On Thursday 10 of June 2010 15:42:38 Michał Górny wrote:
> Hello,
> 
> First of all, I would like to notice I'm not trying to force moving
> Portage-specific features to PMS. I'm just trying to get some
> standarization on one of these features to make it possible for devs to
> use it in gx86 without commiting non-standard files.
> 
> The particular feature I'm talking about is defining repository-wide
> package sets. Currently, this is done through a Portage-specific
> 'sets.conf' file in the repository's root directory. Although such file
> could be considered acceptable for an overlay, I wouldn't like to see
> such a non-standard file commited to gx86.
> 
> On the other hand, many of current Portage users could benefit from
> the 'x11-module-rebuild' set we have introduced in 'x11' overlay [1].
> This particular set quickly aggregates all X11 modules for a rebuild
> after the xorg-server ABI change.
> 
> Portage by default supplies a few more sets which would fit repository-
> -specific set definition file better than the system-wide Portage
> configuration directory -- like the @live-rebuild and @module-rebuild
> sets.
> 
> This is why I suggest considering adding some basic definitions
> for 'sets' in the PMS, keeping that feature fully optional for PMs but
> preparing a standarized ground for those who would like to use it.
> 
> What I would like to see in the PMS is:
> 1) a definition of a 'set',
> 2) a definition of few basic types of sets (Portage currently describes
> them using specific classes but portable names would be much better),
> 3) a specification for repository-wide sets definition file.
> 
> In fact, the specification doesn't really even need to push the 'sets'
> into atom specifications -- as I guess we would rather keep away from
> using them in dependencies, and PM could be free to use any syntax to
> reference them.
> 
> [1] http://tnij.org/g6rl

Please take a look at https://bugs.gentoo.org/show_bug.cgi?id=272488

It contains Zac's PROPERTES=set proposition with sets syntax fitting current 
atom syntax (like metapackages just with a bit different behaviour). By 
definition It supports USE flags and I believe it's also simpler to implement.

-- 
regards
MM