Jeff King <peff@xxxxxxxx> writes: > All of the regressions people have actually _noticed_ stem from my > 136c8c8b8f in v2.14.2. So I think it is a viable option to try to go > back to the pre-v2.14.2 state. I.e.: > ... > That takes us back to the pre-regression state. The ancient bug from > 4c7f1819 still exists, but that would be OK for v2.15. We'd probably > want to bump the -rc cycle a bit to give more confidence that (2) caught > everything. Yes, I think that is the approach I was pushing initially with the jc/ref-filter-colors-fix topic that was later retracted; the result of your 4-patch series more or less matches that one, modulo that I didn't treat for-each-ref as a plumbing. I do share the worry that it is hard to make sure that these post-revert adjustment caught everything; after all, that was a major part of the reason why my earlier attempt was retracted. I still think this is the _right_ direction to go in, even though it is harder to get right. > Post-release, we would either: > ... > But we could punt on that part until after the release. The only thing > we'd need to decide on now is that first set of reversions. What I > really _don't_ want to do is ship v2.15 with "always works like auto" > and then flip that back in v2.16. True. Let's see what others think. I know Jonathan is running the fork at $work with "downgrade always to auto" patches, and while I think both approaches would probably work well in practice, I have preference for this "harder but right" approach, so I'd want to see different views discussed on the list before we decide. Thanks.
