Exceptions vs. Errors (Re: Readline read_history and write_history addition)

[Simon Marlow]

Yitzchak Gale wrote:

> The real fix would be to provide a lower-level primitive
> for block and unblock. The concurrency paper "Asynchronous
> exceptions in Haskell" states (sec. 4.2) that the reason for the type
> 
> block :: IO a -> IO a
> 
> rather than the more obvious
> 
> block :: IO ()
> 
> is clumsiness and fragility. That may be so. But that type
> is too high-level for a primitive. In GHC, the implementation
> of block is:
> 
> block (IO io) = IO $ blockAsyncExceptions# io
> 
> It should really be:
> 
> block x = do
>   IO blockAsyncExceptions#
>   ret <- x
>   IO unblockAsyncExceptions#
>   return ret
> 
> in which case we could then supply implementations for
> other monads as well.

blockAsyncExceptions# has some tricks to restore tail-recursion in some 
cases (see the paper).  But apart from losing that optimisation, I can't 
think of any reasons why the above couldn't work - one thing you have to 
worry about is what happens when x raises an exeption, but I think that is 
handled by the way we save and restore the blocked state in catch.

Cheers,
	Simon