-
Notifications
You must be signed in to change notification settings - Fork 16
/
lib.rs
1760 lines (1563 loc) · 53.2 KB
/
lib.rs
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/*!
This crate provides an implementation of a multi-producer, multi-consumer
channel. Channels come in three varieties:
1. Asynchronous channels. Sends never block. Its buffer is only limited by the
available resources on the system.
2. Synchronous buffered channels. Sends block when the buffer is full. The
buffer is depleted by receiving on the channel.
3. Rendezvous channels (synchronous channels without a buffer). Sends block
until a receive has consumed the value sent. When a sender and receiver
synchronize, they are said to *rendezvous*.
Asynchronous channels are created with `chan::async()`. Synchronous channels
are created with `chan::sync(k)` where `k` is the buffer size. Rendezvous
channels are created with `chan::sync(0)`.
All channels are split into the same two types upon creation: a `Sender` and
a `Receiver`. Additional senders and receivers can be created with reckless
abandon by calling `clone`.
When all senders are dropped, the channel is closed and no other sends are
possible. In a channel with a buffer, receivers continue to consume values
until the buffer is empty, at which point, a `None` value is always returned
immediately.
No special semantics are enforced when all receivers are dropped. Asynchronous
sends will continue to work. Synchronous sends will block indefinitely when
the buffer is full. A send on a rendezvous channel will also block
indefinitely. (**NOTE**: This could be changed!)
All channels satisfy *both* `Send` and `Sync` and can be freely mixed in
`chan_select!`. Said differently, the synchronization semantics of a channel
are encoded upon construction, but are otherwise indistinguishable to the
type system.
Values sent on channels are subject to the normal restrictions Rust has on
values crossing thread boundaries. i.e., Values must implement `Send` and/or
`Sync`. (An `Rc<T>` *cannot* be sent on a channel, but a channel can be sent
on a channel!)
# Example: rendezvous channel
A simple example demonstrating a rendezvous channel:
```
use std::thread;
let (send, recv) = chan::sync(0);
thread::spawn(move || send.send(5));
assert_eq!(recv.recv(), Some(5)); // blocks until the previous send occurs
```
# Example: synchronous channel
Similarly, an example demonstrating a synchronous channel:
```
let (send, recv) = chan::sync(1);
send.send(5); // doesn't block because of the buffer
assert_eq!(recv.recv(), Some(5));
```
# Example: multiple producers and multiple consumers
An example demonstrating multiple consumers and multiple producers:
```
use std::thread;
let r = {
let (s, r) = chan::sync(0);
for letter in vec!['a', 'b', 'c', 'd'] {
let s = s.clone();
thread::spawn(move || {
for _ in 0..10 {
s.send(letter);
}
});
}
// This extra lexical scope will drop the initial
// sender we created. Thus, the channel will be
// closed when all threads spawned above has completed.
r
};
// A wait group lets us synchronize the completion of multiple threads.
let wg = chan::WaitGroup::new();
for _ in 0..4 {
wg.add(1);
let wg = wg.clone();
let r = r.clone();
thread::spawn(move || {
for letter in r {
println!("Received letter: {}", letter);
}
wg.done();
});
}
// If this was the end of the process and we didn't call `wg.wait()`, then
// the process might quit before all of the consumers were done.
// `wg.wait()` will block until all `wg.done()` calls have finished.
wg.wait();
```
# Example: Select on multiple channel sends/receives
An example showing how to use `chan_select!` to synchronize on sends
or receives.
```
#[macro_use]
extern crate chan;
use std::thread;
// Emits the fibonacci sequence on the given channel until `quit` receives
// a sentinel value.
fn fibonacci(s: chan::Sender<u64>, quit: chan::Receiver<()>) {
let (mut x, mut y) = (0, 1);
loop {
// Select will block until at least one of `s.send` or `quit.recv`
// is ready to succeed. At which point, it will choose exactly one
// send/receive to synchronize.
chan_select! {
s.send(x) => {
let oldx = x;
x = y;
y = oldx + y;
},
quit.recv() => {
println!("quit");
return;
}
}
}
}
fn main() {
let (s, r) = chan::sync(0);
let (qs, qr) = chan::sync(0);
// Spawn a thread and ask for the first 10 numbers in the fibonacci
// sequence.
thread::spawn(move || {
for _ in 0..10 {
println!("{}", r.recv().unwrap());
}
// Dropping all sending channels causes the receive channel to
// immediately and always synchronize (because the channel is closed).
drop(qs);
});
fibonacci(s, qr);
}
```
# Example: non-blocking sends/receives
This crate specifically does not expose methods like `try_send` or `try_recv`.
Instead, you should prefer using `chan_select!` to perform a non-blocking
send or receive. This can be done by telling select what to do when no
synchronization events are available.
```
# #[macro_use] extern crate chan; fn main() {
let (s, _) = chan::sync(0);
chan_select! {
default => println!("Send failed."),
s.send("some data") => println!("Send succeeded."),
}
# }
```
When `chan_select!` first runs, it will check if `s.send(...)` can succeed
*without blocking*. If so, `chan_select!` will permit the channels to
rendezvous. However, if there is no `recv` call to accept the send, then
`chan_select!` will immediately execute the `default` arm.
# Example: the sentinel channel idiom
When writing concurrent programs with `chan`, you will often find that you need
to somehow "wait" until some operation is done. For example, let's say you want
to run a function in a separate thread, but wait until it completes. Here's
one way to do it:
```rust
use std::thread;
fn do_work(done: chan::Sender<()>) {
// do something
// signal that we're done.
done.send(());
}
fn main() {
let (sdone, rdone) = chan::sync(0);
thread::spawn(move || do_work(sdone));
// block until work is done, and then quit the program.
rdone.recv();
}
```
In effect, we've created a new channel that sends unit values. When we're
done doing work, we send a unit value and `main` waits for it to be delivered.
Another way of achieving the same thing is to simply close the channel. Once
the channel is closed, any previously blocked receive operations become
immediately unblocked. What's even cooler is that channels are closed
automatically when all senders are dropped. So the new program looks something
like this:
```rust
use std::thread;
fn do_work(_done: chan::Sender<()>) {
// do something
}
fn main() {
let (sdone, rdone) = chan::sync(0);
thread::spawn(move || do_work(sdone));
// block until work is done, and then quit the program.
rdone.recv();
}
```
We no longer need to explicitly do anything with the `_done` channel. We give
`do_work` ownership of the channel, but as soon as the function stops
executing, `_done` is dropped, the channel is closed and `rdone.recv()`
unblocks.
# Example: I want more!
There are some examples in this crate's repository:
https://github.com/BurntSushi/chan/tree/master/examples
Here is a nice example using the `chan-signal` crate to read lines from
stdin while gracefully quitting after receiving a `INT` or `TERM`
signal:
https://github.com/BurntSushi/chan-signal/blob/master/examples/read_names.rs
A non-trivial program for periodically sending email with the output of
running a command: https://github.com/BurntSushi/rust-cmail (The source is
commented more heavily than normal.)
# When are channel operations non-blocking?
Non-blocking in this context means "a send/recv operation can synchronize
immediately." (Under the hood, a mutex may still be acquired, which could
block.)
The following is a list of all cases where a channel operation is considered
non-blocking:
* A send on a synchronous channel whose buffer is not full.
* A receive on a synchronous channel with a non-empty buffer.
* A send on an asynchronous channel.
* A rendezvous send or recv when a corresponding recv or send operation is
already blocked, respectively.
* A receive on any closed channel.
Non-blocking semantics are important because they affect the behavior of
`chan_select!`. In particular, a `chan_select!` with a `default` arm will
execute the `default` case if and only if all other operations are blocked.
# Which channel type should I use?
[From Ken Kahn](http://www.eros-os.org/pipermail/e-lang/2003-January/008183.html):
> About 25 years ago I went to dinner with Carl Hewitt and Robin Milner (of
> CSS and pi calculus fame) and they were arguing about synchronous vs.
> asynchronous communication primitives. Carl used the post office metaphor
> while Robin used the telephone. Both quickly admitted that one can implement
> one in the other.
With three channel types to choose from, it may not always be clear which one
you should use. In fact, there has been a long debate over which are better.
Here are some rough guidelines:
* Historically, asynchronous channels have been associated with the actor
model, which means they're a little out of place in a library inspired by
communicating sequential processes. Nevertheless, an unconstrained buffer can
be occasionally useful.
* Synchronous channels are useful because their stricter synchronization
semantics can make it easier to reason about the flow of your program. In
particular, with a rendezvous channel, one knows that a `send` unblocks only
when a corresponding `recv` consumes the sent value. This makes it *feel*
an awful lot like a function call!
# Warning: leaks
Channels can be leaked! In particular, if all receivers have been dropped,
then any future sends will block. Usually this is indicative of a bug in your
program.
For example, consider a "generator" style pattern where a thread produces
values on a channel and another thread consumes in an iterator.
```no_run
use std::thread;
let (s, r) = chan::sync(0);
thread::spawn(move || {
for val in r {
if val >= 2 {
break;
}
}
});
s.send(1);
s.send(2);
// This will deadlock because the loop in the thread
// above quits after receiving `2`.
s.send(3);
```
If the iterator loop quits early, the channel's buffer could fill up, which
will indefinitely block all future send operations.
(These leaks/deadlocks are detectable in most circumstances, and a `send`
operation could be made to wake up and either return an error or panic. The
semantics here are still experimental.)
# Warning: more leaks
It will always be possible to leak a channel in safe code regardless of the
channel's semantics. For example:
```no_run
use std::mem::forget;
let (s, r) = chan::sync::<()>(0);
forget(s);
// Blocks forever because the channel is never closed.
r.recv();
```
In this case, it is impossible for the channel to close because the internal
reference count will never reach `0`.
# Warning: performance
The primary purpose of this crate is to provide a safe, concurrent abstraction.
Notably, it is *not* a zero-cost abstraction. It is not even a near-zero-cost
abstraction. Throughput on a channel is startlingly low (see the benchmarks
in this crate's repository). Therefore, the channels provided in this crate
are most useful as a means to structure concurrent programs at a coarse level.
If your requirements call for performant synchronization of data, `chan` is not
the crate you're looking for.
# Prior art
The semantics encoded in the channels provided by this crate should mirror or
closely mirror the semantics provided by channels in Go. This includes
select statements! The major difference between concurrent programs written
with `chan` and concurrent programs written with Go is that Go programs can
benefit from being fast and loose with creating goroutines. In `chan`, each
"goroutine" is just an OS thread.
In terms of writing code:
1. Go programs will feature explicit closing of channels. In `chan`, channels
are closed **only** when all senders have been dropped.
2. Since there is no such thing as a "nil" channel in `chan`, the semantics Go
has for nil channels (both sends and receives block indefinitely) do not
exist in `chan`.
3. `chan` does not expose `len` or `cap` methods. (For no reason other than
to start with a totally minimal API. In particular, calling `len` or `cap`
on a channel is often The Wrong Thing. But not always. So this restriction
may be lifted in the future.)
4. In `chan`, all channels are either senders or receivers. There is no
"bidirectional" channel. This is manifest in how channel memory is managed:
channels are closed when all senders are dropped.
Of course, Go is not the origin of these ideas, but it has been the
strongest influence on the design of this library, and at least one of its
authors has done substantial research on the integration of CSP and programming
languages.
*/
#![deny(missing_docs)]
extern crate rand;
use std::collections::VecDeque;
use std::fmt;
use std::hash::{Hash, Hasher};
use std::ops::Drop;
use std::sync::{Arc, Condvar, Mutex, MutexGuard};
use std::sync::atomic::{ATOMIC_USIZE_INIT, AtomicUsize, Ordering};
use std::thread;
use std::time::Duration;
use notifier::Notifier;
pub use select::{Select, SelectRecvHandle, SelectSendHandle};
use tracker::Tracker;
pub use wait_group::WaitGroup;
// This enables us to (in practice) uniquely identify any particular channel.
// A better approach would be to use the pointer's address in memory, but it
// looks like `Arc` doesn't support that (yet?).
//
// Any other ideas? ---AG
//
// N.B. This is combined with ChannelId to distinguish between the sending
// and receiving halves of a channel.
static NEXT_CHANNEL_ID: AtomicUsize = ATOMIC_USIZE_INIT;
mod notifier;
mod select;
mod tracker;
mod wait_group;
/// Create a synchronous channel with a possibly empty buffer.
///
/// When the `size` is zero, the buffer is empty and the channel becomes a
/// rendezvous channel. A rendezvous channel blocks send operations until
/// a corresponding receive operation consumes the sent value.
