Re: canu - not migrating to testing despite several weeks of no excuses

["Adam D. Barratt" <adam@adam-barratt.org.uk>]

On Fri, 2017-10-13 at 00:39 -0400, Afif Elghraoui wrote:
> 
> على الخميس 12 تشرين الأول 2017 ‫01:48، كتب Adam D. Barratt:
> > The britney log says:
> > 
> > trying: canu
> > skipped: canu (0, 2975, 79)
> >     got: 43+0: a-4:i-26:a-2:a-1:a-1:m-1:m-4:m-1:p-1:s-2
> >     * arm64: canu
> > 
> > Which indicates that the binary package "canu" would be
> > uninstallable
> > on (at least) arm64 were the package to migrate. A little
> > investigation
> > leads to this being due to the dependency on "mhap", which in turns
> > depends on "libssw-java", and:
> > 
> > libssw-java | 1.1-1+b1      | testing          | amd64
> > libssw-java | 1.1-1+b1      | unstable         | amd64, kfreebsd-
> > amd64
> > 
> > This also means that your package has an unreported RC bug in
> > unstable
> > right now, as it cannot possibly be installable on any
> > architectures
> > other than amd64 and kfreebsd-amd64.
> 
> Thanks for the explanation.
> 
> libssw uses some x86-specific processor features if I remember
> correctly. Is it not possible to depend on it without incurring an RC
> bug?

Yes, trivially - by depending on it only on architectures where it
actually exists. Your problem is that e.g. your armel packages end up
depending on a package that only exists on amd64.

Your realistic options depend on how strongly canu requires mhap (can
it work without it on non-amd64 architectures?), and how strongly mhap
requires libssw (again, can it work without it on other architctures?).

Regards,

Adam