[GHC] #13144: FoatOut is not floating bottoming expressions properly

17 Jan 2017

      #13144: FoatOut is not floating bottoming expressions properly
-------------------------------------+-------------------------------------
           Reporter:  simonpj        |             Owner:
               Type:  bug            |            Status:  new
           Priority:  normal         |         Milestone:
          Component:  Compiler       |           Version:  8.0.1
           Keywords:                 |  Operating System:  Unknown/Multiple
       Architecture:                 |   Type of failure:  None/Unknown
  Unknown/Multiple                   |
          Test Case:                 |        Blocked By:
           Blocking:                 |   Related Tickets:
Differential Rev(s):                 |         Wiki Page:
-------------------------------------+-------------------------------------
 Consider part of Eric's program in #13143, and pretty much exactly what is
 in `GHC.Arr`.
 {{{
 {-# INLINE index #-}
 index :: (Int, Int) -> Int -> Int
 index b@(l,h) i
   | l <= i && i < h = 0
   | otherwise       = indexError b i 0

 {-# NOINLINE indexError #-}
 indexError :: (Int, Int) -> Int -> Int -> b
 indexError rng i tp = error (show rng)
 }}}
 Before float-out we have
 {{{
 index =
   \ (b_ay8 :: (Int, Int)) (i_ayb :: Int) ->
     case b_ay8 of wild_Xe { (l_ay9, h_aya) ->
     case &&
            (leInt l_ay9 i_ayb) (ltInt i_ayb h_aya)
     of {
       False -> indexError @ Int wild_Xe i_ayb (I# 0#);
       True -> GHC.Types.I# 0#
     } }
 }}}
 and after float-out we see
 {{{
 index =
   \ (b_ay8 :: (Int, Int)) (i_ayb :: Int) ->
     case b_ay8 of wild_Xe { (l_ay9, h_aya) ->
     case &&
            (leInt l_ay9 i_ayb) (ltInt i_ayb h_aya)
     of {
       False -> indexError @ Int b_ay8 i_ayb lvl_s2cd;
       True -> GHC.Types.I# 0#
     } }
 }}}
 We've floated the `(I# 0#)`.  ''But we should have floated the entire call
 to `indexError` thus'':
 {{{
       False -> lvl_xxx b_ay8 i_ayb
 }}}
 with
 {{{
 lvl_xxx b i = indexError @Int b i (I# 0#)
 }}}
 Why?  Because it makes the expression smaller and moves the error handling
 code out of the way.

 Float-out makes some attempt to to this, but it's not right yet.

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13144
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler