[GHC] #4419: dph tests fail with panic

[Simon Peyton-Jones]

|  > But why do the Ids need an unfolding at all?  The simplifier will add one for itself, as
|  it does with ordinary Haskell code
|  
|  Because we want to make some of the bindings we generate INLINE and IIUC, the way
|  to do this is to mkInlineUnfolding.

Ah.  But do *all* of the Ids in a mutually recursive group get INLINE pragamas?  If so that sounds suspicious.  If not, you can cut the loop with the ones that don't

But setting the unf lazily is ok with me, provided it is documented.  These matters are not obvious.

S