bad :). |
467 |
|
468 |
On the plus side, this function is about twice as fast as creating (and |
469 |
destroying) a completely new coro, so if you need a lot of generic |
470 |
coros in quick successsion, use C<async_pool>, not C<async>. |
471 |
|
472 |
The code block is executed in an C<eval> context and a warning will be |
473 |
issued in case of an exception instead of terminating the program, as |
474 |
C<async> does. As the coro is being reused, stuff like C<on_destroy> |
475 |
will not work in the expected way, unless you call terminate or cancel, |
476 |
which somehow defeats the purpose of pooling (but is fine in the |
477 |
exceptional case). |
478 |
|
479 |
The priority will be reset to C<0> after each run, tracing will be |
480 |
disabled, the description will be reset and the default output filehandle |
481 |
gets restored, so you can change all these. Otherwise the coro will |
482 |
be re-used "as-is": most notably if you change other per-coro global |
483 |
stuff such as C<$/> you I<must needs> revert that change, which is most |
484 |
simply done by using local as in: C<< local $/ >>. |
485 |
|
486 |
The idle pool size is limited to C<8> idle coros (this can be |
487 |
adjusted by changing $Coro::POOL_SIZE), but there can be as many non-idle |
488 |
coros as required. |
489 |
|
490 |
If you are concerned about pooled coros growing a lot because a |
491 |
single C<async_pool> used a lot of stackspace you can e.g. C<async_pool |
492 |
{ terminate }> once per second or so to slowly replenish the pool. In |
493 |
addition to that, when the stacks used by a handler grows larger than 32kb |
494 |
(adjustable via $Coro::POOL_RSS) it will also be destroyed. |
495 |
|
496 |
=cut |
497 |
|
498 |
our $POOL_SIZE = 8; |
499 |
our $POOL_RSS = 32 * 1024; |
500 |
our @async_pool; |
501 |
|
502 |
sub pool_handler { |
503 |
while () { |
504 |
eval { |
505 |
&{&_pool_handler} while 1; |
506 |
}; |
507 |
|
508 |
warn $@ if $@; |
509 |
} |
510 |
} |
511 |
|
512 |
=back |
513 |
|
514 |
=head1 STATIC METHODS |
515 |
|
516 |
Static methods are actually functions that implicitly operate on the |
517 |
current coro. |
518 |
|
519 |
=over 4 |
520 |
|
521 |
=item schedule |
522 |
|
523 |
Calls the scheduler. The scheduler will find the next coro that is |
524 |
to be run from the ready queue and switches to it. The next coro |
525 |
to be run is simply the one with the highest priority that is longest |
526 |
in its ready queue. If there is no coro ready, it will call the |
527 |
C<$Coro::idle> hook. |
528 |
|
529 |
Please note that the current coro will I<not> be put into the ready |
530 |
queue, so calling this function usually means you will never be called |
531 |
again unless something else (e.g. an event handler) calls C<< ->ready >>, |
532 |
thus waking you up. |
533 |
|
534 |
This makes C<schedule> I<the> generic method to use to block the current |
535 |
coro and wait for events: first you remember the current coro in |
536 |
a variable, then arrange for some callback of yours to call C<< ->ready |
537 |
>> on that once some event happens, and last you call C<schedule> to put |
538 |
yourself to sleep. Note that a lot of things can wake your coro up, |
539 |
so you need to check whether the event indeed happened, e.g. by storing the |
540 |
status in a variable. |
541 |
|
542 |
See B<HOW TO WAIT FOR A CALLBACK>, below, for some ways to wait for callbacks. |
543 |
|
544 |
=item cede |
545 |
|
546 |
"Cede" to other coros. This function puts the current coro into |
547 |
the ready queue and calls C<schedule>, which has the effect of giving |
548 |
up the current "timeslice" to other coros of the same or higher |
549 |
priority. Once your coro gets its turn again it will automatically be |
550 |
resumed. |
551 |
|
552 |
This function is often called C<yield> in other languages. |
553 |
|
554 |
=item Coro::cede_notself |
555 |
|
556 |
Works like cede, but is not exported by default and will cede to I<any> |
557 |
coro, regardless of priority. This is useful sometimes to ensure |
558 |
progress is made. |
559 |
|
560 |
=item terminate [arg...] |
561 |
|
562 |
Terminates the current coro with the given status values (see |
563 |
L<cancel>). The values will not be copied, but referenced directly. |
564 |
|
565 |
=item Coro::on_enter BLOCK, Coro::on_leave BLOCK |
566 |
|
567 |
These function install enter and leave winders in the current scope. The |
568 |
enter block will be executed when on_enter is called and whenever the |
569 |
current coro is re-entered by the scheduler, while the leave block is |
570 |
executed whenever the current coro is blocked by the scheduler, and |
571 |
also when the containing scope is exited (by whatever means, be it exit, |
572 |
die, last etc.). |
573 |
|
574 |
I<Neither invoking the scheduler, nor exceptions, are allowed within those |
575 |
BLOCKs>. That means: do not even think about calling C<die> without an |
576 |
eval, and do not even think of entering the scheduler in any way. |
577 |
|
578 |
Since both BLOCKs are tied to the current scope, they will automatically |
579 |
be removed when the current scope exits. |
580 |
|
581 |
These functions implement the same concept as C<dynamic-wind> in scheme |
582 |
does, and are useful when you want to localise some resource to a specific |
583 |
coro. |
584 |
|
585 |
They slow down thread switching considerably for coros that use them |
586 |
(about 40% for a BLOCK with a single assignment, so thread switching is |
587 |
still reasonably fast if the handlers are fast). |
588 |
|
589 |
These functions are best understood by an example: The following function |
590 |
will change the current timezone to "Antarctica/South_Pole", which |
591 |
requires a call to C<tzset>, but by using C<on_enter> and C<on_leave>, |
592 |
which remember/change the current timezone and restore the previous |
593 |
value, respectively, the timezone is only changed for the coro that |
594 |
installed those handlers. |
595 |
|
596 |
use POSIX qw(tzset); |
597 |
|
598 |
async { |
599 |
my $old_tz; # store outside TZ value here |
600 |
|
601 |
Coro::on_enter { |
602 |
$old_tz = $ENV{TZ}; # remember the old value |
603 |
|
604 |
$ENV{TZ} = "Antarctica/South_Pole"; |
605 |
tzset; # enable new value |
606 |
}; |
607 |
|
608 |
Coro::on_leave { |
609 |
$ENV{TZ} = $old_tz; |
610 |
tzset; # restore old value |
611 |
}; |
612 |
|
613 |
# at this place, the timezone is Antarctica/South_Pole, |
614 |
# without disturbing the TZ of any other coro. |
615 |
}; |
616 |
|
617 |
This can be used to localise about any resource (locale, uid, current |
618 |
working directory etc.) to a block, despite the existance of other |
619 |
coros. |
620 |
|
621 |
Another interesting example implements time-sliced multitasking using |
622 |
interval timers (this could obviously be optimised, but does the job): |
623 |
|
624 |
# "timeslice" the given block |
625 |
sub timeslice(&) { |
626 |
use Time::HiRes (); |
627 |
|
628 |
Coro::on_enter { |
629 |
# on entering the thread, we set an VTALRM handler to cede |
630 |
$SIG{VTALRM} = sub { cede }; |
631 |
# and then start the interval timer |
632 |
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0.01, 0.01; |
633 |
}; |
634 |
Coro::on_leave { |
635 |
# on leaving the thread, we stop the interval timer again |
636 |
Time::HiRes::setitimer &Time::HiRes::ITIMER_VIRTUAL, 0, 0; |
637 |
}; |
638 |
|
639 |
&{+shift}; |
640 |
} |
641 |
|
642 |
# use like this: |
643 |
timeslice { |
644 |
# The following is an endless loop that would normally |
645 |
# monopolise the process. Since it runs in a timesliced |
646 |
# environment, it will regularly cede to other threads. |
647 |
while () { } |
648 |
}; |
649 |
|
650 |
|
651 |
=item killall |
652 |
|
653 |
Kills/terminates/cancels all coros except the currently running one. |
654 |
|
655 |
Note that while this will try to free some of the main interpreter |
656 |
resources if the calling coro isn't the main coro, but one |
657 |
cannot free all of them, so if a coro that is not the main coro |
658 |
calls this function, there will be some one-time resource leak. |
659 |
|
660 |
=cut |
661 |
|
662 |
sub killall { |
663 |
for (Coro::State::list) { |
664 |
$_->cancel |
665 |
if $_ != $current && UNIVERSAL::isa $_, "Coro"; |
666 |
} |
667 |
} |
668 |
|
669 |
=back |
670 |
|
671 |
=head1 CORO OBJECT METHODS |
672 |
|
673 |
These are the methods you can call on coro objects (or to create |
674 |
them). |
675 |
|
676 |
=over 4 |
677 |
|
678 |
=item new Coro \&sub [, @args...] |
679 |
|
680 |
Create a new coro and return it. When the sub returns, the coro |
681 |
automatically terminates as if C<terminate> with the returned values were |
682 |
called. To make the coro run you must first put it into the ready |
683 |
queue by calling the ready method. |
684 |
|
685 |
See C<async> and C<Coro::State::new> for additional info about the |
686 |
