Re: [gentoo-dev] [RFC] C++ herd proposal

[Kevin F. Quinn]

On 17/9/2005 13:33:30, Christian Parpart (trapni@g.o) wrote:
> On Saturday 17 September 2005 11:36, Kevin F. Quinn wrote:
> > On 17/9/2005 0:20:57, Mark Loeser (halcy0n@g.o) wrote:
> >
> > C++ herd is a good idea, especially with that number of packages.
> >
> > >  I would also like to see many of them, if not all, moved to the
> > > dev-cpp category:
> >
> > Is this bit really necessary?
>
> indeed, it at least helps curious c++ devs to browse through some yet
> unknown c++ libs and he maybe finds something useful.

If the only gain is that one group finds one search criteria a little easier,
then I think that is far from sufficient reason to re-categorise.

What about people searching in the application domain (which to be honest
I think is much more likely)?  Under your approach they have to rummage
around in each dev-<lang> category, hoping that it'll be obvious from the
package name that it's suitable for their application domain.

What happens when the media, games or net herds come along, and want to
pull stuff into a media-libs, dev-games, net-libs?  We end up in a
tug-of-war between competing interests.

Think also of all the work involved in re-categorising stuff; how everyone
with dependencies on these packages in their overlay will have to rework
stuff in their overlay, all because of one group's nice-to-have.  It's
particularly acute for libraries.

I think we should discourage the idea that filesystem categories and herds
are related at all, and think of filesystem categories simply as convenient
buckets preventing lists of packages getting long.  Resist the urge to
re-categorise as much as possible, because in the end it's pointless.

Instead, add to metadata, and use that to find stuff.  In metadata.xml,
we could have as many search criteria as we like; for example source
language(s), library|application, application domain (sound, games, video)
which can happily cope with many->many relationships.

-- 
gentoo-dev@g.o mailing list