Subject: Re: "tfs" and other filesystems with very short names
To: None <rvb@sicily.odyssey.cs.cmu.edu>
From: Greg A. Woods <woods@most.weird.com>
List: source-changes
Date: 09/22/1998 20:24:52
[ On , September 22, 1998 at 17:36:37 (-0400), rvb@sicily.odyssey.cs.cmu.edu wrote: ]
> Subject: Re: "tfs" and other filesystems with very short names
>
> Very good question. But there are two. We call it Coda here so
> the directory is coda. Then I looked at all the fs's and some used
> fs in the name of their methods and some did not; it was pretty evenly
> split.
I should have been specific -- I meant in the kernel's namespace
(including both symbol space and source filesystem naming)
Let's see (this from directories only in the kernel):
With "fs" suffix: Without "fs" suffix:
================= ====================
adosfs coda
filecorefs
isofs
miscfs/
deadfs fdesc
fifofs portal
(genfs) union
kernfs
nullfs
procfs
specfs
umapfs
msdosfs
nfs
ufs/
ex2fs
ffs
lfs
mfs
(ufs)
That's about 16 to 4 if you count real filesystem types. I've not (yet)
collected any stats to compare how consistent the symbol-naming is
inside the code (but if it's as bad as the device driver universe, then
it's pretty bad).
Of course if you look at the various sbin/mount_* directories in the
user-land source:
With "fs" suffix: Without "fs" suffix:
================= ====================
mount_ex2fs mount_ados
mount_ffs mount_cd9660
mount_kernfs mount_fdesc
mount_lfs mount_filecore
mount_nfs mount_msdos
mount_procfs mount_null
mount_portal
mount_umap
mount_union
you get a slightly different perspective.
It's a bit harder to change user-land, and indeed there's less need to
change user-land because it's much more obvious that you're dealing with
a filesystem when you use the "fsck", "newfs", "mount", etc. commands.
(I say only a "bit harder" because users rarely use the
<operation>_<type> commands directly.
If I were in my own universe and cared not for history I'd rename all
the directories and symbol names in the kernel to have the "fs" suffix
(and I'd probably shuffle some of the fs directories around and perhaps
make a sys/fs sub-directory to put them all in), and I'd probably also
add the "fs" suffix to the the filesystem type-name suffixes on the
user-land tools (and I'd move all the <operation>_<fstype> commands into
a directory that's not normally in any user's path (i.e. not even in
root's path), such as /libexec.
But then I'm very draconian about naming conventions and directory
organization.... Unix has always been a bit quirky and inconsistent,
something that comes naturally to a system developed by a community in a
primarily consensus-driven manner.
[Yes the "fs" suffix is redundant when you know you're dealing with
filesystems, but "mount_n" and "mount_f", etc. or "fs/n" and "fs/f" look
pretty funny, at least to me.]
--
Greg A. Woods
+1 416 218-0098 VE3TCP <gwoods@acm.org> <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>