Re: ebuilds in XML

Heya Gerk, 
I wanted to point out some parallel thoughts I've been writing on the wiki.
I'd appreciate a chance to chat with you to give us a chance to reconcile
and perhaps better synthesize thoughts... You are alluding to new
architecture thoughts that I'd like to examine in more detail--any chance
of a link to a wiki page or forum post?  I have some anxiety that
competing/complementary ideas will not make it into a draft that is "born
at 1.0", right?

Here is what I have so far on my wiki page that I'd enjoy discussing (I
linked this in a previous post, but to spur actual discussion, I reprint
here) 
http://wiki.zynot.org/zynot/moin.cgi/PackageFormatIdeas
====

== Goals ==

 1. Be easily hand-editable but also be machine-parseable for 
  1. running scripts (e.g. the install script portion of a package)
  2. dependency graphing and compatibility checking
  3. statistical comparison, other uses of package metadata
 2. Be forward-compatible. Wha?!? Well, an old installer should be able to
cope with newer packages, and it should deal with non-critical metadata
newly added to packages gracefully. This is related to package versioing
but is different.
 3. Be versioned and define a standard naming convention that is human and
machine readable
 4. Be easily extensible for dependency and optional use information. This
is known in the gentoo world as the 'per-package USE variable' problem. A
package should describe what it Needs and what it can Use, in addition to
what it provides 
 5. A corrolary to dependency information is finer granularity in dependency
information; e.g. arbitrary-precision dependency graphs expressed per
package rather than more the simple 'depends on >= foo-4.0'
 
Gerk, the main thing I want to figure out before it gets too late is whether
rdf can be used and can fulfill some difficult objectives; e.g. points 2,
3, 4, 5 above (and 1.3). 

Chris


mark guertin wrote:

> Heya
> 
> I have also been toying with this sort of thing exactly, and i think we
> will have some more detailed specs coming really soon, it's a very
> similar approach, but there are quite a bit more things that will have
> to be taken into account for the metadata to be a proper setup...
> 
> i.e.
> 1) will have to track arches
> 2) will have to track exact pkg revision numbering from server
> 3) will have to support USE like funtionality, i.e. <src_uri
> use=foo>http://blah.com</src_uri>  * there will need to be many
> conditional setups in addition to this *
> 4) will have to be used as transport methods only most likely, the
> general consensus is that devs that I have spoken to do not want to
> write fully in XML, therefor the meat of the builds would need to be
> referenced by the xml files and in a shell-like language, and there
> would have to be tools to assist in building and reading them for both
> dev and user.  Not everyone is (or cares to be) an xml expert, and
> accessibility is a must.
> 
> On that note, will have more info to follow on what we've been throwing
> around very soon as well.
> 
> I do agree that this is a great method to follow, but there are still
> quite a bit of logistics to work out with it.  Once we have a more
> thorough proposal I';ll make sure we start a proper thread here to
> follow it through.
> 
> Gerk
>