About merging patches without tests

[Itamar Heim]

On 05/16/2014 05:38 AM, Sandro Bonazzola wrote:
> Il 16/05/2014 10:37, David Caro ha scritto:
>> On Fri 16 May 2014 09:33:44 AM CEST, Sandro Bonazzola wrote:
>>> Il 15/05/2014 18:46, David Caro ha scritto:
>>>> Hi!
>>>>
>>>>  From time to time we have some patches that are merged to master branches
>>>> without having been tested mostly because the developer merges before having any
>>>> response from the jenkins system.
>>>>
>>>> Merging one of those patches makes any following test run on that branch to
>>>> fail, and creating a lot of noise and trouble around all patches and jobs.
>>>>
>>>> So I wanted to stat a little discussion to bring up ideas on how to prevent
>>>> that. Some random thoughts:
>>>>
>>>> * -1 the patch at jenkins job start, and reset to 0 on success or infra failure,
>>>> and keep -1 if jenkins failure
>>>> * Only send a message to the patch with 'jenkins jobs started'
>>>
>>> I think that something like "jenkins jobs started, please don't merge until they finish" should be enough.
>>>
>>>> * Setup zuul as gateway, and make it block the patches if they do not pass the tests
>>>>
>>>>
>>>>
>>>> Cheers!
>>>>
>>>>
>>>>
>>>> _______________________________________________
>>>> Infra mailing list
>>>> Infra at ovirt.org
>>>> http://lists.ovirt.org/mailman/listinfo/infra
>>>>
>>>
>>>
>>
>> Usually the flow is reabase -> submit, that is done pretty easily from
>> the ui, and it does not give too much time to check any comments (the
>> rebase and submit buttons are on the top, and the comments show at the
>> end).
>> So doing that will not help on that case, as the developer would not
>> see the comment before submitting. I think we should block, at least
>> for a couple minutes, so he has to read the comments to be able to
>> submit.
>>
>> We had that before also, adding a comment when the jobs start, but we
>> removed it by developers request iirc
>
> Also recently you can just press submit and it automatically rebase before merging.
> So maybe yes, just a message isn't enough

its not possible to wait for the job on a simple rebase usually - too 
many collisions possible.