probabilistic approach to tinderboxing

[Bjoern Michaelsen]

Hi,

On Wed, Jun 13, 2012 at 02:03:31PM +0200, Eike Rathke wrote:
> Maybe I misunderstood, but the independency and arbitrary order I see as
> a problem. People do work in iterative steps.. so if we could at least
> assure that patches by the same author are ordered correctly that might
> work.

Well, I simplified a bit here. When using gerrit you would push one patch
separate if it is independant and multiple patches as a series if they are
dependant. gerrit handles that correctly as a 'patch series'. I would not want
to testbuild in the middle of one such series of patches -- so whenever there
are multiple dependant patches in gerrit they would be viewed as 'one' by the
tinderboxes. If one has dependant patches _and_ wants to have them tinderboxed
separately one could push them in multiple sets. However, for now I consider
that scenario a cornercase (its usually good enough to test the final state of
a series of dependant patches).

Best,

Bjoern