///
/// When the `size` is non-zero, the send operations will only block when the
/// buffer is full. Send operations only unblock when a receive operation
/// removes an element from the buffer.
///
/// Values are guaranteed to be received in the same order that they are sent.
///
/// The send and receive values returned can be cloned arbitrarily (i.e.,
/// multi-producer/multi-consumer) and moved to other threads.
///
/// When all senders are dropped, the channel is closed automatically. No
/// more values may be sent on a closed channel. Once a channel is closed and
/// the buffer is empty, all receive operations return `None` immediately.
/// (If a channel is closed and there are still values in the buffer, then
/// receive operations will retrieve those first.)
///
/// When all receivers are dropped, no special action is taken. When the buffer
/// is full, all subsequent send operations will block indefinitely.
///
/// # Examples
///
/// An example of a rendezvous channel:
///
/// ```
/// use std::thread;
///
/// let (send, recv) = chan::sync(0);
/// thread::spawn(move || send.send(5));
/// assert_eq!(recv.recv(), Some(5)); // blocks until the previous send occurs
/// ```
///
/// An example of a synchronous buffered channel:
///
/// ```
/// let (send, recv) = chan::sync(1);
///
/// send.send(5); // doesn't block because of the buffer
/// assert_eq!(recv.recv(), Some(5));
///
/// drop(send); // closes the channel
/// assert_eq!(recv.recv(), None);
/// ```
pub fn sync<T>(size: usize) -> (Sender<T>, Receiver<T>) {
let send = Channel::new(size, false);
let recv = send.clone();
(send.into_sender(), recv.into_receiver())
}
/// Create an asynchronous channel with an unbounded buffer.
///
/// Since the buffer is unbounded, send operations always succeed immediately.
///
/// Receive operations succeed only when there is at least one value in the
/// buffer.
///
/// Values are guaranteed to be received in the same order that they are sent.
///
/// The send and receive values returned can be cloned arbitrarily (i.e.,
/// multi-producer/multi-consumer) and moved to other threads.
///
/// When all senders are dropped, the channel is closed automatically. No
/// more values may be sent on a closed channel. Once a channel is closed and
/// the buffer is empty, all receive operations return `None` immediately.
/// (If a channel is closed and there are still values in the buffer, then
/// receive operations will retrieve those first.)
///
/// When all receivers are dropped, no special action is taken. When the buffer
/// is full, all subsequent send operations will block indefinitely.
///
/// # Example
///
/// Asynchronous channels are nice when you just want to enqueue a bunch
/// of values up front:
///
/// ```
/// let (s, r) = chan::async();
///
/// for i in 0..10 {
/// s.send(i);
/// }
///
/// drop(s); // closing the channel lets the iterator stop
/// let numbers: Vec<i32> = r.iter().collect();
/// assert_eq!(numbers, (0..10).collect::<Vec<i32>>());
/// ```
///
/// (Others should help me come up with more compelling examples of
/// asynchronous channels.)
pub fn async<T>() -> (Sender<T>, Receiver<T>) {
let send = Channel::new(0, true);
let recv = send.clone();
(send.into_sender(), recv.into_receiver())
}
/// Creates a new rendezvous channel that is dropped after a timeout.
///
/// When the channel is dropped, any receive operation on the returned channel
/// will be unblocked.
///
/// # Example
///
/// ```
/// use std::time::Duration;
///
/// let wait = chan::after(Duration::from_millis(1000));
/// // Unblocks after 1 second.
/// wait.recv();
/// ```
pub fn after(duration: Duration) -> Receiver<()> {
let (send, recv) = sync(0);
thread::spawn(move || {
thread::sleep(duration);
drop(send);
});
recv
}
/// Creates a new rendezvous channel that is dropped after a timeout.
///
/// `duration` is specified in milliseconds.
///
/// When the channel is dropped, any receive operation on the returned channel
/// will be unblocked.
///
/// N.B. This will eventually be deprecated when we get a proper duration type.
///
/// # Example
///
/// ```
/// let wait = chan::after_ms(1000);
/// // Unblocks after 1 second.
