From: Simon Strandgaard <0bz63fz3m1qt3001@...>
Date: 2003-05-15T03:25:24+09:00
Subject: Re: RCR for child execution

On Thu, 15 May 2003 03:23:46 +0900, Aredridel wrote:

> On Wed, 2003-05-14 at 04:40, Simon Strandgaard wrote:
>> Besides that I want this 'permanent sandbox' myself. I think that its a 
>> very useful for other people which embeds ruby into c++. There has been
>> very opinions from this embedding-camp so far. Therefore I think this is
>> worth figthing for.
>> I would like to hear opinions: Is this usable ?
> 
> Simon -- I think one thing that would do what you want is a solution
> with dRb and proxy objects.

Agree this is also a possibility that will work. 
But I am very concerned about speed (input-response-repaint). 
Marshalling objects via a pipe  - sounds slow to me.

I know Knuth says: 
  Premature optimization is the root of all evil

But I don't consider my proposal being optimization.

 
> Though it /sounds/ kludgy, I think it may end up being a more stable
> solution that is more portable.

Disagree..  Your proposal will introduce many layers of logic.
Extra code => higher chances for bugs..  
Advanced solutions trend to be more unstable  over simple.


Your idea can be very useful if one wants complete separation of ruby from
c/c++.  So that the c/c++ application can stay unaffected if a segfault
occurs within ruby. It will be a ruby-prison! 
I think ruby is extremly stable so I have no second thoughts integrating
ruby directly into my c++ application. Think of it: every day people
blindly links their applications with 3rd party libraries. Why not link
with ruby too?


I am just searching for the 'simple' intuitive solution (I think we
have found it, see the Kernel#fork talking :-) 

--
Simon Strandgaard