> Are we absolutely sure there's no better solution?

I'm not.

> Where, if anywhere, will this stop?

Only time will tell :-)

> I distrust the additional magic, mostly because it breaks the rule
> that anything but nul and '/' can be in a name.

So don't mount with MAGICLINKS, and the only cost is a couple of
instructions per component when following a symlink.  This does not
seem excessive to me, though I gather you disagree.

Actually, some of you may recall a while ago when I and someone else
(graywolf, I think) were babbling about symlinks, back when unowned
symlinks were a hot topic.  At the time, I proposed semantics for the
06777 mode bits on a symlink, but had nothing for the 01000 bit.  This
seems like a perfect example of a problem in search of a solution
finding a solution in search of a problem: if the 01000 bit is set on a
symlink, the contents of the link are not a simple pathname, but rather
something more complicated, such as a keyword, or a pathname with
embedded keywords, or some such.  (Just please pick some extensible
mechanism, so we don't wind up stuck with something like ANSI C's # and
## operators, which have no extension path...I'd much rather have seen
#string and #concat operators.)

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu