probabilistic approach to tinderboxing

[Norbert Thiebaud]

On Wed, Jun 13, 2012 at 8:53 AM, Bjoern Michaelsen
<bjoern.michaelsen at canonical.com> wrote:
> But we are doing that not now as pushes are atomic (or more precise:
> ref-updates are), so tinderboxes wouldnt see a half-push.

true. but just because we don't, doesn't mean we shouldn't

> Also all patches in a
> series have the same owner, so blaming is easy.

blaming is one thing, quality is another.

> The first goal of the
> tinderboxes is to have the master tip buildable all times -- having it
> buildable at every random commit is secondary.

secondary but important nonetheless.

> Also: bibisect etc. could be
> taught to only look for commits where the next commit date is at least 5 minutes
> off.
No, that would not help. two commit cane within 5 minutes and yet
independant and 2 commit can be 5 minutes apart and yet belong to the
same patch-serie
iow, the timestamp of a commit is not very helpful in general wrt to tinderbox

Norbert