> John S. Dyson wrote:
> > 
> > > Why do major whacking on MFS? Seems to me it'd be much easier to just
> > > write a new fs type from scratch rather than attempting to change code
> > > built around totally different assumptions.
> > > 
> > That would be essentially what it would be.  My initial set of simple
> 
> You: We're going to modify MFS.
> Me: Why not start from scratch instead?
> You: Good idea! We'll start from scratch by taking MFS and modifying
>      it.
  ^^^^  You are misinterpreting my position.
> 
> What, are you taking Clinton lessons in your spare time? :-)
                       ^^^^^^^ I am about as anti-hypocracy and
  accordingly not-for Clinton as one can be.  (I am not actually
  anti-Clinton, but mostly not comfortable with him.)
> 
My initial (read: simple) mods would be to MFS (please read my message.)
Alot of bang for the buck can be gotten by some mods that would
cost minimal amounts of time.  (Est 1-2 days of work for me.)
Right now, I am overloaded with my new LFS implementation, SMP
adaptation of the VM system, Pentium-pro enhancements, etc...

>
> No, that's not essentialy what it would be. "removal [of] much of the
> UFS infrastructure" implies that you're starting with something. What
> I'm suggetsing is that you start with *nothing* (ie, scratch, which is
> what I said).  Start with a clean slate.  Zero, zilch, bubkis. I think,
> as I said, that it'll be easier and produce better results to write a
> ram filesystem by writing a ram filesystem, rather that trying to adapt
> a disk filesystem.
> 
It still needs the naming structure, etc...  I didn't think anyone
would be interested in developing another set of directory/file semantics.
IMO, one of the mistakes in the MSDOSFS is that code wasn't derived using UFS.
(Note that UFS is not the same as FFS, the on-disk FFS filesystem uses
the UFS semantic structure, so does LFS, EXT2FS and MFS currently.)

The low-level code (read/write) could be redone without writing all of
the code for a new filesystem.  Take a look at UFS -- it does do a lot
of good things that can make it easier and less buggy to write and
maintain a filesystem conforming to U**X semantics.  My primary complaint is
a result of current MFS using too much of FFS.  Again, we (FreeBSD) have
been discussing this for the last 2yrs or so also (per my earlier
statement.)  The method that I had suggested would use UFS with a fairly
simple lower end based mostly on the VM system.  (I am actually suprised
that it wasn't done that way anyway!!!)

The MFS "problem" can be solved in various ways.  So, we disagree more
than I had thought, because it just seems to me to be a waste to
reimplement userland filesystem semantics (and I assumed that we would
naturally agree.)

John
dyson@freebsd.org