Bug#924548: gnome-core: does not actually install a desktop environment on s390x

[Andreas Henriksson]

Hi,

On Thu, Mar 14, 2019 at 10:29:06AM +0000, Simon McVittie wrote:
> Package: gnome-core
> Version: 1:3.30+1
> Severity: serious
> Justification: we should make a decision one way or another before release
> 
> I recently changed gnome-core to install gnome-flashback instead of
> GNOME Shell on s390x, because: GNOME Shell is uninstallable on s390x, due
> to mozjs60 not being functional there; testing migration requires task
> packages to be installable; [....]
> Options are:
[...]
> Does anyone have any input on this?

(I am not entirely serious about this being an option, but not entirely
joking either.)

An additional option here which I'd like to propose would be to stop
mucking around with dependencies to try to hide problems on those
architectures. Instead "simply" do arch-specific removal of the broken
packages, including all their reverse dependencies. If the packages are
removed from unstable, they (as far as I can tell from previous
experience) will:
- also automatically be removed from testing
- no longer count as having previously built and thus not qualify
  as an RC bug as far as policy is concerned.
- no longer block testing migration (until they happen to build and then
  fail again, so make sure they continue to fail to build persistently
  unless you want to go through the entire removal dance again)
- likely not affect any real users (as you previously already concluded)
- likely show up in the normal "arch-porting needed" workflow/analysis

The debian social contract says we do not hide our problems so removing
broken packages is in my opinion way more honest than trying to paper
over their problems. Additionally, our users are supposed to be our
priority: since the removal would not affect any users and free up
developer time/resources which could instead be invested in things that
actually affects real users.

The only backlash I can possibly forsee with going down the removal
route would be if you manage to get so many packages that you decrease
the total number of built packages on the affected architecture that
it no longer qualifies as a release arch, which would probably upset
someone but also very much highlight the "arch-porting needed". ;P

The only real downside to this option that I can see is that it might be
hard to find someone who wants to work on pushing the removal through as
getting things removed is probably the hardest task possible in Debian
(even if it's completly non-functional stuff).

Regards,
Andreas Henriksson