1 |
=head1 NAME |
2 |
|
3 |
Coro::State - first class continuations |
4 |
|
5 |
=head1 SYNOPSIS |
6 |
|
7 |
use Coro::State; |
8 |
|
9 |
$new = new Coro::State sub { |
10 |
print "in coro (called with @_), switching back\n"; |
11 |
$new->transfer ($main); |
12 |
print "in coro again, switching back\n"; |
13 |
$new->transfer ($main); |
14 |
}, 5; |
15 |
|
16 |
$main = new Coro::State; |
17 |
|
18 |
print "in main, switching to coro\n"; |
19 |
$main->transfer ($new); |
20 |
print "back in main, switch to coro again\n"; |
21 |
$main->transfer ($new); |
22 |
print "back in main\n"; |
23 |
|
24 |
=head1 DESCRIPTION |
25 |
|
26 |
This module implements coro objects. Coros, similar to threads and |
27 |
continuations, allow you to run more than one "thread of execution" in |
28 |
parallel. Unlike so-called "kernel" threads, there is no parallelism |
29 |
and only voluntary switching is used so locking problems are greatly |
30 |
reduced. The latter is called "cooperative" threading as opposed to |
31 |
"preemptive" threading. |
32 |
|
33 |
This can be used to implement non-local jumps, exception handling, |
34 |
continuation objects and more. |
35 |
|
36 |
This module provides only low-level functionality useful to build other |
37 |
abstractions, such as threads, generators or coroutines. See L<Coro> |
38 |
and related modules for a higher level threads abstraction including a |
39 |
scheduler. |
40 |
|
41 |
=head2 MODEL |
42 |
|
43 |
Coro::State implements two different thread models: Perl and C. The C |
44 |
threads (called cctx's) are basically simplified perl interpreters |
45 |
running/interpreting the Perl threads. A single interpreter can run any |
46 |
number of Perl threads, so usually there are very few C threads. |
47 |
|
48 |
When Perl code calls a C function (e.g. in an extension module) and that C |
49 |
function then calls back into Perl or transfers control to another thread, |
50 |
the C thread can no longer execute other Perl threads, so it stays tied to |
51 |
the specific thread until it returns to the original Perl caller, after |
52 |
which it is again available to run other Perl threads. |
53 |
|
54 |
The main program always has its own "C thread" (which really is |
55 |
*the* Perl interpreter running the whole program), so there will always |
56 |
be at least one additional C thread. You can use the debugger (see |
57 |
L<Coro::Debug>) to find out which threads are tied to their cctx and |
58 |
which aren't. |
59 |
|
60 |
=head2 MEMORY CONSUMPTION |
61 |
|
62 |
A newly created Coro::State that has not been used only allocates a |
63 |
relatively small (a hundred bytes) structure. Only on the first |
64 |
C<transfer> will perl allocate stacks (a few kb, 64 bit architectures |
65 |
use twice as much, i.e. a few kb :) and optionally a C stack/thread |
66 |
(cctx) for threads that recurse through C functions. All this is very |
67 |
system-dependent. On my x86-pc-linux-gnu system this amounts to about 2k |
68 |
per (non-trivial but simple) Coro::State. |
69 |
|
70 |
You can view the actual memory consumption using Coro::Debug. Keep in mind |
71 |
that a for loop or other block constructs can easily consume 100-200 bytes |
72 |
per nesting level. |
73 |
|
74 |
=cut |
75 |
|
76 |
package Coro::State; |
77 |
|
78 |
use common::sense; |
79 |
|
80 |
use Carp; |
81 |
|
82 |
our $DIEHOOK; |
83 |
our $WARNHOOK; |
84 |
|
85 |
BEGIN { |
86 |
$DIEHOOK = sub { }; |
87 |
$WARNHOOK = sub { warn $_[0] }; |
88 |
} |
89 |
|
90 |
sub diehook { &$DIEHOOK } |
91 |
sub warnhook { &$WARNHOOK } |
92 |
|
93 |
use XSLoader; |
94 |
|
95 |
BEGIN { |
96 |
our $VERSION = 6.57; |
97 |
|
98 |
# must be done here because the xs part expects it to exist |
99 |
# it might exist already because Coro::Specific created it. |
100 |
$Coro::current ||= { }; |
101 |
|
102 |
XSLoader::load __PACKAGE__, $VERSION; |
103 |
|
104 |
# major complication: |
105 |
# perl stores a PVMG with sigelem magic in warnhook, and retrieves the |
106 |
# value from the hash, even while PL_warnhook is zero. |
107 |
# Coro can't do that because the value in the hash might be stale. |
108 |
# Therefore, Coro stores a copy, and returns PL_warnhook itself, so we |
109 |
# need to manually copy the existing handlers to remove their magic. |
110 |
# I chose to use "delete", to hopefuly get rid of the remnants, |
111 |
# but (my $v = $SIG{...}) would also work. |
112 |
$SIG{__DIE__} = (delete $SIG{__DIE__} ) || \&diehook; |
113 |
$SIG{__WARN__} = (delete $SIG{__WARN__}) || \&warnhook; |
114 |
} |
115 |
|
116 |
use Exporter; |
117 |
use base Exporter::; |
118 |
|
119 |
=head2 GLOBAL VARIABLES |
120 |
|
121 |
=over 4 |
122 |
|
123 |
=item $Coro::State::DIEHOOK |
124 |
|
125 |
This works similarly to C<$SIG{__DIE__}> and is used as the default die |
126 |
hook for newly created Coro::States. This is useful if you want some generic |
127 |
logging function that works for all threads that don't set their own |
128 |
hook. |
129 |
|
130 |
When Coro::State is first loaded it will install these handlers for the |
131 |
main program, too, unless they have been overwritten already. |
132 |
|
133 |
The default handlers provided will behave like the built-in ones (as if |
134 |
they weren't there). |
135 |
|
136 |
If you don't want to exit your program on uncaught exceptions, you must |
137 |
not return from your die hook - call C<Coro::terminate> instead. |
138 |
|
139 |
Note 1: You I<must> store a valid code reference in these variables, |
140 |
C<undef> will I<not> do. |
141 |
|
142 |
Note 2: The value of this variable will be shared among all threads, so |
143 |
changing its value will change it in all threads that don't have their |
144 |
own die handler. |
145 |
|
146 |
=item $Coro::State::WARNHOOK |
147 |
|
148 |
Similar to above die hook, but augments C<$SIG{__WARN__}>. |
149 |
|
150 |
=back |
151 |
|
152 |
=head2 Coro::State METHODS |
153 |
|
154 |
=over 4 |
155 |
|
156 |
=item $coro = new Coro::State [$coderef[, @args...]] |
157 |
|
158 |
Create a new Coro::State thread object and return it. The first |
159 |
C<transfer> call to this thread will start execution at the given |
160 |
coderef, with the given arguments. |
161 |
|
162 |
Note that the arguments will not be copied. Instead, as with normal |
163 |
function calls, the thread receives passed arguments by reference, so |
164 |
make sure you don't change them in unexpected ways. |
165 |
|
166 |
Returning from such a thread is I<NOT> supported. Neither is calling |
167 |
C<exit> or throwing an uncaught exception. The following paragraphs |
168 |
describe what happens in current versions of Coro. |
169 |
|
170 |
If the subroutine returns the program will be terminated as if execution |
171 |
of the main program ended. |
172 |
|
173 |
If it throws an exception the program will terminate unless the exception |
174 |
is caught, exactly like in the main program. |
175 |
|
176 |
Calling C<exit> in a thread does the same as calling it in the main |
177 |
program, but due to libc bugs on many BSDs, this doesn't work reliable |
178 |
everywhere. |
179 |
|
180 |
If the coderef is omitted this function will create a new "empty" |
181 |
thread, i.e. a thread that cannot be transferred to but can be used |
182 |
to save the current thread state in (note that this is dangerous, as no |
183 |
reference is taken to ensure that the "current thread state" survives, |
184 |
the caller is responsible to ensure that the cloned state does not go |
185 |
away). |
186 |
|
187 |
The returned object is an empty hash which can be used for any purpose |
188 |
whatsoever, for example when subclassing Coro::State. |
189 |
|
190 |
Certain variables are "localised" to each thread, that is, certain |
191 |
"global" variables are actually per thread. Not everything that would |
192 |
sensibly be localised currently is, and not everything that is localised |
193 |
makes sense for every application, and the future might bring changes. |
194 |
|
195 |
The following global variables can have different values per thread, |
196 |
and have the stated initial values: |
197 |
|
198 |
Variable Initial Value |
199 |
@_ whatever arguments were passed to the Coro |
200 |
$_ undef |
201 |
$@ undef |
202 |
$/ "\n" |
203 |
$SIG{__DIE__} aliased to $Coro::State::DIEHOOK(*) |
204 |
$SIG{__WARN__} aliased to $Coro::State::WARNHOOK(*) |
205 |
(default fh) *STDOUT |
206 |
$^H, %^H zero/empty. |
207 |
$1, $2... all regex results are initially undefined |
208 |
|
209 |
(*) reading the value from %SIG is not supported, but local'ising is. |
210 |
|
211 |
If you feel that something important is missing then tell me. Also |
212 |
remember that every function call that might call C<transfer> (such |
213 |
as C<Coro::Channel::put>) might clobber any global and/or special |
214 |
variables. Yes, this is by design ;) You can always create your own |
215 |
process abstraction model that saves these variables. |
216 |
|
217 |
The easiest way to do this is to create your own scheduling primitive like |
218 |
in the code below, and use it in your threads: |
219 |
|
220 |
sub my_cede { |
221 |
local ($;, ...); |
222 |
Coro::cede; |
223 |
} |
224 |
|
225 |
Another way is to use dynamic winders, see C<Coro::on_enter> and |
226 |
C<Coro::on_leave> for this. |
227 |
|
228 |
Yet another way that works only for variables is C<< ->swap_sv >>. |
229 |
|
230 |
=item $prev->transfer ($next) |
231 |
|
232 |
Save the state of the current subroutine in C<$prev> and switch to the |
233 |
thread saved in C<$next>. |
234 |
|
235 |
The "state" of a subroutine includes the scope, i.e. lexical variables and |
236 |
the current execution state (subroutine, stack). |
237 |
|
238 |
=item $state->throw ([$scalar]) |
239 |
|
240 |
=item $state->is_new |
241 |
|
242 |
=item $state->is_zombie |
243 |
|
244 |
See the corresponding method(s) for L<Coro> objects. |
245 |
|
246 |
=item $state->cancel |
247 |
|
248 |
Forcefully destructs the given Coro::State. While you can keep the |
249 |
reference, and some memory is still allocated, the Coro::State object is |
250 |
effectively dead, destructors have been freed, it cannot be transferred to |
251 |
anymore, it's pushing up the daisies. |
252 |
|
253 |
=item $state->call ($coderef) |
254 |
|
255 |
Try to call the given C<$coderef> in the context of the given state. This |
256 |
works even when the state is currently within an XS function, and can |
257 |
be very dangerous. You can use it to acquire stack traces etc. (see the |
258 |
Coro::Debug module for more details). The coderef MUST NOT EVER transfer |
259 |
to another state. |
260 |
|
261 |
=item $state->eval ($string) |
262 |
|
263 |
Like C<call>, but eval's the string. Dangerous. |
264 |
|
265 |
=item $state->swap_defsv |
266 |
|
267 |
=item $state->swap_defav |
268 |
|
269 |
Swap the current C<$_> (swap_defsv) or C<@_> (swap_defav) with the |
270 |
equivalent in the saved state of C<$state>. This can be used to give the |
271 |
coro a defined content for C<@_> and C<$_> before transfer'ing to it. |
272 |
|
273 |
=item $state->swap_sv (\$sv, \$swap_sv) |
274 |
|
275 |
This (very advanced) function can be used to make I<any> variable local to |
276 |
a thread. |
277 |
|
278 |
It works by swapping the contents of C<$sv> and C<$swap_sv> each time the |
279 |
thread is entered and left again, i.e. it is similar to: |
280 |
|
281 |
$tmp = $sv; $sv = $swap_sv; $swap_sv = $tmp; |
282 |
|
283 |
Except that it doesn't make any copies and works on hashes and even more |
284 |
exotic values (code references!). |
285 |
|
286 |
When called on the current thread (i.e. from within the thread that will |
287 |
receive the swap_sv), then this method acts as if it was called from |
288 |
another thread, i.e. after adding the two SV's to the threads swap list |
289 |
their values will be swapped. |
290 |
|
291 |
Needless to say, this function can be very very dangerous: you can easily |
292 |
swap a hash with a reference (i.e. C<%hash> I<becomes> a reference), and perl |
293 |
will not like this at all. |
294 |
|
295 |
It will also swap "magicalness" - so when swapping a builtin perl variable |
296 |
(such as C<$.>), it will lose its magicalness, which, again, perl will |
297 |
not like, so don't do it. |
298 |
|
299 |
Lastly, the C<$swap_sv> itself will be used, not a copy, so make sure you |
300 |
give each thread its own C<$swap_sv> instance. |
301 |
|
302 |
It is, however, quite safe to swap some normal variable with |
303 |
another. For example, L<PApp::SQL> stores the default database handle in |
304 |
C<$PApp::SQL::DBH>. To make this a per-thread variable, use this: |
305 |
|
306 |
my $private_dbh = ...; |
307 |
$coro->swap_sv (\$PApp::SQL::DBH, \$private_dbh); |
308 |
|
309 |
This results in C<$PApp::SQL::DBH> having the value of C<$private_dbh> |
310 |
while it executes, and whatever other value it had when it doesn't |
311 |
execute. |
312 |
|
313 |
You can also swap hashes and other values: |
314 |
|
315 |
my %private_hash; |
316 |
$coro->swap_sv (\%some_hash, \%private_hash); |
317 |
|
318 |
To undo an earlier C<swap_sv> call you must call C<swap_sv> with exactly |
319 |
the same two variables in the same order (the references can be different, |
320 |
it's the variables that they point to that count). For example, the |
321 |
following sequence will remove the swap of C<$x> and C<$y>, while keeping |
322 |
the swap of C<$x> and C<$z>: |
323 |
|
324 |
$coro->swap_sv (\$x, \$y); |
325 |
$coro->swap_sv (\$x, \$z); |
326 |
$coro->swap_sv (\$x, \$y); |
327 |
|
328 |
=item $bytes = $state->rss |
329 |
|
330 |
Returns the memory allocated by the coro (which includes static |
331 |
structures, various perl stacks but NOT local variables, arguments or any |
332 |
C context data). This is a rough indication of how much memory it might |
333 |
use. |
334 |
|
335 |
=item ($real, $cpu) = $state->times |
336 |
|
337 |
Returns the real time and cpu times spent in the given C<$state>. See |
338 |
C<Coro::State::enable_times> for more info. |
339 |
|
340 |
=item $state->trace ($flags) |
341 |
|
342 |
Internal function to control tracing. I just mention this so you can stay |
343 |
away from abusing it. |
344 |
|
345 |
=back |
346 |
|
347 |
=head3 METHODS FOR C CONTEXTS |
348 |
|
349 |
Most coros only consist of some Perl data structures - transferring to a |
350 |
coro just reconfigures the interpreter to continue somewhere else. |
351 |
|
352 |
However. this is not always possible: For example, when Perl calls a C/XS function |
353 |
(such as an event loop), and C then invokes a Perl callback, reconfiguring |
354 |
the interpreter is not enough. Coro::State detects these cases automatically, and |
355 |
attaches a C-level thread to each such Coro::State object, for as long as necessary. |
356 |
|
357 |
The C-level thread structure is called "C context" (or cctxt for short), |
358 |
and can be quite big, which is why Coro::State only creates them as needed |
359 |
and can run many Coro::State's on a single cctxt. |
360 |
|
361 |
This is mostly transparent, so the following methods are rarely needed. |
362 |
|
363 |
=over 4 |
364 |
|
365 |
=item $state->has_cctx |
366 |
|
367 |
Returns whether the state currently uses a cctx/C context. An active |
368 |
state always has a cctx, as well as the main program. Other states only |
369 |
use a cctxts when needed. |
370 |
|
371 |
=item Coro::State::force_cctx |
372 |
|
373 |
Forces the allocation of a private cctxt for the currently executing |
374 |
Coro::State even though it would not normally ned one. Apart from |
375 |
benchmarking or testing Coro itself, there is little point in doing so, |
376 |
however. |
377 |
|
378 |
=item $ncctx = Coro::State::cctx_count |
379 |
|
380 |
Returns the number of C contexts allocated. If this number is very high |
381 |
(more than a dozen) it might be beneficial to identify points of C-level |
382 |
recursion (Perl calls C/XS, which calls Perl again which switches coros |
383 |
- this forces an allocation of a C context) in your code and moving this |
384 |
into a separate coro. |
385 |
|
386 |
=item $nidle = Coro::State::cctx_idle |
387 |
|
388 |
Returns the number of allocated but idle (currently unused and free for |
389 |
reuse) C contexts. |
390 |
|
391 |
=item $old = Coro::State::cctx_max_idle [$new_count] |
392 |
|
393 |
Coro caches C contexts that are not in use currently, as creating them |
394 |
from scratch has some overhead. |
395 |
|
396 |
This function returns the current maximum number of idle C contexts and |
397 |
optionally sets the new amount. The count must be at least C<1>, with the |
398 |
default being C<4>. |
399 |
|
400 |
=item $old = Coro::State::cctx_stacksize [$new_stacksize] |
401 |
|
402 |
Returns the current C stack size and optionally sets the new I<minimum> |
403 |
stack size to C<$new_stacksize> (in units of pointer sizes, i.e. typically |
404 |
4 on 32 bit and 8 on 64 bit hosts). Existing stacks will not be changed, |
405 |
but Coro will try to replace smaller stacks as soon as possible. Any |
406 |
Coro::State that starts to use a stack after this call is guaranteed this |
407 |
minimum stack size. |
408 |
|
409 |
Please note that coros will only need to use a C-level stack if the |
410 |
interpreter recurses or calls a function in a module that calls back into |
411 |
the interpreter, so use of this feature is usually never needed. |
412 |
|
413 |
=back |
414 |
|
415 |
=head2 FUNCTIONS |
416 |
|
417 |
=over 4 |
418 |
|
419 |
=item @states = Coro::State::list |
420 |
|
421 |
Returns a list of all Coro::State objects currently allocated. This |
422 |
includes all derived objects (such as L<Coro> threads). |
423 |
|
424 |
=item $was_enabled = Coro::State::enable_times [$enable] |
425 |
|
426 |
Enables/disables/queries the current state of per-thread real and |
427 |
cpu-time gathering. |
428 |
|
429 |
When enabled, the real time and the cpu time (user + system time) |
430 |
spent in each thread is accumulated. If disabled, then the accumulated |
431 |
times will stay as they are (they start at 0). |
432 |
|
433 |
Currently, cpu time is only measured on GNU/Linux systems, all other |
434 |
systems only gather real time. |
435 |
|
436 |
Enabling time profiling slows down thread switching by a factor of 2 to |
437 |
10, depending on platform on hardware. |
438 |
|
439 |
The times will be displayed when running C<Coro::Debug::command "ps">, and |
440 |
can be queried by calling C<< $state->times >>. |
441 |
|
442 |
=back |
443 |
|
444 |
=head3 CLONING |
445 |
|
446 |
=over 4 |
447 |
|
448 |
=item $clone = $state->clone |
449 |
|
450 |
This exciting method takes a Coro::State object and clones it, i.e., it |
451 |
creates a copy. This makes it possible to restore a state more than once, |
452 |
and even return to states that have returned or have been terminated. |
453 |
|
454 |
Since its only known purpose is for intellectual self-gratification, and |
455 |
because it is a difficult piece of code, it is not enabled by default, and |
456 |
not supported. |
457 |
|
458 |
Here are a few little-known facts: First, coros *are* full/true/real |
459 |
continuations. Secondly Coro::State objects (without clone) *are* first |
460 |
class continuations. Thirdly, nobody has ever found a use for the full |
461 |
power of call/cc that isn't better (faster, easier, more efficiently) |
462 |
implemented differently, and nobody has yet found a useful control |
463 |
construct that can't be implemented without it already, just much faster |
