Re: utnubu-desktop for the masses
On Tue, 2006-04-25 at 08:48, Joey Hess wrote:
> FWIW, I advocated using packages with collections of recommends as task
> packages when the task system was being designed, but the idea was
> discarded for various reasons. You can find the full details of that
> decision in the threads starting here, but I don't see much point in
> rehashing it since very little of relevance has changed since:
The decision to go with tasks rather than metapackage
recommendations seems to have happened largely by default.
One proto-DPL wrote that task fields were "the only way
around this I can think of" and provided sample code.
Another developer pointed out the many advantages of
metapackage recommendations, including heirarchical
structure and that it would not be difficult to deploy
as the infrastructure already existed.
AJ was correct when he wrote that "the complexity of
dependency specifications just isn't warranted" for
tasks but the unintended side-effect was that we now
have two dependency mechanisms to track and, although
tasks started out simple, the magic hooks that have been
have caused task dependencies to become opaque to the
point that programs cannot reason about them without
emulating tasksel in a simulated environment.
Metapackage recommendations are not then "a solution in
search of a problem" but rather a proposed solution to
the opacity and complexity which now bedevil automated
reasoning about tasks.
So Mr Hess, you were right when you espoused metapackage
recommendations six years ago. As for what has changed
* We now have time to consider the problem logically
rather than conceding defeat to the first sample of
* We now have tools such as wajig[6,7] and aag[8,9]
which can install a metapackage with its
recommendations without changing the meaning of
"apt-get install". (And such functionality could
be added to apt-get if desired.)