probabilistic approach to tinderboxing

[Stephan Bergmann]

On 06/14/2012 04:57 PM, Norbert Thiebaud wrote:
> On Thu, Jun 14, 2012 at 1:59 AM, Stephan Bergmann<sbergman at redhat.com>  wrote:
>> On 06/13/2012 03:00 PM, Norbert Thiebaud wrote:
>>
>>
>> I think you are confusing buildability after each patch of a series has been
>> applied in sequence (which is a desirable property indeed) vs. buildable
>> with each patch of a series applied individually ("out of context," what
>> Eike's concern was).
>
> When you apply a series in sequence, at any point the result must be
> buildable...
> 1/ it is good practice
> 2/ if that is not the case bisection is much harder
>
> If you pushed directly to master and happened to push something that
> is not buildable, then of course you can't go back and fix it...
> but with gerrit, as long as your patches series is not 'merged' then
> you can git rebase interactive to fix the offending commit.
>
> Yes, you cannot pick a random patch in a series, apply just that and
> expect it to be buildable
> but if my series has 3 patches 1,2,3
>
> the If I apply (1) (1,2) or (1,2,3) I expect the result to be buildable

Yes sure (and I think nobody is arguing with that).  Its just that I 
felt you guys where talking past each other, one talking about an "out 
of context buildability" problem while one discussing "sequential 
buildability."  So, best forget that I said anything, anyway---seems to 
produce more confusion than clarification...  ;)

Stephan