Re: [GHC] #14356: "Main: thread blocked indefinitely in an MVar operation" in fixIO

19 Oct 2017

      #14356: "Main: thread blocked indefinitely in an MVar operation" in fixIO
-------------------------------------+-------------------------------------
        Reporter:  nickkuk           |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Runtime System    |              Version:  8.2.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Incorrect result  |  Unknown/Multiple
  at runtime                         |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by alexbiehl):

 By using a lazy pattern match it is working again:

 {{{
 main = fixIO (\ ~(x, _) -> return (1, print x)) >>= print . fst
 }}}

 Looking at the core for the reported version we find: (compiled with -O2
 to better understand the flow)

 {{{#!haskell
 main1
   = \ s_a1AP ->
       case newMVar# s_a1AP of { (# ipv_a1AZ, ipv1_a1B0 #) ->
       case ((unsafeDupableInterleaveIO
                ((\ eta_a1B8 -> readMVar# ipv1_a1B0 eta_a1B8) `cast`
 Co:6))
             `cast` Co:5)
              ipv_a1AZ
       of
       { (# ipv2_X1B5, ipv3_X1B7 #) ->

       -- ipv3_X1B7 is our chunk which delays the readMVar#
       -- entering ipv3_X1B7 causes a deadlock!
       case ipv3_X1B7 of { (x_aV8, ds2_d1zX) ->
       case putMVar#
              ipv1_a1B0
              (main3,
               (hPutStr2 stdout ($fShowInteger_$cshow x_aV8) True) `cast`
 Co:3)
              ipv2_X1B5
       of s2#_a25J
       { __DEFAULT ->
       hPutStr2 stdout main2 True s2#_a25J
       }
       }
       }
       }

 }}}

 By using a strict pattern match on the tuple we trigger the `readMVar#`
 which blocks and deadlocks as we can't make further progress.

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14356#comment:4
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler