[GHC] #3808: piping binary files sometimes fail

[GHC]

#3808: piping  binary files sometimes fail
-------------------------------+--------------------------------------------
    Reporter:  paolino         |       Owner:                
        Type:  bug             |      Status:  new           
    Priority:  normal          |   Milestone:                
   Component:  Compiler        |     Version:  6.12.1        
  Resolution:                  |    Keywords:  pipe binary IO
          Os:  Linux           |    Testcase:                
Architecture:  x86_64 (amd64)  |     Failure:  None/Unknown  
-------------------------------+--------------------------------------------
Changes (by Baughn):

 * cc: sveina at gmail.com (added)

Comment:

 I'd like to add that
 - If you write the output of prod to a file, and then read it in, cons
 doesn't fail.
 - If you write the output of prod to a file, and read it in through pipe
 as in 'cat output | ./cons', cons doesn't fail.
 - If you simplify prod, using [0..] instead of fibs, cons hasn't failed
 either.

 This suggests that the bug is timing-dependent, which doesn't seem
 plausible on the face of it. However, one way in which pipes are timing-
 dependent is that the read() system call can return incomplete data when
 reading from pipes (although not when reading from files, at least on
 linux). Therefore, it would be useful to consider what happens when read
 does return incomplete data.

 Regardless of the exact circumstances, the bug itself is one that should
 never happen. Reading bytestrings uses the binary hGetBuf function, and
 should always ignore encoding; thus, no encoding error should be produced.

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/3808#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler