On Tue, Apr 26, 2016 at 9:37 AM, Peter Geoghegan <pg@heroku.com> wrote:
> Only the 9.5 backpatch was a simple, conflict-free "git cherry-pick".
> Most of the effort here involved producing a clean 9.4 patch. This was
> largely mechanical, if a little tricky. In release branches for
> releases that preceded 9.4, there were a few further merge conflicts
> as I worked backwards through the branches, but those were trivial.
Looking again at this thread, the general agreement was to clear the
error stack before calling any SSL routine. Those patches are doing
so, and they look in good shape to me. Note: there is
SSL_do_handshake() on back-branches for the SSL renegotiation but we
don't need to bother about clearing the error queue as any error
occurring in those cases just stops the session, and we've never
bothered calling ERR_get_error there to get more details about the
errors.
> I'm not sure if project policy around backpatching (that commit
> messages and so on should match exactly) has anything to say about the
> case where backpatching follows several weeks after commit to the
> master branch. In the absence of any clear direction on that, I've
> created commits that look like what Peter E might have pushed in early
> April, had he decided to do that backpatch leg-work up front.
It seems to me that we definitely want to get this stuff backpatched
at the end. So +1 for this move.
--
Michael