Equality constraint in superclass not supported

[rodprice]

The code in "bug.hs" causes a GHC panic; the nearly identical code in "unbug.hs" does not.

[rodprice]

Causes a GHC panic

[rodprice]

This code compiles fine.

[rodprice]

Here's the error message:

stm$ ghci bug.hs GHCi, version 6.10.0.20081007: http://www.haskell.org/ghc/ :? for help Loading package ghc-prim ... linking ... done. Loading package integer ... linking ... done. Loading package base ... linking ... done. [1 of 1] Compiling Main ( bug.hs, interpreted ) ghc: panic! (the 'impossible' happened)

(GHC version 6.10.0.20081007 for i386-unknown-linux):

applyTypeToArgs

main:Main.:DIIDomain{v rmQ} [gid]

@ main:Main.Interval{tc rfz} @ main:Main.Interval{tc rfz} $dIDomain{v an6} [lid] $dIDomain{v an7} [lid] @ (main:Main.Value{tc rfv} main:Main.Interval{tc rfz})

(main:Main.Value{tc rfv} main:Main.Interval{tc rfz}

~

main:Main.Value{tc rfv} main:Main.Interval{tc rfz}, main:Main.IDomain{tc rfr} main:Main.Interval{tc rfz}, main:Main.IDomain{tc rfr} main:Main.Interval{tc rfz}) =>

(main:Main.Domain{tc rfx}

main:Main.Interval{tc rfz} (main:Main.Value{tc rfv} main:Main.Interval{tc rfz})

-> main:Main.Domain{tc rfx}

main:Main.Interval{tc rfz} (main:Main.Value{tc rfv} main:Main.Interval{tc rfz})

-> ghc-prim:GHC.Bool.Bool{(w) tc 3c})

-> main:Main.Interval{tc rfz}

main:Main.:TIIDomain{tc rmP} main:Main.Interval{tc rfz}

[simonpj]

Ah, you are using an equality predicate in the superclass context

class (Value d1 ~ Value d2) => IIDomain d1 d2 where

This specifically does not work in 6.10; it's the major unfinished piece in the type-family story.

Really GHC 6.10 should simply reject such programs. As it stands, it has a half-baked go at it.

Simon

[simonmar]

Adding more descriptive description.

[igloo]

See also: #2686

[chak]

There is no panic anymore. Programs with super class equalities are outright rejected with an error message explaining that superclass equalities are a currently unsupported feature.

[porges]

Just wondering if this is still slated to be in 6.12... :)

[simonpj]

Yes, that's still the plan. But there's still a fair bit of upheaval going on in the type inference for equalities, so don't hold your breath.

Simon

[Andrey.Sisoyev]

Didn't notice it in release info (for GHC 6.12.1). How is the progress? =)

[simonpj]

Hurrah! We now have superclass equalities. I've added this example as a test.

commit 940d1309e58382c889c2665227863fd790bdb21c
Author: Simon Peyton Jones <simonpj@microsoft.com>
Date:   Wed Jun 22 17:37:47 2011 +0100

    Add equality superclasses
    
    Hurrah.  At last we can write
    
       class (F a ~ b) => C a b where { ... }
    
    This fruit of the fact that equalities are now values,
    and all evidence is handled uniformly.
    
    The main tricky point is that when translating to Core
    an evidence variable 'v' is represented either as
      either   Var v
      or       Coercion (CoVar v)
    depending on whether or not v is an equality.  This leads
    to a few annoying calls to 'varToCoreExpr'.

 compiler/basicTypes/MkId.lhs      |    4 +-
 compiler/deSugar/DsExpr.lhs       |   18 +++++++-
 compiler/iface/BuildTyCl.lhs      |   50 ++++++++++------------
 compiler/typecheck/TcInstDcls.lhs |   85 +++++++++++++++++-------------------
 compiler/typecheck/TcMType.lhs    |    9 +----
 compiler/types/Class.lhs          |   21 ++++------
 6 files changed, 92 insertions(+), 95 deletions(-)