/// wait.recv();
/// ```
pub fn after_ms(duration: u32) -> Receiver<()> {
#![allow(deprecated)]
let (send, recv) = sync(0);
thread::spawn(move || {
thread::sleep_ms(duration);
drop(send);
});
recv
}
/// Creates a new rendezvous channel that is "ticked" every duration.
///
/// When `duration` is `0`, no ticks are ever sent.
///
/// When `duration` is non-zero, then a new channel is created and sent at
/// every duration. When the sent channel is dropped, the timer is reset
/// and the process repeats after the duration.
///
/// This is especially convenient because it keeps the ticking in sync with
/// the code that uses it. Namely, the ticks won't "build up."
///
/// N.B. There is no way to reclaim the resources used by this function.
/// If you stop receiving on the channel returned, then the thread spawned by
/// `tick_ms` will block indefinitely.
///
/// # Examples
///
/// This is most useful when used in `chan_select!` because the received
/// sentinel channel gets dropped only after the correspond arm has
/// executed. At which point, the ticker is reset and waits to tick until
/// `duration` milliseconds lapses *after* the `chan_select!` arm is executed.
///
/// ```
/// # #[macro_use] extern crate chan; fn main() {
/// use std::thread;
/// use std::time::Duration;
///
/// let tick = chan::tick(Duration::from_millis(100));
/// let boom = chan::after(Duration::from_millis(500));
/// loop {
/// chan_select! {
/// default => {
/// println!(" .");
/// thread::sleep(Duration::from_millis(50));
/// },
/// tick.recv() => println!("tick."),
/// boom.recv() => { println!("BOOM!"); return; },
/// }
/// }
/// # }
/// ```
pub fn tick(duration: Duration) -> Receiver<Sender<()>> {
let (send, recv) = sync(0);
if duration.as_secs() == 0 && duration.subsec_nanos() == 0 {
// Leak the send channel so that it never gets closed and
// `recv` never synchronizes.
::std::mem::forget(send);
} else {
thread::spawn(move || {
loop {
thread::sleep(duration);
let (sdone, rdone) = sync(0);
send.send(sdone);
// Block until `sdone` gets closed by the caller.
rdone.recv();
}
});
}
recv
}
/// Creates a new rendezvous channel that is "ticked" every duration.
///
/// `duration` is specified in milliseconds.
///
/// When `duration` is `0`, no ticks are ever sent.
///
/// When `duration` is non-zero, then a new channel is created and sent at
/// every duration. When the sent channel is dropped, the timer is reset
/// and the process repeats after the duration.
///
/// This is especially convenient because it keeps the ticking in sync with
/// the code that uses it. Namely, the ticks won't "build up."
///
/// N.B. There is no way to reclaim the resources used by this function.
/// If you stop receiving on the channel returned, then the thread spawned by
/// `tick_ms` will block indefinitely.
///
/// # Examples
///
/// This is most useful when used in `chan_select!` because the received
/// sentinel channel gets dropped only after the correspond arm has
/// executed. At which point, the ticker is reset and waits to tick until
/// `duration` milliseconds lapses *after* the `chan_select!` arm is executed.
///
/// ```
/// # #[macro_use] extern crate chan; fn main() {
/// use std::thread;
/// use std::time::Duration;
///
/// let tick = chan::tick_ms(100);
/// let boom = chan::after_ms(500);
/// loop {
/// chan_select! {
/// default => {
/// println!(" .");
/// thread::sleep(Duration::from_millis(50));
/// },
/// tick.recv() => println!("tick."),
/// boom.recv() => { println!("BOOM!"); return; },
/// }
/// }
/// # }
/// ```
pub fn tick_ms(duration: u32) -> Receiver<Sender<()>> {
#![allow(deprecated)]
let (send, recv) = sync(0);
if duration == 0 {
// Leak the send channel so that it never gets closed and
// `recv` never synchronizes.
::std::mem::forget(send);
} else {
thread::spawn(move || {
loop {
thread::sleep_ms(duration);
let (sdone, rdone) = sync(0);
send.send(sdone);
// Block until `sdone` gets closed by the caller.
rdone.recv();
}
});
}
recv
}
/// A value that uniquely identifies one half of a channel.
///
/// For any `s: Sender<T>`, `s.id() == s.clone().id()`. Similarly for
/// any `r: Receiver<T>`.
#[doc(hidden)]
#[derive(Copy, Clone, Debug, Eq, Hash, Ord, PartialEq, PartialOrd)]
pub struct ChannelId(ChannelKey);
#[derive(Copy, Clone, Debug, Eq, Hash, Ord, PartialEq, PartialOrd)]
enum ChannelKey {
Sender(u64),
Receiver(u64),
}
impl ChannelId {
fn sender(id: u64) -> ChannelId {
ChannelId(ChannelKey::Sender(id))
}
fn receiver(id: u64) -> ChannelId {
ChannelId(ChannelKey::Receiver(id))
}
}
/// An iterator over values received in a channel.
pub struct Iter<T> {
chan: Receiver<T>,
}
impl<T> Iterator for Iter<T> {
type Item = T;
fn next(&mut self) -> Option<T> { self.chan.recv() }
}
impl<T> IntoIterator for Receiver<T> {
type Item = T;
type IntoIter = Iter<T>;
fn into_iter(self) -> Iter<T> { Iter { chan: self } }
}
impl<'a, T> IntoIterator for &'a Receiver<T> {
type Item = T;
type IntoIter = Iter<T>;
fn into_iter(self) -> Iter<T> { self.iter() }
}
/// The sending half of a channel.
///
/// Senders can be cloned any number of times and sent to other threads.
///
/// Senders also implement `Sync`, which means they can be shared among threads
/// without cloning if the channels can be proven to outlive the execution
/// of the threads.
///
/// When all sending halves of a channel are dropped, the channel is closed
/// automatically. When a channel is closed, no new values can be sent on the
/// channel. Also, all receive operations either return any values left in the
/// buffer or return immediately with `None`.
#[derive(Debug)]
pub struct Sender<T>(Channel<T>);
/// The receiving half of a channel.
///
/// Receivers can be cloned any number of times and sent to other threads.
///
/// Receivers also implement `Sync`, which means they can be shared among
/// threads without cloning if the channels can be proven to outlive the
/// execution of the threads.
///
/// When all receiving halves of a channel are dropped, no special action is
/// taken. If the buffer in the channel is full, all sends will block
/// indefinitely.
#[derive(Debug)]
pub struct Receiver<T>(Channel<T>);
/// All senders and receivers are just newtypes around a more base channel.
///
/// i.e., All senders and receivers have direct access to any underlying
/// buffer.
#[derive(Debug)]
struct Channel<T>(Arc<Inner<T>>);
#[derive(Clone, Copy, Debug, Eq, PartialEq)]
enum ChannelType {
Async,
Rendezvous,
Buffered,
}
struct Inner<T> {
/// An auto-incrementing id.
id: u64,
/// Manages subscriptions to channels (e.g., from a `chan_select!`).
notify: Notifier,
/// Tracks reference counts of senders and receivers.
track: Tracker,
/// A condition variable on the contents of `data`.
cond: Condvar,
/// The capacity of a synchronous buffer. This corresponds to the number
/// of elements allowed in the buffer before send operations block.
///
/// For asynchronous and rendezvous channels, this is always 0.
cap: usize,
/// The type of the channel.
ty: ChannelType,
/// Synchronized data in the channel. e.g., the queued values.
data: Mutex<Data<T>>,
}
#[derive(Debug)]
struct Data<T> {
/// Whether the channel is closed or not. Once set to `true` it can never
/// be changed.
closed: bool,
/// The number of senders waiting. (Currently only used in rendezvous
/// channels.)
waiting_send: usize,
/// The number of receivers waiting. (Currently only used in rendezvous
/// channels.)
waiting_recv: usize,
/// The actual data stored by the user.
user: UserData<T>,
}
#[derive(Debug)]
enum UserData<T> {
/// Used for rendezvous channels. We only need to ever store one value.
One(Option<T>),
/// A ring buffer for synchronous channels.
/// There's definitely a more efficient representation, but I don't think
/// we really care.
Ring { queue: Vec<Option<T>>, pos: usize, len: usize },
/// An unbounded queue for asynchronous channels.
Queue(VecDeque<T>),
}
// The SendOp and RecvOp types unify the return values of all channel
// operations. Their primary purpose is to permit the caller to retrieve the
// channel's lock after the channel operation is done without the lock ever
// being released. (This is critical functionality for `Select`.)
