[petsc-dev] does next model mess up our histories

[Barry Smith]

  Sure, but this means we need a “test if I broke everything” infrastructure in addition to next.  Either running nightly or “continuous integration”

   Maybe we should start designing a “continuous integration” capability right now.

   Barry

On Oct 2, 2014, at 10:58 AM, Jed Brown <jed at jedbrown.org> wrote:

> Barry Smith <bsmith at mcs.anl.gov> writes:
> 
>> On Oct 2, 2014, at 9:44 AM, Jed Brown <jed at jedbrown.org> wrote:
>> 
>>> Barry Smith <bsmith at mcs.anl.gov> writes:
>>>>  Yup. Jed is saying our histories should be terrible.
>>> 
>>> ;-)
>>> 
>>> There is a balancing act between "quality" of long-term history, utility
>>> of near-term history, and development effort/opportunity for mistakes.
>>> I personally think the sweet spot is to prefer not to rebase that which
>>> has been merged to 'next’.
>> 
>>   But this is just a fluke of our testing approach. If I had 15
>>   machines I could test all myself and not put major broken things
>>   into next.
> 
> The unique thing that 'next' enables, but continuous integration cannot,
> is having external users actually use the features.
> 
> If your view is that the test suite is a complete and unambiguous
> specification for correct execution, then passing the test suite by
> definition means that a feature is bug-free.  But if you view the test
> suite as a proxy for "does consistent and useful things in user
> applications", the test suite will never be perfect.  Merging to 'next'
> allows "eager" users (including developers working with applications) to
> experiment and give feedback.