From: Brian Candler <B.Candler@...>
Date: 2003-05-14T18:31:57+09:00
Subject: Re: RCR for child execution

On Wed, May 14, 2003 at 07:38:27AM +0900, Simon Strandgaard wrote:
> On Wed, 14 May 2003 00:44:36 +0900, ts wrote:
> 
> >>>>>> "S" == Simon Strandgaard <0bz63fz3m1qt3001@sneakemail.com> writes:
> > 
> >   I really don't understand what you want
> 
> The refined description of what I want: a permanent sandbox
> http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/71364

Let me try to state your requirements more precisely, and you tell me if
this is correct.

You want to call the Ruby interpreter from C++, subject to the following
conditions:

1. Absolutely *all* output generated on stdout/stderr as a side-effect of
   running the Ruby code is captured.

2. It is captured directly into a *string* in memory (i.e. capturing into
   a temporary file is forbidden)

3. It works for *all* possible Ruby programs (in particular we are
   including Ruby programs which call Kernel#exec or which call C libraries)

4. It must run as the *same* process as the C++ program (i.e. we are not
   allowed to do an additional fork for the purposes of capturing the
   output; any forks which would have been done anyway by the Ruby code as
   part of its normal operation are still allowed of course)

Is that a correct description of the constraints of the problem? Please tell
me, or please modify them. I have highlighted the key words.

If so, I can demonstrate (prove?) that they are mutually exclusive, by
reference to the Unix execution model: it's nothing to do with Ruby itself,
and therefore cannot be changed by modifying Ruby.

If you are happy to lift any one of those four constraints, I can give you a
solution - one of those posted already.

Regards,

Brian.