Todd Vierling wrote:

> But the more I think about this, a failure is OK.  The whole fail/skip thing
> probably needs a rework anyway.  :-/

Yeah, it sucks. Something similar came up at pkgsrcCon that led me
to offhandedly remark that it needed to be thought about again,
and more carefully. (The current state of affairs is my doing, so
I'll criticize it as mercilessly as needed. :-)

Initial thoughts (please help flesh these out):

If I'm a package, I decide whether I have what I need to build or
not. Obviously, we'll continue to need a way to denote this.

If I'm a package that decides not to build, it's not reasonable
for me to dictate how fatal this should be for who/whatever asked
for me. For one thing, I can't know; for another, maybe it's fine
for one package and fatal for another; for another, bulk builds
might want to handle these situations still differently; likewise
for pkg_chk and similar tools.

The decision of how to proceed when a requested (by user at command
line, by dependency, by bulk build, by pkg_chk...) needs to be able
to be made by the requester.