Commit | Line | Data |
43d3ddbe |
1 | package Thread; |
2 | |
3 | $VERSION = '2.00'; |
4 | |
4038bebf |
5 | use strict; |
6 | |
7 | our $ithreads; |
8 | our $othreads; |
9 | |
43d3ddbe |
10 | BEGIN { |
11 | use Config; |
4038bebf |
12 | $ithreads = $Config{useithreads}; |
13 | $othreads = $Config{use5005threads}; |
43d3ddbe |
14 | } |
15 | |
16 | require Exporter; |
17 | use XSLoader (); |
4038bebf |
18 | our($VERSION, @ISA, @EXPORT, @EXPORT_OK); |
43d3ddbe |
19 | |
20 | @ISA = qw(Exporter); |
21 | |
22 | BEGIN { |
23 | if ($ithreads) { |
24 | @EXPORT = qw(share cond_wait cond_broadcast cond_signal unlock) |
25 | } elsif ($othreads) { |
26 | @EXPORT_OK = qw(cond_signal cond_broadcast cond_wait); |
27 | } |
28 | push @EXPORT_OK, qw(async yield); |
29 | } |
30 | |
31 | =head1 NAME |
32 | |
33 | Thread - manipulate threads in Perl |
34 | |
35 | =head1 CAVEAT |
36 | |
37 | Perl has two thread models. |
38 | |
39 | In Perl 5.005 the thread model was that all data is implicitly shared |
40 | and shared access to data has to be explicitly synchronized. |
41 | This model is called "5005threads". |
42 | |
43 | In Perl 5.6 a new model was introduced in which all is was thread |
44 | local and shared access to data has to be explicitly declared. |
45 | This model is called "ithreads", for "interpreter threads". |
46 | |
47 | In Perl 5.6 the ithreads model was not available as a public API, |
48 | only as an internal API that was available for extension writers, |
49 | and to implement fork() emulation on Win32 platforms. |
50 | |
51 | In Perl 5.8 the ithreads model became available through the C<threads> |
52 | module. |
53 | |
54 | Neither model is configured by default into Perl (except, as mentioned |
55 | above, in Win32 ithreads are always available.) |
56 | |
57 | For backwards compatibility, the Thread module has been reworked |
58 | to function as a frontend for both 5005threads and ithreads. |
59 | Note that the compatibility is not complete: because the data sharing |
60 | models are directly opposed, anything to do with data sharing has to |
61 | be thought differently. With the ithreads you must explicitly share() |
62 | variables between the threads. |
63 | |
64 | Finally, note that there are many known serious problems with the |
65 | 5005threads, one of the least of which is that regular expression |
66 | match variables like $1 are not threadsafe, that is, they easily get |
67 | corrupted by competing threads. Other problems include more insidious |
68 | data corruption and mysterious crashes. You are seriously urged to |
69 | use ithreads instead. |
70 | |
71 | =head1 SYNOPSIS |
72 | |
73 | use Thread; |
74 | |
75 | my $t = Thread->new(\&start_sub, @start_args); |
76 | |
77 | $result = $t->join; |
78 | $result = $t->eval; |
79 | $t->detach; |
80 | |
81 | if ($t->done) { |
82 | $t->join; |
83 | } |
84 | |
85 | if($t->equal($another_thread)) { |
86 | # ... |
87 | } |
88 | |
89 | yield(); |
90 | |
91 | my $tid = Thread->self->tid; |
92 | |
93 | lock($scalar); |
94 | lock(@array); |
95 | lock(%hash); |
96 | |
97 | lock(\&sub); # not available with ithreads |
98 | |
99 | $flags = $t->flags; # not available with ithreads |
100 | |
101 | my @list = Thread->list; # not available with ithreads |
102 | |
103 | unlock(...); # not available with the 5.005 threads |
104 | |
105 | use Thread 'async'; |
106 | |
107 | =head1 DESCRIPTION |
108 | |
109 | The C<Thread> module provides multithreading support for perl. |
110 | |
111 | =head1 FUNCTIONS |
112 | |
113 | =over 8 |
114 | |
115 | =item $thread = Thread->new(\&start_sub) |
116 | |
117 | =item $thread = Thread->new(\&start_sub, LIST) |
118 | |
119 | C<new> starts a new thread of execution in the referenced subroutine. The |
120 | optional list is passed as parameters to the subroutine. Execution |
121 | continues in both the subroutine and the code after the C<new> call. |
122 | |
123 | C<Thread->new> returns a thread object representing the newly created |
124 | thread. |
125 | |
126 | =item lock VARIABLE |
127 | |
128 | C<lock> places a lock on a variable until the lock goes out of scope |
129 | (with ithreads you can also explicitly unlock()). |
130 | |
131 | If the variable is locked by another thread, the C<lock> call will |
132 | block until it's available. C<lock> is recursive, so multiple calls |
133 | to C<lock> are safe--the variable will remain locked until the |
134 | outermost lock on the variable goes out of scope. |
135 | |
136 | Locks on variables only affect C<lock> calls--they do I<not> affect normal |
137 | access to a variable. (Locks on subs are different, and covered in a bit.) |
138 | If you really, I<really> want locks to block access, then go ahead and tie |
139 | them to something and manage this yourself. This is done on purpose. |
140 | While managing access to variables is a good thing, Perl doesn't force |
141 | you out of its living room... |
142 | |
143 | If a container object, such as a hash or array, is locked, all the |
144 | elements of that container are not locked. For example, if a thread |
145 | does a C<lock @a>, any other thread doing a C<lock($a[12])> won't |
146 | block. |
147 | |
148 | With 5005threads you may also C<lock> a sub, using C<lock &sub>. |
149 | Any calls to that sub from another thread will block until the lock |
150 | is released. This behaviour is not equivalent to declaring the sub |
151 | with the C<locked> attribute. The C<locked> attribute serializes |
152 | access to a subroutine, but allows different threads non-simultaneous |
153 | access. C<lock &sub>, on the other hand, will not allow I<any> other |
154 | thread access for the duration of the lock. |
155 | |
156 | Finally, C<lock> will traverse up references exactly I<one> level. |
157 | C<lock(\$a)> is equivalent to C<lock($a)>, while C<lock(\\$a)> is not. |
158 | |
159 | =item async BLOCK; |
160 | |
161 | C<async> creates a thread to execute the block immediately following |
162 | it. This block is treated as an anonymous sub, and so must have a |
163 | semi-colon after the closing brace. Like C<Thread->new>, C<async> |
164 | returns a thread object. |
165 | |
166 | =item Thread->self |
167 | |
168 | The C<Thread-E<gt>self> function returns a thread object that represents |
169 | the thread making the C<Thread-E<gt>self> call. |
170 | |
171 | =item cond_wait VARIABLE |
172 | |
173 | The C<cond_wait> function takes a B<locked> variable as |
174 | a parameter, unlocks the variable, and blocks until another thread |
175 | does a C<cond_signal> or C<cond_broadcast> for that same locked |
176 | variable. The variable that C<cond_wait> blocked on is relocked |
177 | after the C<cond_wait> is satisfied. If there are multiple threads |
178 | C<cond_wait>ing on the same variable, all but one will reblock waiting |
179 | to reaquire the lock on the variable. (So if you're only using |
180 | C<cond_wait> for synchronization, give up the lock as soon as |
181 | possible.) |
182 | |
183 | =item cond_signal VARIABLE |
184 | |
185 | The C<cond_signal> function takes a locked variable as a parameter and |
186 | unblocks one thread that's C<cond_wait>ing on that variable. If more than |
187 | one thread is blocked in a C<cond_wait> on that variable, only one (and |
188 | which one is indeterminate) will be unblocked. |
189 | |
190 | If there are no threads blocked in a C<cond_wait> on the variable, |
191 | the signal is discarded. |
192 | |
193 | =item cond_broadcast VARIABLE |
194 | |
195 | The C<cond_broadcast> function works similarly to C<cond_signal>. |
196 | C<cond_broadcast>, though, will unblock B<all> the threads that are |
197 | blocked in a C<cond_wait> on the locked variable, rather than only |
198 | one. |
199 | |
200 | =item yield |
201 | |
202 | The C<yield> function allows another thread to take control of the |
203 | CPU. The exact results are implementation-dependent. |
204 | |
205 | =back |
206 | |
207 | =head1 METHODS |
208 | |
209 | =over 8 |
210 | |
211 | =item join |
212 | |
213 | C<join> waits for a thread to end and returns any values the thread |
