On Sat, Feb 16, 2002 at 09:42:26AM +0000, Andy Wardley wrote:
 
> It should allow you to remove the boundaries between directories, files
> and file contents.  All file systems suck.  They must die.

I'm not that revolutionary as you here :).  The filesystem is quite
basic structure and nobody will change it if it works well enough.  And
I really believe it works well enough for most of the tasks.  And I
really think it is basic - so the modifications need much training
ground.  And we can supply some training ground by implementing library
abstraction layer.
I've already coded such layer for TT - perhaps a bit crude (I'm beginner
to object oriented perl), but quite simple and I believe useful for
templating.
Generally you can write: 
[% for article in directory.files %]
[% for section in article.sections %]
[% section.title %]
[% end %]
[% end %]
Where the directory is a path in the filesystem with '.' substituted for
'/' and 'par' (like parent) for '..'.

For my current goals this is enough, but there are many ways to
improve it and if you would like to incorporate it into something more
general I promise to be cooperative. 

> And no more hierachies!  Why should we be forced to think in such unnatural
> ways?  The property based path is much preferable, IMHO, to hierarchical
> filesystem/xpath like paths.  e.g.
> 
>     /module=FooBar::Blam/language=Perl/author
> 
>     /language=Perl/module=FooBar::Blam/author
> 
> or for another example:
> 
>    /appointments/year=2002/month=feb/day=16/time=1300-1900
> 
>    /date=today/afternoon/appointments
> 
I don't get it - for me it is still a hierarchy (you can even take your
examples literary and create such directories with '=' sings in the
names).  But actually for some purposes even more elaborate schemas for
addressing might be usefull.  And when you think abouth the most general
abstraction - you come to the conclusion that the storage can be just
any object that you provide with some persistency - and you are lost in
possibillities.  You can wade through them only by experimentation.
Perhaps HURD would be good for that.

Zbigniew Lukasiak


Reply via email to