From: ara.t.howard@...
Date: 2006-04-18T03:19:14+09:00
Subject: Re: IO not closed by GC

On Tue, 18 Apr 2006, Stephan Maka wrote:

> Gennady Bystritsky wrote:
>> But you do not even know when GC will be run, so in your case it might
>
> Inserting 'GC.start; sleep 1' won't help...
>
>> not even have kicked in. On the other hand, IO.popen has a nice block
>> form that gurantees that an underlying IO object will be closed
>> properly. You can rewrite your code as:
>>
>> os = IO.popen('uname -sr') { |io| io.readlines.to_s.strip }
>
> That works, but is longer. Shorter code is more readable in general...
>
> Anyway, I think an IO object should be closed when not being needed
> anylonger.


this doesn't fit the posix process model.  under posix fd's are managed per
process-group, not per process.  consider this simple quine:



   harp:~ > cat a.rb
   open __FILE__

   IO.new(3).close if ARGV.delete '--close'

   fork ? Process.wait : print(IO.new(3).read)



   harp:~ > ruby a.rb
   open __FILE__

   IO.new(3).close if ARGV.delete '--close'

   fork ? Process.wait : print(IO.new(3).read)



   harp:~ > ruby a.rb --close
   a.rb:6:in `initialize': Bad file descriptor (Errno::EBADF)
           from a.rb:6



read this

   http://www.opengroup.org/onlinepubs/000095399/functions/fork.html

in particular bullet 4.  gc'd io object would be incompatible with the posix
process model.

i suppose one could have

   io.dont_gc

but i seems the best POLS approach (not to gc) is the posix route.

regards.


-a
-- 
be kind whenever possible... it is always possible.
- h.h. the 14th dali lama