[GHC] #9718: Avoid TidyPgm predicting what CorePrep will do

[GHC]

#9718: Avoid TidyPgm predicting what CorePrep will do
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  (none)
            Type:  task              |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.8.3
      Resolution:                    |             Keywords:  CodeGen, CAFs
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by osa1):

 I now built GHC with some debug prints to see what's going on. Here are
 some
 numbers:

 - The new analysis never returns `MayHaveCafRefs` for a binder that
 previous
   passes assigned `NoCafRefs`.
 - There are 9604 binders for which the new analysis returned `NoCafRefs`,
 but
   they were assigned `MayHaveCafRefs` by previous passes.

 I looked at a few of the definitions that were assigned `NoCafRefs` by the
 new
 analysis but were previously `MayHaveCafRefs`. Here are a few examples:
 (see also
 the example in comment:19)

 ----

 utils/hpc/HpcMarkup.hs:

 {{{
 sat_s7Je :: GHC.Types.Int
 [LclId] =
     CCS_DONT_CARE GHC.Types.I#! [68#];
 }}}

 (I don't know why this was previously `MayHaveCafRefs`)

 ----

 libraries/base/GHC/Conc/Sync.hs

 {{{
 sat_s8GJ
   :: forall a.
      GHC.Conc.Sync.STM a -> GHC.Conc.Sync.STM a -> GHC.Conc.Sync.STM a
 [LclId] =
     \r [eta_B3 eta_B2 void_0E]
         catchRetry# [eta_B3 eta_B2 GHC.Prim.void#];
 }}}

 This itself is not a CAF. Only free variable is `GHC.Prim.void#` (we don't
 consider `catchRetry#` as a free variable because it's a primop) which is
 defined like this

 {{{
 voidPrimId  = pcMiscPrelId voidPrimIdName voidPrimTy
                 (noCafIdInfo `setUnfoldingInfo` evaldUnfolding     -- Note
 [evaldUnfoldings]
                              `setNeverLevPoly`  voidPrimTy)
 }}}

 So this is not CAFFY (again, I don't know why this was previously
 `MayHaveCafRefs`).

 ----

 I also checked a few other definitions, one interesting binder that was
 previously `MayHaveCafRefs` and is now `NoCafRefs` is this foreign import:

 {{{
 foreign import ccall unsafe "static stdlib.h &free" c_free_finalizer
     :: FunPtr (Ptr Word8 -> IO ())
 }}}

 Once again I have no idea why this was previously `MayHaveCafRefs`...

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9718#comment:20>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler