[meego-commits] Too fast acceptance!

[Arjan van de Ven]

On 8/11/2010 6:42 AM, Marko Saukko wrote:
> On 08/11/2010 04:21 PM, Arjan van de Ven wrote:
>> On 8/11/2010 6:09 AM, Marko Saukko wrote:
>>> On 08/11/2010 03:55 PM, Arjan van de Ven wrote:
>>>> On 8/11/2010 5:45 AM, Marko Saukko wrote:
>>>>>
>>>>> So what I'm asking here is that please give us AT LEAST 24 HOURS to
>>>>> check that everything works before moving from Testing to Trunk.
>>>>
>>>>
>>>> 24 hours is totally unacceptable.
>>>> it should be below 8 hours for everything!
>>>>
>>>> trunk:testing ->   trunk migration rule is "migrate unless there is a
>>>> counter indication"...
>>>>
>>>> we can't slow the world down by making everything take 24 hours; we'll
>>>> never be able to ship anything.
>>>>
>>>> (for one, I want to do a new kernel build at least once, but 
>>>> preferably
>>>> twice a day so that we can get smaller steps that make it easier to
>>>> narrow down issues)
>>>>
>>>
>>> There is only one sanity test done during one day to Trunk:Testing so
>>> how can you be sure that all you changes are tested by the sanity test
>>> if you submit more than once per day?
>>>
>>> I agree that 24 hours is long time but currently because we have only
>>> ONE sanity test per day this is the only way to be sure that both
>>> Trunk and Trunk:Testing will not end up broken, right?
>>
>> actually it's not.
>> due to this long window, the sanity check is no longer really
>> representative; too many changes at the same time.
>>
>> the sanity check needs to be continuous... not "once a day". I'll settle
>> for "once an hour" I suppose.
>>
>>
>
> Compiling the kernel can take more than 1 hour, e.g.,
>
> osc jobhistory Trunk:Testing kernel-netbook standard i586
>
> 2010-08-10 03:36:18  kernel-netbook source change    succeeded         
> 1h  1m 33s      build09/1

I'm ok with the clock starting when the package has finished compiling.
(although hopefully when the new nokia hardware arrives we don't 
actually have to wait forever anymore on builds like this that normally 
only take 10 minutes)

>
> and after this has been compiled the dependencies start to compile. 
> Should be also noted that testing image can be done only after the 
> repository has been published so after all packages are compiled, or 
> has there been change to this?

it looks like this whole thing is a bit iffy as a whole. A process is 
put in place, but the backing infrastructure for that process isn't 
quite there yet.... and hasn't been for some time.
I would really like to say that... to a large degree "if you don't test 
enough, don't complain if the bug doesn't get caught early enough".
Yes now it's caught, and I'll spend time today finding out what is 
happening.
but... seriously, if you don't want such things to slip through, test 
frequently enough.
(and this is both for x86 and arm)