On Tue, 11 Feb 1997 14:01:19 PST,
 Jason Thorpe <thorpej@nas.nasa.gov> wrote:

>  > That's a bug I encountered when I started my POSIX work on NetBSD.
>  > In general, POSIX.1 section 2.4 (pathname resolution) states:
>  > "A null pathname is invalid.", so I simply added a short check:
> 
> ...I agree we should make this change... However, I'd like to also
> know how many programs this breaks :-/

That was my worry, too, but I've been running this change for ~10
days now and have yet to encounter problems in the NetBSD userland
related to this. Of course I'll try to make some clean-sweep to be sure.


>  > + 	if (!error && ndp->ni_pathlen == 1)
>  > + 		error = ENOENT ;
>  > + 
> 
> Hm, would a more obvious test be:
> 
> 	if (!error && cnp->cn_pnbuf[0] == '\0')
> 		error = ENOENT;

Yes, that was my first attempt... I changed over to check pathlen
just for the (possible ?) sake of performance, without readability
in mind. Well, "either way does it", and we should to "something".


-klaus