Sven Luther <[EMAIL PROTECTED]> writes: > On Tue, Aug 26, 2003 at 11:17:48AM +0200, Goswin von Brederlow wrote: > > Sven Luther <[EMAIL PROTECTED]> writes: > > > > > On Mon, Aug 25, 2003 at 07:42:13PM +0100, Mark Howard wrote: > Hello, > > > If we don't rebuild every dependant package in the experimental pool, > > > we need then to have a way to eventually retrigger the build of every > > > dependant package once it reaches unstable. > > > > Binaries from experimental should not move into sid but be rebuild for > > it. That can happen automatically by the buildds if source-only > > uploads could be handled. Move the source to sid killing all binaries > > and let the autobuilder crunch away. > > Well, but you need to build all the dependant packages (and use them) to > catch bugs caused by it.
Only those contained in the sections enabled on the users system. If you have experimental-core and experimental-gnome all gnome debs should be comnpiled against the experimental glibc for example. Having all combinations on the mirrors would overflow them. > > > That said, ideally all packages would be rebuilt in the experimental > > > pool with in an automatic triggered way and then only can the library > > > component migrate to unstable, in the same way as stuff migrate from > > > unstable to testing in this way. > > > > > > Special care would need to be taken for uploads of new stuff when one of > > > the library pools is waiting to move to unstable. > > > > Maybe experimental could be a source only section with a proper > > build-experimental package thats preconfigured to build on a local > > maschine and install. > > No, you loose the benefit of the autobuilder, and would result in > experimenting only on x86 this way. Some base set, like experimental-core + experimental-<section this is in>, could be autobuild and any other combinations build on the clients side. MfG Goswin