[GHC] #11731: Simplifier: Inlining trivial let can lose sharing

[GHC]

#11731: Simplifier: Inlining trivial let can lose sharing
-------------------------------------+-------------------------------------
        Reporter:  nomeata           |                Owner:
            Type:  bug               |               Status:  patch
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):  Phab:D2073
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by nomeata):

 With the late demand analysis, suddenly demand signatures involving free
 variables turn up in `-ddump-simpl`:

 {{{
  Roman.foo_$s$wgo [Occ=LoopBreaker]
    :: GHC.Prim.Int# -> GHC.Prim.Int# -> GHC.Prim.Int#
 -[GblId, Arity=2, Caf=NoCafRefs, Str=<L,U><L,U>]
 +[GblId, Arity=2, Caf=NoCafRefs, Str=<L,U><S,U>]
  Roman.foo_$s$wgo =
    \ (sc :: GHC.Prim.Int#) (sc1 :: GHC.Prim.Int#) ->
      let {
        m :: GHC.Prim.Int#
 -      [LclId]
 +      [LclId, Str={s1Yc-><S,A>}]
 ...
 }}}

 As this contains uniques, this breaks a few test cases. So it seems that
 previously, this information has been removed somewhere (in the simplifier
 or so). With DmdAnal running right at the end, this is not happening. So I
 wondered if `CoreTidy` should remove them, but there it says
 {{{
         -- Note [Tidy IdInfo]
         -- We need to keep around any interesting strictness and
         -- demand info because later on we may need to use it when
         -- converting to A-normal form.
         -- eg.
         --      f (g x),  where f is strict in its argument, will be
 converted
         --      into  case (g x) of z -> f z  by CorePrep, but only if f
 still
         --      has its strictness info.
         --
         -- Similarly for the demand info - on a let binder, this tells
         -- CorePrep to turn the let into a case.
 }}}

 Questions:
  * At CorePrep time, should there still be a demand environment in the
 strictness signatures? (Note that the `Binary` instance does _not_
 serialize the environment, but discards it).
  * If not, should I remove the environment in TidyCore?
  * Where was it removed before?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11731#comment:41>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler