Build and Integration schedule -- skip weeks

[Joe Darcy]

On 05/08/09 02:12 PM, Xiomara Jayasena wrote:
>
>
> Joseph D. Darcy wrote:
>> Xiomara Jayasena wrote:
>>>
>>> There is no really point in doing a promotion if nothing has 
>>> changed.  Would a link for the skipped build to the previous build 
>>> number suffice, instead?
>>
>> The appropriate state of the source code in the repositories should 
>> also be tagged for both builds in this situation.
>>
>> Basically, places where people get the build from, source, binaries, 
>> etc., should have a conceptual link from the skipped build to the 
>> prior one.
>
> In the long term the above could become quite confusing.
> I do not quite understand the need to skip build numbers or re-build.  
> I believe in the past RE has done neither, we just update whatever 
> documents need to be updated.
>
> I can see that publishing a calendar in advance and knowing what build 
> number to target for, is very useful for gatekeepers, so if we must do 
> one of the two options above then skipping numbers maybe the best 
> alternative, from RE's perspective.

I find skipped build numbers to have a very high long-term cognitive 
cost.  Nine months, or a year or two years after JDK7 m3 when someone is 
trying to track down in which build a bug was introduced, how likely is 
it that he or she will remember, "Ah yes, b60 was skipped because that 
was the stopper build for JDK 7 m3 and there were no problems to fix!"  
Missing build numbers create questions rather than answers and 
complicate attempts to perform things like binary search on the builds.

I think the set of build numbers should be a dense sequence of 
consecutive integers.  There are multiple ways of achieving this and I 
don't have a strong preference between:

* Rename b61 as b60 if m03 doesn't need a stopped build
* Make b60 a duplicate of b59 if no stopper fixes are needed

-Joe

>
> -Xiomara
>
>>
>> -Joe
>>
>>>
>>> -Xiomara
>>>
>>>
>>> Paul Hohensee wrote:
>>>> Could also just build b60 to be identical to b59.  I.e., the only 
>>>> difference
>>>> between b59 and b60 bundles would be the build number.
>>>>
>>>> Paul
>>>>
>>>> Joseph D. Darcy wrote:
>>>>> Mark Reinhold wrote:
>>>>>>> Date: Fri, 08 May 2009 08:20:13 -0700
>>>>>>> From: xiomara.jayasena at sun.com
>>>>>>>     
>>>>>>
>>>>>>  
>>>>>>> I was under the impression that b60 was going to be the last 
>>>>>>> build for
>>>>>>> M3 and according to the Calendar here:
>>>>>>> http://openjdk.java.net/projects/jdk7/calendar/
>>>>>>> it shows that b60 is the last build for M3.  Are we saying that 
>>>>>>> b59 is
>>>>>>> now going to be the last build in M3 then?
>>>>>>>     
>>>>>>
>>>>>> Build 60 is the last scheduled build for M3.  It's the 
>>>>>> showstopper build.
>>>>>> If we need it, we'll do it; if we don't, then we'll skip it.  I 
>>>>>> suggest
>>>>>> you leave 60 in the schedule until we make that decision.
>>>>>>
>>>>>> If we skip 60 then that does raise the question of whether what's 
>>>>>> now
>>>>>> called 61 should be renamed to 60, and so forth for all following 
>>>>>> builds.
>>>>>> Personally I'd prefer to keep the present numbering and just 
>>>>>> document the
>>>>>> fact that we skipped 60.  I tend to view build numbers as part of 
>>>>>> the
>>>>>> calendar.  If you don't do something on a particular day then that
>>>>>> doesn't mean you remove the day from the calendar, it just means 
>>>>>> that
>>>>>> you do it on some later day.
>>>>>>
>>>>>> What do others think?
>>>>>>
>>>>>>
>>>>>>   
>>>>>
>>>>> I'd prefer to renumber b61 to b60 is b60 is skipped.
>>>>>
>>>>> I think it is less confusing long term to have the build numbers 
>>>>> be a dense consective sequence of integers.
>>>>>
>>>>> -Joe
>>>>>
>>