464 |
and with fewer resources. And lastly, Scheme's call/cc doesn't support |
465 |
using call/cc to implement threads. |
466 |
|
467 |
Among the games you can play with this is implementing a scheme-like |
468 |
call-with-current-continuation, as the following code does (well, with |
469 |
small differences). |
470 |
|
471 |
# perl disassociates from local lexicals on frame exit, |
472 |
# so use a global variable for return values. |
473 |
my @ret; |
474 |
|
475 |
sub callcc($@) { |
476 |
my ($func, @arg) = @_; |
477 |
|
478 |
my $continuation = new Coro::State; |
479 |
$continuation->transfer (new Coro::State sub { |
480 |
my $escape = sub { |
481 |
@ret = @_; |
482 |
Coro::State->new->transfer ($continuation->clone); |
483 |
}; |
484 |
$escape->($func->($escape, @arg)); |
485 |
}); |
486 |
|
487 |
my @ret_ = @ret; @ret = (); |
488 |
wantarray ? @ret_ : pop @ret_ |
489 |
} |
490 |
|
491 |
Which could be used to implement a loop like this: |
492 |
|
493 |
async { |
494 |
my $n; |
495 |
my $l = callcc sub { $_[0] }; |
496 |
|
497 |
$n++; |
498 |
print "iteration $n\n"; |
499 |
|
500 |
$l->($l) unless $n == 10; |
501 |
}; |
502 |
|
503 |
If you find this confusing, then you already understand the coolness of |
504 |
call/cc: It can turn anything into spaghetti code real fast. |
505 |
|
506 |
Besides, call/cc is much less useful in a Perl-like dynamic language (with |
507 |
references, and its scoping rules) then in, say, scheme. |
508 |
|
509 |
Now, the known limitations of C<clone>: |
510 |
|
511 |
It probably only works on perl 5.10; it cannot clone a coro inside |
512 |
the substition operator (but windows perl can't fork from there either) |
513 |
and some other contexts, and C<abort ()> is the preferred mechanism to |
514 |
signal errors. It cannot clone a state that has a c context attached |
515 |
(implementing clone on the C level is too hard for me to even try), |
516 |
which rules out calling call/cc from the main coro. It cannot |
517 |
clone a context that hasn't even been started yet. It doesn't work with |
518 |
C<-DDEBUGGING> (but what does). It probably also leaks, and sometimes |
519 |
triggers a few assertions inside Coro. Most of these limitations *are* |
520 |
fixable with some effort, but that's pointless just to make a point that |
521 |
it could be done. |
522 |
|
523 |
The current implementation could without doubt be optimised to be a |
524 |
constant-time operation by doing lazy stack copying, if somebody were |
525 |
insane enough to invest the time. |
526 |
|
527 |
=cut |
528 |
|
529 |
# used by Coro::Debug only atm. |
530 |
sub debug_desc { |
531 |
$_[0]{desc} |
532 |
} |
533 |
|
534 |
# for very deep reasons, we must initialise $Coro::main here. |
535 |
|
536 |
{ |
537 |
package Coro; |
538 |
|
539 |
our $main; # main coro |
540 |
our $current; # current coro |
541 |
|
542 |
$main = Coro::new Coro::; |
543 |
|
544 |
$main->{desc} = "[main::]"; |
545 |
|
546 |
# maybe some other module used Coro::Specific before... |
547 |
$main->{_specific} = $current->{_specific} |
548 |
if $current; |
549 |
|
550 |
_set_current $main; |
551 |
} |
552 |
|
553 |
# we also make sure we have Coro::AnyEvent when AnyEvent is used, |
554 |
# without loading or initialising AnyEvent |
555 |
if (defined $AnyEvent::MODEL) { |
556 |
require Coro::AnyEvent; |
557 |
} else { |
558 |
push @AnyEvent::post_detect, sub { require Coro::AnyEvent }; |
559 |
} |
560 |
|
561 |
1; |
562 |
|
563 |
=back |
564 |
|
565 |
=head1 BUGS |
566 |
|
567 |
This module is not thread-safe. You must only ever use this module from |
568 |
the same thread (this requirement might be removed in the future). |
569 |
|
570 |
=head1 SEE ALSO |
571 |
|
572 |
L<Coro>. |
573 |
|
574 |
=head1 AUTHOR/SUPPORT/CONTACT |
575 |
|
576 |
Marc A. Lehmann <schmorp@schmorp.de> |
577 |
http://software.schmorp.de/pkg/Coro.html |
578 |
|
579 |
=cut |
580 |
|