Control.Concurrent.Async

This module provides a set of operations for running IO operations asynchronously and waiting for their results. It is a thin layer over the basic concurrency operations provided by Control.Concurrent. The main additional functionality it provides is the ability to wait for the return value of a thread, but the interface also provides some additional safety and robustness over using threads and MVar directly.

The basic type is Async a, which represents an asynchronous IO action that will return a value of type a, or die with an exception. An Async corresponds to a thread, and its ThreadId can be obtained with asyncThreadId, although that should rarely be necessary.

For example, to fetch two web pages at the same time, we could do this (assuming a suitable getURL function):

   do a1 <- async (getURL url1)
      a2 <- async (getURL url2)
      page1 <- wait a1
      page2 <- wait a2
      ...

where async starts the operation in a separate thread, and wait waits for and returns the result. If the operation throws an exception, then that exception is re-thrown by wait. This is one of the ways in which this library provides some additional safety: it is harder to accidentally forget about exceptions thrown in child threads.

A slight improvement over the previous example is this:

      withAsync (getURL url1) $ \a1 -> do
      withAsync (getURL url2) $ \a2 -> do
      page1 <- wait a1
      page2 <- wait a2
      ...

withAsync is like async, except that the Async is automatically killed (using uninterruptibleCancel) if the enclosing IO operation returns before it has completed. Consider the case when the first wait throws an exception; then the second Async will be automatically killed rather than being left to run in the background, possibly indefinitely. This is the second way that the library provides additional safety: using withAsync means we can avoid accidentally leaving threads running. Furthermore, withAsync allows a tree of threads to be built, such that children are automatically killed if their parents die for any reason.

The pattern of performing two IO actions concurrently and waiting for their results is packaged up in a combinator concurrently, so we can further shorten the above example to:

      (page1, page2) <- concurrently (getURL url1) (getURL url2)
      ...

The Functor instance can be used to change the result of an Async. For example:

ghci> a <- async (return 3)
ghci> wait a
3
ghci> wait (fmap (+1) a)
4

Asynchronous actions

data Async a

An asynchronous action spawned by async or withAsync. Asynchronous actions are executed in a separate thread, and operations are provided for waiting for asynchronous actions to complete and obtaining their results (see e.g. wait).

Instances

Functor Async

Methods

fmap :: (a -> b) -> Async a -> Async b

(<$) :: a -> Async b -> Async a

Eq (Async a)

Methods

(==) :: Async a -> Async a -> Bool

(/=) :: Async a -> Async a -> Bool

Ord (Async a)

Methods

compare :: Async a -> Async a -> Ordering

(<) :: Async a -> Async a -> Bool

(<=) :: Async a -> Async a -> Bool

(>) :: Async a -> Async a -> Bool

(>=) :: Async a -> Async a -> Bool

max :: Async a -> Async a -> Async a

min :: Async a -> Async a -> Async a

Spawning

async :: IO a -> IO (Async a)

Spawn an asynchronous action in a separate thread.

asyncBound :: IO a -> IO (Async a)

Like async but using forkOS internally.

asyncOn :: Int -> IO a -> IO (Async a)

Like async but using forkOn internally.

asyncWithUnmask :: ((forall b. IO b -> IO b) -> IO a) -> IO (Async a)

Like async but using forkIOWithUnmask internally. The child thread is passed a function that can be used to unmask asynchronous exceptions.

asyncOnWithUnmask :: Int -> ((forall b. IO b -> IO b) -> IO a) -> IO (Async a)

Like asyncOn but using forkOnWithUnmask internally. The child thread is passed a function that can be used to unmask asynchronous exceptions.

Spawning with automatic cancelation

withAsync :: IO a -> (Async a -> IO b) -> IO b

Spawn an asynchronous action in a separate thread, and pass its Async handle to the supplied function. When the function returns or throws an exception, uninterruptibleCancel is called on the Async.

withAsync action inner = bracket (async action) uninterruptibleCancel inner

This is a useful variant of async that ensures an Async is never left running unintentionally.

withAsyncBound :: IO a -> (Async a -> IO b) -> IO b

Like withAsync but uses forkOS internally.

withAsyncOn :: Int -> IO a -> (Async a -> IO b) -> IO b

Like withAsync but uses forkOn internally.

withAsyncWithUnmask :: ((forall c. IO c -> IO c) -> IO a) -> (Async a -> IO b) -> IO b

Like withAsync but uses forkIOWithUnmask internally. The child thread is passed a function that can be used to unmask asynchronous exceptions.

withAsyncOnWithUnmask :: Int -> ((forall c. IO c -> IO c) -> IO a) -> (Async a -> IO b) -> IO b

Like withAsyncOn but uses forkOnWithUnmask internally. The child thread is passed a function that can be used to unmask asynchronous exceptions

Querying Asyncs

wait :: Async a -> IO a

Wait for an asynchronous action to complete, and return its value. If the asynchronous action threw an exception, then the exception is re-thrown by wait.

wait = atomically . waitSTM

poll :: Async a -> IO (Maybe (Either SomeException a))

Check whether an Async has completed yet. If it has not completed yet, then the result is Nothing, otherwise the result is Just e where e is Left x if the Async raised an exception x, or Right a if it returned a value a.

poll = atomically . pollSTM

waitCatch :: Async a -> IO (Either SomeException a)

Wait for an asynchronous action to complete, and return either Left e if the action raised an exception e, or Right a if it returned a value a.

waitCatch = atomically . waitCatchSTM

cancel :: Async a -> IO ()

Cancel an asynchronous action by throwing the ThreadKilled exception to it, and waiting for the Async thread to quit. Has no effect if the Async has already completed.

cancel a = throwTo (asyncThreadId a) ThreadKilled <* waitCatch a

Note that cancel will not terminate until the thread the Async refers to has terminated. This means that cancel will block for as long said thread blocks when receiving an asynchronous exception.

For example, it could block if:

• It's executing a foreign call, and thus cannot receive the asynchronous exception;
• It's executing some cleanup handler after having received the exception, and the handler is blocking.

uninterruptibleCancel :: Async a -> IO ()

Cancel an asynchronous action

This is a variant of cancel, but it is not interruptible.

cancelWith :: Exception e => Async a -> e -> IO ()

Cancel an asynchronous action by throwing the supplied exception to it.

cancelWith a x = throwTo (asyncThreadId a) x

The notes about the synchronous nature of cancel also apply to cancelWith.

asyncThreadId :: Async a -> ThreadId

Returns the ThreadId of the thread running the given Async.

STM operations

waitSTM :: Async a -> STM a

A version of wait that can be used inside an STM transaction.

pollSTM :: Async a -> STM (Maybe (Either SomeException a))

A version of poll that can be used inside an STM transaction.

waitCatchSTM :: Async a -> STM (Either SomeException a)

A version of waitCatch that can be used inside an STM transaction.

Waiting for multiple Asyncs

waitAny :: [Async a] -> IO (Async a, a)

Wait for any of the supplied Asyncs to complete. If the first to complete throws an exception, then that exception is re-thrown by waitAny.

If multiple Asyncs complete or have completed, then the value returned corresponds to the first completed Async in the list.

waitAnyCatch :: [Async a] -> IO (Async a, Either SomeException a)

Wait for any of the supplied asynchronous operations to complete. The value returned is a pair of the Async that completed, and the result that would be returned by wait on that Async.

If multiple Asyncs complete or have completed, then the value returned corresponds to the first completed Async in the list.

waitAnyCancel :: [Async a] -> IO (Async a, a)

Like waitAny, but also cancels the other asynchronous operations as soon as one has completed.

waitAnyCatchCancel :: [Async a] -> IO (Async a, Either SomeException a)

Like waitAnyCatch, but also cancels the other asynchronous operations as soon as one has completed.

waitEither :: Async a -> Async b -> IO (Either a b)

Wait for the first of two Asyncs to finish. If the Async that finished first raised an exception, then the exception is re-thrown by waitEither.

waitEitherCatch :: Async a -> Async b -> IO (Either (Either SomeException a) (Either SomeException b))

Wait for the first of two Asyncs to finish.

waitEitherCancel :: Async a -> Async b -> IO (Either a b)

Like waitEither, but also cancels both Asyncs before returning.

waitEitherCatchCancel :: Async a -> Async b -> IO (Either (Either SomeException a) (Either SomeException b))

Like waitEitherCatch, but also cancels both Asyncs before returning.

waitEither_ :: Async a -> Async b -> IO ()

Like waitEither, but the result is ignored.

waitBoth :: Async a -> Async b -> IO (a, b)

Waits for both Asyncs to finish, but if either of them throws an exception before they have both finished, then the exception is re-thrown by waitBoth.

Waiting for multiple Asyncs in STM

waitAnySTM :: [Async a] -> STM (Async a, a)

A version of waitAny that can be used inside an STM transaction.

Since: 2.1.0

waitAnyCatchSTM :: [Async a] -> STM (Async a, Either SomeException a)

A version of waitAnyCatch that can be used inside an STM transaction.

Since: 2.1.0

waitEitherSTM :: Async a -> Async b -> STM (Either a b)

A version of waitEither that can be used inside an STM transaction.

Since: 2.1.0

waitEitherCatchSTM :: Async a -> Async b -> STM (Either (Either SomeException a) (Either SomeException b))

A version of waitEitherCatch that can be used inside an STM transaction.

Since: 2.1.0

waitEitherSTM_ :: Async a -> Async b -> STM ()

A version of waitEither_ that can be used inside an STM transaction.

Since: 2.1.0

waitBothSTM :: Async a -> Async b -> STM (a, b)

A version of waitBoth that can be used inside an STM transaction.

Since: 2.1.0

Linking

link :: Async a -> IO ()

Link the given Async to the current thread, such that if the Async raises an exception, that exception will be re-thrown in the current thread.

link2 :: Async a -> Async b -> IO ()

Link two Asyncs together, such that if either raises an exception, the same exception is re-thrown in the other Async.

Convenient utilities

race :: IO a -> IO b -> IO (Either a b)

Run two IO actions concurrently, and return the first to finish. The loser of the race is cancelled.

race left right =
  withAsync left $ \a ->
  withAsync right $ \b ->
  waitEither a b

race_ :: IO a -> IO b -> IO ()

Like race, but the result is ignored.

concurrently :: IO a -> IO b -> IO (a, b)

Run two IO actions concurrently, and return both results. If either action throws an exception at any time, then the other action is cancelled, and the exception is re-thrown by concurrently.

concurrently left right =
  withAsync left $ \a ->
  withAsync right $ \b ->
  waitBoth a b

concurrently_ :: IO a -> IO b -> IO ()

concurrently, but ignore the result values

Since: 2.1.1

mapConcurrently :: Traversable t => (a -> IO b) -> t a -> IO (t b)

maps an IO-performing function over any Traversable data type, performing all the IO actions concurrently, and returning the original data structure with the arguments replaced by the results.

For example, mapConcurrently works with lists:

pages <- mapConcurrently getURL ["url1", "url2", "url3"]

forConcurrently :: Traversable t => t a -> (a -> IO b) -> IO (t b)

forConcurrently is mapConcurrently with its arguments flipped

pages <- forConcurrently ["url1", "url2", "url3"] $ \url -> getURL url

Since: 2.1.0

mapConcurrently_ :: Foldable f => (a -> IO b) -> f a -> IO ()

mapConcurrently_ is mapConcurrently with the return value discarded, just like @mapM_

forConcurrently_ :: Foldable f => f a -> (a -> IO b) -> IO ()

forConcurrently_ is forConcurrently with the return value discarded, just like @forM_

replicateConcurrently :: Int -> IO a -> IO [a]

Perform the action in the given number of threads.

Since: 2.1.1

replicateConcurrently_ :: Int -> IO a -> IO ()

Same as replicateConcurrently, but ignore the results.

Since: 2.1.1

newtype Concurrently a

A value of type Concurrently a is an IO operation that can be composed with other Concurrently values, using the Applicative and Alternative instances.

Calling runConcurrently on a value of type Concurrently a will execute the IO operations it contains concurrently, before delivering the result of type a.

For example

(page1, page2, page3)
    <- runConcurrently $ (,,)
    <$> Concurrently (getURL "url1")
    <*> Concurrently (getURL "url2")
    <*> Concurrently (getURL "url3")

Constructors

Concurrently

Fields

• runConcurrently :: IO a

Instances

Functor Concurrently

Methods

fmap :: (a -> b) -> Concurrently a -> Concurrently b

(<$) :: a -> Concurrently b -> Concurrently a

Applicative Concurrently

Methods

pure :: a -> Concurrently a

(<*>) :: Concurrently (a -> b) -> Concurrently a -> Concurrently b

(*>) :: Concurrently a -> Concurrently b -> Concurrently b

(<*) :: Concurrently a -> Concurrently b -> Concurrently a

Alternative Concurrently

Methods

empty :: Concurrently a

(<|>) :: Concurrently a -> Concurrently a -> Concurrently a

some :: Concurrently a -> Concurrently [a]

many :: Concurrently a -> Concurrently [a]

Semigroup a => Semigroup (Concurrently a)

Only defined by async for base >= 4.9

Since: 2.1.0

Methods

(<>) :: Concurrently a -> Concurrently a -> Concurrently a

sconcat :: NonEmpty (Concurrently a) -> Concurrently a

stimes :: Integral b => b -> Concurrently a -> Concurrently a

(Semigroup a, Monoid a) => Monoid (Concurrently a)

Since: 2.1.0

Methods

mempty :: Concurrently a

mappend :: Concurrently a -> Concurrently a -> Concurrently a

mconcat :: [Concurrently a] -> Concurrently a