Move MonadIO to base

[wren ng thornton]

wren ng thornton wrote:
> Anders Kaseorg wrote:
>> On Tue, 13 Apr 2010, Anders Kaseorg wrote:
>>> The striking similarity between instances of MonadCatchIO suggests to 
>>> me that something deeper is going on.  Is there a cleaner abstraction 
>>> that captures this idea?
>>
>> Here a possible answer.  I haven’t entirely figured out what it 
>> “means” yet, but maybe someone who knows more category theory will be 
>> able to figure that out.  :-)
>>
>> class Monad m => MonadMorphIO m where
>>     morphIO :: (forall b. (m a -> IO b) -> IO b) -> m a
> 
> [...]
> Putting these together, your class means: if you can construct/eliminate 
> an (F m a) in the IO category, then you can construct an (m a) in the 
> Hask category. I.e., IO(m a) is a subset of (m a).

Which is more general than liftIO which only says (IO a) is a subset of 
(m a). That is, liftIO requires that the non-IO layers of m are 
constructed via `return`, whereas morphIO permits non-pointed use of IO. 
In other words, liftIO only allows lifting "effectful" values, whereas 
morphIO also allows lifting "structural" values.

-- 
Live well,
~wren