d1b6875e66285ebcd132e2f0ac69d16949eac690
[packages/base.git] / Control / Concurrent.hs
1 {-# OPTIONS_GHC -fno-warn-unused-imports #-}
2 -----------------------------------------------------------------------------
3 -- |
4 -- Module : Control.Concurrent
5 -- Copyright : (c) The University of Glasgow 2001
6 -- License : BSD-style (see the file libraries/base/LICENSE)
7 --
8 -- Maintainer : libraries@haskell.org
9 -- Stability : experimental
10 -- Portability : non-portable (concurrency)
11 --
12 -- A common interface to a collection of useful concurrency
13 -- abstractions.
14 --
15 -----------------------------------------------------------------------------
16
17 module Control.Concurrent (
18 -- * Concurrent Haskell
19
20 -- $conc_intro
21
22 -- * Basic concurrency operations
23
24 ThreadId,
25 #ifdef __GLASGOW_HASKELL__
26 myThreadId,
27 #endif
28
29 forkIO,
30 #ifdef __GLASGOW_HASKELL__
31 forkIOUnmasked,
32 killThread,
33 throwTo,
34 #endif
35
36 -- * Scheduling
37
38 -- $conc_scheduling
39 yield, -- :: IO ()
40
41 -- ** Blocking
42
43 -- $blocking
44
45 #ifdef __GLASGOW_HASKELL__
46 -- ** Waiting
47 threadDelay, -- :: Int -> IO ()
48 threadWaitRead, -- :: Int -> IO ()
49 threadWaitWrite, -- :: Int -> IO ()
50 #endif
51
52 -- * Communication abstractions
53
54 module Control.Concurrent.MVar,
55 module Control.Concurrent.Chan,
56 module Control.Concurrent.QSem,
57 module Control.Concurrent.QSemN,
58 module Control.Concurrent.SampleVar,
59
60 -- * Merging of streams
61 #ifndef __HUGS__
62 mergeIO, -- :: [a] -> [a] -> IO [a]
63 nmergeIO, -- :: [[a]] -> IO [a]
64 #endif
65 -- $merge
66
67 #ifdef __GLASGOW_HASKELL__
68 -- * Bound Threads
69 -- $boundthreads
70 rtsSupportsBoundThreads,
71 forkOS,
72 isCurrentThreadBound,
73 runInBoundThread,
74 runInUnboundThread
75 #endif
76
77 -- * GHC's implementation of concurrency
78
79 -- |This section describes features specific to GHC's
80 -- implementation of Concurrent Haskell.
81
82 -- ** Haskell threads and Operating System threads
83
84 -- $osthreads
85
86 -- ** Terminating the program
87
88 -- $termination
89
90 -- ** Pre-emption
91
92 -- $preemption
93 ) where
94
95 import Prelude
96
97 import Control.Exception.Base as Exception
98
99 #ifdef __GLASGOW_HASKELL__
100 import GHC.Exception
101 import GHC.Conc ( ThreadId(..), myThreadId, killThread, yield,
102 threadDelay, forkIO, forkIOUnmasked, childHandler )
103 import qualified GHC.Conc
104 import GHC.IO ( IO(..), unsafeInterleaveIO, unsafeUnmask )
105 import GHC.IORef ( newIORef, readIORef, writeIORef )
106 import GHC.Base
107
108 import System.Posix.Types ( Fd )
109 import Foreign.StablePtr
110 import Foreign.C.Types ( CInt )
111 import Control.Monad ( when )
112
113 #ifdef mingw32_HOST_OS
114 import Foreign.C
115 import System.IO
116 #endif
117 #endif
118
119 #ifdef __HUGS__
120 import Hugs.ConcBase
121 #endif
122
123 import Control.Concurrent.MVar
124 import Control.Concurrent.Chan
125 import Control.Concurrent.QSem
126 import Control.Concurrent.QSemN
127 import Control.Concurrent.SampleVar
128
129 #ifdef __HUGS__
130 type ThreadId = ()
131 #endif
132
133 {- $conc_intro
134
135 The concurrency extension for Haskell is described in the paper
136 /Concurrent Haskell/
137 <http://www.haskell.org/ghc/docs/papers/concurrent-haskell.ps.gz>.
138
139 Concurrency is \"lightweight\", which means that both thread creation
140 and context switching overheads are extremely low. Scheduling of
141 Haskell threads is done internally in the Haskell runtime system, and
142 doesn't make use of any operating system-supplied thread packages.
143
144 However, if you want to interact with a foreign library that expects your
145 program to use the operating system-supplied thread package, you can do so
146 by using 'forkOS' instead of 'forkIO'.
147
148 Haskell threads can communicate via 'MVar's, a kind of synchronised
149 mutable variable (see "Control.Concurrent.MVar"). Several common
150 concurrency abstractions can be built from 'MVar's, and these are
151 provided by the "Control.Concurrent" library.
152 In GHC, threads may also communicate via exceptions.
153 -}
154
155 {- $conc_scheduling
156
157 Scheduling may be either pre-emptive or co-operative,
158 depending on the implementation of Concurrent Haskell (see below
159 for information related to specific compilers). In a co-operative
160 system, context switches only occur when you use one of the
161 primitives defined in this module. This means that programs such
162 as:
163
164
165 > main = forkIO (write 'a') >> write 'b'
166 > where write c = putChar c >> write c
167
168 will print either @aaaaaaaaaaaaaa...@ or @bbbbbbbbbbbb...@,
169 instead of some random interleaving of @a@s and @b@s. In
170 practice, cooperative multitasking is sufficient for writing
171 simple graphical user interfaces.
172 -}
173
174 {- $blocking
175 Different Haskell implementations have different characteristics with
176 regard to which operations block /all/ threads.
177
178 Using GHC without the @-threaded@ option, all foreign calls will block
179 all other Haskell threads in the system, although I\/O operations will
180 not. With the @-threaded@ option, only foreign calls with the @unsafe@
181 attribute will block all other threads.
182
183 Using Hugs, all I\/O operations and foreign calls will block all other
184 Haskell threads.
185 -}
186
187 #ifndef __HUGS__
188 max_buff_size :: Int
189 max_buff_size = 1
190
191 mergeIO :: [a] -> [a] -> IO [a]
192 nmergeIO :: [[a]] -> IO [a]
193
194 -- $merge
195 -- The 'mergeIO' and 'nmergeIO' functions fork one thread for each
196 -- input list that concurrently evaluates that list; the results are
197 -- merged into a single output list.
198 --
199 -- Note: Hugs does not provide these functions, since they require
200 -- preemptive multitasking.
201
202 mergeIO ls rs
203 = newEmptyMVar >>= \ tail_node ->
204 newMVar tail_node >>= \ tail_list ->
205 newQSem max_buff_size >>= \ e ->
206 newMVar 2 >>= \ branches_running ->
207 let
208 buff = (tail_list,e)
209 in
210 forkIO (suckIO branches_running buff ls) >>
211 forkIO (suckIO branches_running buff rs) >>
212 takeMVar tail_node >>= \ val ->
213 signalQSem e >>
214 return val
215
216 type Buffer a
217 = (MVar (MVar [a]), QSem)
218
219 suckIO :: MVar Int -> Buffer a -> [a] -> IO ()
220
221 suckIO branches_running buff@(tail_list,e) vs
222 = case vs of
223 [] -> takeMVar branches_running >>= \ val ->
224 if val == 1 then
225 takeMVar tail_list >>= \ node ->
226 putMVar node [] >>
227 putMVar tail_list node
228 else
229 putMVar branches_running (val-1)
230 (x:xs) ->
231 waitQSem e >>
232 takeMVar tail_list >>= \ node ->
233 newEmptyMVar >>= \ next_node ->
234 unsafeInterleaveIO (
235 takeMVar next_node >>= \ y ->
236 signalQSem e >>
237 return y) >>= \ next_node_val ->
238 putMVar node (x:next_node_val) >>
239 putMVar tail_list next_node >>
240 suckIO branches_running buff xs
241
242 nmergeIO lss
243 = let
244 len = length lss
245 in
246 newEmptyMVar >>= \ tail_node ->
247 newMVar tail_node >>= \ tail_list ->
248 newQSem max_buff_size >>= \ e ->
249 newMVar len >>= \ branches_running ->
250 let
251 buff = (tail_list,e)
252 in
253 mapIO (\ x -> forkIO (suckIO branches_running buff x)) lss >>
254 takeMVar tail_node >>= \ val ->
255 signalQSem e >>
256 return val
257 where
258 mapIO f xs = sequence (map f xs)
259 #endif /* __HUGS__ */
260
261 #ifdef __GLASGOW_HASKELL__
262 -- ---------------------------------------------------------------------------
263 -- Bound Threads
264
265 {- $boundthreads
266 #boundthreads#
267
268 Support for multiple operating system threads and bound threads as described
269 below is currently only available in the GHC runtime system if you use the
270 /-threaded/ option when linking.
271
272 Other Haskell systems do not currently support multiple operating system threads.
273
274 A bound thread is a haskell thread that is /bound/ to an operating system
275 thread. While the bound thread is still scheduled by the Haskell run-time
276 system, the operating system thread takes care of all the foreign calls made
277 by the bound thread.
278
279 To a foreign library, the bound thread will look exactly like an ordinary
280 operating system thread created using OS functions like @pthread_create@
281 or @CreateThread@.
282
283 Bound threads can be created using the 'forkOS' function below. All foreign
284 exported functions are run in a bound thread (bound to the OS thread that
285 called the function). Also, the @main@ action of every Haskell program is
286 run in a bound thread.
287
288 Why do we need this? Because if a foreign library is called from a thread
289 created using 'forkIO', it won't have access to any /thread-local state/ -
290 state variables that have specific values for each OS thread
291 (see POSIX's @pthread_key_create@ or Win32's @TlsAlloc@). Therefore, some
292 libraries (OpenGL, for example) will not work from a thread created using
293 'forkIO'. They work fine in threads created using 'forkOS' or when called
294 from @main@ or from a @foreign export@.
295
296 In terms of performance, 'forkOS' (aka bound) threads are much more
297 expensive than 'forkIO' (aka unbound) threads, because a 'forkOS'
298 thread is tied to a particular OS thread, whereas a 'forkIO' thread
299 can be run by any OS thread. Context-switching between a 'forkOS'
300 thread and a 'forkIO' thread is many times more expensive than between
301 two 'forkIO' threads.
302
303 Note in particular that the main program thread (the thread running
304 @Main.main@) is always a bound thread, so for good concurrency
305 performance you should ensure that the main thread is not doing
306 repeated communication with other threads in the system. Typically
307 this means forking subthreads to do the work using 'forkIO', and
308 waiting for the results in the main thread.
309
310 -}
311
312 -- | 'True' if bound threads are supported.
313 -- If @rtsSupportsBoundThreads@ is 'False', 'isCurrentThreadBound'
314 -- will always return 'False' and both 'forkOS' and 'runInBoundThread' will
315 -- fail.
316 foreign import ccall rtsSupportsBoundThreads :: Bool
317
318
319 {- |
320 Like 'forkIO', this sparks off a new thread to run the 'IO'
321 computation passed as the first argument, and returns the 'ThreadId'
322 of the newly created thread.
323
324 However, 'forkOS' creates a /bound/ thread, which is necessary if you
325 need to call foreign (non-Haskell) libraries that make use of
326 thread-local state, such as OpenGL (see "Control.Concurrent#boundthreads").
327
328 Using 'forkOS' instead of 'forkIO' makes no difference at all to the
329 scheduling behaviour of the Haskell runtime system. It is a common
330 misconception that you need to use 'forkOS' instead of 'forkIO' to
331 avoid blocking all the Haskell threads when making a foreign call;
332 this isn't the case. To allow foreign calls to be made without
333 blocking all the Haskell threads (with GHC), it is only necessary to
334 use the @-threaded@ option when linking your program, and to make sure
335 the foreign import is not marked @unsafe@.
