>...
>> -----Original Message-----
>> From: tech-kern-owner@NetBSD.org [mailto:tech-kern-owner@NetBSD.org]On
>> Behalf Of Daniel Carosone
>> Sent: Wednesday, June 22, 2005 11:27 PM
>> To: Jason Thorpe
>> Cc: YAMAMOTO Takashi; thorpej@netbsd.org; tech-kern@netbsd.org
>> Subject: Re: CVS commit: src
>> 
>> 
>> On Wed, Jun 22, 2005 at 10:35:26PM -0700, Jason Thorpe wrote:
>> > On Jun 22, 2005, at 8:20 PM, YAMAMOTO Takashi wrote:
>> > 
>> > >is there any os which uses the compatible semantics with this?
>> > >why don't you follow the existing conventions like tru64 cdsl or
>> > >dragonfly varsym?
>> > 
>> > I will look at varsym.  Dragonfly's "where to fetch our sources"  
>> > instructions leave a lot to be desired.
>> > 
>> > Tru64 CDSL, according to the documentation I read, seemed to be  
>> > pretty lame.
>> 
>> Apollo Domain/OS and (iirc) Amoeba both have prior art here too.  I'm
>> not sure that compatibility with either is a major priority, though :)
>
>Take this with a grain of salt.....
>
>I saw a demo of AFS (I believe) that had conditional
>symbolic links. The selectors were actually
>part of the proc rather than system wide.
>
>This allowed an AFS file server to, for example, have bin/
>and lib/ directories for a variety of architectures. The
>workstations could be very different yet use the same
>file server for executables. It was something like
>this. A conditional symlink "bin" pointed to bin.i386/,
>bin.68k/, bin.vax/, etc. The process on the workstation
>contained the selector for which actual link to follow.
>
>That's about all I remember. It was a brief demo.
>They did other tricky stuff not related to CPU
>architecture, but I don't remember that clearly.
>
>How do these magic symlink things work with NFS?
>
>Regards,
>	-gww
>

	That was indeed a version of AFS from the original Andrew Project
at CMU.  I don't believe it was every generally released and was intended
to be part of the TCF initiative from IBM which incorporated the LOCUS
features (funded by the now defunct ACIS group of IBM in Palo Alto, who
also paid for most of the Mach development at CMU).  It did work under
both the BSD4.3 port for the RT and the pre-release version of AIX/PS2
(but was removed from the product before ship - there may also have been
some pre-release AIX3 version).  It also was in heavy use at CMU running
on Sun 50s or 60s (the on 68020 boxes) running on many OSs, but primarily
BSD4.3 and early Mach3 (but not Mach2).


	Paul Shupak
	track@plectere.com