Todd V. submitted a PR 21097 in April:

 http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=21097

I am using this PACKAGE_DEPENDS_RECURSE idea (set to "false" in my
mk.conf) on a Linux system using LOCALBASE as /usr and with xpkgwedge.

It has reduced the number packages that are listed as "required by".

An example: my wip/gramps package has 80
packages listed as "Requires:". And with using PACKAGE_DEPENDS_RECURSE as
false then it reduces it slightly by removing:
 db
 docbook-xml
 docbook
 gnome2-dirs
 iso8879
 ncurses
 perl
 pkg_install-info
 python23-pth
 xmlcatmgr

This is good.

But I still have problems with wrong registrations. For example, as far as
I know, gramps (genealogy program) should not be dependent on metacity (a
window manager).

metacity is brought in as a buildlink for x11/gnome2-control-center.
gnome2-control-center is brought in with x11/gnome-panel/buildlink2.mk.
And that gnome-panel is used by py-gnome2 which gramps really uses.

pkg_info metacity says:
 Required by:
 gnome2-control-center-2.2.1nb5
 gnome-panel-2.2.2.2nb3
 py23pth-gnome2-2.0.0nb1
 gramps-0.9.5nb1

I want that reduced more so it only says "gnome-panel-2.2.2.2nb3" (and
only if that really uses metacity).

Another example is gnome2-dirs. With new PACKAGE_DEPENDS_RECURSE set to
false, it removed gnome-vfs, control-center, gal, gnumeric,
xscreensaver-gnome, and several others. Yeah :)

And another good example is xmlcatmgr. It got rid of numerous "Required
by:" entries. And now it looks like it only is "Required by:" packages
that actually use it.

My extra registration problems may be caused by extra stuff in my mk/. I
will try this also with a clean mk/ (with the PACKAGE_DEPENDS_RECURSE
idea).

Anyone else using PACKAGE_DEPENDS_RECURSE?

   Jeremy C. Reed
   http://bsd.reedmedia.net/