Il giorno gio, 19/01/2017 alle 15.19 +0000, Ghislain Vaillant ha scritto: > On Thu, 2017-01-19 at 15:55 +0100, Pietro Battiston wrote: > > Il giorno mar, 17/01/2017 alle 23.26 +0000, Ghislain Vaillant ha > > scritto: > > [...] > > > Assuming, numpy stays in this version for the release, i.e. > > > without > > > the > > > fix for the regression, then we'd be effectively releasing a > > > buggy > > > bottleneck package for Stretch. > > > > I don't consider a software "buggy" just because its tests expose a > > regression of some other software :-) > > This is true if you consider the software *alone*. Debian is a > *distribution*, so we should consider the software and its > environment > as a whole, which includes its runtime dependencies.
We are in the Debian bug tracker, I gave it as granted that the choice depends on what's best for Debian. On the other hand, your reasoning is nice because if it held, it would mean that we'd be better off if upstream had never released the tests in the first place, since we wouldn't even be discussing this and Stretch users wouldn't be at risk of not having bottleneck (which works fine) included. Instead it is in the best interest of Stretch if bottleneck is included _and_ it is in the best interest of Stretch if the regression in numpy is fixed, and the two aspects are pretty orthogonal. Oh, and it is even in the best interest of Stretch if by failing (even just in the logs), its tests help exposing (possible future) regressions in other Debian packages. Then, the optimum is clearly that numpy is fixed and tests are enabled, so let's wait and see. Pietro
