Time to abandon recursive pulling of dependencies?

[Ulrich Spoerlein]

On 5/16/07, Alexander Leidinger <Alexander at leidinger.net> wrote:
> Quoting "J. Porter Clark" <jpc at porterclark.com> (from Wed, 16 May 2007
> 06:25:32 -0500):
> > What I don't like about the flattening of the dependencies is
> > that there seems to be information loss; that is, I can't figure
> > out why one port (e.g., gweled) requires another port (e.g.,
> > cdrtools).  Is there any tool to unflatten the dependencies?
> No. And just recording the first order dependencies would be a
> sensible approach to get this information. But this is just one more
> reason why first order dependencies would be better than the current
> recording of everything.

I wrote a tool, that will generate a Makefile and dot(1) files for
given ports or the whole tree, using only first order dependencies.
Sadly, dot(1)/gs(1) cannot process EPS files with 16.000 nodes (or I'm
doing something stupid).

> The problem not discussed so far is: some ports may not have all first
> order dependencies. So anyone wanting to change this should install a
> tinderbox and start testing fixing those ports.

Hmmm, this is a red herring, no? A first order dependency is
everything the port specifies in it's _DEPENDS variables. If you
change the internal representation of the tree, keeping the transitive
hull intact (!!) then there should be no user visible change in how
package dependencies are pulled in.

Uli