[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

re-thinking architecture qualification for jessie

I'm not happy how the architecture qualification for wheezy did go (as
communicated in the session about the status of the release at DebConf 2012).  I
did criticize the attitude of the release team as overly optimistic ("green
light attitude"), and I do see that at least GCC and binutils don't have any
support from some ports in Debian.  I would like to see this qualification to
happen earlier in the jessie release cycle, and to be more honest, maybe in that
members of the release team which are port maintainers as well, are not involved
with the release qualification for that port. For wheezy, there are some green
crosses in the architecture qualification matrix which from my point of view are
not correct.

 - At the time when the mips porter box was marked as available, it
   was not available, and then again, during the freeze, it was about
   five months down, without anything happening (yes, contacted debian-admin).

 - The number of porters was marked as green for every architecture, but
   when asking for help in GCC and binutils I never got this help.  From
   my point of view this includes addressing architecture specific issues
   in the toolchain.  Some people for ports do have access to the GCC
   packaging, and do use it.  Other port specific issues are addressed by
   patches, and afaics no issues are outstanding.  I do consider this
   involvement as sufficient for non-release architectures like alpha,
   hppa, m68k, x32, sh4, powerpcspe.

   It's easy to remove java support for some architectures, but as long
   as you don't have a replacement for the toolchain, it's difficult to
   maintain such a port (haven't seen clang supported on these archs yet).

 - Release critical issues are almost only searched on x86, and I assume
   that some of these are ignored for other architectures if not found on
   x86.  A full archive rebuild for every release architecture doesn't
   seem feasible, however defining a subset of packages which have to
   be buildable on every architecture seems like a doable idea, at least
   for some architectures. There seems to be fast enough hardware available
   at least for s390x, ppc64, powerpc, and the arm architectures could
   compensate with the numbers of build machines for such a test rebuilds.

gabrielli as the porter box is now up again, but I don't see any real support
for mips, mipsel, s390, sparc, and maybe powerpc within Debian.  Please consider
toolchain maintenance when starting the architecture qualification for jessie.
As Roger Leigh noted, non-recent toolchain versions may cause extra work for
Debian maintainers.


Reply to: