Commit b3ef6457 authored by pcapriotti's avatar pcapriotti
Browse files

Improve definition of forever (#5205)

The previous implementation was:

    forever a = a >> forever a

which can create a space leak in some cases, even with optimizations.
The current implementation:

    forever a = let a' = a >> a' in a'

prevents repeated thunk allocations by creating a single thunk for the
final result, even without optimizations.
parent 2df687f7
......@@ -190,25 +190,10 @@ f >=> g = \x -> f x >>= g
-- | @'forever' act@ repeats the action infinitely.
forever :: (Monad m) => m a -> m b
{-# INLINABLE forever #-} -- See Note [Make forever INLINABLE]
forever a = a >> forever a
{- Note [Make forever INLINABLE]
If you say x = forever a
you'll get x = a >> a >> a >> a >> ... etc ...
and that can make a massive space leak (see Trac #5205)
In some monads, where (>>) is expensive, this might be the right
thing, but not in the IO monad. We want to specialise 'forever' for
the IO monad, so that eta expansion happens and there's no space leak.
To achieve this we must make forever INLINABLE, so that it'll get
specialised at call sites.
Still delicate, though, because it depends on optimisation. But there
really is a space/time tradeoff here, and only optimisation reveals
the "right" answer.
{-# INLINE forever #-}
forever a = let a' = a >> a' in a'
-- Use explicit sharing here, as it is prevents a space leak regardless of
-- optimizations.
-- | @'void' value@ discards or ignores the result of evaluation, such as the return value of an 'IO' action.
void :: Functor f => f a -> f ()
......@@ -27,7 +27,6 @@ module GHC.ST (
import GHC.Base
import GHC.Show
import Control.Monad( forever )
default ()
......@@ -82,9 +81,6 @@ instance Monad (ST s) where
data STret s a = STret (State# s) a
{-# SPECIALISE forever :: ST s a -> ST s b #-}
-- See Note [Make forever INLINABLE] in Control.Monad
-- liftST is useful when we want a lifted result from an ST computation. See
-- fixST below.
liftST :: ST s a -> State# s -> STret s a
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment