Re: [GHC] #14779: Compiling with -g fails -lint-core checks

20 Feb 2018

      #14779: Compiling with -g fails -lint-core checks
-------------------------------------+-------------------------------------
        Reporter:  niteria           |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:
  (Debugging)                        |
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #14122, #14123,   |  Differential Rev(s):  phab:D4423
  #8472, #14406                      |
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 I'm afraid I'm still not happy.  The offending case is this.  Before, we
 have:
 {{{
 T14779.conMkFixed
   = srcT14779.hs:78:1-49
     T14779.mkConstr
       (srcT14779.hs:78:23-29 T14779.tyFixed)
       (srcT14779.hs:78:31-39 GHC.CString.unpackCString# "MkFixed"#)
       ((srcT14779.hs:78:41-42 GHC.Types.[]) @ GHC.Base.String)
       (srcT14779.hs:78:44-49 T14779.Prefix)
 }}}
 The simplifier inlines `mkConstr`, let-binding its arguments like this
 {{{
 conMkFixed = let str_s2UM :: Addr#
                  str_s2UM = srcT14779.hs:78:31-39 "MkFixed"#

                  str_a2jk :: String
                  str_a2jk = srcT14779.hs:78:31-39 unpackCString#
 str_s2UM
              in
              ...str_a2jk...
 }}}
 and then the simplifier floats those two bindings outward.

 Ben/Bartosz's patch messes with `prepareRhs` in a slightly delicate way,
 to remove the tick on the string, but I think the bug is earlier.  '''I
 believe we should never have a tick wrapped around a literal string'''.
 That's certainly what Phab:D3925 seems to say.

 If that is true (is it?) then that invariant is already broken in the
 above intermediate form.  We should instead have
 {{{
 conMkFixed = let str_s2UM :: Addr#
                  str_s2UM = "MkFixed"#

                  str_a2jk :: String
                  str_a2jk = srcT14779.hs:78:31-39 unpackCString#
 str_s2UM
              in
              ...str_a2jk...
 }}}
 Let's first establish whether we ever want a tick wrapped around a literal
 string. If we don't, let's adopt Phab:D3925, and make the smart
 constructor `mkTick` guarantee it. (Unfortunately `CoreUtils.mkTick` is a
 jolly complicated function and I don't understand it.)

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14779#comment:10
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler