Well, I think his proposal was that for a constructor C with n slots, the RHS of the case alt would be expected to be an n-ary function where arg*i* for *i..n* would have type T*i* for the types of each slot in C. I think it adds up logically... but it is rather odd. On 10 April 2015 at 20:35, Frank Staals wrote:

Tom Ellis writes:

...

Has anyone ever considered permitting case branches to be eta reduced? For example, it is often nice to rewrite

foo x = bar baz x

as

foo = bar baz

Likewise, I have often wanted to rewrite

case m of Nothing -> n Just x -> quux x

as case m of Nothing -> n Just -> quux

Am I missing an obvious reason this wouldn't work?

Tom

I would think that is a bit weird since Nothing and Just have different types.

--

- Frank _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe

On Sat, Apr 11, 2015 at 01:22:22AM -0700, Erik de Castro Lopo wrote:

Tom Ellis wrote:

...

Likewise, I have often wanted to rewrite

case m of Nothing -> n Just x -> quux x

Why not emply the maybe function (b -> (a -> b) -> Maybe a -> b)

maybe n quux m

This is a fine idea which I often use on simple examples like this. My proposal has some additional benefits * Closer connection between the names of the constructors and the alternatives (I always think "is it `maybe n quux` or `maybe quux n`"?. In fact I made this mistake just today.) * No need to hand write such a function for each ADT you define (arguably they could be autogenerated, but that's a different story). * You could nest matches: data Foo = Foo (Maybe Int) Bool | Bar String case m of Foo (Just 1) -> f Foo Nothing -> g Bar "Hello" -> h