From: Sam Roberts <sroberts@...>
Date: 2004-03-21T04:19:16+09:00
Subject: Re: Can $LOAD_PATH be relative?

Quoteing cez-misc.ruby-talk@necrofish.org.uk, on Sun, Mar 21, 2004 at 03:51:04AM +0900:
> > Executables in unix don't have a knowledge of "where they are". They
> > don't, in fact, have to "be" anywhere. Some are even built into the boot
> > image, so don't in fact have to have ever "been" anywhere.
> 
> Well, on some operating systems it's certainly possible to find the
> executable image corresponding to a process (eg: Linux) and on others

Even if its been unlinked? How would one do this, exactly, read it from
/proc? It would be a handy trick to know.

> you can find the device number / inode no. of each text section within
> the process. 

That's pretty cool, I didn't know that.

But a (device#,inode#) pair isn't a path, and can be reachable through 0
(if unlinked), 1, or more (if links are present) paths, and wouldn't it
be a fairly time consuming process, searching from /, to try and find a
path that points to that dev/inod pair?.

Its not only about security, its also about predictability. Even if you
could find a path, when I choose which of the rubys in my system to use
by putting a symlink to the correct version in my ~/bin directory,
wouldn't that mean that ruby wouldn't run, when it found that ~/lib/ruby
didn't exist?

Maybe its a really great idea, but my point is that its not just unix
people being stuck in their ways (though we notoriously are :-), it is
not obvious how/whether this could be implemented on a unix box.

Cheers,
Sam