Re: rm patch suggestion

[Roland McGrath]

> I'm just not sure.  It creates one inconsistency in a rather core area
> as a way of saving an inconsistency in another area.

Well, the consistency of "." and ".." with other names in path name
translation is not what I'd call pure to begin with.  I did point out that
the change would be consistent with the treatment of ".." at the root of a
filesystem.  But that consistency is not really my argument in favor of the
change.  The argument in favor of the change is the principle of least
astonishment, and the venerable principle (one of my favorites) of "how the
hell else would you do that?"  The "that" being to sensically access an
underlying directory tree that you'd either opened with O_NOTRANS, or
gotten into at some point before its translator was set.  I just bet that
you too never thought that "cd ." would be starting a translator and moving
you into a different filesystem, or that "cd .." should naturally do the
same when done inside a filesystem but something different at the root of a
mount point.  You think some screwy things sometimes, but I don't think you
were planning on that.