coro environment. |
687 |
|
688 |
=cut |
689 |
|
690 |
sub _coro_run { |
691 |
terminate &{+shift}; |
692 |
} |
693 |
|
694 |
=item $success = $coro->ready |
695 |
|
696 |
Put the given coro into the end of its ready queue (there is one |
697 |
queue for each priority) and return true. If the coro is already in |
698 |
the ready queue, do nothing and return false. |
699 |
|
700 |
This ensures that the scheduler will resume this coro automatically |
701 |
once all the coro of higher priority and all coro of the same |
702 |
priority that were put into the ready queue earlier have been resumed. |
703 |
|
704 |
=item $coro->suspend |
705 |
|
706 |
Suspends the specified coro. A suspended coro works just like any other |
707 |
coro, except that the scheduler will not select a suspended coro for |
708 |
execution. |
709 |
|
710 |
Suspending a coro can be useful when you want to keep the coro from |
711 |
running, but you don't want to destroy it, or when you want to temporarily |
712 |
freeze a coro (e.g. for debugging) to resume it later. |
713 |
|
714 |
A scenario for the former would be to suspend all (other) coros after a |
715 |
fork and keep them alive, so their destructors aren't called, but new |
716 |
coros can be created. |
717 |
|
718 |
=item $coro->resume |
719 |
|
720 |
If the specified coro was suspended, it will be resumed. Note that when |
721 |
the coro was in the ready queue when it was suspended, it might have been |
722 |
unreadied by the scheduler, so an activation might have been lost. |
723 |
|
724 |
To avoid this, it is best to put a suspended coro into the ready queue |
725 |
unconditionally, as every synchronisation mechanism must protect itself |
726 |
against spurious wakeups, and the one in the Coro family certainly do |
727 |
that. |
728 |
|
729 |
=item $state->is_new |
730 |
|
731 |
Returns true iff this Coro object is "new", i.e. has never been run |
732 |
yet. Those states basically consist of only the code reference to call and |
733 |
the arguments, but consumes very little other resources. New states will |
734 |
automatically get assigned a perl interpreter when they are transfered to. |
735 |
|
736 |
=item $state->is_zombie |
737 |
|
738 |
Returns true iff the Coro object has been cancelled, i.e. |
739 |
it's resources freed because they were C<cancel>'ed, C<terminate>'d, |
740 |
C<safe_cancel>'ed or simply went out of scope. |
741 |
|
742 |
The name "zombie" stems from UNIX culture, where a process that has |
743 |
exited and only stores and exit status and no other resources is called a |
744 |
"zombie". |
745 |
|
746 |
=item $is_ready = $coro->is_ready |
747 |
|
748 |
Returns true iff the Coro object is in the ready queue. Unless the Coro |
749 |
object gets destroyed, it will eventually be scheduled by the scheduler. |
750 |
|
751 |
=item $is_running = $coro->is_running |
752 |
|
753 |
Returns true iff the Coro object is currently running. Only one Coro object |
754 |
can ever be in the running state (but it currently is possible to have |
755 |
multiple running Coro::States). |
756 |
|
757 |
=item $is_suspended = $coro->is_suspended |
758 |
|
759 |
Returns true iff this Coro object has been suspended. Suspended Coros will |
760 |
not ever be scheduled. |
761 |
|
762 |
=item $coro->cancel (arg...) |
763 |
|
764 |
Terminates the given Coro thread and makes it return the given arguments as |
765 |
status (default: an empty list). Never returns if the Coro is the |
766 |
current Coro. |
767 |
|
768 |
This is a rather brutal way to free a coro, with some limitations - if |
769 |
the thread is inside a C callback that doesn't expect to be canceled, |
770 |
bad things can happen, or if the cancelled thread insists on running |
771 |
complicated cleanup handlers that rely on it'S thread context, things will |
772 |
not work. |
773 |
|
774 |
Any cleanup code being run (e.g. from C<guard> blocks) will be run without |
775 |
a thread context, and is not allowed to switch to other threads. On the |
776 |
plus side, C<< ->cancel >> will always clean up the thread, no matter |
777 |
what. If your cleanup code is complex or you want to avoid cancelling a |
778 |
C-thread that doesn't know how to clean up itself, it can be better to C<< |
779 |
->throw >> an exception, or use C<< ->safe_cancel >>. |
780 |
|
781 |
The arguments to C<< ->cancel >> are not copied, but instead will |
782 |
be referenced directly (e.g. if you pass C<$var> and after the call |
783 |
change that variable, then you might change the return values passed to |
784 |
e.g. C<join>, so don't do that). |
785 |
|
786 |
The resources of the Coro are usually freed (or destructed) before this |
787 |
call returns, but this can be delayed for an indefinite amount of time, as |
788 |
in some cases the manager thread has to run first to actually destruct the |
789 |
Coro object. |
790 |
|
791 |
=item $coro->safe_cancel ($arg...) |
792 |
|
793 |
Works mostly like C<< ->cancel >>, but is inherently "safer", and |
794 |
consequently, can fail with an exception in cases the thread is not in a |
795 |
cancellable state. |
796 |
|
797 |
This method works a bit like throwing an exception that cannot be caught |
798 |
- specifically, it will clean up the thread from within itself, so |
799 |
all cleanup handlers (e.g. C<guard> blocks) are run with full thread |
800 |
context and can block if they wish. The downside is that there is no |
801 |
guarantee that the thread can be cancelled when you call this method, and |
802 |
therefore, it might fail. It is also considerably slower than C<cancel> or |
803 |
C<terminate>. |
804 |
|
805 |
A thread is in a safe-cancellable state if it either hasn't been run yet, |
806 |
or it has no C context attached and is inside an SLF function. |
807 |
|
808 |
The latter two basically mean that the thread isn't currently inside a |
809 |
perl callback called from some C function (usually via some XS modules) |
810 |
and isn't currently executing inside some C function itself (via Coro's XS |
811 |
API). |
812 |
|
813 |
This call returns true when it could cancel the thread, or croaks with an |
814 |
error otherwise (i.e. it either returns true or doesn't return at all). |
815 |
|
816 |
Why the weird interface? Well, there are two common models on how and |
817 |
when to cancel things. In the first, you have the expectation that your |
818 |
coro thread can be cancelled when you want to cancel it - if the thread |
819 |
isn't cancellable, this would be a bug somewhere, so C<< ->safe_cancel >> |
820 |
croaks to notify of the bug. |
821 |
|
822 |
In the second model you sometimes want to ask nicely to cancel a thread, |
823 |
but if it's not a good time, well, then don't cancel. This can be done |
824 |
relatively easy like this: |
825 |
|
826 |
if (! eval { $coro->safe_cancel }) { |
827 |
warn "unable to cancel thread: $@"; |
828 |
} |
829 |
|
830 |
However, what you never should do is first try to cancel "safely" and |
831 |
if that fails, cancel the "hard" way with C<< ->cancel >>. That makes |
832 |
no sense: either you rely on being able to execute cleanup code in your |
833 |
thread context, or you don't. If you do, then C<< ->safe_cancel >> is the |
834 |
only way, and if you don't, then C<< ->cancel >> is always faster and more |
835 |
direct. |
836 |
|
837 |
=item $coro->schedule_to |
838 |
|
839 |
Puts the current coro to sleep (like C<Coro::schedule>), but instead |
840 |
of continuing with the next coro from the ready queue, always switch to |
841 |
the given coro object (regardless of priority etc.). The readyness |
842 |
state of that coro isn't changed. |
843 |
|
844 |
This is an advanced method for special cases - I'd love to hear about any |
845 |
uses for this one. |
846 |
|
847 |
=item $coro->cede_to |
848 |
|
849 |
Like C<schedule_to>, but puts the current coro into the ready |
850 |
queue. This has the effect of temporarily switching to the given |
851 |
coro, and continuing some time later. |
852 |
|
853 |
This is an advanced method for special cases - I'd love to hear about any |
854 |
uses for this one. |
855 |
|
856 |
=item $coro->throw ([$scalar]) |
857 |
|
858 |
If C<$throw> is specified and defined, it will be thrown as an exception |
859 |
inside the coro at the next convenient point in time. Otherwise |
860 |
clears the exception object. |
861 |
|
862 |
Coro will check for the exception each time a schedule-like-function |
863 |
returns, i.e. after each C<schedule>, C<cede>, C<< Coro::Semaphore->down |
864 |
>>, C<< Coro::Handle->readable >> and so on. Most of those functions (all |
865 |
that are part of Coro itself) detect this case and return early in case an |
866 |
exception is pending. |
867 |
|
868 |
The exception object will be thrown "as is" with the specified scalar in |
869 |
C<$@>, i.e. if it is a string, no line number or newline will be appended |
870 |
(unlike with C<die>). |
871 |
|
872 |
This can be used as a softer means than either C<cancel> or C<safe_cancel |
873 |
>to ask a coro to end itself, although there is no guarantee that the |
874 |
exception will lead to termination, and if the exception isn't caught it |
875 |
might well end the whole program. |
876 |
|
877 |
You might also think of C<throw> as being the moral equivalent of |
878 |
C<kill>ing a coro with a signal (in this case, a scalar). |
879 |
|
880 |
=item $coro->join |
881 |
|
882 |
Wait until the coro terminates and return any values given to the |
883 |
C<terminate> or C<cancel> functions. C<join> can be called concurrently |
884 |
from multiple threads, and all will be resumed and given the status |
885 |
return once the C<$coro> terminates. |
886 |
|
887 |
=item $coro->on_destroy (\&cb) |
888 |
|
889 |
Registers a callback that is called when this coro thread gets destroyed, |
890 |
that is, after it's resources have been freed but before it is joined. The |
891 |
callback gets passed the terminate/cancel arguments, if any, and I<must |
892 |
not> die, under any circumstances. |
893 |
|
894 |
There can be any number of C<on_destroy> callbacks per coro, and there is |
895 |
no way currently to remove a callback once added. |
896 |
|
897 |
=item $oldprio = $coro->prio ($newprio) |
898 |
|
899 |
Sets (or gets, if the argument is missing) the priority of the |
900 |
coro thread. Higher priority coro get run before lower priority |
901 |
coros. Priorities are small signed integers (currently -4 .. +3), |
902 |
that you can refer to using PRIO_xxx constants (use the import tag :prio |
903 |
to get then): |
904 |
|
905 |
PRIO_MAX > PRIO_HIGH > PRIO_NORMAL > PRIO_LOW > PRIO_IDLE > PRIO_MIN |
906 |
3 > 1 > 0 > -1 > -3 > -4 |
907 |
|
908 |
# set priority to HIGH |
909 |
current->prio (PRIO_HIGH); |
910 |
|
911 |
The idle coro thread ($Coro::idle) always has a lower priority than any |
912 |
existing coro. |
913 |
|
914 |
Changing the priority of the current coro will take effect immediately, |
915 |
but changing the priority of a coro in the ready queue (but not running) |
916 |
will only take effect after the next schedule (of that coro). This is a |
917 |
bug that will be fixed in some future version. |
918 |
|
919 |
=item $newprio = $coro->nice ($change) |
920 |
|
921 |
Similar to C<prio>, but subtract the given value from the priority (i.e. |
922 |
higher values mean lower priority, just as in UNIX's nice command). |
923 |
|
924 |
=item $olddesc = $coro->desc ($newdesc) |
925 |
|
926 |
Sets (or gets in case the argument is missing) the description for this |
927 |
coro thread. This is just a free-form string you can associate with a |
928 |
coro. |
929 |
|
930 |
This method simply sets the C<< $coro->{desc} >> member to the given |
931 |
string. You can modify this member directly if you wish, and in fact, this |
932 |
is often preferred to indicate major processing states that cna then be |
933 |
seen for example in a L<Coro::Debug> session: |
934 |
|
935 |
sub my_long_function { |
936 |
local $Coro::current->{desc} = "now in my_long_function"; |
937 |
... |
938 |
$Coro::current->{desc} = "my_long_function: phase 1"; |
939 |
... |
940 |
$Coro::current->{desc} = "my_long_function: phase 2"; |
941 |
... |
942 |
} |
943 |
|
944 |
=cut |
945 |
|
946 |
sub desc { |
947 |
my $old = $_[0]{desc}; |
948 |
$_[0]{desc} = $_[1] if @_ > 1; |
949 |
$old; |
950 |
} |
951 |
|
952 |
sub transfer { |
953 |
require Carp; |
954 |
Carp::croak ("You must not call ->transfer on Coro objects. Use Coro::State objects or the ->schedule_to method. Caught"); |
955 |
} |
956 |
|
957 |
=back |
958 |
|
959 |
=head1 GLOBAL FUNCTIONS |
960 |
|
961 |
=over 4 |
962 |
|
963 |
=item Coro::nready |
964 |
|
965 |
Returns the number of coro that are currently in the ready state, |
966 |
i.e. that can be switched to by calling C<schedule> directory or |
967 |
indirectly. The value C<0> means that the only runnable coro is the |
968 |
currently running one, so C<cede> would have no effect, and C<schedule> |
969 |
would cause a deadlock unless there is an idle handler that wakes up some |
970 |
coro. |
971 |
|
972 |
=item my $guard = Coro::guard { ... } |
973 |
|
974 |
This function still exists, but is deprecated. Please use the |
975 |
C<Guard::guard> function instead. |
976 |
|
977 |
=cut |
978 |
|
979 |
BEGIN { *guard = \&Guard::guard } |
980 |
|
981 |
=item unblock_sub { ... } |
982 |
|
983 |
This utility function takes a BLOCK or code reference and "unblocks" it, |
