release: Improve error handling and option logic

[ivy]

Summary

Based on the user-provided options, define three internal variables,
_USE_PKG, _USE_PKGBASE and _USE_PORTS, to avoid having to repeat the
hard-to-read "!defined(X) || empty(X)" conditional in many places.

Add some early errors for things we know don't work, like using pkg
on non-FreeBSD platforms.

Stop ignoring errors when installing packages onto the media; if this
fails and the user is okay with that, they should set NOPKG to avoid
trying it in the first place. This avoids silently building media
which is missing expected components.

[emaste]

This seems like a good idea to me; will take a detailed look soon. Patch needs to be rebased.

[ivy]

In D54548#1286203, @emaste wrote:

This seems like a good idea to me; will take a detailed look soon. Patch needs to be rebased.

this entire stack is stalled until we can get some consensus on D54282. i'm working on other pkgbase things in the mean time.

[ivy]

rebase

[emaste]

release/Makefile

103

probably worth being explicit about _USE_PKG referring to pkg(8), since it seems _USE_PKG and NOPKG ought to be related.

222

I think it was originally intentional to have only the dependencies in the conditional, but I guess this will be clearer for when distset support is eventually removed.

270

Maybe a

#
# Base system packages
#

comment to mirror Legacy distribution tarballs above?

378

This is not necessarily true is it? Can the user set both NODISTSETS and NOPKGBASE (to produce a mostly useless installer)?

Or, is this just intended to explain why we're copying the MANIFEST even if building pkgbase installer? The comment a few lines below (Copy MANIFEST to provide legacy dist checksums in both modes) was intended to indicate that, but is admittedly terse.

Maybe just Even if we're building pkgbase media... and then remove the terse comment below.

386

This precludes us producing an installer with both pkgbase packages and dist sets that allows the user to choose either from the media. We'll likely get there (and remove dist set logic altogether) but we might not want to do it yet?