No strong opinion.

Just throwing out my view of the world.

I think that builds testing CLs should use "cros_workon" instead of "local uprevs" for the test build. This would make it more explicit that the versions built weren't "real" ebuild versions.

Then cros_mark_as_stable should uprev and push all as atomic (ish) step.

to the original comment: sure, pulling out "push" functionality sounds fine.

to Don's thoughts: the argument for doing an uprev and building that is two fold:
- when the bots are publishing prebuilts, we need to do this.  with post submit publishing prebuilts, that's no longer an issue.
- keeping the stuff we test in the CQ as close as possible to what everyone else sees (postsubmit/release/etc...).  we do some things differently when it's a 9999 ebuild.

we could arguably focus on making the 9999 code path more like non-9999 ...

Fair. I'm not sure if the benefits of using "cros_workon" are worth the effort, it just seems to me that it would make things conceptually cleaner.