[Releasing] rethinking the branching scheme
oswald.buddenhagen at digia.com
Tue Feb 25 15:04:19 CET 2014
On Tue, Feb 25, 2014 at 12:58:07PM +0100, Simon Hausmann wrote:
> On Tuesday 25. February 2014 12.49.23 Sergio Ahumada wrote:
> > I absolutely agree. revdeps should be removed.
> I for one like them. They've helped us numerous times from changes landing for
> example in qtdeclarative that - had they gone in - would have blocked
> qtquickcontrols development. And we've had the same situation in qtbase - for
> build and test failures.
i also think we should not remove revdeps. sadly, they are the only
coverage we have for a lot of code.
in fact, i want *full* revdep coverage - that would be a precondition
for making the qt5 updates a side effect of module integrations. yes,
that's effectively de-modularization at the CI level. this shouldn't be
a problem when we implement my optimization ideas properly.
however, we need some way to make atomic integrations. a
Linked-Change-Id commit message footer and the corresponding logic in
the CI system might be the way to go.
this could be actually done to solve the current issues, irrespective of
it means *quite* some reshuffling in the CI system, though.
More information about the Releasing