<div dir="ltr"><div>Coming back on this thread. I am sure that Gijs and others will think about it but even with the approval-by-default, we still need the status flags (<i>affected</i> => <i>fixed/verified</i>) to be updated once the patch reaches aurora.<br></div>Release management relies on the values of the status flags.<br><br></div><div class="gmail_extra"><br><div class="gmail_quote">2015-12-10 20:43 GMT+01:00 Sylvestre Ledru <span dir="ltr"><<a href="mailto:sylvestre@mozilla.com" target="_blank">sylvestre@mozilla.com</a>></span>:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
  
    
  
  <div bgcolor="#FFFFFF" text="#000000"><span class="">
    Hello,<br>
    <br>
    First, I understand your pain and frustration. We have plan to
    simplify the aurora/devedition uplift processes for next year.<br>
    <br>
    I think we can give a try to your proposition in 46. However, we
    think we will need two things:<br>
    * usage of a a=css-image-only approval<br>
    * a hook to verify the list of files being touched in the commit.<br>
    <br>
    I reported this bug:<br>
    <a href="https://bugzilla.mozilla.org/show_bug.cgi?id=1231768" target="_blank">https://bugzilla.mozilla.org/show_bug.cgi?id=1231768</a><br></span>
    to get this hook.<br>
    <br>
    Thanks for this proposition,<span class=""><br>
    Sylvestre<br>
    <br>
    <br>
    <br>
    <div>Le 27/11/2015 11:21, Lawrence Mandel a
      écrit :<br>
    </div>
    </span><div><div class="h5"><blockquote type="cite">
      <div dir="ltr">
        <div>
          <div>
            <div>
              <div>
                <div>Good discussion. A few thoughts on top and in
                  summary of what's already been said:<br>
                  <br>
                </div>
                - I think experimenting is good. 46 seems like a good
                time in that we have some time to work out all of the
                details.<br>
              </div>
              - The relman team agrees that while we do get value from
              Aurora approvals there is a lot of overhead here - and I
              would say ideally there are better activities on which to
              spend our time.<br>
            </div>
            <div>- Much of the value that relman gets out of approvals
              should be obtainable from other sources. We should
              identify the real value (I think we can do this quickly)
              and figure out how to get this information from the
              pushlog and other sources.<br>
            </div>
            - There is an issue of trust. While many (in my experience
            the majority) of Mozilla devs have demonstrated that they
            make good decisions wrt uplifts, some don't. Having an open
            m-a branch is scary. We really need to keep this branch
            stable. This is the one piece of value, control over
            landings, that will be hard to replace. I agree with
            experimenting as we don't know how much actual value we're
            getting with our current model.<br>
            <br>
          </div>
          Sylvestre - Can we please put this on our list to discuss in
          Orlando and come back with either a commitment to do what Gijs
          has proposed or an alternative proposal?<br>
          <br>
        </div>
        Lawrence<br>
        <div>
          <div>
            <div>
              <div>
                <div>
                  <div>
                    <div>
                      <div>
                        <div>
                          <div>
                            <div class="gmail_extra"><br>
                              <div class="gmail_quote">On Thu, Nov 26,
                                2015 at 7:28 AM, Gijs Kruitbosch <span dir="ltr"><<a href="mailto:gijskruitbosch@gmail.com" target="_blank"></a><a href="mailto:gijskruitbosch@gmail.com" target="_blank">gijskruitbosch@gmail.com</a>></span>
                                wrote:<br>
                                <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Time for a
                                  recap now this discussion has died
                                  down a little.<br>
                                  <br>
                                  TL;DR: I would like to experiment with
                                  this for the 46 cycle to see how this
                                  works in practice. If we find it has
                                  negative effects, we can stop doing it
                                  (immediately; we don't need to wait
                                  for the end of the cycle...). If we
                                  find it has broadly positive effects
                                  (ie doesn't lead to more breakage than
                                  usual), we can look at broadening the
                                  set of things we uplift without going
                                  through the formal approval process we
                                  currently use.<span><br>
                                    <br>
                                    On 20/11/2015 19:22, Gijs Kruitbosch
                                    wrote:<br>
                                    <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                      I would like to propose that we
                                      eliminate the explicit aurora
                                      approval requirement for
                                      changesets that meet ALL of the
                                      following criteria:<br>
                                      - only change CSS and/or image
                                      files<br>
                                      - are landing in the first 4 weeks
                                      of the aurora cycle;<br>
                                      - have landed on m-c and stuck<br>
                                    </blockquote>
                                    <br>
                                  </span><span>
                                    On 20/11/2015 19:33, Robert Kaiser
                                    wrote:<br>
                                    <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                      How many of those do we have every
                                      cycle (i.e. is it enough that it
                                      gives us a significant benefit and
                                      even warrants discussion)?<br>
                                    </blockquote>
                                  </span>
                                  I don't know. I tried to find out, and
                                  it's hard to get that information. I'm
                                  fairly sure it's a comparatively small
                                  (<10%) subset of the patches that
                                  gets uplift, though.<br>
                                  <br>
                                  I think it warrants discussion because
                                  I think there is broad agreement from
                                  relmgmt, QE and engineers that the
                                  current status quo is not ideal, but
                                  less agreement on what we should be
                                  aiming for instead. I'd like us to
                                  experiment and evolve our strategy
                                  here. This is not me personally going
                                  "ugh, I wish the patches I wrote
                                  didn't have to do this approval
                                  dance". This is me going "I wonder if
                                  this is a good set of things to use to
                                  evolve how we deal with approvals"
                                  (though full disclosure, because of
                                  the type of patches I often write,
                                  this probably does benefit me
                                  personally more than it will some
                                  other people). I'll get back to this
                                  point replying to some of the other
                                  posts further downthread.<span><br>
                                    <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                      What's the definition of "have
                                      stuck"? For how long?<br>
                                    </blockquote>
                                  </span>
                                  "landed on m-c without being backed
                                  out". I don't really want to use extra
                                  criteria like "landed and stuck for X
