Re: Newtype wrappers

On Mon, Jan 14, 2013 at 1:45 PM, Roman Cheplyaka wrote:

* Johan Tibell [2013-01-14 13:32:54-0800]

...

On Mon, Jan 14, 2013 at 1:19 PM, Simon Peyton-Jones wrote:

...

Have you considered the effect on types like Data.Set that use the uniqueness of typeclass instances to maintain invariants? e.g. even when we have "newtype X = X Y" coercing "Set X" to "Set Y" can produce a tree with the wrong shape for the Ord instance of Y.

Good point. I should add this. The wrapper should only work if the relevant data constructors are in scope; rather like GHC’s existing auto-unwrapping on foreign calls (http://www.haskell.org/ghc/docs/latest/html/users_guide/ffi.html#ffi-newtype...)

I don't follow. Are you saying that adding an import, even if nothing from that import is used, can effect if the program compiles?

Does that mean if we ever add Data.Map.Internal that exposes the data constructors to users who "know what they're doing" (i.e. are willing to take it upon themselves to maintain the invariants) then code that used to compile will stop to do so?

Now I don't follow you. Why will it stop compiling?

If you define wrappers/unwrappers involving Data.Map, then they will compile if Data.Map.Internal is imported and will not compile if it isn't.

Let me rephrase: how will Simon's proposed "data constructors are in scope" mechanism work? For example, will let xs :: Map = ... in map MyNewtype xs behave differently if the constructors of Map are in scope or not?