Re: [9fans] Re: 9fans Digest, Vol 27, Issue 52

[John Pritchard <jdp@syntelos.com>]

[-- Attachment #1: Type: text/plain, Size: 3843 bytes --]


Bakul Shah wrote:
> This thread has me wondering....  Could it be that Plan 9's
> "everything is a file" model works extremely well in a narrow
> band of designs but does not offer much of an advantage for
> larger scale designs?  May be one needs so many other things
> that the advantage offered by plan 9 gets lost in the noise.
>
> Or is it that no one has pushed use of this model hard enough
> for things like browsers, compilers and such?  And may be
> what is needed are Plan 9 analogues of books like The Unix
> Programming Environment, The Unix Network Programming, The
> Haskell School of Expression, Design Patterns, Structure and
> Interpretation of Computer Programs and so on?
>
> Or may be its way is most suited for some future where
> computers as public transport for a variety of unrelated
> programs are a thing of the past?
>
>
>

Hi,

Can't resist to write on this subject...

Consider this...

With Unix the OS was created in a much less complex world where time and
space were orders of magnitude less expansive.  As the computing world
evolved through the 80's and 90's this OS (Unix, the founding world view
for many of us) came to need "application layer" software (libs,
interpreters, etc) to bring high level symbolic primitives to
application development.  An interesting example may be the Java VM in
that a fat layer like that would have been entirely unfathomable in 1970
and even in 1980 (Objects seem to be an appendage to the Oak instruction
set), but today this level of abstraction from the system is good for a
lot of application development (performance aside).

Inferno's Dis (Limbo) is on the same level of abstraction (but different
of course), while Plan9's Alef is another approach.

The Plan9 and Inferno OSes have sought to provide clear application
platforms in terms of system model and development language for this
networked world.  But it's easy to go too far in pushing the system
model on applications.  These modern OSes don't have the need for
application layers like Unix/C did, but developing in oneself the
experience of interesting application architectures over these system
architectures has to begin somewhere -- and so it begins with our
current concepts.

Ideally one gets to specialize in an application domain where the
interesting problems in an application domain are studied and become
known.  Then finding intersections between system architectures and
application architectures for more compelling applications is the art
one may spend decades developing.

If we were to formulate mozilla for an OS like Plan9 or Inferno, what
would be different in that software architecture?  One can either
descend into the endless void of great possibilities, or truncate that
space at a first stage (it's just an HTTP client) and then  start over
from scratch for a second (it's a world of multi-protocol clients).
Maintain sanity.

It's not for nothing that the world of software has gone through many
object and networked object models in the past two or three decades ---
one perspective.  Or if we go back to John Holland's Universal Spaces
(1960's, partially available in Arthur Burk's "Essays") we can see this
history from another perspective in which models are differentiated into
universal and specific (very interesting).

When is a universal model what an application wants, and when is it a
specific one.  Perhaps this question is interesting when it helps to
define "stage one" and "stage two" in our broad prototyping cycles.
Because models tend toward the universal in the scope of any one
(networked, multi) application, but are always compromised.  The term
Data Structures can evoke the opposing or bottom- up view of specific
models.

..book truncated here..


cheers

/j



[-- Attachment #2: Type: text/html, Size: 4323 bytes --]