[webkit-dev] Comment on the bug & email author/reviewer before reverting a patch

[Ryosuke Niwa]

On Wed, Jul 9, 2014 at 7:25 PM, Maciej Stachowiak <mjs at apple.com> wrote:

>
> On Jul 9, 2014, at 4:45 PM, Ryosuke Niwa <rniwa at webkit.org> wrote:
>
> On Wednesday, July 9, 2014, Brady Eidson <beidson at apple.com> wrote:
>
>>
>> On Jul 9, 2014, at 4:15 PM, Ryosuke Niwa <rniwa at webkit.org> wrote:
>>
>>  Again, im not requesting anything new here. The consensus on webkit-dev
>> has been to ping the author/reviewer on IRC or via email and comment in the
>> original bug PRIOR to using webkitbot to start reverting the patch.
>>
>>
>> I went through the first handful of emails on that thread.  The original
>> request that wasn't meeting a lot of opposition before I stopped digging
>> through the thread was:
>> "Please contact the author/reviewer and give them a reasonable amount of
>> time *before rolling out their patch*."
>>
>> I did not reach the message where the consensus was "contact the author
>> and reviewer manually, *do not use webkitbot*"
>>
>> I believe that using webkitbot:
>> 1 - Comments in a new bugzilla created specifically because there's an
>> issue
>> 2 - Comments in the original bugzilla notifying of an issue
>>
>
> It doesn't. The bot only files a new bug, make it a blocker of the
> original bug, and then reopen the bug.
>
> It doesn't copy over any comments made in the new bug for example.
>
>  Assuming my webkitbot command contains a description of the reason this
>> patch is suspect, including a URL to the failure, can you further explain
>> why using webkitbot is unreasonable?
>>
>
> I'm not saying that using webkitbot is unreasonable. I'm saying that the
> person trying to revert a patch should first inform the author/reviewer
> first BEFORE start reverting the patch.
>
> Since webkitbot doesn't automatically post the details as to what failures
> the patch caused, and one line description is almost never adequate (e.g.
> needs a hyperlink to buildbot page, test failure diff or error log, et
> c...), I don't see how using webkitbot in its current state could ever be
> adequate.
>
> Of course, I'm not saying that webkitbot could never be improved to do
> these things.
>
>
> What things need to be done in addition to using 'webkitbot rollout' to
> meet a sufficient standard of notification? I assume based on your comments
> that it should:
>
> (1) Add a comment to the original bug that caused the regression (maybe
> something like "this caused regression bug XXX" where XXX is the rollout
> bug).
> (2) Add links to diagnostic information about the problem (e.g. buildbot
> results page showing the failure, or website URL illustrating a
> regression). They should probably go in the bug reporting the regression,
> not the original bug.
>
> Anything else? It seems like (1) and (2) could be done manually while also
> using 'webkitbot rollout', and (1) could in principle be done by the bot.
> Would you object if someone used 'webkitbot rollout' and then did (1) and
> (2)?
>

It is my understanding that doing (1), (2), and 'webkit rollout' doesn't
contradict the previously reached consensus.  All I'm stating is doing (1)
and (2) before reverting the patch has been the consensus.

- R. Niwa
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.webkit.org/pipermail/webkit-dev/attachments/20140709/2ad9d6ba/attachment.html>