Copyright | (c) The University of Glasgow 2001 |
---|---|
License | BSD-style (see the file libraries/base/LICENSE) |
Maintainer | libraries@haskell.org |
Stability | stable |
Portability | portable |
Safe Haskell | Safe |
Language | Haskell2010 |
Mutable references in the IO monad.
Synopsis
- data IORef a
- newIORef :: a -> IO (IORef a)
- readIORef :: IORef a -> IO a
- writeIORef :: IORef a -> a -> IO ()
- modifyIORef :: IORef a -> (a -> a) -> IO ()
- modifyIORef' :: IORef a -> (a -> a) -> IO ()
- atomicModifyIORef :: IORef a -> (a -> (a, b)) -> IO b
- atomicModifyIORef' :: IORef a -> (a -> (a, b)) -> IO b
- atomicWriteIORef :: IORef a -> a -> IO ()
- mkWeakIORef :: IORef a -> IO () -> IO (Weak (IORef a))
IORefs
A mutable variable in the IO
monad.
>>>
import GHC.Internal.Data.IORef
>>>
r <- newIORef 0
>>>
readIORef r
0>>>
writeIORef r 1
>>>
readIORef r
1>>>
atomicWriteIORef r 2
>>>
readIORef r
2>>>
modifyIORef' r (+ 1)
>>>
readIORef r
3>>>
atomicModifyIORef' r (\a -> (a + 1, ()))
>>>
readIORef r
4
readIORef :: IORef a -> IO a Source #
Read the value of an IORef
.
Beware that the CPU executing a thread can reorder reads or writes to independent locations. See Data.IORef for more details.
writeIORef :: IORef a -> a -> IO () Source #
Write a new value into an IORef
.
This function does not create a memory barrier and can be reordered
with other independent reads and writes within a thread, which may cause issues
for multithreaded execution. In these cases, consider using atomicWriteIORef
instead. See Data.IORef for more details.
modifyIORef :: IORef a -> (a -> a) -> IO () Source #
Mutate the contents of an IORef
, combining readIORef
and writeIORef
.
This is not an atomic update, consider using atomicModifyIORef
when
operating in a multithreaded environment.
Be warned that modifyIORef
does not apply the function strictly. This
means if the program calls modifyIORef
many times, but seldom uses the
value, thunks will pile up in memory resulting in a space leak. This is a
common mistake made when using an IORef as a counter. For example, the
following will likely produce a stack overflow:
ref <- newIORef 0 replicateM_ 1000000 $ modifyIORef ref (+1) readIORef ref >>= print
To avoid this problem, use modifyIORef'
instead.
modifyIORef' :: IORef a -> (a -> a) -> IO () Source #
Strict version of modifyIORef
.
This is not an atomic update, consider using atomicModifyIORef'
when
operating in a multithreaded environment.
Since: base-4.6.0.0
atomicModifyIORef :: IORef a -> (a -> (a, b)) -> IO b Source #
Atomically modifies the contents of an IORef
.
This function is useful for using IORef
in a safe way in a multithreaded
program. If you only have one IORef
, then using atomicModifyIORef
to
access and modify it will prevent race conditions.
Extending the atomicity to multiple IORef
s is problematic, so it
is recommended that if you need to do anything more complicated
then using MVar
instead is a good idea.
Conceptually,
atomicModifyIORef ref f = do -- Begin atomic block old <-readIORef
ref let r = f old new = fst rwriteIORef
ref new -- End atomic block case r of (_new, res) -> pure res
The actions in the section labeled "atomic block" are not subject to
interference from other threads. In particular, it is impossible for the
value in the IORef
to change between the readIORef
and writeIORef
invocations.
The user-supplied function is applied to the value stored in the IORef
,
yielding a new value to store in the IORef
and a value to return. After
the new value is (lazily) stored in the IORef
, atomicModifyIORef
forces
the result pair, but does not force either component of the result. To force
both components, use atomicModifyIORef'
.
Note that
atomicModifyIORef ref (_ -> undefined)
will raise an exception in the calling thread, but will also
install the bottoming value in the IORef
, where it may be read by
other threads.
This function imposes a memory barrier, preventing reordering around the "atomic block"; see Data.IORef for details.
atomicModifyIORef' :: IORef a -> (a -> (a, b)) -> IO b Source #
A strict version of atomicModifyIORef
. This forces both the
value stored in the IORef
and the value returned.
Conceptually,
atomicModifyIORef' ref f = do -- Begin atomic block old <-readIORef
ref let r = f old new = fst rwriteIORef
ref new -- End atomic block case r of (!_new, !res) -> pure res
The actions in the "atomic block" are not subject to interference
by other threads. In particular, the value in the IORef
cannot
change between the readIORef
and writeIORef
invocations.
The new value is installed in the IORef
before either value is forced.
So
atomicModifyIORef' ref (x -> (x+1, undefined))
will increment the IORef
and then throw an exception in the calling
thread.
atomicModifyIORef' ref (x -> (undefined, x))
and
atomicModifyIORef' ref (_ -> undefined)
will each raise an exception in the calling thread, but will also
install the bottoming value in the IORef
, where it may be read by
other threads.
This function imposes a memory barrier, preventing reordering around the "atomic block"; see Data.IORef for details.
Since: base-4.6.0.0
atomicWriteIORef :: IORef a -> a -> IO () Source #
Variant of writeIORef
. The prefix "atomic" relates to a fact that
it imposes a reordering barrier, similar to atomicModifyIORef
.
Such a write will not be reordered with other reads
or writes even on CPUs with weak memory model.
Since: base-4.6.0.0
Memory Model
Most modern CPU achitectures (e.g. x86/64, ARM) have a memory model which allows threads to reorder reads with earlier writes to different locations, e.g. see the x86/64 architecture manual, 8.2.3.4 Loads May Be Reordered with Earlier Stores to Different Locations.
Because of that, in a concurrent program, IORef
operations may appear out-of-order
to another thread. In the following example:
import GHC.Internal.Data.IORef import GHC.Internal.Control.Monad (unless) import Control.Concurrent (forkIO, threadDelay) maybePrint :: IORef Bool -> IORef Bool -> IO () maybePrint myRef yourRef = do writeIORef myRef True yourVal <- readIORef yourRef unless yourVal $ putStrLn "critical section" main :: IO () main = do r1 <- newIORef False r2 <- newIORef False forkIO $ maybePrint r1 r2 forkIO $ maybePrint r2 r1 threadDelay 1000000
it is possible that the string "critical section"
is printed
twice, even though there is no interleaving of the operations of the
two threads that allows that outcome. The memory model of x86/64
allows readIORef
to happen before the earlier writeIORef
.
The ARM memory order model is typically even weaker than x86/64, allowing any reordering of reads and writes as long as they are independent from the point of view of the current thread.
The implementation is required to ensure that reordering of memory
operations cannot cause type-correct code to go wrong. In
particular, when inspecting the value read from an IORef
, the
memory writes that created that value must have occurred from the
point of view of the current thread.
atomicWriteIORef
, atomicModifyIORef
and atomicModifyIORef'
act
as a barrier to reordering. Multiple calls to these functions
occur in strict program order, never taking place ahead of any
earlier (in program order) IORef
operations, or after any later
IORef
operations.