[Haskell-cafe] Re: type synonym liberalization (was "class []" proposal)

[Arie Peterson]

Hello Brandon,


> This could be handled with existential wrappers, except that the
> wrapping is annoying, and probably interferes with optimizing when a
> concrete type is known. Instead, a few changes to type synonyms handle
> Bulat's cases.
>
> With the proper interpretation, type synonyms like
> type ABlockStream = BlockStream b => b
> type AMemoryStream = MemoryStream m => m
>
> support some of Bulat's signatures like
> copyStream :: ABlockStream -> ABlockStream -> Integer -> IO ()
> saveToFile :: AMemoryStream -> FilePath -> IO ()
>
> This requires two changes to the interpretation of type synonyms
>
> 1) Free variables in the definition of the type synonym are allowed, and
> become fresh (wobbly) variables when the synonym is applied
>
> 2) Class constraints in the definition of a type synonym float up to the
> closest binding of a constrained type.

I find those free variables a bit scary. It is not clear to me what it
means for a value to have type 'AnAuto'. I like to think about type
synonyms as only that, synonyms - so you can always substitute the
definition for an occurrence.

How does your proposal compare to introducing existential types proper? As in

  type ABlockStream = exists b. BlockStream b => b

. This is a known extension, supported partly by jhc and fully by ehc. It
seems to be exactly what you need for

> copyStream :: ABlockStream -> ABlockStream -> Integer -> IO ()


Greetings,

Arie

-- 

Mr. Pelican Shit may be Willy.

  ^
 /e\
 ---