From mboxrd@z Thu Jan  1 00:00:00 1970
Date: Fri, 29 Aug 1997 10:34:16 -0500
From: Brandon Black photon@nol.net
Subject: [9fans] Re: Anyone still running plan9?
Topicbox-Message-UUID: 60828120-eac8-11e9-9e20-41e7f4b1d025
Message-ID: <19970829153416.loJfziEz68W91ogixsohnqnxE40f7bDqelH_uWkbHGk@z>



In response yo this all the other responses that went like "Much too
difficult, you must not have looked at this before", I have looked.  It
would be very painful.  Implementing per-process namespace would only be a
third of the work, compared to hacking the rest of the system to make it
continue to work (like the security example).  You'd have to change the
whole security paradigm.

However, I disagree about the effort/benefit ratio.  I see every day in
our Unix development environment where I work how much benefit we would
gain if our commercial unices had per-process inherited namespaces.  If
you develop C/C++ in a multi-platform unix environment.. and use all of
the GNu tools, and DejaGNu testing suite, and database clients, and
syb/ora-tcl, and perl, and RogueWave, and the list goes one..... And
different projects (sharing generic development boxes) want to use
different versions of all these tools, and might be compiling to different
targets.... And throw in the use of IRMS as an intelligent source code
repository that knows about your tools and generates your makefiles, and
runs your nightly builds and tests....

In this situation.. everything we do is a major hack, and it would all be
clean and perfect if we had Plan9's namespacing to work with.

Brandon

On Wed, 27 Aug 1997, Boyd Roberts wrote:

>     From: Brandon Black <photon@nol.net>
>
>     It really irks me a little that so many people want *nix ports of Plan 9's
>     window manager, or editor, etc... but nobody wants to take the really
>     great ideas.  WHy won't somebody take things like the namespacing stuff,
>     the 9P and IL protocols, the fileserver/cpuserver/terminal concepts...
>
> i would have loved to have put 9p into ultrix several years ago, but
> decided that it was just not worth the pain.  anyone who's keen on
> retrofitting cool stuff into modern unix kernels either a) hasn't seen
> one, b) doesn't understand the problem, or c) has too much time on
> their hands (cue denis leary).  maybe all three.
>
> btw:  i class that ghastly mess, known as linux, as a 'modern unix kernel'.
>       you may have the code to it, but have you actually read it?
>
> bbtw: ultrix supported nfs mounts by mortals, so i decided to implement
>       ftpfs with nfs.  it just wasn't worth the effort.  i wound up with
>       a sort of neat toy, but useless.  i spent several weeks on it, much
>       of it wasted trying to understand why nfs did what it did, trying
>       to turn it into some sort of reliable tool.  i was no stranger to
>       nfs, having ported it back in '86.  but how it actually behaved
>       was pretty sad.
>