Re: Slight example about Essential packages
On Wed, 19 Nov 1997, Bruce Perens wrote:
> From: Jason Gunthorpe <jgg@gpu.srv.ualberta.ca>
> > You will note that even though the policy manual says dependanices are not
> > needed it does -NOT- say they are implicitly present (the point of my
> > post). The subtle difference is that implicitly present depends will mean
> > essential packages will try to be configured first.
>
> Essential packages have already been configured before the first time Deity
> runs on a system. That is the point of the "base archive". Forget about
> putting together a system by installing packages on an empty disk, it doesn't
> work and is not likely to work soon.
>
> Just when essential packages configure themselves during an _upgrade_ is less
> important.
I must admit I haven't been following the libc6 upgrade too closely, but
doesn't part of the problem with upgrading bash have to do with it and
libreadline2g being configured too late in the process?
Right now it seems you are correct though, the essential packages don't
generaly need to be configured before they can be used, so it is less
important.
However, since we are experimenting with ordering I would like to be
certain that however Manoj implements the algorithm he will allow for some
future growth in areas like this. Also, I keep suggesting the property of
highly predictable ordering. The smaller the number of possible orders the
easier it is for the testing team and the maintainers.
So by applying a few usefull things (like essentials installed first) and
some unusefull things (like by name sorting), Deity's algorithm should
generate one and only one order for any set of packages.
At least it seems like a good idea..
Jason
Reply to: