Re: [9fans] long long whining

Re: [9fans] long long whining

[anothy]

// Also, heard of Hurd?

i can't tell if this is a rhetorical question or not, but if not, it's quite
funny (unless the name is just an amazing coincidence...). see:
	http://www.gnu.org/software/hurd/docs.html
in particular, the name attached to the first paper.
ア

Re: [9fans] long long whining

[James A. Robinson]

On skimming the article, he says that on one hand any BSD developer is
going to tell you that a transparent ftp client doesn't belong in the
kernel. He writes "in a sense, this is correct," but then proceeds to
argue that such "design methodology (which is based on preventing users
from changing things they don't like) is being used to prevent system
designers from making things better."  I don't follow what he's trying
to say.

How is a user level transparent ftp client going to be worse than a
system level transparent ftp client?  The per-process namespace that the
Plan 9 folks developed made transparent ftp clients and other services
very flexible.  I'm not clear what the argument for such services in a
micro kernel would be.  But this is probably too off-topic, and rob is
going to yell at us soon.


Jim

Re: [9fans] long long whining

[Thomas Bushnell, BSG]

jim.robinson@stanford.edu (James A. Robinson) writes:

> On skimming the article, he says that on one hand any BSD developer is
> going to tell you that a transparent ftp client doesn't belong in the
> kernel. He writes "in a sense, this is correct," but then proceeds to
> argue that such "design methodology (which is based on preventing users
> from changing things they don't like) is being used to prevent system
> designers from making things better."  I don't follow what he's trying
> to say.

That BSD will probably never have a transparent ftp client.

> How is a user level transparent ftp client going to be worse than a
> system level transparent ftp client?  The per-process namespace that the
> Plan 9 folks developed made transparent ftp clients and other services
> very flexible.  I'm not clear what the argument for such services in a
> micro kernel would be.  But this is probably too off-topic, and rob is
> going to yell at us soon.

I have no objection to the Plan 9 way, certainly.  I never said a user
level transparent FTP client is worse than a system level one.  

Putting one in the *kernel* is architecturally bad.

For Plan 9, and the Hurd, this is still true, but we have a perfectly
good way (a better way!) than putting it in the kernel.

At the time that article was written, BSD had no such way.

Thomas

Re: [9fans] long long whining

[Boyd Roberts]

"Thomas Bushnell, BSG" wrote:
> At the time that article was written, BSD had no such way.

I read 1996.  4.3BSD had the DEADFS and ample functionality
to do a file-system switch (well, they told us 'we already
got one... it's very nice') as early as 1991-1992.

Re: [9fans] long long whining

[Thomas Bushnell, BSG]

boyd@strakt.com (Boyd Roberts) writes:

> "Thomas Bushnell, BSG" wrote:
> > At the time that article was written, BSD had no such way.
> 
> I read 1996.  4.3BSD had the DEADFS and ample functionality
> to do a file-system switch (well, they told us 'we already
> got one... it's very nice') as early as 1991-1992.

BSD had the ability to put one *in* the kernel.  Not out of, except by
hacks like using the NFS protocol.

Thomas

Q [Re: [9fans] long long whining]

[ozan s yigit]

"Thomas Bushnell, BSG" <tb+usenet@becket.net> writes:

> For Plan 9, and the Hurd, this is still true, but we have a perfectly
> good way (a better way!) than putting it in the kernel.

since you mention both in the same breath, i'm curious what ideas, if any
you borrow from plan9 for the current(?) version. i see some bits and
pieces of work from roland mcgrath, but no indication of any important
architectural work. anything happened since 1.2 that in any way
relates to plan9?

oz
-- 
One o'clock pee em! Hello, Insert Name Here! -- The Dis-organizer

Re: Q [Re: [9fans] long long whining]

[Thomas Bushnell, BSG]

ozan s yigit <oz@blue.cs.yorku.ca> writes:

> "Thomas Bushnell, BSG" <tb+usenet@becket.net> writes:
> 
> > For Plan 9, and the Hurd, this is still true, but we have a perfectly
> > good way (a better way!) than putting it in the kernel.
> 
> since you mention both in the same breath, i'm curious what ideas, if any
> you borrow from plan9 for the current(?) version. i see some bits and
> pieces of work from roland mcgrath, but no indication of any important
> architectural work. anything happened since 1.2 that in any way
> relates to plan9?

The basic architecture is pretty much fixed.  There wasn't a great
deal of conscious borrowing from Plan 9 in designing the architecture,
but both share the property of putting filesystems outside the kernel
and making them able to easily support all the various file system
operations.

I first thought of these ideas when I was working for UNM, in
conversations with Lee Ward there, in the late 1980s.  I don't know
whether he had been familiar with Plan 9 [or other work at Bell Labs]
or not, but I was not at the time.

Thomas

Re: [9fans] long long whining

[Thomas Bushnell, BSG]

forsyth@caldo.demon.co.uk writes:

> that's not what i meant.  given the existence of the vnode switch,
> and the primitives contained therein, there is no big obstacle to
> writing a vfs/vnode implementation that converts the operations
> within its scope into messages on a file descriptor in much the same
> way as network file systems were implemented by 8th edition, or by
> me in my re-ported Unix kernels for the VAX and Sun-3, using a
> simpler file system switch.  do just one of those and you can have
> as many things as you like outside the kernel.

There's already in BSD a "portal" concept, though it had not been
implemented when I wrote the essay in question.

Once you've done that, however, you are no longer a monolithic kernel,
however.  That's kinda the point of the essay.

Re: [9fans] long long whining

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 671 bytes --]

that's not what i meant.  given the existence of the vnode switch, and the
primitives contained therein, there is no big obstacle to writing a vfs/vnode implementation
that converts the operations within its scope into messages on a file
descriptor in much the same way as network file systems were implemented
by 8th edition, or by me in my re-ported Unix kernels for the VAX and Sun-3,
using a simpler file system switch.    do just one of those and you can have as many
things as you like outside the kernel. 

having said that, it's hard to know quite where to start with some of these
new improved switches.  whew!   just look at the size of that one...


[-- Attachment #2: Type: message/rfc822, Size: 2247 bytes --]

To: 9fans@cse.psu.edu
Subject: Re: [9fans] long long whining
Date: Mon, 25 Mar 2002 16:21:49 GMT
Message-ID: <874rj43aiq.fsf@becket.becket.net>

forsyth@caldo.demon.co.uk writes:

> you could use their vnode switch--possibly called something else but
> i haven't got the old documentation here--to convert file system
> operations to messages in much the same way as the 8th edition file
> system switch, although the latter was simpler.  (the RFS changes
> were something else again.)  NFS wasn't completely useful because
> the close wasn't transmitted.

BSD doesn't *have* messages.  Eventualy, you end up implementing Mach
2 when you add them.

The vnode switch is purely *internal* to the kernel, which is the
point of that paragraph.

Thomas

Re: [9fans] long long whining

[Thomas Bushnell, BSG]

forsyth@caldo.demon.co.uk writes:

> you could use their vnode switch--possibly called something else but
> i haven't got the old documentation here--to convert file system
> operations to messages in much the same way as the 8th edition file
> system switch, although the latter was simpler.  (the RFS changes
> were something else again.)  NFS wasn't completely useful because
> the close wasn't transmitted.

BSD doesn't *have* messages.  Eventualy, you end up implementing Mach
2 when you add them.

The vnode switch is purely *internal* to the kernel, which is the
point of that paragraph.

Thomas

Re: [9fans] long long whining

[Boyd Roberts]

forsyth@caldo.demon.co.uk wrote:
> ... --to convert file system operations
> to messages in much the same way as the 8th edition file system switch,
> although the latter was simpler.

Depended on what you wanted to do.  For some things it was well
suited, but for others it wasn't.  I had to cheat to add in
512 byte block file-systems to it.  It turned out that the
file-system switch was not useful, in this case, but it certainly
was simple.

Fortunately, I had enough spare bits to do a bitfs style hack
on the dev_t.

RFS & NFS:  The less said the better.

RFS was way to complex and a dreadful piece of design/engineering.

NFS's lack of close and the non-idempotency of some operations are
real killers.

I did a NFS version of ftpfs -- seriously unpleasant.

Re: [9fans] long long whining

[forsyth]

[-- Attachment #1: Type: text/plain, Size: 370 bytes --]

you could use their vnode switch--possibly called something else
but i haven't got the old documentation here--to convert file system operations
to messages in much the same way as the 8th edition file system switch,
although the latter was simpler.  (the RFS changes were something else again.)
NFS wasn't completely useful because the close wasn't transmitted.


[-- Attachment #2: Type: message/rfc822, Size: 2101 bytes --]

To: 9fans@cse.psu.edu
Subject: Re: [9fans] long long whining
Date: Mon, 25 Mar 2002 09:57:35 GMT
Message-ID: <87n0x0h1uw.fsf@becket.becket.net>

boyd@strakt.com (Boyd Roberts) writes:

> "Thomas Bushnell, BSG" wrote:
> > At the time that article was written, BSD had no such way.
> 
> I read 1996.  4.3BSD had the DEADFS and ample functionality
> to do a file-system switch (well, they told us 'we already
> got one... it's very nice') as early as 1991-1992.

BSD had the ability to put one *in* the kernel.  Not out of, except by
hacks like using the NFS protocol.

Thomas

Re: [9fans] long long whining

[Dan Cross]

In article <0a60f5849450f1b29f8f17584cbcb7eb@plan9.bell-labs.com> you write:
>Leave Thomas alone.  Someone has to whack the hornets
>nest with a branch every now and again just to
>remind the hornets that they're still alive.

I live in New York City.  I've had enough whacking for a while.

	- Dan C.

Re: [9fans] long long whining

[presotto]

[-- Attachment #1: Type: text/plain, Size: 147 bytes --]

Leave Thomas alone.  Someone has to whack the hornets
nest with a branch every now and again just to
remind the hornets that they're still alive.

[-- Attachment #2: Type: message/rfc822, Size: 1831 bytes --]

From: Dan Cross <cross@math.psu.edu>
To: 9fans@cse.psu.edu
Subject: Re: [9fans] long long whining
Date: Fri, 22 Mar 2002 13:37:47 -0500
Message-ID: <200203221837.NAA14221@math.psu.edu>

> Support != interest
> Support != desire to learn about

That's true, but there's a fine line between interest, desire to learn,
and antagonism.  And while you *say* you're not trying to be
antagonistic, that's started to become unbelievable.

Okay, I'm done talking about it; I have little hope that anything
productive would come out of the discussion.

	- Dan C.

Re: [9fans] long long whining

[Dan Cross]

> Support != interest
> Support != desire to learn about

That's true, but there's a fine line between interest, desire to learn,
and antagonism.  And while you *say* you're not trying to be
antagonistic, that's started to become unbelievable.

Okay, I'm done talking about it; I have little hope that anything
productive would come out of the discussion.

	- Dan C.

Re: [9fans] long long whining

[Thomas Bushnell, BSG]

dhog@plan9.bell-labs.com (David Gordon Hogan) writes:

> tb+usenet@becket.net writes:
> > Well, I don't support non-free systems anymore (where "free" has the
> > definition used by the FSF or the Debian Project).  
> 
> Ummm, why are you here?

Support != interest
Support != desire to learn about

thomas

Re: [9fans] long long whining

[Mike Haertel]

>This is a great suggestion. Perhaps Mr. Bushnell would find it
>more rewarding than banging heads here.  Also, heard of Hurd?

Thomas Bushnell is one of the principal authors of the Hurd.

Re: [9fans] long long whining

[skipt]

This is a great suggestion. Perhaps Mr. Bushnell would find it
more rewarding than banging heads here.  Also, heard of Hurd?

BTW, Andy Valencia (a very nice guy) was kind enough to demo VSTa to me a
few months back, complete with a port of MGR for a window system.  It was
complete and very functional. 

At 06:58 PM 3/21/2002 +0100, Fco.J.Ballesteros wrote:
>:  I don't know, but I was thinking, quite seriously, that Bushnell might
>:  be more interested in the VSTa project, which has a license he should
>:  like and uses GCC as it's compiler, while trying to do some of the same
>:  things as Plan 9.  http://www.vsta.org/
>
>And has it's own mailing list as well.
>
>The vsta kernel source is quite nice, btw.

Re: [9fans] long long whining

[Fco.J.Ballesteros]

:  I don't know, but I was thinking, quite seriously, that Bushnell might
:  be more interested in the VSTa project, which has a license he should
:  like and uses GCC as it's compiler, while trying to do some of the same
:  things as Plan 9.  http://www.vsta.org/

And has it's own mailing list as well.

The vsta kernel source is quite nice, btw.

Re: [9fans] long long whining

[Dan Cross]

> Ummm, why are you here?

I don't know, but I was thinking, quite seriously, that Bushnell might
be more interested in the VSTa project, which has a license he should
like and uses GCC as it's compiler, while trying to do some of the same
things as Plan 9.  http://www.vsta.org/

	- Dan C.

Re: [9fans] long long whining

[David Gordon Hogan]

tb+usenet@becket.net writes:
> Well, I don't support non-free systems anymore (where "free" has the
> definition used by the FSF or the Debian Project).  

Ummm, why are you here?

---
Elsewhere,
ggm@apnic.net writes:
> I would be regrettable if we all started discussing personal anguish
> of various s/w peoples (sex or non-sex) lives on email wouldn't it?

Now there's a non-sequitur.

Perhaps we could all sublimate our (alleged) anguish and be
catty and obnoxious to each other instead?

One should be aware that alternate sources of anguish may
be present, for instance the presence of vast fields of Negative
Elegance, propagating virally, disguising themselves as `software'.
(Shades of Wilhelm Reich ;-).

---
2 1/2 weeks away, and ~400 messages to read, mostly 9fans
mail.  Could be worse, I guess...