336 -}
337
338 forkOS :: IO () -> IO ThreadId
339
340 foreign export ccall forkOS_entry
341 :: StablePtr (IO ()) -> IO ()
342
343 foreign import ccall "forkOS_entry" forkOS_entry_reimported
344 :: StablePtr (IO ()) -> IO ()
345
346 forkOS_entry :: StablePtr (IO ()) -> IO ()
347 forkOS_entry stableAction = do
348 action <- deRefStablePtr stableAction
349 action
350
351 foreign import ccall forkOS_createThread
352 :: StablePtr (IO ()) -> IO CInt
353
354 failNonThreaded :: IO a
355 failNonThreaded = fail $ "RTS doesn't support multiple OS threads "
356 ++"(use ghc -threaded when linking)"
357
358 forkOS action0
359 | rtsSupportsBoundThreads = do
360 mv <- newEmptyMVar
361 b <- Exception.getMaskingState
362 let
363 -- async exceptions are masked in the child if they are masked
364 -- in the parent, as for forkIO (see #1048). forkOS_createThread
365 -- creates a thread with exceptions masked by default.
366 action1 = case b of
367 Unmasked -> unsafeUnmask action0
368 MaskedInterruptible -> action0
369 MaskedUninterruptible -> uninterruptibleMask_ action0
370
371 action_plus = Exception.catch action1 childHandler
372
373 entry <- newStablePtr (myThreadId >>= putMVar mv >> action_plus)
374 err <- forkOS_createThread entry
375 when (err /= 0) $ fail "Cannot create OS thread."
376 tid <- takeMVar mv
377 freeStablePtr entry
378 return tid
379 | otherwise = failNonThreaded
380
381 -- | Returns 'True' if the calling thread is /bound/, that is, if it is
382 -- safe to use foreign libraries that rely on thread-local state from the
383 -- calling thread.
384 isCurrentThreadBound :: IO Bool
385 isCurrentThreadBound = IO $ \ s# ->
386 case isCurrentThreadBound# s# of
387 (# s2#, flg #) -> (# s2#, not (flg ==# 0#) #)
388
389
390 {- |
391 Run the 'IO' computation passed as the first argument. If the calling thread
392 is not /bound/, a bound thread is created temporarily. @runInBoundThread@
393 doesn't finish until the 'IO' computation finishes.
394
395 You can wrap a series of foreign function calls that rely on thread-local state
396 with @runInBoundThread@ so that you can use them without knowing whether the
397 current thread is /bound/.
398 -}
399 runInBoundThread :: IO a -> IO a
400
401 runInBoundThread action
402 | rtsSupportsBoundThreads = do
403 bound <- isCurrentThreadBound
404 if bound
405 then action
406 else do
407 ref <- newIORef undefined
408 let action_plus = Exception.try action >>= writeIORef ref
409 bracket (newStablePtr action_plus)
410 freeStablePtr
411 (\cEntry -> forkOS_entry_reimported cEntry >> readIORef ref) >>=
412 unsafeResult
413 | otherwise = failNonThreaded
414
415 {- |
416 Run the 'IO' computation passed as the first argument. If the calling thread
417 is /bound/, an unbound thread is created temporarily using 'forkIO'.
418 @runInBoundThread@ doesn't finish until the 'IO' computation finishes.
419
420 Use this function /only/ in the rare case that you have actually observed a
421 performance loss due to the use of bound threads. A program that
422 doesn't need it's main thread to be bound and makes /heavy/ use of concurrency
423 (e.g. a web server), might want to wrap it's @main@ action in
424 @runInUnboundThread@.
425
426 Note that exceptions which are thrown to the current thread are thrown in turn
427 to the thread that is executing the given computation. This ensures there's
428 always a way of killing the forked thread.
429 -}
430 runInUnboundThread :: IO a -> IO a
431
432 runInUnboundThread action = do
433 bound <- isCurrentThreadBound
434 if bound
435 then do
436 mv <- newEmptyMVar
437 mask $ \restore -> do
438 tid <- forkIO $ Exception.try (restore action) >>= putMVar mv
439 let wait = takeMVar mv `Exception.catch` \(e :: SomeException) ->
440 Exception.throwTo tid e >> wait
441 wait >>= unsafeResult
442 else action
443
444 unsafeResult :: Either SomeException a -> IO a
445 unsafeResult = either Exception.throwIO return
446 #endif /* __GLASGOW_HASKELL__ */
447
448 #ifdef __GLASGOW_HASKELL__
449 -- ---------------------------------------------------------------------------
450 -- threadWaitRead/threadWaitWrite
451
452 -- | Block the current thread until data is available to read on the
453 -- given file descriptor (GHC only).
454 --
455 -- This will throw an 'IOError' if the file descriptor was closed
456 -- while this thread was blocked. To safely close a file descriptor
457 -- that has been used with 'threadWaitRead', use
458 -- 'GHC.Conc.closeFdWith'.
459 threadWaitRead :: Fd -> IO ()
460 threadWaitRead fd
461 #ifdef mingw32_HOST_OS
462 -- we have no IO manager implementing threadWaitRead on Windows.
463 -- fdReady does the right thing, but we have to call it in a
464 -- separate thread, otherwise threadWaitRead won't be interruptible,
465 -- and this only works with -threaded.
466 | threaded = withThread (waitFd fd 0)
467 | otherwise = case fd of
468 0 -> do _ <- hWaitForInput stdin (-1)
469 return ()
470 -- hWaitForInput does work properly, but we can only
471 -- do this for stdin since we know its FD.
472 _ -> error "threadWaitRead requires -threaded on Windows, or use System.IO.hWaitForInput"
473 #else
474 = GHC.Conc.threadWaitRead fd
475 #endif
476
477 -- | Block the current thread until data can be written to the
478 -- given file descriptor (GHC only).
479 --
480 -- This will throw an 'IOError' if the file descriptor was closed
481 -- while this thread was blocked. To safely close a file descriptor
482 -- that has been used with 'threadWaitWrite', use
483 -- 'GHC.Conc.closeFdWith'.
484 threadWaitWrite :: Fd -> IO ()
485 threadWaitWrite fd
486 #ifdef mingw32_HOST_OS
487 | threaded = withThread (waitFd fd 1)
488 | otherwise = error "threadWaitWrite requires -threaded on Windows"
489 #else
490 = GHC.Conc.threadWaitWrite fd
491 #endif
492
493 #ifdef mingw32_HOST_OS
494 foreign import ccall unsafe "rtsSupportsBoundThreads" threaded :: Bool
495
496 withThread :: IO a -> IO a
497 withThread io = do
498 m <- newEmptyMVar
499 _ <- mask_ $ forkIO $ try io >>= putMVar m
500 x <- takeMVar m
501 case x of
502 Right a -> return a
503 Left e -> throwIO (e :: IOException)
504
505 waitFd :: Fd -> CInt -> IO ()
506 waitFd fd write = do
507 throwErrnoIfMinus1_ "fdReady" $
508 fdReady (fromIntegral fd) write iNFINITE 0
509
510 iNFINITE :: CInt
511 iNFINITE = 0xFFFFFFFF -- urgh
512
513 foreign import ccall safe "fdReady"
514 fdReady :: CInt -> CInt -> CInt -> CInt -> IO CInt
515 #endif
516
517 -- ---------------------------------------------------------------------------
518 -- More docs
519
520 {- $osthreads
521
522 #osthreads# In GHC, threads created by 'forkIO' are lightweight threads, and
523 are managed entirely by the GHC runtime. Typically Haskell
524 threads are an order of magnitude or two more efficient (in
525 terms of both time and space) than operating system threads.
526
527 The downside of having lightweight threads is that only one can
528 run at a time, so if one thread blocks in a foreign call, for
529 example, the other threads cannot continue. The GHC runtime
530 works around this by making use of full OS threads where
531 necessary. When the program is built with the @-threaded@
532 option (to link against the multithreaded version of the
533 runtime), a thread making a @safe@ foreign call will not block
534 the other threads in the system; another OS thread will take
535 over running Haskell threads until the original call returns.
536 The runtime maintains a pool of these /worker/ threads so that
537 multiple Haskell threads can be involved in external calls
538 simultaneously.
539
540 The "System.IO" library manages multiplexing in its own way. On
541 Windows systems it uses @safe@ foreign calls to ensure that
542 threads doing I\/O operations don't block the whole runtime,
543 whereas on Unix systems all the currently blocked I\/O requests
544 are managed by a single thread (the /IO manager thread/) using
545 @select@.
546
547 The runtime will run a Haskell thread using any of the available
548 worker OS threads. If you need control over which particular OS
549 thread is used to run a given Haskell thread, perhaps because
550 you need to call a foreign library that uses OS-thread-local
551 state, then you need bound threads (see "Control.Concurrent#boundthreads").
552
553 If you don't use the @-threaded@ option, then the runtime does
554 not make use of multiple OS threads. Foreign calls will block
555 all other running Haskell threads until the call returns. The
556 "System.IO" library still does multiplexing, so there can be multiple
557 threads doing I\/O, and this is handled internally by the runtime using
558 @select@.
559 -}
560
561 {- $termination
562
563 In a standalone GHC program, only the main thread is
564 required to terminate in order for the process to terminate.
565 Thus all other forked threads will simply terminate at the same
566 time as the main thread (the terminology for this kind of
567 behaviour is \"daemonic threads\").
568
569 If you want the program to wait for child threads to
570 finish before exiting, you need to program this yourself. A
571 simple mechanism is to have each child thread write to an
572 'MVar' when it completes, and have the main
573 thread wait on all the 'MVar's before
574 exiting:
575
576 > myForkIO :: IO () -> IO (MVar ())
577 > myForkIO io = do
578 > mvar <- newEmptyMVar
579 > forkIO (io `finally` putMVar mvar ())
580 > return mvar
581
582 Note that we use 'finally' from the
583 "Control.Exception" module to make sure that the
584 'MVar' is written to even if the thread dies or
585 is killed for some reason.
586
587 A better method is to keep a global list of all child
588 threads which we should wait for at the end of the program:
589
590 > children :: MVar [MVar ()]
591 > children = unsafePerformIO (newMVar [])
592 >
593 > waitForChildren :: IO ()
594 > waitForChildren = do
595 > cs <- takeMVar children
596 > case cs of
597 > [] -> return ()
598 > m:ms -> do
599 > putMVar children ms
600 > takeMVar m
601 > waitForChildren
602 >
603 > forkChild :: IO () -> IO ThreadId
604 > forkChild io = do
605 > mvar <- newEmptyMVar
606 > childs <- takeMVar children
607 > putMVar children (mvar:childs)
608 > forkIO (io `finally` putMVar mvar ())
609 >
610 > main =
611 > later waitForChildren $
612 > ...
613
614 The main thread principle also applies to calls to Haskell from
615 outside, using @foreign export@. When the @foreign export@ed
616 function is invoked, it starts a new main thread, and it returns
617 when this main thread terminates. If the call causes new
618 threads to be forked, they may remain in the system after the
619 @foreign export@ed function has returned.
620 -}
621
622 {- $preemption
623
624 GHC implements pre-emptive multitasking: the execution of
625 threads are interleaved in a random fashion. More specifically,
626 a thread may be pre-empted whenever it allocates some memory,
627 which unfortunately means that tight loops which do no
628 allocation tend to lock out other threads (this only seems to
629 happen with pathological benchmark-style code, however).
630
631 The rescheduling timer runs on a 20ms granularity by
632 default, but this may be altered using the
633 @-i\<n\>@ RTS option. After a rescheduling
634 \"tick\" the running thread is pre-empted as soon as
635 possible.
636
637 One final note: the
638 @aaaa@ @bbbb@ example may not
639 work too well on GHC (see Scheduling, above), due
640 to the locking on a 'System.IO.Handle'. Only one thread
641 may hold the lock on a 'System.IO.Handle' at any one
642 time, so if a reschedule happens while a thread is holding the
643 lock, the other thread won't be able to run. The upshot is that
644 the switch from @aaaa@ to
645 @bbbbb@ happens infrequently. It can be
646 improved by lowering the reschedule tick period. We also have a
647 patch that causes a reschedule whenever a thread waiting on a
648 lock is woken up, but haven't found it to be useful for anything
649 other than this example :-)
650 -}
651 #endif /* __GLASGOW_HASKELL__ */