[feature-proposal] Natural transformation type class with covariance support

[Alois Cochard]

Hi hackers without shape,

Here is a proposal for a variant of the natural transformation typeclass that support covariance,

In short it make possible to support 'Some' instead of only 'Option' when doing type 'transformation', here is the implementation:

https://github.com/aloiscochard/shapeless/commit/6e6924191a3bd502f12a1397776e90df0205d90e

Of course this apply to any type constructor that have sub-type.

It's quite usefull for me when designing DSL, let me know what you think of that.

Cheers,

Alois Cochard

http://twitter.com/aloiscochard

http://aloiscochard.blogspot.com

http://aloiscochard.github.com

[Alois Cochard]

Sorry I posted a link to wrong commit!

Here is the good one -> https://github.com/aloiscochard/shapeless/commit/96ed0c38e29f4eb27a05c84949666132ab87b7e6

[Miles Sabin]

On Mon, Jul 2, 2012 at 9:02 PM, Alois Cochard <alois....@gmail.com> wrote:
> Sorry I posted a link to wrong commit!
>
> Here is the good one ->
> https://github.com/aloiscochard/shapeless/commit/96ed0c38e29f4eb27a05c84949666132ab87b7e6

I have the same general comments here as on NatTRel.

I'm also not completely convinced that we need any additional support
for variance beyond the fact that HLists are themselves already
covariant. Could you show us an example of something that works with
smoothly with CoNatTRel and awkwardly with NatTRel?

Cheers,


Miles

--
Miles Sabin
tel: +44 7813 944 528
skype: milessabin
gtalk: mi...@milessabin.com
g+: http://www.milessabin.com
http://twitter.com/milessabin

[Alois Cochard]

Hi Miles,

If I created that variation, it's because (as a simple example) I wasn't able to give a HList with a TC of Some to function that accept TC of Option:

https://github.com/aloiscochard/shapeless/commit/96ed0c38e29f4eb27a05c84949666132ab87b7e6#L1R70

Even if HList are covariant, I need the TC constrain to be too, otherwise I don't see how it's possible to cover this use-case?

Cheers,

Alois