214 | exited with. C<join> will block until the thread has ended, though |
215 | it won't block if the thread has already terminated. |
216 | |
217 | If the thread being C<join>ed C<die>d, the error it died with will |
218 | be returned at this time. If you don't want the thread performing |
219 | the C<join> to die as well, you should either wrap the C<join> in |
220 | an C<eval> or use the C<eval> thread method instead of C<join>. |
221 | |
222 | =item eval |
223 | |
224 | The C<eval> method wraps an C<eval> around a C<join>, and so waits for |
225 | a thread to exit, passing along any values the thread might have returned. |
226 | Errors, of course, get placed into C<$@>. (Not available with ithreads.) |
227 | |
228 | =item detach |
229 | |
230 | C<detach> tells a thread that it is never going to be joined i.e. |
231 | that all traces of its existence can be removed once it stops running. |
232 | Errors in detached threads will not be visible anywhere - if you want |
233 | to catch them, you should use $SIG{__DIE__} or something like that. |
234 | |
235 | =item equal |
236 | |
237 | C<equal> tests whether two thread objects represent the same thread and |
238 | returns true if they do. |
239 | |
240 | =item tid |
241 | |
242 | The C<tid> method returns the tid of a thread. The tid is |
243 | a monotonically increasing integer assigned when a thread is |
244 | created. The main thread of a program will have a tid of zero, |
245 | while subsequent threads will have tids assigned starting with one. |
246 | |
247 | =item flags |
248 | |
249 | The C<flags> method returns the flags for the thread. This is the |
250 | integer value corresponding to the internal flags for the thread, |
251 | and the value may not be all that meaningful to you. |
252 | (Not available with ithreads.) |
253 | |
254 | =item done |
255 | |
256 | The C<done> method returns true if the thread you're checking has |
257 | finished, and false otherwise. (Not available with ithreads.) |
258 | |
259 | =back |
260 | |
261 | =head1 LIMITATIONS |
262 | |
263 | The sequence number used to assign tids is a simple integer, and no |
264 | checking is done to make sure the tid isn't currently in use. If a |
265 | program creates more than 2**32 - 1 threads in a single run, threads |
266 | may be assigned duplicate tids. This limitation may be lifted in |
267 | a future version of Perl. |
268 | |
269 | =head1 SEE ALSO |
270 | |
271 | L<threads::shared> (not available with 5005threads) |
272 | |
273 | L<attributes>, L<Thread::Queue>, L<Thread::Semaphore>, |
274 | L<Thread::Specific> (not available with ithreads) |
275 | |
276 | =cut |
277 | |
278 | # |
279 | # Methods |
280 | # |
281 | |
282 | # |
283 | # Exported functions |
284 | # |
285 | |
286 | sub async (&) { |
287 | return Thread->new($_[0]); |
288 | } |
289 | |
290 | sub eval { |
291 | return eval { shift->join; }; |
292 | } |
293 | |
294 | sub unimplemented { |
295 | print $_[0], " unimplemented with ", |
296 | $Config{useithreads} ? "ithreads" : "5005threads", "\n"; |
297 | |
298 | } |
299 | |
300 | sub unimplement { |
301 | for my $m (@_) { |
4038bebf |
302 | no strict 'refs'; |
43d3ddbe |
303 | *{"Thread::$m"} = sub { unimplemented $m }; |
304 | } |
305 | } |
306 | |
307 | BEGIN { |
308 | if ($ithreads) { |
309 | XSLoader::load 'threads'; |
310 | for my $m (qw(new join detach yield self tid equal)) { |
4038bebf |
311 | no strict 'refs'; |
43d3ddbe |
312 | *{"Thread::$m"} = \&{"threads::$m"}; |
313 | } |
314 | XSLoader::load 'threads::shared'; |
315 | for my $m (qw(cond_signal cond_broadcast cond_wait unlock share)) { |
4038bebf |
316 | no strict 'refs'; |
43d3ddbe |
317 | *{"Thread::$m"} = \&{"threads::shared::${m}_enabled"}; |
318 | } |
319 | unimplement(qw(list done eval flags)); |
320 | } elsif ($othreads) { |
321 | XSLoader::load 'Thread'; |
322 | unimplement(qw(unlock)); |
323 | } else { |
324 | require Carp; |
325 | Carp::croak("This Perl has neither ithreads not 5005threads"); |
326 | } |
327 | } |
328 | |
329 | 1; |