Re: [9fans] Typesetting

[Steve Kilbane <steve@whitecrow.demon.co.uk>]

> ...But how the data is used might give insight into the most efficient
> way to transmit it.  For instance, it wouldn't be efficient, in terms
> of computational resources or in terms of people's time, to use file
> transfer protocols in distributed computing.

That depends on the file transfer protocol. Microsoft's aren't great
in this area. :-)

> Ie, get my home directory
> via FTP each time I want to access it.

That's application-specific. Literally - if you want to get the whole
directory every time, then maybe. Like you say - how the data is
used has an effect. File systems sevve applications that want to
access their contents as filesystems. Data streaming is different.

> I think it's largely the same
> thing with the web; in particular, I want to use the pantheon of useful
> tools that currently exists on my system for dealing with disk files to
> deal with data on the web.

As I've already mentioned - it would be nice, but most of those tools
assume a file system and files containing text.


> I mean categorial structure.  Data on the web is, for all practical
> purposes, completely random.

Ah. Sigh. Again, not only is it effectively impossible to break the
data down, but adding an ordering is even further. The ordering required
varies according to how you want to view the data. The data no longer
belongs to category X, but to category X with a given attribute.


> >But that's never going to change, at least until Microsoft (or any other
> >company) achieve their goal of being the universal platform. While there
> >is heterogeny, you'll need some way to insulate the data from the destination
> >platform's weirdness.
> 
> Yes, but text, presented in a standard encoding, isolates me rather
> well.  There's nothing inherent in the browser model which makes it the
> only tool or the best tool for cross platform data sharing.

Except that it's effectively ubiquitous. Which is a large part of the
problem.


> Well, I disagree that it simplifies something which is already simple.
> Consider creating ``sessions'' over HTTP as an example; I think the
> methods to do this are ad hoc and complex, if not entirely broken.
> Using file semantics makes this much easier.

[ open a file to do a session ]

That's not quite the same thing. Opening a connection to a file *server*
is one thing, while opening a file is something else (and much simpler,
because the server connection has been done). Again, it depends on the
protocol. 9P works much better in this circumstance than most, because
it's designed to operate over a pipe regardless of the medium, and it
doesn't assume the other end is a Real Disk. I wouldn't call 9P authentication
simpler than cookies, though - just *much* better defined.

> Well, Yahoo! and companies like Yahoo! have (until recently... :-) made
> a lot of money by organizing web content into a ``hierarchy.'' Sure,
> not everyone would be happy, but not everyone is happy with the
> filesystem layout of modern operating systems.

Indeed, but pay attention to who's doing what: users on Yahoo are storing
data references in certain categories. There's nothing inherent in the
data itself that causes that, or that supports it. It's an entirely
subjective viewpoint. Moreover, the categorisation is stored separately
from the data itself, and it's a web, not a tree. I don't see how putting
a filesystem on top of that gives you anything different.

> My point is more, do I want the web to be an online brochure (or maybe
> data sheet, whatever), or do I want it to be something that I can
> interact with?
> 
> People seem bent on making the thing interactive, even though it wasn't
> built with that in mind.

That's true. That's because they're trying to attract your attention, to
make you buy something. The old marketing approach of using style to
disguise lack of substance.


> No, I appreciate the comments....  Indeed the problem is difficult.  I
> think that the general issues are solveable, though.  Oh well, time
> will tell....

I don't know that the general issues are solvable, given that they contain
such a high level of subjectivity, but I do think that individual components
of the whole system can be improved. For example, while a hierarchy can't
support all the categorisation necessary, Plan 9 can at least offer multiple
categorisations in the form of multiple hierarchies onto the same data, at
a cheaper cost than most systems. This isn't necessarily a good thing, though:
consider the connection server and dns on Plan 9, as opposited to just
making the whole internet a tree onto the dns.

steve