#5448: GHC stuck in infinite loop compiling with optimizations
-------------------------------+--------------------------------------------
Reporter: ronwalf | Owner:
Type: bug | Status: new
Priority: normal | Milestone:
Component: Compiler | Version: 7.0.3
Keywords: | Testcase:
Blockedby: | Difficulty:
Os: MacOS X | Blocking:
Architecture: x86_64 (amd64) | Failure: Compile-time crash
-------------------------------+--------------------------------------------
Comment(by simonpj):
Actually, yes, it ''is'' an example of the classic inliner-looping bug.
Here's a very slightly simpler version (no need for `deriving Eq`):
{{{
module GHCLoop where
newtype Expr f = In (f (Expr f))
class FuncEq f where
funcEq :: FuncEq g => f (Expr g) -> f (Expr g) -> Bool
instance (FuncEq f) => Eq (Expr f) where
(In x) == (In y) = funcEq x y
data Not e = Not e
instance FuncEq Not where
funcEq (Not x) (Not y) = x == y -- x,y :: Expr g
-- Needs Eq (Expr g)
-- = $c== (funcEq g)
foo :: Expr Not -> Bool
foo x = x == x
}}}
The classic bug shows up when you have a type declared with a
contravariant occurrence of itself. Something like:
{{{
data T = MkT (T -> Int)
get (MkT f) = f
foo t = get t t -- Loops
}}}
And that is what we get here: the dictionary for `FunEq` looks like this:
{{{
data FunEq f = MkD (forall g. FuncEq g -> f (Expr g) -> f (Expr g) ->
Bool)
}}}
When `f` and `g` both get instantiated to the same type, we are in the
contravariant situation, and that's exactly what happens here. Here are
my notes on what happens (mainly for me):
{{{
Rule fired: Class op funcEq
Inlining done: $cfuncEq{v akX} [lid]
Inlining done: $c=={v alm} [lid]
$cfuncEq $dfuncEq = ...$c== $dfuncEq...
$c== $dfuncEq = ...funcEq $dfuncEq $dfuncEq...
$funcEqNot = MkD $cfuncEq
$sc== = ...funcEq $funcEqNot $funcEqNot...
funcEq $funcEqNot $funcEqNot
--> $cfuncEq $funcEqnot
--> $c== $funcEqNot
--> funcEq $funcEqNot $funcEqNot
}}}
So this is the first time I've seen a ''real'' example of the bug. You
can probably work around it with a judicious NOINLINE pragma. But really
we need something better.
Your example is a more convincing version of #5400.
See also #3400, #3872.
--
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/5448#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler
_______________________________________________
Glasgow-haskell-bugs mailing list
[email protected]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
