> > No, because POSIX doesn't mandate it.  I just checked my copy of the
> > standard (ISO/IEC 9945-1; IEEE 1003.1-1990), and it seems that the
> > only place "symbolic link" is mentioned in the normative section is in
> > the descriptions of the tar and cpio archive format.
> 
> As i've siad before, to others, in private mail:
> 
> The behaviour that is implemented in 4BSD (and NetBSD) is described in
> a draft of an amendment to .1 (i think).  It's not yet a "real"
> standard, but it's likely that it will be.

The actual document involved is the draft for the 1003.1a amendment. This
is the amendment that, among other things, adds readlink(), symlink(),
lstat(), and fixes up all the errno lists to deal with symbolic links in
pathnames. 

Now what was it we were discussing? I think the implemented behavior 
conforms to the draft specifications. I am not sure what in the traditional
implementation was not conforming. I suspect though that there are other
factors outside of conformance that provide a motivation for this
particular change. Just guessing here, because I wasn't privy to the
design discussions.

/sjd