Date:        Sun, 11 Nov 2001 00:42:08 -0800
    From:        Johnny Lam <jlam@jgrind.org>
    Message-ID:  <20011111004208.A27538@c35.jgrind.org>

  | This does look like a bug in the implementation of "make update".

I disagree - though it certainly depends upon how you read Packages.txt

In this case, t1lib was being updated, t1lib has no dependencies on
gmake  ...

	jade$ pkg_info -R gmake-3.79.1
	Information for gmake-3.79.1:


	jade$ 

in fact, nothing does, which is exactly as it should be (barring some
very unusual package that wanted to run gmake at run time).

(If there's an option to pkg_info to list the dependencies of a package
instead of the packages that depend upon it, I didn't find it...)

I think it is entirely reasonable to interpret Packages.txt as correctly
describing what "make update" does.   (I also expect it is reasonable to
interpret it the other way ... it could perhaps be clearer).


The real problem here is that nothing records dependencies on the base
system - had there been, then when the old libutil was removed, gmake
would have gone with it, and then when t1lib was to be rebuilt, gmake
would have been noticed to be gone, and rebuilt.


Until NetBSD gets full dependency recording, deleting anything from the
base system without simply nuking all installed packages is just dangerous,
and can't be expected to work (or at least not without going and checking
everything installed to see if it matters, and then deleting the ones where
it does).

kre