                                  days" because I think the criteria are
                                  already pretty narrow, and adding
                                  verification burden here is not going
                                  to help materially.<span><br>
                                    <br>
                                    On 20/11/2015 19:34, Chris Peterson
                                    wrote:<br>
                                    <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                      Why not allow any code change to
                                      be uplifted to Aurora based on the
                                      developer's discretion? Or allow a
                                      patch's reviewer to a+ for Aurora
                                      uplift?<br>
                                    </blockquote>
                                  </span>
                                  I think "developer's discretion" alone
                                  is not likely to be conducive to
                                  stability. I think patch reviewer
                                  and/or third person review by another
                                  person (not necessarily relman) might
                                  be interesting to explore, but still
                                  more high-risk than what I am
                                  suggesting. As an experiment, I'd like
                                  to start with what I originally
                                  suggested. We can evaluate after 1
                                  cycle (or before if it blows up in our
                                  face).<span><br>
                                    <br>
                                    On 20/11/2015 19:55, Liz Henry
                                    (:lizzard) wrote:<br>
                                    <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
                                      I could be ok with us outlining
                                      this kind of criteria and letting
                                      developers know they can land
                                      their own changes in some cases. 
                                       Verifying the fix on m-c (by
                                      someone other than the developer)
                                      might help too.<br>
                                      <br>
                                      But, what those cases are, I'm not
                                      entirely sure.  I think there have
                                      been css only uplifts that cause
                                      significant regressions.   Now, me
                                      reviewing them isn't going to
                                      catch those regressions. But my
                                      being aware of a cluster of uplift
                                      requests can mean I know to ask QE
                                      for extra testing in particular
                                      areas. I factor the uplift action
                                      into a more holistic view of
                                      what's happening across the board
                                      for a release and the amount of
                                      change and risk we're looking at.
                                      That isn't always visible to the
                                      developers requesting uplift.<br>
                                      <br>
                                      Gijs you mention devs not being
                                      "required" to do this. I wonder if
                                      this process might normalize
                                      uplifting things that could just
                                      as well ride the trains.<br>
                                      <br>
                                      The value of asking people to ask
                                      for uplift is sometimes in that
                                      they take their changes more
                                      seriously. Often, people talk to
                                      me first on irc before even
                                      asking, and during that
                                      conversation they find more work
                                      to do, they think about testing or
                                      verifying fixes, or I uncover more
                                      uplifts that need to happen for
                                      their work to land on aurora.<br>
                                    </blockquote>
                                    <br>
                                  </span>
                                  I think all of these paragraphs point
                                  to a single (valid!) concern:
                                  uplifting things increases
                                  instability. We can contain the risk
                                  of instability by not uplifting
                                  things, or at least by asking for
                                  automated tests and/or manual QE, or
                                  restricting what things we uplift
                                  (interface changes, patch size, l10n
                                  concerns, etc. (l10n being a bit
                                  special because there are other
                                  reasons besides stability)).<br>
                                  <br>
                                  But as dbaron and past have pointed
                                  out (and as I've discussed with Liz
                                  privately), realistically part of the
                                  issue is that we often don't get QE
                                  verification or crash-rate-combatting
                                  until we get to beta/release. That
                                  means nightly and, to a slightly
                                  lesser degree, aurora, are both not
                                  particularly stable, and that that
                                  stability is not completely dependent
                                  on how strict we are with uplifts.<br>
                                  <br>
                                  It's very hard to assess the success
                                  of the current system at increasing
                                  stability because there is no way to
                                  know "what would have been". We don't
                                  know if developers or relman are
                                  overly cautious or not cautious/strict
                                  enough, unless we experiment. Because
                                  nobody seems particularly happy with
                                  the current state of things, I would
                                  like us to experiment. :-)<br>
                                  <br>
                                  <br>
                                  So perhaps my original phrasing was
                                  clumsy. I would like us to experiment
                                  with being "more permissive" about
                                  uplifts, and I'm purposefully starting
                                  out with a subset of patches that I
                                  suspect are lower risk, and where
                                  there is a clear secondary benefit (ie
                                  devedition theming).<br>
                                  <br>
                                  I believe it's probably equally useful
                                  to take a hard look at improving
                                  quality on nightly and aurora through
                                  other means. I'll open a separate
                                  thread about that.<span><font color="#888888"><br>
                                      <br>
                                      ~ Gijs<br>
                                    </font></span></blockquote>
                              </div>
                              <br>
                            </div>
                          </div>
                        </div>
                      </div>
                    </div>
                  </div>
                </div>
              </div>
            </div>
          </div>
        </div>
      </div>
    </blockquote>
    <br>
  </div></div></div>

</blockquote></div><br></div>