allow stacking of stages - build implies fetch unless told not to about coreos-assembler HOT 5 CLOSED

I chose the existing semantics intentionally because I found when I was using rpm-ostree compose tree a lot locally I'd end up passing --cache-only a lot by hand. Not entirely opposed to changing but...I like what we have now too.

How about coreos-assembler fetch -b, which would work exactly like how rpmdistro-gitoverlay fetch -b does: do a fetch, and if something changed, do a build.

(Though in the c-a case "something changed" does includes the images, so it'd probably be easier to just have fetch -b just run a build afterwards regardless)

from coreos-assembler.

Not entirely opposed to changing but...I like what we have now too.

yeah. i actually am not advocating for changing it right now. I think there are a lot of "paper cut" type things that we can address with a re-write/design exercise. This is mostly food for thought for when that happens.

from coreos-assembler.

Proposal:
Break the build process into as many small pieces as possible. For example init would be broken into getting the config and downloading anaconda. build would be broken into composing the ostree, building the disk image, and running gf-oemid. Each of these stages would have their own build script (or program) as well. Each stage would be responsible for knowing its inputs (including itself), knowing what other stage(s) generate those inputs, and caching its results. If you run gf-oemid without initializing it would call diskimage, which would run get-anaconda and ostree which would call init.

The hard part would be figuring out how to plumb various options to the stages. I.e. how do you pass an argument to init when you called gf-oemid.

This of course, is reinventing make, so it might be possible to actually just use make in conjunction with our scripts.

from coreos-assembler.

This of course, is reinventing make, so it might be possible to actually just use make in conjunction with our scripts.

May be interesting to look at https://github.com/apenwarr/redo
(this bit of the docs was making the rounds on social media recently; it has some inaccuracies but is still interesting)

from coreos-assembler.

stale.. closing

from coreos-assembler.