//
// N.B. The `WouldBlock` variants are only constructed if a non-blocking
// operation is used (i.e., try_send or try_recv).
struct SendOp<'a, T: 'a> {
lock: MutexGuard<'a, Data<T>>,
kind: SendOpKind<T>,
}
#[derive(Debug)]
enum SendOpKind<T> {
Ok,
Closed(T),
WouldBlock(T),
}
struct RecvOp<'a, T: 'a> {
lock: MutexGuard<'a, Data<T>>,
kind: RecvOpKind<T>,
}
#[derive(Debug)]
enum RecvOpKind<T> {
Ok(T),
Closed,
WouldBlock,
}
impl<T> Sender<T> {
/// Send a value on this channel.
///
/// If this is an asnychronous channel, `send` never blocks.
///
/// If this is a synchronous channel, `send` only blocks when the buffer
/// is full.
///
/// If this is a rendezvous channel, `send` blocks until a corresponding
/// `recv` retrieves `val`.
///
/// Values are guaranteed to be received in the same order that they
/// are sent.
///
/// This operation will never `panic!` but it can deadlock.
pub fn send(&self, val: T) {
self.send_op(self.inner().lock(), false, val).unwrap()
}
fn try_send(&self, val: T) -> Result<(), T> {
self.send_op(self.inner().lock(), true, val).into_result()
}
fn send_op<'a>(
&'a self,
data: MutexGuard<'a, Data<T>>,
try: bool,
val: T,
) -> SendOp<'a, T> {
match self.inner().ty {
ChannelType::Async => self.inner().async_send(data, val),
ChannelType::Rendezvous => {
self.inner().rendezvous_send(data, try, val)
}
ChannelType::Buffered => {
self.inner().buffered_send(data, try, val)
}
}
}
fn inner(&self) -> &Inner<T> {
&(self.0).0
}
fn id(&self) -> ChannelId {
ChannelId::sender(self.inner().id)
}
}
impl<T> Receiver<T> {
/// Receive a value on this channel.
///
/// If this is an asnychronous channel, `recv` only blocks when the
/// buffer is empty.
///
/// If this is a synchronous channel, `recv` only blocks when the buffer
/// is empty.
///
/// If this is a rendezvous channel, `recv` blocks until a corresponding
/// `send` sends a value.
///
/// For all channels, if the channel is closed and the buffer is empty,
/// then `recv` always and immediately returns `None`. (If the buffer is
/// non-empty on a closed channel, then values from the buffer are
/// returned.)
///
/// Values are guaranteed to be received in the same order that they
/// are sent.
///
/// This operation will never `panic!` but it can deadlock if the channel
/// is never closed.
pub fn recv(&self) -> Option<T> {
self.recv_op(self.inner().lock(), false).unwrap()
}
fn try_recv(&self) -> Result<Option<T>, ()> {
self.recv_op(self.inner().lock(), true).into_result()
}
fn recv_op<'a>(
&'a self,
data: MutexGuard<'a, Data<T>>,
try: bool,
) -> RecvOp<'a, T> {
self.inner().recv(data, try)
}
/// Return an iterator for receiving values on this channel.
///
/// This iterator yields values (blocking if necessary) until the channel
/// is closed.
pub fn iter(&self) -> Iter<T> { Iter { chan: self.clone() } }
fn inner(&self) -> &Inner<T> {
&(self.0).0
}
fn id(&self) -> ChannelId {
ChannelId::receiver(self.inner().id)
}
}
impl<T> Channel<T> {
fn new(size: usize, async: bool) -> Channel<T> {
let (user, ty) = if async {
(
UserData::Queue(VecDeque::with_capacity(1024)),
ChannelType::Async,
)
} else if size == 0 {
(UserData::One(None), ChannelType::Rendezvous)
} else {
let mut queue = Vec::with_capacity(size);
for _ in 0..size { queue.push(None); }
(
UserData::Ring { queue: queue, pos: 0, len: 0 },
ChannelType::Buffered,
)
};
Channel(Arc::new(Inner {
id: NEXT_CHANNEL_ID.fetch_add(1, Ordering::SeqCst) as u64,
notify: Notifier::new(),
track: Tracker::new(),
cond: Condvar::new(),
cap: size,
ty: ty,
data: Mutex::new(Data {
closed: false,
waiting_send: 0,
waiting_recv: 0,