Re: First steps towards source-only uploads
On Tue, 19 Aug 2014, Hector Oron wrote:
> 2014-08-15 16:04 GMT+02:00 Ondřej Surý <firstname.lastname@example.org>:
> > I have encountered a situation where the FTBFS bug was caused
> > by segfault in other package. This has forced me to split opendnssec-doc
> > to arch:all package (which was good thing anyway), so there are cases
> > where you want to build the arch:all on most stable and fastest arch.
> > Could we just pick amd64 and be done? :)
> We currently got powerpc or s390x machines producing faster builds
> than amd64, or would you instead base your arch selection on archive
> build coverage which amd64 is probably closer to 100% than any other
The *use* coverage is the major reason, IMO. AMD64 is the arch most likely
to have the best quality assurance, because it is directly used by the vast
majority of the packagers.
s390X is not generally available and we cannot own or replace them easily,
so using it would be a bad idea IMHO: we need to keep our independence.
AFAIK, very powerful AMD64 boxes are easier and cheaper to buy/replace than
very powerful powerpc ones. Especially outside the USA and EU.
> If amd64 was to be picked, what would happen to packages producing
> arch:all packages that do not build on such architecture?
It is a FTBFS error that must be fixed. Nothing new there.
I do think we should ignore FTBFS bugs for arch:all on arches where the
build failure happens due to insufficient resources (memory, storage,
address space, etc).
So, if the FTBFS bug for the arch:all package happens in some other arch
than AMD64 *and* it is not due to resource restrictions on that arch, it is
a FTBFS bug that has to be fixed just the same. This does not discriminate
against other arches.
"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