I also make no claim to genius, but I have long felt that a system that
was based on ideas of relational database with a multiple descriptor
attributes that could be searched using a query language would be great
for the user. The present system has a purely hierarchical structure.
This structure is repeated almost everywhere, and in many cases whole
trees of directories are empty, but, apparently, must be present for the
correct functioning of some part of the support software. I don't have a
well formed suggestion for change, but it does seem that a review of
existing experience in the light of know principals of database design
would lead to a design that is much more accessable to the user. But I'm
a newbie and don't have knowledge of the existing experience. I guess
I'm offering moral support. At least moral support for continued
discussion. On the other hand, maybe there are people talking about
these issues on another list that Evan and I have not yet discovered.
Paul
Evan Prodromou wrote:
"AH" == Alexander Hvostov <[EMAIL PROTECTED]> writes:
AH> That should be a good start. Anyone care to comment?
Yes.
I'm no genius, but even to a dunderhead like me, it seems like the
Debian package mechanism is getting quite creaky with age. Something
that scaled well for a package catalog in the hundreds does terribly for
package catalogs in the tens of thousands. For example:
* Loading package catalogs takes a long time on loaded or
low-memory machines.
* Browsing package catalogs for useful packages is practically
impossible.
It seems like we need a revamp of the Debian catalog and package
format to allow at least the following:
* Sub-architectures. Allow some binary packages that are
CPU-bound to be optimized for particular chips. For example,
a Pentium II computer might be able to install packages
tagged for the i386/i686 architecture, the i386/i586
architecture, and the i386 architecture.
* Multiple versions of the same package. Instead of having 10
jillion python2.2, python2.1, and python1.5 library
packages, allow multiple binaries tagged with the
appropriate version.
* Hierarchical categories. We have <10 categories of software
in the Debian system right now. This is just laughable. We
need to be able to tag packages according to real useful
categories, like "Network/Internet/Clients/Chat/IRC" or
"Libraries/DataFormat/XML/Parser". This should make browsing
for appropriate software a lot easier than "apt-cache search
irc" or "apt-cache search xml".
* Optimizations of the catalog as stored at the leaf node
machine.
My main comment is this: where does this discussion go on? On
debian-devel? -private? Who do I talk to to get Debian working better
for me.
~ESP
