Re: Bug#791230: GCC5 transition and jessie-backports
2015-08-30 15:46 GMT+02:00 Andreas Metzler <firstname.lastname@example.org>:
> In gmane.linux.debian.backports.general Daniel Pocock <email@example.com> wrote:
>> As part of the GCC5 update, one of the binary packages built from
>> openrpt has been renamed from libopenrpt1 to libopenrpt1v5 - see bug
>> #791230 - and that is the version that will be in testing.
>> If I understand the rules for backports correctly, backports is supposed
>> to contain the same thing that is in testing.
>> Given that unstable and testing will use GCC5 and jessie-backports
>> should be using the GCC 4.9 from jessie, what is supposed to happen?
>> Will GCC5 be available to compile for backports and should such packages
>> declare a dependency on the newer libstdc++? Or is it permitted to make
>> uploads to jessie-backports without the "v5" suffix, so the package name
>> in backports won't be the same as the package name in testing? Or
>> something else?
> afaict the only way to handle this is to undo any -v5 renaming for
> Backporting gcc5 and using it to compile selective backports is not an
> option since one cannot mix (all) c++ code built with gcc49 and with
> gcc5. (Except for those libraries that do not need a gcc5 transition).
> Later on it might even become necessary to actually make a reverse gcc
> transition (v4) for backports of C++ libraries which had a soname bump
> in sid after gcc5 and therefore do not carry a v5 name.
> That is all just afaict, and needs confirmation by backports admin
> with better C++ foo than me.
It seems reasonable to me. It think we would also need to add
Breaks/Replaces to the unstable/testing versions:
testing: libfooN (Breaks: libfooNv4; Replaces: libfooNv4)
I plan back-porting some affected packages but it would be nice if
Matthias and backports admins could agree on the best practice first
and add this to the transition documentation: