[GHC] #15493: Elide empty dictionaries

[GHC]

#15493: Elide empty dictionaries
-------------------------------------+-------------------------------------
        Reporter:  mnoonan           |                Owner:  (none)
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.4.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------
Description changed by mnoonan:

Old description:

> Suppose you define a type class `C` with no methods, and use it as a
> constraint on a function `f :: C => Foo`. GHC's generated Core for `f`
> currently passes an empty dictionary for `C`. With optimizations, it
> seems to be true that at use sites, calls to `f` will be replaced with
> calls to an inner function that skips the typeclass dictionary. But can I
> count on this optimization always working?
>
> Would it possible to get a guarantee that the `C =>` constraint will have
> no run-time overhead, by dropping it entirely from the generated Core? Or
> is there a subtlety that prevents this from being sound?
>
> For context, I would like to implicitly pass compile-time evidence that
> various properties hold (e.g. such-and-such a list is non-empty), while
> retaining a guarantee that the evidence will have no run-time cost. Like
> this:
>
> {{{
> newtype Named a name = Named a
>
> class Fact p
>
> -- Ideally, we'd have a guarantee that the following function
> --   compiles to exactly the same code as `Prelude.head`
> head :: Fact (IsCons xs) => (a `Named` xs) -> a
> head xs = Prelude.head (coerce xs)
> }}}

New description:

 Suppose you define a type class `C` with no methods, and use it as a
 constraint on a function `f :: C => Foo`. GHC's generated Core for `f`
 currently passes an empty dictionary for `C`. With optimizations, it seems
 to be true that at use sites, calls to `f` will be replaced with calls to
 an inner function that skips the typeclass dictionary. But can I count on
 this optimization always working?

 Would it possible to get a guarantee that the `C =>` constraint will have
 no run-time overhead, by dropping it entirely from the generated Core? Or
 is there a subtlety that prevents this from being sound?

 For context, I would like to implicitly pass compile-time evidence that
 various properties hold (e.g. such-and-such a list is non-empty), while
 retaining a guarantee that the evidence will have no run-time cost. Like
 this:

 {{{
 newtype Named a name = Named a

 class Fact p

 -- Ideally, we'd have a guarantee that the following function
 --   compiles to exactly the same code as `Prelude.head`
 head :: Fact (IsCons xs) => ([a] `Named` xs) -> a
 head xs = Prelude.head (coerce xs)
 }}}

--

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/15493#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler