Optimization + adding an INLINE pragma triggers Core Lint error (Type of case alternatives not the same as the annotation on case)

changed weight to 5

Attached file ghc-8.6.3.dump-core-lint (download).

Result of running /opt/ghc/8.6.3/bin/ghc -fforce-recomp -dcore-lint Bug.hs -O

Definite bug here! I investigated a bit. Here's a simpler example that crashes in the same way.

{-# LANGUAGE GADTs #-}
{-# LANGUAGE RankNTypes, KindSignatures #-}
module Bug where
import GHC.Base

data Curried (g :: * -> *) (h :: * -> *) a

instance Functor (Curried g h) where
  fmap = error "urk2"

instance (g ~ h) => Applicative (Curried g h) where
  pure   = error "urk"
  liftA2 = error "urk"
  (<*)   = error "urk"
  (*>)   = error "urk"

  (<*>) = grstargr

wib :: (g ~ h) => Curried g h (b->b) -> Curried g h b -> Curried g h b
wib a1 a2 =  a1 <*> a2

grstargr :: Curried g h (a->b) -> Curried g h a -> Curried g h b
{-# NOINLINE grstargr #-}
grstargr = error "urk"

I'm pretty sure that the wrong-ness is that the result type attached to a Case in Core is getting out of date. I don't yet know how though.

added compiler crash label

The following one-line patch to the simplifier makes the Core Lint error go away:

diff --git a/compiler/simplCore/Simplify.hs b/compiler/simplCore/Simplify.hs
index c6bd413c51..ca9a716e95 100644
--- a/compiler/simplCore/Simplify.hs
+++ b/compiler/simplCore/Simplify.hs
@@ -2621,7 +2621,7 @@ simplAlts env0 scrut case_bndr alts cont'
         ; alts' <- mapM (simplAlt alt_env' (Just scrut') imposs_deflt_cons case_bndr' cont') in_alts
         ; -- pprTrace "simplAlts" (ppr case_bndr $$ ppr alts_ty $$ ppr alts_ty' $$ ppr alts $$ ppr cont') $
 
-        ; let alts_ty' = contResultType cont'
+        ; let alts_ty' = substTy alt_env' (contResultType cont')
         -- See Note [Avoiding space leaks in OutType]
         ; seqType alts_ty' `seq`
           mkCase (seDynFlags env0) scrut' case_bndr' alts_ty' alts' }

I can't claim to know if this is the correct solution, however. Can a simplifier expert double-check my work?

Thoughts on #16312 (comment 199221), @simonpj?

Good ping, sorry to have missed this.

Your fix is plausible but wrong: alts_ty has already had the substitution applied (it's an OutType), so should not have it applied again.

The right fix is this

@@ -1269,9 +1269,13 @@ simplCast env body co0 cont0
 
         addCoerce co cont@(ApplyToTy { sc_arg_ty = arg_ty, sc_cont = tail })
           | Just (arg_ty', m_co') <- pushCoTyArg co arg_ty
+          , Pair hole_ty _ <- coercionKind co
           = {-#SCC "addCoerce-pushCoTyArg" #-}
             do { tail' <- addCoerceM m_co' tail
-               ; return (cont { sc_arg_ty = arg_ty', sc_cont = tail' }) }
+               ; return (cont { sc_arg_ty  = arg_ty'
+                              , sc_hole_ty = hole_ty  -- NB!  As the cast goes past, the
+                                                      -- type of the hole changes (#16312)
+                              , sc_cont    = tail' }) }
 
         addCoerce co cont@(ApplyToVal { sc_arg = arg, sc_env = arg_se
                                       , sc_dup = dup, sc_cont = tail })

I was simply failing to update sc_hole_ty as the coercion went past. Yikes.

This fixes the bug. Could you possibly do the rest (test case, etc)? Here is a possible commit message

Add a missing update of sc_hole_ty

In simplCast I totally failed to keep the sc_hole_ty field of
ApplyToTy (see Note [The hole type in ApplyToTy]) up to date.
When a cast goes by, of course the hole type changes.

Amazingly this has not bitten us before, but #16312 finally 
triggered it.  Fortunately the fix is simple.

mentioned in merge request !1063 (closed)

Thanks for the patch, @simonpj! I've submitted a corresponding MR in !1063 (closed).

added 1 deleted label

added typed holes label and removed 1 deleted label

added 1 deleted label

mentioned in issue #16704 (closed)

mentioned in commit d09a05d7

mentioned in commit b99077b5

mentioned in commit 589ad602

added Pnormal label

mentioned in issue #16925 (closed)

removed 1 deleted label

mentioned in commit 3ed6cf8a

closed via commit c3eaaca6

mentioned in commit c3eaaca6

Regression test: typecheck/should_compile/T16312

marked #16925 (closed) as a duplicate of this issue

marked this issue as related to #16925 (closed)

Trac field	Value
Version	8.6.3
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

Optimization + adding an INLINE pragma triggers Core Lint error (Type of case alternatives not the same as the annotation on case)

Child items ...

Activity

Optimization + adding an INLINE pragma triggers Core Lint error (Type of case alternatives not the same as the annotation on case)

Relates to

Activity