Re: [ANNOUNCE] : Gerrit Patch Verification Buildbot

Am 29.01.2013 00:43, schrieb Norbert Thiebaud:
> On Mon, Jan 28, 2013 at 5:16 PM, Eike Rathke<erack@redhat.com>  wrote:
> > >Hi Norbert,
> > >
> > >On Saturday, 2013-01-26 01:48:17 -0600, Norbert Thiebaud wrote:
> > >
> > > >>The goal being to test build gerrit patch on Linux, Mac and Windows,
> > > >>to verify them before they get integrated.
> > >
> > >First of all: Great stuff! Thanks!
> > >
> > > >>Fourth we will ask people to leave the Verify field of the patches to
> > > >>the care of the tinderboxes, unless there are pressing reason not to,
> > > >>and wait until they pass the buildbot verification before pushing them
> > >
> > >I have one request here. Could we keep the Verify field for manual "I
> > >tried this patch and can verify it fixes the problem or implements the
> > >feature", and have an additional field for the automated builds, for
> > >example "Builds"?
Well you are talking about a work around for current workflow:
every commiter commits directly to master, bypassing gerrit patch set 
verification -- master is broken -- we have false negative.
In ideal world, where we have unlimited  hardware resources, that 
wouldn't be needed any more.
Norbert's MAC TB 21 (full ccache on ssd) required only 30 min. to check 
that patch yesterday:
https://gerrit.libreoffice.org/#/c/1901/

But because we are not there today, it is indeed a problem...

> Well everything is possible but it would be messy.
> The 'Submit' button enforce rules wrt to the state of the label.
> for instance a patch can only be submitted if tehre is at least a +1
> verify and no -1 verify and if there is a +2 review and no -2 review.
> if you add a new 'verify' field then
> 1/ that field will need to_have_  a value before one can submit
> 2/ the other verify would have to have a value +1 too before one can submit...
> that is kind of awkard.
>
> Otoh, you can still add your own +1 verify today... and we wil now it
> is a 'manual' one because of the user associated with that +1 verify
> all the automated +1 verify would come from the the 'gerrit' user...
> your own would be assocaited with 'Eike'
> iow there can be already more than one verify value recorded.
> seehttps://gerrit.libreoffice.org/#/c/1885/   for an example.
>
> Now I think the real issue is : what if the builbot assigned a -1 and
> I know that was a false positive... how do I get to override that -1
> so that I can still 'Publish' that patch. the answer is: we have a
> command line to do just that... eventually a group of committers will
> be granted 'Buildbot admin' that will allow them to manually overrule
> the Verify status that the gerrit user had set.
Yes, ssh command line [1] is the only way to go for now, but we are 
thinking about to find a way
to put buildbot-plugin own's buttons on the gerrit's core patch view.
I. .e. put a button 'Schedule' left to 'Review' button. (If the 
identified user was granted the ACL, of course.)
Once we can do that, we could provide syntactic sugar to override 
buildbot verification score:
put another button 'Reset buildbot verification', resetting -1 to +0, so 
the patch is not blocked any more.

[1] 
https://gerrit.libreoffice.org/plugins/buildbot/Documentation/cmd-verify.html

David