On December 9, 2004 [sic], Sergey Reva wrote:
> term%: pwd
> term%: /n/dump/2004/xxxx
> term%: lc
> xxxxxxxxxxxxxx
> xxxxxxxxxxxxxx
> xxxxxxxxxxxxxx
> term%: unmount /n/dump                  # how I can do this?
>                                         # I mean, Linux in this case
>                                         # say: ... busy ...

Linux is broken.  Not being able to unmount a file system
(thus stopping people from accessing it in the future)
shouldn't be held up by one or two processes that have 
open files on that file system.  It's not hard to get the reference
counts right, and Plan 9 does.  The 9P connection corresponding
to /n/dump won't be dropped until all the references go away,
but it can be removed from the name space at any time.

> term%: cd ..
>                                         # now I in /n/dump/2004
> term%: cd ..
>                                         # now I in /n/dump
> term%: cd ..
>                                         # now I in /n
> term%: lc                               # but, what I see?
> active archive snapshot
> term%:                                  # now I can browse all subfolders

Plan 9 is broken too.  The implementation of dotdot consults
the name space to answer the question "I just walked ..; what 
directory am I in?", but you removed the necessary entry
from the name space!

This is a nice example why the dotdot algorithm isn't quite right.
(I have previously posted much more complicated and less
understandable examples of the same.)  Thanks.  I missed this
email the first time it went by (I was out of town and only skimming
mail), but it was a joy to discover, even nine months late.

There is a new dotdot algorithm in the works.
Russ

Reply via email to