Newtype wrappers

[Roman Cheplyaka]

* Johan Tibell <johan.tibell at gmail.com> [2013-01-14 15:11:25-0800]
> On Mon, Jan 14, 2013 at 2:57 PM, Roman Cheplyaka <roma at ro-che.info> wrote:
> > It's described here:
> > http://hackage.haskell.org/trac/ghc/wiki/NewtypeWrappers
> 
> We seem to be talking past each other. There's a specific problem
> related to type classes and invariants on data types mentioned earlier
> on this thread. Simon's solution here seems to be that we only coerce
> a structure from one newtype to the base type if the constructors are
> exposed, hence my question if the code changes semantics due to adding
> imports.

Yes, but it is an additional condition. For coercion to be even
considered, the coercion function has to be defined somewhere.

So Simon's proposal, as I understand it, is to allow compilation of that
coercion function only when the relevant data constructors are in scope
in the module where the coercion function is defined.

In the code you showed in an earlier message, there's no coercion
function (just the newtype constructor used as a function), hence the
semantics of that code would not change.

Here's an example of the code whose compilation would depend on the
constructors availability:

  newtype Age = MkAge Int

  newtype wrap ageMapWrapper :: Map Int a -> Map Age a

  f ... =
    let xs :: Map Int String = ...
    in ageMapWrapper xs

This code is currently impossible to write, if only for the reason that
"newtype wrap" is not a valid declaration yet. After the extension is
introduced, but before you expose Data.Map.Internal, this code will
parse (assuming the relevant extension is turned on) but fail
(presumably at the renaming stage) when it is discovered that the
coercion requires access to the internal structure of Map.

Finally, when you expose Data.Map.Internal, and the author of the above
code imports it, the code starts to compile, but the correctness of the
Map operations is now contingent on the Age's Ord instance and is the
responsibility of the code's author, as we would expect.

Roman