On Mon, 3 Oct 2005 14:11:59 +0100
segv@netctl.net wrote:
> On Mon, 3 Oct 2005 15:02:41 +0200
> Jeroen Ruigrok/asmodai <asmodai@in-nomine.org> wrote:
> > Wouldn't it be better to do:
> > 
> > build D, deinstall D, install D,
> > build C, deinstall C, install C,
> > et cetera?
> 
> OK
> 
> A_v1 -> B_v1 -> C_v1 -> D_v1
> 
> where v<num> is the version number of a package
> If C_v1 depends on D_v1 and you build D_v2 then remove D_v1 and install D_v2,
> then C_v1 will break (possibly breaking dependant packages B_v1 and A_v1)

The last time this came up I pointed out that this is the worst case
scenario if you do as Jeroen suggests.  That is, in the pathological
case where building one package causes an existing installed package to
fail, you lose the dependent package until it rebuilds.  This is
currently the best case situation.  By changing the order we lose
nothing in the pathological case and gain in the usual case.

I understand that we can do chroot jail builds, etc. but wouldn't it be
nice if the average user could just do a "make update" and have it just
work without being down for any length of time in the usual case?

-- 
D'Arcy J.M. Cain <darcy@NetBSD.org>
http://www.NetBSD.org/