From: Gary Wright <gwtmp01@...>
Date: 2009-07-16T02:57:08+09:00
Subject: Re: Process.fork weirdness


On Jul 14, 2009, at 4:29 PM, Joel VanderWerf wrote:
> That leaves batch filtering of files, which is most often performed  
> on stdin/stdout, and these descriptors are flushed in ruby's #fork  
> wrapper.

Interesting.  I didn't realize that Kernel#fork flushes stdout and  
stderr.  That should probably be in the rdoc text for fork.

> One thing left to watch out for is race conditions in (for example):
>
> File.open(..., "w") do |f|
>  YAML.dump(..., f)
>  ### before the file is closed, suppose another thread forks and exits
> end

I can think of three ways to avoid these sorts of race conditions:

-- do all forking before opening files
-- do all forking before creating any threads
-- use a synchronization mechanism to control the timing of
    forks in a multi-threaded process

In all cases I think it requires library code to be aware of these  
constraints.  I'm not sure there is a way to protect buffered IO  
objects against a library that can fork at indeterminate times.

Gary Wright