Am Mittwoch, den 06.01.2016, 01:27 +0100 schrieb Samuel Thibault: > Rene Engelhard, on Tue 05 Jan 2016 22:15:31 +0100, wrote: > > On Tue, Jan 05, 2016 at 09:58:03PM +0100, Rene Engelhard wrote: > > > On Tue, Jan 05, 2016 at 09:50:58PM +0100, Samuel Thibault wrote: > > > > Rene Engelhard, on Tue 05 Jan 2016 21:43:33 +0100, wrote: > > > > > So that it will immediately fail when something eventually > > > > > picks > > > > > up the experimental packages (as experimental buildds right > > > > > now do in some > > > > > situations)? > > > > > > > > ? No, experimental buildds only pick from experimental what is > > > > not > > > > available from sid. If something is available from sid, they > > > > will pick > > > > it up from there, even if experimental has a newer version. > > > > > > That is demonstrateable untrue since the resolver switch a few > > > times ago. > > > > See e.g. > > > > https://buildd.debian.org/status/fetch.php?pkg=libreoffice&arch=i38 > > 6&ver=1%3A5.1.0~alpha1-3&stamp=1446915107 > > > > firebird-dev in sid satifies the build-dep and still it was > > installed from > > experimental. > > Uh, I'm surprised, that's not what I thought was supposed to > happen... > > Samuel >
My observation during the mass-rebuild was to be really careful with Provided packages (here the libpng-dev Package): If another B-D requests explictly libpng12-dev, for example, the resolver WILL install this as it thinks it will make everyone happy. That's why I have a hook in my pbulder setup to *really ensure* that libpng16-dev is installed before starting the real build. Howver, I leave the exact layout of the new package up to the libpng maintainers: They should decide how they want to do it... (IMHO I'd love to have a real libpng-dev package which depends on the real thing; this would enable to version B-Ds on libpng-dev..) -- tobi