On 7/5/06, Daniel Iliev <[EMAIL PROTECTED]> wrote:
> Why not just merge the
> top-level package, and if you don't like it, unmerge and use
> --depclean --pretend to figure out what can safely be removed?
>
Because if I decide to keep it, all dependencies it pulls-in don't get
updated until the top-level package starts depending on a different
version of those packages. Actually this is the main reason I started
this practice.
Not if you use --deep on your updates. Then dependancies are also
considered for updates. Some people here will tell you that --deep is
troublesome, but I am not one of them, and it seems like what you want
to do.
"emerge --depclean" yells a big warning that it is broken.
There are 2 "problems" with --depclean:
1. it takes your current use flags into account, rather than those
that were in effect at the time a package was merged. So if you
modify USE flags, it can report things can be removed, when in reality
that would break something. But if you do an "emerge -DNvp world",
and it doesn't report anything needing to be [re]merged, then this
doesn't apply.
2. it can remove packages that you really do want. As an example,
let's say you are programming something that uses the "boost" c++
library. If you were to remove everything in portage that depended on
boost, and it wasn't in your world file, then depclean would want to
remove it. The solution here is to add boost to your world file,
since you want that no matter what else is installed.
IMO neither of the above 'problems' are particularly serious, or a
good reason to add every dependancy to world.
> And I don't necessarily believe that having everything in world
> results in a significantly faster scan time than having only top-level
> packages there. I would like to see actual proof of this assertion.
No, no! I'm saying just the opposite - the more packages you have
recorded in the world list, the slower scanning you get.
Yeah, well, I don't necessarily believe the reverse either! :-)
Regards,
-Richard
--
gentoo-user@gentoo.org mailing list
