Re: r-base-core upload to unstable does not respect freeze policy
On Tue, 11 Nov 2014, Ian Jackson wrote:
> Simon McVittie writes ("Re: r-base-core upload to unstable does not respect freeze policy"):
> > On 11/11/14 15:55, Ian Jackson wrote:
> > > perhaps we should in general make more use of testing chroots for RC
> > > bugfixes to testing during the freeze.
> > For build-time dependencies, I'm not sure that actually helps very much:
> > the buildds for unstable take build-dependencies from unstable.
> This is a good point. (Although if your package has only arch:all
> debs you can build them all yourself against testing.)
> Perhaps the buildds should be reconfigured, during the freeze, to
> build against testing ?
That has some nasty implications as well. If this needs to be addressed,
the likely safest answer would be to have t-p-u built on buildds with
up-to-date testing chroots.
 If you even have to ask why something like this needs to be explicitly
stated, well, you're a lucky one...
AFAIK, the whole policy of not updating to unstable something that is not
supposed to migrate to testing during a freeze is a way to cope with the
extremely non-trivial cost of deploying and managing a fully independent
t-p-u (which is more than an entire new set of autobuilder instances).
This reminds me that, AFAIK, we still don't have a good answer to this
scenario: "what do I do when I notice the toolchain was generating broken
code", which actually means "we should binNMU everything that was built with
the broken toolchain, as well as anything that lifted code from the broken
binaries (static linking, etc)".
"One disk to rule them all, One disk to find them. One disk to bring
them all and in the darkness grind them. In the Land of Redmond
where the shadows lie." -- The Silicon Valley Tarot