Version française
Home     About     Download     Resources     Contact us    
Browse thread
Unix.system returns "no child processes"
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Christopher Conway <conway@c...>
Subject: Unix.system returns "no child processes"
My employer recently upgraded our development machines from relatively 
old Pentiums to AMD Opterons (x86_64, easily an order of magnitude 
faster). This naturally required a re-compile of my Ocaml installation 
(3.09.0) and the supporting libraries for my application.

I am using Unix.system to invoke external commands from within Ocaml. On 
the old machines (with the 32-bit version of Ocaml), I would 
occasionally get the exception Unix_error(ECHILD,"waitpid","") from 
Unix.system. With the new machines, I'm seeing this at every call to 
Unix.system, every time. I have investigate the behavior of the 
sub-processes, and they are terminating normally, with no indication of 
any error.

I have a theory why this is happening, but no supporting evidence. It 
goes like this: system is basically fork+execv+waitpid; on the new 
machines, fork+execv is terminating so quickly that the process is gone 
before the call to waitpid. (NOTE: The external commands are relatively 
small, and take on the order of 0.05s of user time.)

There's one rather large piece of evidence cutting against this theory: 
I can't reproduce this behavior using the simple test case below, even 
using precisly the same command arguments that get used in my full app. 
I don't know how anything else going on in my code could interfere with 
the output of Unix.system. Maybe the context-switching time with a 
bigger executable makes a difference?

This isn't the biggest deal in the world, but I'd like to know for sure 
that this behavior is harmless before I write code that ignores the 
exception. If my theory is correct, it would be nice to see this fixed 
(since this behavior will become more and more common with time).

Thanks,
Chris

system.ml:
let n = ref "echo 'no command'" in
   Arg.parse [] (fun s -> n := s) "Usage: system <cmd>"  ;
   try
     ignore (Unix.system !n); (* returns normally *)
     print_endline "Success"
   with
     | Unix.Unix_error(err,fn,arg) ->
         print_endline
           ("Unix_error: \"" ^ (Unix.error_message err) ^ "\" in "
            ^ fn ^ "(" ^ arg ^ ")")