[Haskell-cafe] (no subject)

[Tom Ellis]

On Sun, Apr 24, 2016 at 08:07:40PM +0000, haskell-cafe-bounces at haskell.org wrote:
> >From int-e at gmx.de Sun Apr 24 21:13:18 2016
> Date: Sun, 24 Apr 2016 21:13:18 +0200
> From: Bertram Felgenhauer <bertram.felgenhauer at googlemail.com>
> To: haskell-cafe at haskell.org
> Subject: Re: [Haskell-cafe] Is it possible to make lazy combinators for IO?
> Message-ID: <20160424191319.GA32509 at 24f89f8c-e6a1-4e75-85ee-bb8a3743bb9f>
> References: <CAMgWh9t4MbzbbdvnU8zz6Lm-p9Gtf2UO-+GaW=ofPsNLmt4aCA at mail.gmail.com>
>         <CAErCyQ-J+iYo2tXVs9S4VtGtuurFaNp=zsrY_58=7EeL7q0HVQ at mail.gmail.com>
>         <CAMgWh9s3+ksQOh52oU=azKcdLDyuFy2sYKPbz30xaXnTkLK9cw at mail.gmail.com>
>         <CAErCyQ8OVMF0fMDuxLpuvHimCt7RBijjzp1wEBMxRFyyyCXZVg at mail.gmail.com>
>         <CAMgWh9vsxijoCuNrMxeAZmVxML+A8iNUwpTy9Es1BF0z_+pryQ at mail.gmail.com>
>         <CAMgWh9tV_aq2=NkKZ79QZX6Xn6V953RwWoFqTeUMDggF0=ZZ+g at mail.gmail.com>
>         <CAErCyQ_zUqn53x1=g3RKDjQc3+GXfXkj1VvaTxo8ONPeTfym0A at mail.gmail.com>
>         <CAMgWh9s=0bAsjGEgGwSr8bPL6vDqNL8yyY=G+qANoz35G0NKuw at mail.gmail.com>
> MIME-Version: 1.0
> Content-Type: text/plain; charset=us-ascii
> Content-Disposition: inline
> In-Reply-To: <CAMgWh9s=0bAsjGEgGwSr8bPL6vDqNL8yyY=G+qANoz35G0NKuw at mail.gmail.com>
> User-Agent: Mutt/1.5.24 (2015-08-30)
> Status: RO
> Content-Length: 1283
> Lines: 37
> 
> David Feuer wrote:
> > What I'm looking for is more limited than lazy IO or unsafeInterleaveIO,
> > but it seems quite possible that there's no way to get just what I'm
> > looking for with the IO type proper using GHC's implementation of IO. Lazy
> > IO allows evaluation to drive action. When a thunk is forced, it may
> > trigger I/O (spooky action at a distance). What I'm talking about is
> > separating what actions are performed from what values are calculated from
> > them. Here's a partial concept which won't actually compile because of the
> > lazy pattern matches:
> >
> > data MyIO a = forall b . MyIO (b -> a) (IO b)
> > instance Functor MyIO where
> >    fmap f ~(MyIO t m) = MyIO (f . t) m
> > instance Applicative MyIO where
> >    pure a = MyIO (const a) (pure ())
> >    MyIO t1 m1 <*> ~(MyIO t2 m2) =
> >      MyIO (\(r1, r2) -> t1 r1 (t2 r2)) ((,) <$> m1 <*> m2)
> > instance Monad MyIO where
> >   ???
> > instance MonadFix MyIO where
> >   ???
> 
> I believe that using this interface `unsafeInterleaveIO` could be
> implemented as follows, making it just as powerful as lazy IO:
> 
>     data Box a = Box a
> 
>     unsafeInterleaveMyIO :: MyIO a -> MyIO a
>     unsafeInterleaveMyIO act = do
>         act' <- Box `fmap` act
>         return $ case act' of Box !r -> r
> 
> Have I missed anything?

Since MyIO and its associate functions don't contain any unsafe primitives
it seems highly unlikely you can implement unsafeInterleaveIO with them!

In fact I can't see how MyIO is any different to IO.  All you can do with
the function field is fmap it over the IO action field.

Tom