[GHC] #13031: Bogus calculation of bottoming arity

[GHC]

#13031: Bogus calculation of bottoming arity
-------------------------------------+-------------------------------------
           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:
-------------------------------------+-------------------------------------
 In `CoreArity` we see
 {{{
 andArityType (ABot n1) (ABot n2) = ABot (n1 `min` n2)
 }}}
 If you look at the documentation for `ABot` (which is about functions that
 are ''guaranteed'' to diverge) this is obviously bogus. The `min` should
 be `max`!

 Here's a test case:
 {{{
 {-# LANGUAGE MagicHash #-}

 module Foo( f ) where
 import GHC.Prim

 f True  = raise# True
 f False = \p q -> raise# False
 }}}
 If you compile (with GHC 8.0) `-ddump-simpl` you'll see
 {{{
 f :: forall a p1 p2. Bool -> p2 -> p1 -> a
 [GblId, Arity=1, Caf=NoCafRefs, Str=<L,U>x]
 }}}
 That `Str=<L,U>x` is totally bogus; it claims that `f` will always diverge
 given one argument.

 The fix is easy.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13031>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler