1 {-# OPTIONS_GHC -fno-warn-unused-imports #-}
2 -----------------------------------------------------------------------------
4 -- Module : Control.Concurrent
5 -- Copyright : (c) The University of Glasgow 2001
6 -- License : BSD-style (see the file libraries/base/LICENSE)
8 -- Maintainer : libraries@haskell.org
9 -- Stability : experimental
10 -- Portability : non-portable (concurrency)
12 -- A common interface to a collection of useful concurrency
15 -----------------------------------------------------------------------------
17 module Control.Concurrent (
18 -- * Concurrent Haskell
22 -- * Basic concurrency operations
25 #ifdef __GLASGOW_HASKELL__
30 #ifdef __GLASGOW_HASKELL__
45 #ifdef __GLASGOW_HASKELL__
47 threadDelay, -- :: Int -> IO ()
48 threadWaitRead, -- :: Int -> IO ()
49 threadWaitWrite, -- :: Int -> IO ()
52 -- * Communication abstractions
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,
60 -- * Merging of streams
62 mergeIO, -- :: [a] -> [a] -> IO [a]
63 nmergeIO, -- :: [[a]] -> IO [a]
67 #ifdef __GLASGOW_HASKELL__
70 rtsSupportsBoundThreads,
77 -- * GHC's implementation of concurrency
79 -- |This section describes features specific to GHC's
80 -- implementation of Concurrent Haskell.
82 -- ** Haskell threads and Operating System threads
86 -- ** Terminating the program
97 import Control.Exception.Base as Exception
99 #ifdef __GLASGOW_HASKELL__
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 )
108 import System.Posix.Types ( Fd )
109 import Foreign.StablePtr
110 import Foreign.C.Types ( CInt )
111 import Control.Monad ( when )
113 #ifdef mingw32_HOST_OS
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
135 The concurrency extension for Haskell is described in the paper
137 <http://www.haskell.org/ghc/docs/papers/concurrent-haskell.ps.gz>.
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.
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'.
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.
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
165 > main = forkIO (write 'a') >> write 'b'
166 > where write c = putChar c >> write c
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.
175 Different Haskell implementations have different characteristics with
176 regard to which operations block /all/ threads.
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.
183 Using Hugs, all I\/O operations and foreign calls will block all other
191 mergeIO :: [a] -> [a] -> IO [a]
192 nmergeIO :: [[a]] -> IO [a]
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.
199 -- Note: Hugs does not provide these functions, since they require
200 -- preemptive multitasking.
203 = newEmptyMVar >>= \ tail_node ->
204 newMVar tail_node >>= \ tail_list ->
205 newQSem max_buff_size >>= \ e ->
206 newMVar 2 >>= \ branches_running ->
210 forkIO (suckIO branches_running buff ls) >>
211 forkIO (suckIO branches_running buff rs) >>
212 takeMVar tail_node >>= \ val ->
217 = (MVar (MVar [a]), QSem)
219 suckIO :: MVar Int -> Buffer a -> [a] -> IO ()
221 suckIO branches_running buff@(tail_list,e) vs
223 [] -> takeMVar branches_running >>= \ val ->
225 takeMVar tail_list >>= \ node ->
227 putMVar tail_list node
229 putMVar branches_running (val-1)
232 takeMVar tail_list >>= \ node ->
233 newEmptyMVar >>= \ next_node ->
235 takeMVar next_node >>= \ y ->
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
246 newEmptyMVar >>= \ tail_node ->
247 newMVar tail_node >>= \ tail_list ->
248 newQSem max_buff_size >>= \ e ->
249 newMVar len >>= \ branches_running ->
253 mapIO (\ x -> forkIO (suckIO branches_running buff x)) lss >>
254 takeMVar tail_node >>= \ val ->
258 mapIO f xs = sequence (map f xs)
259 #endif /* __HUGS__ */
261 #ifdef __GLASGOW_HASKELL__
262 -- ---------------------------------------------------------------------------
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.
272 Other Haskell systems do not currently support multiple operating system threads.
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
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@
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.
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@.
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.
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.
312 -- | 'True' if bound threads are supported.
313 -- If @rtsSupportsBoundThreads@ is 'False', 'isCurrentThreadBound'
314 -- will always return 'False' and both 'forkOS' and 'runInBoundThread' will
316 foreign import ccall rtsSupportsBoundThreads :: Bool
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.
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").
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@.
338 forkOS :: IO () -> IO ThreadId
340 foreign export ccall forkOS_entry
341 :: StablePtr (IO ()) -> IO ()
343 foreign import ccall "forkOS_entry" forkOS_entry_reimported
344 :: StablePtr (IO ()) -> IO ()
346 forkOS_entry :: StablePtr (IO ()) -> IO ()
347 forkOS_entry stableAction = do
348 action <- deRefStablePtr stableAction
351 foreign import ccall forkOS_createThread
352 :: StablePtr (IO ()) -> IO CInt
354 failNonThreaded :: IO a
355 failNonThreaded = fail $ "RTS doesn't support multiple OS threads "
356 ++"(use ghc -threaded when linking)"
359 | rtsSupportsBoundThreads = do
361 b <- Exception.getMaskingState
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.
367 Unmasked -> unsafeUnmask action0
368 MaskedInterruptible -> action0
369 MaskedUninterruptible -> uninterruptibleMask_ action0
371 action_plus = Exception.catch action1 childHandler
373 entry <- newStablePtr (myThreadId >>= putMVar mv >> action_plus)
374 err <- forkOS_createThread entry
375 when (err /= 0) $ fail "Cannot create OS thread."
379 | otherwise = failNonThreaded
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
384 isCurrentThreadBound :: IO Bool
385 isCurrentThreadBound = IO $ \ s# ->
386 case isCurrentThreadBound# s# of
387 (# s2#, flg #) -> (# s2#, not (flg ==# 0#) #)
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.
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/.
399 runInBoundThread :: IO a -> IO a
401 runInBoundThread action
402 | rtsSupportsBoundThreads = do
403 bound <- isCurrentThreadBound
407 ref <- newIORef undefined
408 let action_plus = Exception.try action >>= writeIORef ref
409 bracket (newStablePtr action_plus)
411 (\cEntry -> forkOS_entry_reimported cEntry >> readIORef ref) >>=
413 | otherwise = failNonThreaded
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.
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@.
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.
430 runInUnboundThread :: IO a -> IO a
432 runInUnboundThread action = do
433 bound <- isCurrentThreadBound
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
444 unsafeResult :: Either SomeException a -> IO a
445 unsafeResult = either Exception.throwIO return
446 #endif /* __GLASGOW_HASKELL__ */
448 #ifdef __GLASGOW_HASKELL__
449 -- ---------------------------------------------------------------------------
450 -- threadWaitRead/threadWaitWrite
452 -- | Block the current thread until data is available to read on the
453 -- given file descriptor (GHC only).
454 threadWaitRead :: Fd -> IO ()
456 #ifdef mingw32_HOST_OS
457 -- we have no IO manager implementing threadWaitRead on Windows.
458 -- fdReady does the right thing, but we have to call it in a
459 -- separate thread, otherwise threadWaitRead won't be interruptible,
460 -- and this only works with -threaded.
461 | threaded = withThread (waitFd fd 0)
462 | otherwise = case fd of
463 0 -> do _ <- hWaitForInput stdin (-1)
465 -- hWaitForInput does work properly, but we can only
466 -- do this for stdin since we know its FD.
467 _ -> error "threadWaitRead requires -threaded on Windows, or use System.IO.hWaitForInput"
469 = GHC.Conc.threadWaitRead fd
472 -- | Block the current thread until data can be written to the
473 -- given file descriptor (GHC only).
474 threadWaitWrite :: Fd -> IO ()
476 #ifdef mingw32_HOST_OS
477 | threaded = withThread (waitFd fd 1)
478 | otherwise = error "threadWaitWrite requires -threaded on Windows"
480 = GHC.Conc.threadWaitWrite fd
483 #ifdef mingw32_HOST_OS
484 foreign import ccall unsafe "rtsSupportsBoundThreads" threaded :: Bool
486 withThread :: IO a -> IO a
489 _ <- mask_ $ forkIO $ try io >>= putMVar m
493 Left e -> throwIO (e :: IOException)
495 waitFd :: Fd -> CInt -> IO ()
497 throwErrnoIfMinus1_ "fdReady" $
498 fdReady (fromIntegral fd) write iNFINITE 0
501 iNFINITE = 0xFFFFFFFF -- urgh
503 foreign import ccall safe "fdReady"
504 fdReady :: CInt -> CInt -> CInt -> CInt -> IO CInt
507 -- ---------------------------------------------------------------------------
512 #osthreads# In GHC, threads created by 'forkIO' are lightweight threads, and
513 are managed entirely by the GHC runtime. Typically Haskell
514 threads are an order of magnitude or two more efficient (in
515 terms of both time and space) than operating system threads.
