From: Simon Strandgaard <0bz63fz3m1qt3001@...>
Date: 2003-05-14T06:56:51+09:00
Subject: Re: RCR for child execution

On Tue, 13 May 2003 21:11:08 +0000, ahoward wrote:

>> >> >         pid = fork do
>> >> >           exec("ls /nonexistent")
>> >> >         end
>> >> >         Process.waitpid(pid)
>> >> >
> 
> why must this be done in ruby simon?  
> this can all be done from within c or c--:

The code you see here is Brian Candlers.. not mine. 

 
> i think this does exactly what you want - it captures all output, including
> output from any child, from a ruby script run from within c.

[snip CODE]

Yes your code does *most* of what I want.. but unfortunatly not all.
I want a permanent sandbox,  not a setup/eval/restore sandbox.
The reason why I want a permanent sandbox:
I have instances of Ruby classes owned by C++ which I access over long
time. 

Exactly "a permanent sandbox" is what all this trouble is about, I should
have wrote that in the RCR.  The above is the best description of my problem so
far. Does anyone understand me at all?

 
>> Kernel#fork  does spawn a child.. changes to Rubys Core is necessary.
>> unix-fork inherit stdin/stdout/stderr from parent to the child process.  I
>> want Rubys Kernel#fork to inherit rb_stdin/rb_stdout/rb_stderr from the
>> parent to the child process.  But hey.. unix-fork() does not take any
>> arguments.  Is this possible?  My guess - Yes this should be possible. Here
>> is some pseudo code.
>>
>> void Kernel#fork() {
>>     backup_stdout = stdout;      /* what i want */
>>     stdout = rb_stdout;          /* what i want */
>>     pid = fork();
>>     if(pid == 0)
>>         stdout = backup_stdout;  /* what i want */
>> }
> <snip>
> 
> it is possible, but requires no changes to ruby's core, the above accomplishes
> this in c.

A permanent sandbox requires changes to ruby's core (what I want).

 
>> Again assignment to stdout is illegal.. freopen should be used. I just use
>> assignment in this pseudo-code in order to keep it simple. Further more
>> stderr/stdin is left out.. they should of cause also be redirected.
> 
> if this is done using dup(2) - all is well in the world.

Agree.. There is many examples around in this single thread.

 
--
Simon Strandgaard