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

Re: Packages built for arm64 with old toolchain, Aug 3-9



+++ Edmund Grimley Evans [2015-09-04 11:56 +0100]:
> The buildd arm-arm-03 failed to update its chroot on Aug 2. As a
> result, in the period Aug 3-9 it built various packages using g++-4.9
> when people were expecting it to use g++-5. So there are library
> packages, such as libsigc++-2.0-0v5, which are intended to handle the
> transition to GCC 5 but were unexpectedly built with the old compiler.
> The consequences are sometimes unpleasant and non-trivial to diagnose.
> For example, the package synfig was successfully built, but the
> command synfig crashes in some cases and that prevented synfigstudio
> from building.
> 
> I'm not going to copy the commands I used to get this list, because
> they're just too embarrassing, but there's a list below of the 37
> source packages that were (earlier today) last built using
> build-essential_11 on arm-arm-03 in the period Aug 3-10. Most are
> probably harmless, because they don't use C++, don't use libraries, or
> don't have any of the critical types in their APIs, but some have been
> shown to be problematic.
> 
> What to do, if anything?

Look to see which ones do in fact have some c++ and rebuild those, I
think. Thanks for generating the list. 

> Does anyone know for sure how to detect which packages really should
> be rebuilt?

I've read the gcc c++ transition materials but am still quite vague
about exactly how to tell. I've also been told that quite a lot of
affected packages are going to get rebuilt once the 'inner' libraries
are transitionned, so there is no need to worry about stuff that will
get rebuilt later anyway.

I'm not sure how this is being co-ordinated. I have also failed to
determine whether a new sourceful upload or a binnmu is preferred for
any reason. Anyone know?

> Obviously one could do nothing for now and just check again in 6
> months' time, when many of the packages will have been rebuilt anyway,
> but during those 6 months some developer time may have been wasted in
> investigating weird failures.

I think we should look through the list and actively bin-nmu anything
that looks like it might be affected, to avoid knock-on breakage.

Wookey
-- 
Principal hats:  Linaro, Debian, Wookware, ARM
http://wookware.org/


Reply to: