Re: [gentoo-portage-dev] dispatch-conf

[Jason Stubbs]

On Thursday 02 June 2005 21:42, Ciaran McCreesh wrote:
> On Thu, 2 Jun 2005 21:31:17 +0900 Jason Stubbs <jstubbs@g.o>
> wrote:
> | On Thursday 02 June 2005 21:12, Ciaran McCreesh wrote:
> | > Make it a PDEPEND of portage. Much cleaner.
> |
> | I won't argue on the merits of virtuals compared to direct
> | dependencies, but  why do you choose PDEPEND?
>
> Because portage currently always pulls in RDEPENDs before the main
> package, and because it makes figuring out system / bootstrap deps much
> easier if portage has as little as possible being pulled in before it.

I don't really like that either. PDEPEND is a hack to work around the
"A RDEPENDs on B DEPENDs on A" case. Seeing that's not the case here, it would 
appear to be an abuse of a pre-existing abuse. The order that the packages 
are emerged doesn't affect compilation or runtime. It's even questionable 
whether portage need depend on it at all.

As for the virtuals thing, hardcoding the packages into the portage ebuild 
means that tools-portage has to bother dev-portage every time a new provider 
is to be added or removed. It's easier to convey what is expected of a 
virtual rather than what a set of packages have in common. It also allows a 
smoother migration when one of the packages floats to the top. I'm not sure 
how others feel, but it seems to me that adding a package name into portage's 
ebuild implies a stamp of approval. Evaluating new tools is something I'd 
rather not be doing.

So I'll meet you halfway. No DEPEND at all will leave lots of users asking 
where they're config tools went - and not all users -pv system for that route 
to be truly viable. And there is one reason why abusing a PDEPEND is better 
than an RDEPEND in this case. Users with collision-protect will get a nasty 
shock unless PDEPEND is used.

Unless anybody disagrees with the above...

Regards,
Jason Stubbs