984 |
returning a new coderef. Unblocking means that calling the new coderef |
985 |
will return immediately without blocking, returning nothing, while the |
986 |
original code ref will be called (with parameters) from within another |
987 |
coro. |
988 |
|
989 |
The reason this function exists is that many event libraries (such as |
990 |
the venerable L<Event|Event> module) are not thread-safe (a weaker form |
991 |
of reentrancy). This means you must not block within event callbacks, |
992 |
otherwise you might suffer from crashes or worse. The only event library |
993 |
currently known that is safe to use without C<unblock_sub> is L<EV> (but |
994 |
you might still run into deadlocks if all event loops are blocked). |
995 |
|
996 |
Coro will try to catch you when you block in the event loop |
997 |
("FATAL:$Coro::IDLE blocked itself"), but this is just best effort and |
998 |
only works when you do not run your own event loop. |
999 |
|
1000 |
This function allows your callbacks to block by executing them in another |
1001 |
coro where it is safe to block. One example where blocking is handy |
1002 |
is when you use the L<Coro::AIO|Coro::AIO> functions to save results to |
1003 |
disk, for example. |
1004 |
|
1005 |
In short: simply use C<unblock_sub { ... }> instead of C<sub { ... }> when |
1006 |
creating event callbacks that want to block. |
1007 |
|
1008 |
If your handler does not plan to block (e.g. simply sends a message to |
1009 |
another coro, or puts some other coro into the ready queue), there is |
1010 |
no reason to use C<unblock_sub>. |
1011 |
|
1012 |
Note that you also need to use C<unblock_sub> for any other callbacks that |
1013 |
are indirectly executed by any C-based event loop. For example, when you |
1014 |
use a module that uses L<AnyEvent> (and you use L<Coro::AnyEvent>) and it |
1015 |
provides callbacks that are the result of some event callback, then you |
1016 |
must not block either, or use C<unblock_sub>. |
1017 |
|
1018 |
=cut |
1019 |
|
1020 |
our @unblock_queue; |
1021 |
|
1022 |
# we create a special coro because we want to cede, |
1023 |
# to reduce pressure on the coro pool (because most callbacks |
1024 |
# return immediately and can be reused) and because we cannot cede |
1025 |
# inside an event callback. |
1026 |
our $unblock_scheduler = new Coro sub { |
1027 |
while () { |
1028 |
while (my $cb = pop @unblock_queue) { |
1029 |
&async_pool (@$cb); |
1030 |
|
1031 |
# for short-lived callbacks, this reduces pressure on the coro pool |
1032 |
# as the chance is very high that the async_poll coro will be back |
1033 |
# in the idle state when cede returns |
1034 |
cede; |
1035 |
} |
1036 |
schedule; # sleep well |
1037 |
} |
1038 |
}; |
1039 |
$unblock_scheduler->{desc} = "[unblock_sub scheduler]"; |
1040 |
|
1041 |
sub unblock_sub(&) { |
1042 |
my $cb = shift; |
1043 |
|
1044 |
sub { |
1045 |
unshift @unblock_queue, [$cb, @_]; |
1046 |
$unblock_scheduler->ready; |
1047 |
} |
1048 |
} |
1049 |
|
1050 |
=item $cb = rouse_cb |
1051 |
|
1052 |
Create and return a "rouse callback". That's a code reference that, |
1053 |
when called, will remember a copy of its arguments and notify the owner |
1054 |
coro of the callback. |
1055 |
|
1056 |
See the next function. |
1057 |
|
1058 |
=item @args = rouse_wait [$cb] |
1059 |
|
1060 |
Wait for the specified rouse callback (or the last one that was created in |
1061 |
this coro). |
1062 |
|
1063 |
As soon as the callback is invoked (or when the callback was invoked |
1064 |
before C<rouse_wait>), it will return the arguments originally passed to |
1065 |
the rouse callback. In scalar context, that means you get the I<last> |
1066 |
argument, just as if C<rouse_wait> had a C<return ($a1, $a2, $a3...)> |
1067 |
statement at the end. |
1068 |
|
1069 |
See the section B<HOW TO WAIT FOR A CALLBACK> for an actual usage example. |
1070 |
|
1071 |
=back |
1072 |
|
1073 |
=cut |
1074 |
|
1075 |
for my $module (qw(Channel RWLock Semaphore SemaphoreSet Signal Specific)) { |
1076 |
my $old = defined &{"Coro::$module\::new"} && \&{"Coro::$module\::new"}; |
1077 |
|
1078 |
*{"Coro::$module\::new"} = sub { |
1079 |
require "Coro/$module.pm"; |
1080 |
|
1081 |
# some modules have their new predefined in State.xs, some don't |
1082 |
*{"Coro::$module\::new"} = $old |
1083 |
if $old; |
1084 |
|
1085 |
goto &{"Coro::$module\::new"}; |
1086 |
}; |
1087 |
} |
1088 |
|
1089 |
1; |
1090 |
|
1091 |
=head1 HOW TO WAIT FOR A CALLBACK |
1092 |
|
1093 |
It is very common for a coro to wait for some callback to be |
1094 |
called. This occurs naturally when you use coro in an otherwise |
1095 |
event-based program, or when you use event-based libraries. |
1096 |
|
1097 |
These typically register a callback for some event, and call that callback |
1098 |
when the event occured. In a coro, however, you typically want to |
1099 |
just wait for the event, simplyifying things. |
1100 |
|
1101 |
For example C<< AnyEvent->child >> registers a callback to be called when |
1102 |
a specific child has exited: |
1103 |
|
1104 |
my $child_watcher = AnyEvent->child (pid => $pid, cb => sub { ... }); |
1105 |
|
1106 |
But from within a coro, you often just want to write this: |
1107 |
|
1108 |
my $status = wait_for_child $pid; |
1109 |
|
1110 |
Coro offers two functions specifically designed to make this easy, |
1111 |
C<Coro::rouse_cb> and C<Coro::rouse_wait>. |
1112 |
|
1113 |
The first function, C<rouse_cb>, generates and returns a callback that, |
1114 |
when invoked, will save its arguments and notify the coro that |
1115 |
created the callback. |
1116 |
|
1117 |
The second function, C<rouse_wait>, waits for the callback to be called |
1118 |
(by calling C<schedule> to go to sleep) and returns the arguments |
1119 |
originally passed to the callback. |
1120 |
|
1121 |
Using these functions, it becomes easy to write the C<wait_for_child> |
1122 |
function mentioned above: |
1123 |
|
1124 |
sub wait_for_child($) { |
1125 |
my ($pid) = @_; |
1126 |
|
1127 |
my $watcher = AnyEvent->child (pid => $pid, cb => Coro::rouse_cb); |
1128 |
|
1129 |
my ($rpid, $rstatus) = Coro::rouse_wait; |
1130 |
$rstatus |
1131 |
} |
1132 |
|
1133 |
In the case where C<rouse_cb> and C<rouse_wait> are not flexible enough, |
1134 |
you can roll your own, using C<schedule>: |
1135 |
|
1136 |
sub wait_for_child($) { |
1137 |
my ($pid) = @_; |
1138 |
|
1139 |
# store the current coro in $current, |
1140 |
# and provide result variables for the closure passed to ->child |
1141 |
my $current = $Coro::current; |
1142 |
my ($done, $rstatus); |
1143 |
|
1144 |
# pass a closure to ->child |
1145 |
my $watcher = AnyEvent->child (pid => $pid, cb => sub { |
1146 |
$rstatus = $_[1]; # remember rstatus |
1147 |
$done = 1; # mark $rstatus as valud |
1148 |
}); |
1149 |
|
1150 |
# wait until the closure has been called |
1151 |
schedule while !$done; |
1152 |
|
1153 |
$rstatus |
1154 |
} |
1155 |
|
1156 |
|
1157 |
=head1 BUGS/LIMITATIONS |
1158 |
|
1159 |
=over 4 |
1160 |
|
1161 |
=item fork with pthread backend |
1162 |
|
1163 |
When Coro is compiled using the pthread backend (which isn't recommended |
1164 |
but required on many BSDs as their libcs are completely broken), then |
1165 |
coro will not survive a fork. There is no known workaround except to |
1166 |
fix your libc and use a saner backend. |
1167 |
|
1168 |
=item perl process emulation ("threads") |
1169 |
|
1170 |
This module is not perl-pseudo-thread-safe. You should only ever use this |
1171 |
module from the first thread (this requirement might be removed in the |
1172 |
future to allow per-thread schedulers, but Coro::State does not yet allow |
1173 |
this). I recommend disabling thread support and using processes, as having |
1174 |
the windows process emulation enabled under unix roughly halves perl |
1175 |
performance, even when not used. |
1176 |
|
1177 |
Attempts to use threads created in another emulated process will crash |
1178 |
("cleanly", with a null pointer exception). |
1179 |
|
1180 |
=item coro switching is not signal safe |
1181 |
|
1182 |
You must not switch to another coro from within a signal handler (only |
1183 |
relevant with %SIG - most event libraries provide safe signals), I<unless> |
1184 |
you are sure you are not interrupting a Coro function. |
1185 |
|
1186 |
That means you I<MUST NOT> call any function that might "block" the |
1187 |
current coro - C<cede>, C<schedule> C<< Coro::Semaphore->down >> or |
1188 |
anything that calls those. Everything else, including calling C<ready>, |
1189 |
works. |
1190 |
|
1191 |
=back |
1192 |
|
1193 |
|
1194 |
=head1 WINDOWS PROCESS EMULATION |
1195 |
|
1196 |
A great many people seem to be confused about ithreads (for example, Chip |
1197 |
Salzenberg called me unintelligent, incapable, stupid and gullible, |
1198 |
while in the same mail making rather confused statements about perl |
1199 |