517 The downside of having lightweight threads is that only one can
518 run at a time, so if one thread blocks in a foreign call, for
519 example, the other threads cannot continue. The GHC runtime
520 works around this by making use of full OS threads where
521 necessary. When the program is built with the @-threaded@
522 option (to link against the multithreaded version of the
523 runtime), a thread making a @safe@ foreign call will not block
524 the other threads in the system; another OS thread will take
525 over running Haskell threads until the original call returns.
526 The runtime maintains a pool of these /worker/ threads so that
527 multiple Haskell threads can be involved in external calls
530 The "System.IO" library manages multiplexing in its own way. On
531 Windows systems it uses @safe@ foreign calls to ensure that
532 threads doing I\/O operations don't block the whole runtime,
533 whereas on Unix systems all the currently blocked I\/O requests
534 are managed by a single thread (the /IO manager thread/) using
537 The runtime will run a Haskell thread using any of the available
538 worker OS threads. If you need control over which particular OS
539 thread is used to run a given Haskell thread, perhaps because
540 you need to call a foreign library that uses OS-thread-local
541 state, then you need bound threads (see "Control.Concurrent#boundthreads").
543 If you don't use the @-threaded@ option, then the runtime does
544 not make use of multiple OS threads. Foreign calls will block
545 all other running Haskell threads until the call returns. The
546 "System.IO" library still does multiplexing, so there can be multiple
547 threads doing I\/O, and this is handled internally by the runtime using
553 In a standalone GHC program, only the main thread is
554 required to terminate in order for the process to terminate.
555 Thus all other forked threads will simply terminate at the same
556 time as the main thread (the terminology for this kind of
557 behaviour is \"daemonic threads\").
559 If you want the program to wait for child threads to
560 finish before exiting, you need to program this yourself. A
561 simple mechanism is to have each child thread write to an
562 'MVar' when it completes, and have the main
563 thread wait on all the 'MVar's before
566 > myForkIO :: IO () -> IO (MVar ())
568 > mvar <- newEmptyMVar
569 > forkIO (io `finally` putMVar mvar ())
572 Note that we use 'finally' from the
573 "Control.Exception" module to make sure that the
574 'MVar' is written to even if the thread dies or
575 is killed for some reason.
577 A better method is to keep a global list of all child
578 threads which we should wait for at the end of the program:
580 > children :: MVar [MVar ()]
581 > children = unsafePerformIO (newMVar [])
583 > waitForChildren :: IO ()
584 > waitForChildren = do
585 > cs <- takeMVar children
589 > putMVar children ms
593 > forkChild :: IO () -> IO ThreadId
595 > mvar <- newEmptyMVar
596 > childs <- takeMVar children
597 > putMVar children (mvar:childs)
598 > forkIO (io `finally` putMVar mvar ())
601 > later waitForChildren $
604 The main thread principle also applies to calls to Haskell from
605 outside, using @foreign export@. When the @foreign export@ed
606 function is invoked, it starts a new main thread, and it returns
607 when this main thread terminates. If the call causes new
608 threads to be forked, they may remain in the system after the
609 @foreign export@ed function has returned.
614 GHC implements pre-emptive multitasking: the execution of
615 threads are interleaved in a random fashion. More specifically,
616 a thread may be pre-empted whenever it allocates some memory,
617 which unfortunately means that tight loops which do no
618 allocation tend to lock out other threads (this only seems to
619 happen with pathological benchmark-style code, however).
621 The rescheduling timer runs on a 20ms granularity by
622 default, but this may be altered using the
623 @-i\<n\>@ RTS option. After a rescheduling
624 \"tick\" the running thread is pre-empted as soon as
628 @aaaa@ @bbbb@ example may not
629 work too well on GHC (see Scheduling, above), due
630 to the locking on a 'System.IO.Handle'. Only one thread
631 may hold the lock on a 'System.IO.Handle' at any one
632 time, so if a reschedule happens while a thread is holding the
633 lock, the other thread won't be able to run. The upshot is that
634 the switch from @aaaa@ to
635 @bbbbb@ happens infrequently. It can be
636 improved by lowering the reschedule tick period. We also have a
637 patch that causes a reschedule whenever a thread waiting on a
638 lock is woken up, but haven't found it to be useful for anything
639 other than this example :-)
641 #endif /* __GLASGOW_HASKELL__ */