Replace explicit wrapping and unwrapping of newtypes by coercions

[kl0tl]

This adds Coercible superclasses to Newtype in order to implement most combinators with coerce.

Newtype constructors have to be exported for them to be unwrapped when solving Coercible constraints so this means that Newtype instances for types with hidden constructors won’t be usable anymore outside of their module of definition. This is mitigated by purescript/purescript#3907 which emits a warning for such types (this warning would be turned into an error by the pull request updating the compiler support for Newtype instances deriving) and purescript/purescript#3927 which suggests to export hidden newtype constructors when a Coercible constraint fails to solve because of that.

This also means that Newtype instances can only de declared for representationally equal types.

I reckon that releasing this along with the next compiler release is perhaps too reckless but at least it should provide more context and motivation for the mentioned warnings.

[rhendric]

Isn't Coercible symmetric? What's the thinking behind giving two Coercible superclasses instead of just one and letting the compiler solve the other as needed—or does that not work?

[kl0tl]

Computing subgoals is indeed symmetric, the subgoals of Coercible a b are the same than the subgoals of Coercible b a, but the compiler doesn’t deduce Coercible b a from Coercible a b.

This means that we need Newtype t a to imply both Coercible t a and Coercible a t for unwrap and wrap to compile:

module Example where

import Safe.Coerce (coerce)
import Prim.Coerce (class Coercible)

class Coercible t a <= Newtype t a | t -> a

wrap :: forall t a. Newtype t a => a -> t
wrap = coerce

Error found:
in module Example
at Example.purs:9:8 - 9:14 (line 9, column 8 - line 9, column 14)

  No type class instance was found for
                            
    Prim.Coerce.Coercible a0
                          t1
                            

while checking that type forall (a :: Type) (b :: Type). Coercible a b => a -> b
  is at least as general as type a0 -> t1
while checking that expression coerce
  has type a0 -> t1
in value declaration wrap

where t1 is a rigid type variable
        bound at (line 9, column 8 - line 9, column 14)
      a0 is a rigid type variable
        bound at (line 9, column 8 - line 9, column 14)

See https://github.com/purescript/documentation/blob/master/errors/NoInstanceFound.md for more information,
or to contribute content related to this error.

This is something we might be able to do though, I’ll look into it.

[hdgarrood]

Is there a particular reason not to do this now? I think if we're going to do it at all, then the sooner the better.

[kl0tl]

I’d like to release this with the v0.14.0 too but this depends on and implies a few things:

• We should merge Turn Coercible into a symmetric relation purescript#3930 first so we can drop one of the Coercible superclasses.
• We should also merge Warn against exported types with hidden constructors but Generic or Newtype instances purescript#3907 and Suggest to export newtype constructors when the unwrapping rule fails purescript#3927 (which depends on the more critical Require newtype constructors to be imported for coercing them purescript#3937) for helpful warnings.
• We should then update the compiler support for deriving Newtype instances. I have prepared a pull request but it depends on Warn against exported types with hidden constructors but Generic or Newtype instances purescript#3907 (to turn the warning into an error for types with hidden constructors but Newtype instances) and a new version of this package to be released (to update tests/support/bower.json in the compiler repository).

[hdgarrood]

Right, ok, I didn't realise we'd have to do all that stuff too. In that case I think we should leave this out for now. If we manage to do all those things before we're ready to release v0.14.0 then that's fine, but otherwise I don't think this is an important enough change to hold v0.14.0 up.

[JordanMartinez]

Of the PRs @kl0tl mentioned above that need to be merged before this PR can be merged, only purescript/purescript#3927 hasn't yet been merged (but looks like it will be soon).

[hdgarrood]

I think purescript/purescript#3927 isn't even a hard requirement for merging this, it's just a change to make certain compiler errors more helpful, so I'd be in favour of going ahead and merging this now.

[JordanMartinez]

Do we need to rebase this on top of current master to make GH Actions run rather than Travis?

[hdgarrood]

I think so, yes.

[JordanMartinez]

CI is failing with this error:

[1/1 NoInstanceFound] src/Data/Newtype.purs:42:1

  42  instance newtypeMultiplicative :: Newtype (Multiplicative a) a
      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  
  No type class instance was found for
  
    Prim.Coerce.Coercible (Multiplicative a0)
                          a0
  
  while checking that expression #dict Coercible
    has type Coercible$Dict @Type (Multiplicative a0) a0
  in value declaration newtypeMultiplicative
  
  where a0 is a rigid type variable
          bound at (line 0, column 0 - line 0, column 0)

           Src   Lib   All
Warnings   0     0     0  
Errors     1     0     1  

[hdgarrood]

That will probably be because the Multiplicative constructor is not in scope, we’ll need to import it in that module.