ithreads (for example, that memory or files would be shared), showing his |
1200 |
lack of understanding of this area - if it is hard to understand for Chip, |
1201 |
it is probably not obvious to everybody). |
1202 |
|
1203 |
What follows is an ultra-condensed version of my talk about threads in |
1204 |
scripting languages given on the perl workshop 2009: |
1205 |
|
1206 |
The so-called "ithreads" were originally implemented for two reasons: |
1207 |
first, to (badly) emulate unix processes on native win32 perls, and |
1208 |
secondly, to replace the older, real thread model ("5.005-threads"). |
1209 |
|
1210 |
It does that by using threads instead of OS processes. The difference |
1211 |
between processes and threads is that threads share memory (and other |
1212 |
state, such as files) between threads within a single process, while |
1213 |
processes do not share anything (at least not semantically). That |
1214 |
means that modifications done by one thread are seen by others, while |
1215 |
modifications by one process are not seen by other processes. |
1216 |
|
1217 |
The "ithreads" work exactly like that: when creating a new ithreads |
1218 |
process, all state is copied (memory is copied physically, files and code |
1219 |
is copied logically). Afterwards, it isolates all modifications. On UNIX, |
1220 |
the same behaviour can be achieved by using operating system processes, |
1221 |
except that UNIX typically uses hardware built into the system to do this |
1222 |
efficiently, while the windows process emulation emulates this hardware in |
1223 |
software (rather efficiently, but of course it is still much slower than |
1224 |
dedicated hardware). |
1225 |
|
1226 |
As mentioned before, loading code, modifying code, modifying data |
1227 |
structures and so on is only visible in the ithreads process doing the |
1228 |
modification, not in other ithread processes within the same OS process. |
1229 |
|
1230 |
This is why "ithreads" do not implement threads for perl at all, only |
1231 |
processes. What makes it so bad is that on non-windows platforms, you can |
1232 |
actually take advantage of custom hardware for this purpose (as evidenced |
1233 |
by the forks module, which gives you the (i-) threads API, just much |
1234 |
faster). |
1235 |
|
1236 |
Sharing data is in the i-threads model is done by transfering data |
1237 |
structures between threads using copying semantics, which is very slow - |
1238 |
shared data simply does not exist. Benchmarks using i-threads which are |
1239 |
communication-intensive show extremely bad behaviour with i-threads (in |
1240 |
fact, so bad that Coro, which cannot take direct advantage of multiple |
1241 |
CPUs, is often orders of magnitude faster because it shares data using |
1242 |
real threads, refer to my talk for details). |
1243 |
|
1244 |
As summary, i-threads *use* threads to implement processes, while |
1245 |
the compatible forks module *uses* processes to emulate, uhm, |
1246 |
processes. I-threads slow down every perl program when enabled, and |
1247 |
outside of windows, serve no (or little) practical purpose, but |
1248 |
disadvantages every single-threaded Perl program. |
1249 |
|
1250 |
This is the reason that I try to avoid the name "ithreads", as it is |
1251 |
misleading as it implies that it implements some kind of thread model for |
1252 |
perl, and prefer the name "windows process emulation", which describes the |
1253 |
actual use and behaviour of it much better. |
1254 |
|
1255 |
=head1 SEE ALSO |
1256 |
|
1257 |
Event-Loop integration: L<Coro::AnyEvent>, L<Coro::EV>, L<Coro::Event>. |
1258 |
|
1259 |
Debugging: L<Coro::Debug>. |
1260 |
|
1261 |
Support/Utility: L<Coro::Specific>, L<Coro::Util>. |
1262 |
|
1263 |
Locking and IPC: L<Coro::Signal>, L<Coro::Channel>, L<Coro::Semaphore>, |
1264 |
L<Coro::SemaphoreSet>, L<Coro::RWLock>. |
1265 |
|
1266 |
I/O and Timers: L<Coro::Timer>, L<Coro::Handle>, L<Coro::Socket>, L<Coro::AIO>. |
1267 |
|
1268 |
Compatibility with other modules: L<Coro::LWP> (but see also L<AnyEvent::HTTP> for |
1269 |
a better-working alternative), L<Coro::BDB>, L<Coro::Storable>, |
1270 |
L<Coro::Select>. |
1271 |
|
1272 |
XS API: L<Coro::MakeMaker>. |
1273 |
|
1274 |
Low level Configuration, Thread Environment, Continuations: L<Coro::State>. |
1275 |
|
1276 |
=head1 AUTHOR |
1277 |
|
1278 |
Marc Lehmann <schmorp@schmorp.de> |
1279 |
http://home.schmorp.de/ |
1280 |
|
1281 |
=cut |
1282 |
|