[Haskell-cafe] Typeclasses -- if does not satisfy

[adam vogt]

Hi Dmitry,

It's pretty straightforward to use OverlappingInstances:

class Runnable a b where
      run :: a -> b

instance (r ~ (s -> (m (a,s)))) => Runnable (StateT s m a) r where
      run m = runStateT m

instance (r ~ (s -> (a,s))) => Runnable (StateT s Identity a) r where
      run m = runState m

Involving a type family doesn't simplify things, since you still need
two class instances to choose between runStateT and runState.


With TH you could avoid overlapping instances by looking at the names
in (reify ''Monad) to replace the type variable `m' with a particular
instance of Monad. In other words, with TH you can generate many many
instances that look like:

instance (r ~ (s -> (IO (a,s)))) => Runnable (StateT s IO a) r where
      run m = runStateT m

instance (r ~ (s -> (ReaderT t (a,s)))) => Runnable (StateT s (ReaderT
t) a) r where
      run m = runStateT m

And then users that define their own instances of Monad would have run
the same TH for Runnable to work. Is that the unpleasant scenario you
mean by "even TH does not helps here"?

Regards,
Adam

On Sat, May 24, 2014 at 10:53 AM, Dmitry Bogatov <KAction at gnu.org> wrote:
> Hello! I am working on package to generalize
> common `runSomeThing` and `runSomethingT` functions.
>
> It is very easy to generalize runStateT and runReaderT
> like
>
>         class Runnable a where
>               type RunResult a
>               run :: a -> RunResult a
>         instance Runnable (StateT s m a) where
>               --- Why ScopedTypeVariables do not work here?
>               type RunResult (StateT s m a) = s -> m (a, s)
>               run = runStateT
>         --- ReaderT absolutely the same
>
> But problem is that `State s a` is in fact `StateT s Identity a` and
> I get type family conflicts.
>
> In fact, it boils down to following:
>
>    type family TestIdentity (m :: * -> *)
>    type instance TestIdentity Identity = Int
>    type instance TestIdentity m = () -- Anything but Identity.
>
> In general, with all this TypeClass magic, I have a lot of ways that
> something belong to some class, and no (aside OverlappingInstances)
> to tell that it does NOT belong. I belive, there is reason to, but I do
> not see it. Also, seems even TH does not helps here.
>
> PS. Please, keep me in To:
>
> --
> Best regards, Dmitry Bogatov <KAction at gnu.org>,
> Free Software supporter, esperantisto and netiquette guardian.
>         git://kaction.name/rc-files.git
>         GPG: 54B7F00D
> _______________________________________________
> Haskell-Cafe mailing list
> Haskell-Cafe at haskell.org
> http://www.haskell.org/mailman/listinfo/haskell-cafe