[GHC] #11715: Constraint vs *

[GHC]

#11715: Constraint vs *
-------------------------------------+-------------------------------------
        Reporter:  bgamari           |                Owner:
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Compiler (Type    |              Version:  8.0.1-rc1
  checker)                           |
      Resolution:                    |             Keywords:  Typeable
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by int-index):

 Replying to [comment:41 goldfire]:
 > Now, understanding `with 3 $ with 4 $ ...` is easy: the `...` will have
 access to `4`, which shadows `3`.

 I am very suspicious about shadowing semantics. I'd rather see `Int`
 treated like any other constraint, no different than `Given Int`. We don't
 have shadowing semantics for `Given Int`: the example from comment:24
 demonstrates that the first dictionary was picked, not the second one (and
 this behavior can change with a minor compiler version).

 Once again, it moves us further down the road where we admit that there
 can be more than one value of one type to the left of `=>`. There
 shouldn't be. Well-defined semantics for using nested `incoherentWith`
 will encourage people to do so. In Edward's parlance, the `Constraint`
 category is thin; let's not change it.

 I also couldn't find any mention of shadowing in the "Implicit parameters"
 section of the GHC user guide. Could you point me where I can read how it
 gets resolved? As far as I know, implicit parameters use the `IP` class
 under the hood, but GHC seems to consistently choose the innermost binding
 (in contrast to `Given`, where the outermost binding prevails):

 {{{#!hs
 GHCi, version 8.1.20160813: http://www.haskell.org/ghc/  :? for help
 Prelude> :set -XImplicitParams
 Prelude> :set -XGADTs
 Prelude> data X where X :: (?x :: Int) => X
 Prelude> f :: X -> X -> Int; f X X = ?x
 Prelude> f (let ?x = 1 in X) (let ?x = 2 in X)
 2
 }}}

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11715#comment:42>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler