Re: [9fans] Re: plan 9

Re: [9fans] Re: plan 9

[Russ Cox]

Ron's comment makes me think I should have explained
myself a little more, and anyway I have a fun story.

The whole fad OO argument saddens me, since Plan 9 probably
pushes the real point of OO -- consistent and reused
interfaces -- farther than any other system.  The problem
with the current fad OO world is that there are hardly
any consistently-used interfaces, so you lose all the
potential reuse.  Plan 9 may only have one real interface
but we sure do reuse it a LOT.  And we really do have
many interfaces, such as the one presented to a cpu
server by a terminal (and by drawterm), or the one
presented to clients by kernel graphics drivers
(and by rio, and by vncs, and by drawterm), or the
authentication files presented by the 3e kernels
(and by authfs), or the auth files presented by
auth/factotum (and by auth/factotum, whenever you
care to reinvoke it!), and on and on.  I would very much
like to hear about any systems that are more object
oriented.

I got an iPod a few days ago, one of the spiffy new
Windows ones that have a FAT32 file system.  It came
with MusicMatch Jukebox, which was exhibiting some
bugs in the actual downloading to the iPod.  There
were various people who'd written software for
Windows to talk to the Mac iPod (which has an HFS+
file system), software that was supposed to be quite
good, and I mourned the fact that I couldn't use it.
Then I remembered that all the Windows software for
talking with the Mac iPod was layered on top of a
general Mac file system driver, so that the
iPod was actually mounted by Windows and manipulated
via file system operations.  Well, I reasoned, my iPod
is already mounted, so I'll install the iPod software,
ignore the Mac FS drivers, and point the software at
my already-mounted FAT32 iPod.  It worked the first
time.

Imagine if everything behaved like that, presenting
good interfaces so that only the interface rather than
the actual details of the implementation mattered.
You'd have Plan 9.

Russ

Re: [9fans] OO plan 9

[matt]

I would contend that the OO concepts of polymorphism, inheritance can be
expressed through mount and bind.

It would be farily trivial to write class factories with rc.

It just so happens that most of the supplied 'objects' are application
sized.

Besides OO is just a struct with a pointer to itself and all of it's
functions, the rest is just syntax and macros ;)

m


---
Outgoing mail is certified as a Virus.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.384 / Virus Database: 216 - Release Date: 21/08/2002

Re: [9fans] Re: plan 9

[Eric Dorman]

Russ Cox wrote:

>Ron's comment makes me think I should have explained
>myself a little more, and anyway I have a fun story.
>
>The whole fad OO argument saddens me, since Plan 9 probably
>pushes the real point of OO -- consistent and reused
>interfaces -- farther than any other system.  The problem
>with the current fad OO world is that there are hardly
>any consistently-used interfaces, so you lose all the
>potential reuse.  Plan 9 may only have one real interface
>but we sure do reuse it a LOT.  And we really do have
>many interfaces, such as the one presented to a cpu
>server by a terminal (and by drawterm), or the one
>presented to clients by kernel graphics drivers
>(and by rio, and by vncs, and by drawterm), or the
>authentication files presented by the 3e kernels
>(and by authfs), or the auth files presented by
>auth/factotum (and by auth/factotum, whenever you
>care to reinvoke it!), and on and on.  I would very much
>like to hear about any systems that are more object
>oriented.
>[xxx]
>Imagine if everything behaved like that, presenting
>good interfaces so that only the interface rather than
>the actual details of the implementation mattered.
>You'd have Plan 9.
>
>Russ
>
This all is precisely why I've pretty much setttled on a 9p2000/virtual
mount
architecture for my high-performance computing stuff rather than some
distributed-object protocol.  Avoid CORBA bloat and still be able to
rendezvous in Java, Smalltalk, C and C++.

Now if I could only run Plan9 on HPPA-MP, I'd be set :)

Eric Dorman