Would this be fixed by having --clobber not skip .cache?

Yeah, --clobber should wipe all of .cache. That makes is safer, but also more expensive (slower for the next build).

This should be okay, since we clobber a lot less than we used too.

Chris, tracking down the --clobber behavior for cbuildbot shouldn't be TOO hard.

--clobber doesn't clean /b/git-cache, not sure if it affects the build.

/b/git-cache is different from .cache. Though, we have seen at least one bug caused by bad contents in /b/git-cache. Maybe --clobber should wipe the git cache directory as well.

Chris, not that both such directories can be specified on the command line.

I'd recommend against cleaning /b/git-cache under --clobber, at least until we can prove it's needed. We'll lose a lot of the benefits of that cache, and we don't quite control who can hit the clobber check mark when forcing a build. (I've seen multiple clobber builds in parallel launched by team members outside infra).
Secondly, the bug that I refer to specifically here only needs .cache to be cleaned, because we bindmount source from there into the chroot directly.

/b/git-cache is fetched from GS, not GoB, so re-fetching won't cause quota problems.

Refetching it will slow things some, but won't break anything.

So do we want a 2nd argument for clobbering the git-cache or roll that into the existing clobber?

Same argument is best.

In practice, this should be fixed now.

1) cbuildbot_launch clobbers that source tree when a builder changes branches.
2) recent cbuildbot clobbers that directory if gclient sync fails (then tries again).

Are there any known issues currently?

Though, I agree that clobber should wipe the cache directory.

According to OP (and my memory), #12 is enough, until proven otherwise.