[this is getting a bit afield of the original intent...]

Thus spake Peter Seebach ("PS> ") sometime Today...

PS> Right.  I was just toying with the idea that specifying "size" of a "plain
PS> file" is no weirder than specifying "major and minor number" of a "device
PS> special".  It's a LOT like mknod.  I want a file of a given type with the
PS> following trivially-specified attributes...

...but mknod(2) would need to support it.  Specifying "size" may not appear
weirder than specifying major/minor, but in fact when you specify major/minor,
those are properties germane to devices, and that information gets passed
in to mknod(2) and applied directly to the node at creation time.  Currently
we can't do that with size (or, incidentally, uid, gid, atime or mtime).
The set of attributes directly controlled by mknod(2) is very small
(smaller than it used to be -- I remember being able to mknod(2) a directory
(which had only one link, no ".", no "..", and it wasn't useful compared
to mkdir(2)).

Until relatively recently, mknod(8) didn't unlink existing files.

mknod(2) also wouldn't handle actual zero-filling of the file if that
was what was needed.

				--*greywolf;
--
NetBSD: It spanks the knickers off those other operating systems.