[Haskell-cafe] Strange problem with inference

[Nikolay Amiantov]

Thank you all for responses! There are always more things to learn ^_^.
I'll also take a look at the paper to understand better what's going on.

On 09/11/2015 02:58 AM, Nikolay Amiantov wrote:
> Hi Cafe,
> 
> I've been playing around with lens and stumbled upon strange GHC
> behaviour. Consider this source (using lens package and GHC 7.10.2):
> 
> {-# LANGUAGE TypeFamilies #-}
> 
> import Control.Lens
> 
> class Test a where
>   type TestT a
>   myiso :: Iso' a (TestT a)
> 
> test1 :: Test a => a -> TestT a
> test1 = view myiso
> 
> test2 :: Test a => TestT a -> a
> test2 = view (from myiso)
> 
> GHC would emit this error:
> 
> /tmp/test.hs:13:9:
>     Could not deduce (Control.Monad.Reader.Class.MonadReader
>                         (TestT a) ((->) (TestT a)))
>       arising from a use of ‘view’
>     from the context (Test a)
>       bound by the type signature for test2 :: Test a => TestT a -> a
>       at /tmp/test.hs:12:10-31
>     In the expression: view (from myiso)
>     In an equation for ‘test2’: test2 = view (from myiso)
> Failed, modules loaded: none.
> 
> However, `MonadReader r ((->) r)` is defined for any and all `r`!
> Furthermore, `test1` has no problem with this and `view` there uses this
> instance too. The only difference that I see is the presence of a type
> family:
> 
> * `test1` needs `MonadReader a ((->) a)`
> * `test2` needs `MonadReader (TestT a) ((->) (TestT a))`
> 
> , but I don't understand how can this result in a different behavior.
> Notice that this likely may be reproduced somehow without lens -- I've
> spent some time trying to minify this example further but alas to no avail.
> 
> Thanks in advance!
> 

-- 
Nikolay.