[Haskell-cafe] Can this be improved?

[Chris Kuklewicz]

Bruno Oliveira wrote:
> On Tue, 27 Dec 2005 16:39:34 +0000, Chris Kuklewicz wrote:
>>I was wondering if it this small snippet of code could be altered to
>>require fewer OPTIONS -fallow-... switches.
> 
> 
> Here is a partial solution using only -fglasgow-exts:

...deleted...

> The problem of this solution is that, because we are using show in the 
> definition of swim for (a -> t), the strings are beeing printed with extra 
> "\"". 
> 
> *MyShow> main
> Hello" ""World #"[17,18,19]"!"
> I also think [4,5,6]" and "7" are ""cool""."

The extra double quotes being what I am trying to avoid with MyShow

> 
>>instance MyShow t => MyShow (String -> t) where
>>    swim s x = swim (s ++ x)

This is interesting.  It does remove the need for
-fallow-undecidable-instances.  Thanks.


>>Could this be improved by TypeEq machinery?  Is there a Haskell 98 way
>>to make myShow work this way?
> 
> 
> The problem of the previous solution is that you do not want, in the general case,
> use the default instance of Show for Strings. An alternative is to just introduce 
> a newtype S (isomorphic to String) that allows you to handle Strings differently:
> 
... deleted...
> 
> By using the newtype S instead of Haskell String, you obtain an Haskell 98 
> solution. I am not sure if this is good enough for you but it seems a good compromise.
> 
> Hope it helps!
> 
> Cheers,
> 
> Bruno

I did find the newtype solution, and trying to avoid that is what
motivated such ugly switch flipping usage of GHC.  My intuition was that
this example of creating "something like a sub-type-class of Show" which
treats String differently ought to be possible.  And it is possible, and
I would hazard a guess that your solution with "MyShow (String->t)" is
safe in spite of the extra switches.

The sad thing is that the two -fallow switches infect client code that
uses the MyShow module.  So allowing those tricks for class MyShow must
turn on for classes in code that imports MyShow.  If GHC could -fallow
those flags only for certain typeclasses then it would be confinable.

In principle, could GHC ever be extended to confine -fallow this way?

-- 
Chris