Re: Branch for netbsd 9 upcoming, please help and test -current

[Greg Troxel <gdt%lexort.com@localhost>]

Andreas Gustafsson <gson%gson.org@localhost> writes:

> And if we can't get all those post-February regressions fixed, there's
> always the option of branching retroactively, for example from source
> date 2019.02.08.09.17.12 which was the last one to achieve zero
> unexpected ATF test failures on real amd64 hardware.  It's never too
> late to branch early :)

I think this idea has a lot of merit, as I tend to favor stability over
features.  But I can see how it would be really nice to avoid this
outcome.

I wonder about a multi-pronged approach:

  a freeze in -current, much like we have in pkgsrc: bug fixes only, no
  features.  Basically only things that are almost provable not to cause
  regressions.  releng approval for changes that don't meet the above

  the usual work on fixing regressions and other problems

  using the ATF history and perhaps bisection to identify problematic
  commits (those that increased test failures, or introduce things we
  know broke even if they didn't cause a test failure).  For each such
  problematic commit, see if it can be fixed relatively quickly, and if
  not consider reverting it.  (For things like switching to newer
  versions, a tunable change may well be enough.)