"Preserving user bottoms" was found to be better behavior for us with Void as well back in the day. Evaluating such a term to get the bottom out is better than making up one that loses information for the user about precisely what bottom it is they had. We do so with absurd and the like for Void. 

This way if you map over a structure with errors at the leaves you get a new structure with those same errors at the leaves.

tl;dr +1 from me. 

-Edward

On Sun, Jan 15, 2017 at 11:00 PM, Kevin Cotrone <kevincotrone@gmail.com> wrote:
That seems to have a surprising strictness.

I'm not sure if it would be the best idea to try and evaluate a type with no inhabitants.

On Sun, Jan 15, 2017 at 2:37 PM, David Feuer <david.feuer@gmail.com> wrote:
Currently, if you write

data V a deriving Functor

GHC generates

fmap _ _ = error "Void fmap"

This seems quite unfortunate, because it loses potentially useful error information:

fmap (+ 3) (error "Too many snozzcumbers!")

throws "Void fmap", rather than the much more precise "Too many snozzcumbers!" I've opened Trac #13117 to fix this, but I figured I should double check that no one is opposed.

David Feuer

_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries



_______________________________________________
Libraries mailing list
Libraries@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/libraries