[GHC] #14816: Missed Called Arity opportunity?

[GHC]

#14816: Missed Called Arity opportunity?
-------------------------------------+-------------------------------------
        Reporter:  dfeuer            |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.6.1
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Runtime           |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by nomeata):

 It works if you don't export `insertModifyingArr`. Then it gets inlined
 into `test`, and CallArity, when looking at the binding of `g_s9E` in your
 original snippet, sees all the uses, sees that they are called at most
 once, and is happy to eta-expand it!

 But without inlining `insertModifyingArr`, this is beyond the reach of
 Call Arity, because it is not a higher-order analysis.

 Now, why does the demand analyser (which is higher-order, i.e. knows how
 functions call their arguments) not fix this? Because the demand analyser
 does not see that `insertModifyingArr` calls its argument only once,
 because the call is in a recursive loop.

 Sebastian Graf (@sgraf812) has thoughts on combining the analyses to give
 us the best of both worlds, maybe he can comment.

 For now, does
 {{{
 import GHC.Magic

 blink :: (a -> b) -> a -> (# b #)
 blink g = oneShot $ \a -> (# g a #)
 }}}
 do what you want?

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14816#comment:4>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler