From: Simon Strandgaard <0bz63fz3m1qt3001@...>
Date: 2003-05-14T20:21:07+09:00
Subject: Re: RCR for child execution

On Wed, 14 May 2003 20:30:55 +0900, Brian Candler wrote:

> On Wed, May 14, 2003 at 08:38:44AM +0900, Simon Strandgaard wrote:
>> > No, because it cannot work. If I write
>> > 
>> >      s = ""
>> >      $stdout = StringIO.new(s)
>> >      pid = fork do
>> >        exec("ls *.txt")
>> >      end
>> >      Process.wait(pid)
>> > 
>> > how are you going to modify fork or exec so that it captures the output of
>> > 'ls' into the StringIO object created earlier?
>> 
>> I luckyly forgot all about this senario... I you can forgive me :-)
>> 
>> StringIO has no filedescriptor that fork() can parse down to the child.
>> So we must create such filedescriptor.
>> 
>> A pipe and a receiver thread in the parent process should do it. 
>> Some semaphore protection is probably necessary, But I don't have enough
>> insight into the core of ruby, so I cannot tell where semaphore protection should
>> go.  I think this is possible.
> 
> So I think you're saying you want Kernel#fork to do the following:
>  - create a pipe
>  - fork
>  - in the child, dup2 the writer end to stdout/stderr then continue as
>    normal
>  - in the parent, start a new Ruby thread which takes data from the reader
>    end of this pipe and writes it to $stdout (or $defout)
> 
> and make sure the Ruby IO system is thread-safe such that this child thread
> can write to the StringIO object at the same time as the parent reads from
> it (it may already be OK to do that). The thread dies when the pipe is
> closed.

Yes Kernel#fork should do something like the above.
If I know you correct you are very sceptic about this, thats OK with me.
Do you agree with me that this feature *is possible* to implement
into Ruby's core ???   
 

> OK, now fix the following one-line Ruby program which you call from C++:
> 
>       exec("ls /tmp")
> 
> Note that right now it will not only write to Unix stdout, it will cause
> your entire C++ program to exit.

I said earlier (sorry no link this time) that I don't want to build a
Ruby-prison. If this is what they want then they are free to do it.
If I wanted a Ruby-prison then I would use a pedantic $SAFE level :-)

--
Simon Strandgaard