Re: [boost] Re: Filesystem Library: Final version for Formal Review

From: Beman Dawes (bdawes_at_[hidden])
Date: 2002-09-13 15:16:46

At 02:02 PM 9/13/2002, Eric Woodruff wrote:

 >Is it possible to get a virtual 'root' similar to what chroot in does?
 >
 >Though, this would require an instance of a file system instead of using
 >static methods/functions in the namespace.

In earlier designs there were filesystem objects and thus they could have 
that sort of behavior.

But the degree of complication introduced was completely out of proportion 
to the small increase in functionality. Furthermore, it was hard to come up 
with real world use cases where there would be any practical benefit to 
portable programs.

In past discussions people asked for functionality that would make the 
Filesystem Library better for use with a specific operating system, but 
would not be portable to other O/S's.  The intent of the library isn't to 
provide a better API to a specific operating system, so I resist 
suggestions that don't seem to be portable.  Creating some kind of abstract 
filesystem on top of the native filesystem is well beyond the current 
effort.

That being said, I think there is room for additional functionality in the 
area of the "root" concept, the "current directory" concept, "absolute" vs 
"relative" paths, and probably other areas.  But because of the portability 
issues, these features can't be designed in a vaccuum.  They require 
research.  I'd like to see the basic library nailed down first.

 >Maybe a process would want to globally change it's root so that it can 
work
 >with relative directories only.

I'm pretty much dead set against global variables, as are a lot of other 
programmers. I would want to see the root associated with some object.

But a