Simplifier infinite loop regression

[ronwalf]

GHC gets stuck in the simplifier when compiling the attached file, eventually exhausting all ram. Compiling gives:

$ ghc -v3 -O --make ghcloop.hs 
Glasgow Haskell Compiler, Version 7.4.2, stage 2 booted by GHC version 7.0.3
...
*** Simplifier:
^C*** Deleting temp files:
...

Running with with all the dump flags suggests the fault is pddlDocExpr's. I'm finding it hard to create a smaller example, as even silly things like removing the container (or not using Data.Generics) restore termination.

[ronwalf]

Infinite loop test case

[ronwalf]

It compiles fine on 6.12.3, and I have reports it does not on GHC 7.0.4 (thank you ben).

[simonpj]

Thanks. This yet another example of #5448, the Russell-paradox bug.

It is easily fixed by adding {-# NOINLINE #-} on the pddlDocExpr definition in the instance declaration, and in general on every such instance declaration. The threads attached to #5448 explain the background. So that's a workaround, but not a satisfactory one.

Since the "live" examples of this bug all involve a class used contravariantly in one of its methods, I'm beginning to think that an effective counter-measure might be to NOINLINE such methods. It's a bit disgusting, and does not solve the whole problem, but it would fix all the real occurrences in the wild.