Re: Named streams, Re: Cleaning up namei

[David Holland <dholland-tech%netbsd.org@localhost>]

On Tue, Mar 25, 2008 at 01:54:52AM -0400, der Mouse wrote:
 > > But there's no reason they should be prohibited by the VFS layer.
 > 
 > I spent the second half of 2002 writing impedance-matching glue code so
 > that a research project could be mounted as a filesystem.  It ended up
 > with something very similar: files were, well, files, but if you
 > treated one like a directory it worked like a directory, exposing some
 > of the internal data structures that were important to the underlying
 > research project but which didn't fit the Unix filesystem model any
 > better than MacOS named forks do.

I tried to write a fs like that once a long time ago (1995? 1996?), in
Linux. I never did manage to get it working even after fairly drastic
surgery to Linux's then-idea of VFS. It was one of the early
experiences informing my ideas about what should and should not be
done at the VFS level.

Research on file naming semantics has always been difficult because of
issues like this.

IMO, we should enable FS research where we can, provided it doesn't
cost much. Cost is mostly measured in (un)maintainability...

-- 
David A. Holland
dholland%netbsd.org@localhost