Re: [Qemu-devel] We need more reviewers/maintainers!!

[Blue Swirl]

On Tue, Mar 13, 2012 at 14:41, Anthony Liguori <address@hidden> wrote:
> On 03/13/2012 09:38 AM, Avi Kivity wrote:
>>
>> On 03/13/2012 04:00 PM, Anthony Liguori wrote:
>>>
>>> On 03/13/2012 08:40 AM, Avi Kivity wrote:
>>>>
>>>> On 03/12/2012 10:27 PM, Anthony Liguori wrote:
>>>>>>
>>>>>> I agree that more maintainers would be good, but we also need
>>>>>> more people with commit rights.
>>>>>
>>>>>
>>>>> I disagree strongly.  Having multiple pushers makes things difficult
>>>>> and encourages people to push without testing.  Part of what makes
>>>>> pushing take longer than it should today is that my test cycle takes
>>>>> at least 1-2 hours and it's not uncommon to have to go through 3-4
>>>>> cycles of rebasing before being able to push.
>>>>
>>>>
>>>> This really sucks.
>>>>
>>>> If testing was automated, we could have a staging branch where
>>>> maintainers would push patches, they'd get tested automatically and then
>>>> graduate to master.  The workflow would look something like
>>>>
>>>>    git fetch
>>>>    git checkout staging
>>>>    git rebase origin/staging
>>>>    <apply patches, pull trees>
>>>>    git push staging
>>>>    <wait>
>>>>    <staging gets merged into master autoamatically, or you get an email
>>>> from the test system>
>>>
>>>
>>> The problem for me with this is that I test before I do a thorough
>>> review.  I do a quick review, but not a line-by-line review.  So I
>>> don't necessarily want to queue for push.
>>
>>
>> Seems to me it's better to review before testing, no?
>
>
> I typically do a high level review before queuing for testing, but I don't
> do a line-by-line review for coding style or minor issues.
>
> The later must be done before committing no matter how many revisions are
> sent.  It's more time efficient to catch a functional problem without doing
> the line-by-line review.  Best case scenario is that the line-by-line review
> happens only once for a patch before it's committed.

Perhaps patchwork is not the right tool for this. Coreboot uses
Gerrit: http://review.coreboot.org/#/q/status:open,n,z combined with
Jenkins build bot. This looks much more professional.

>>>> If testing cannot be automated, perhaps a lock around the tree would
>>>> help.
>>>
>>>
>>> I think merging qemu-test into make check would help a lot.  If all
>>> committers are running the same test suite before pushing, then this
>>> problem would become less common.  It's livable now because most
>>> committers commit infrequently.
>>>
>>> But if we added more committers, it would become pretty problematic.
>>
>>
>> I'm not arguing either for or against that, just trying to make the
>> commit process more efficient.
>
>
> Yup, and appreciate the suggestions.
>
> Regards,
>
> Anthony Liguori
>
>
>