Bug#726165: Acknowledgement (transition: protobuf)
Julien Cristau wrote:
> On Sun, Jan 26, 2014 at 12:19:49 +0100, Julien Cristau wrote:
>
> > On Sat, Jan 25, 2014 at 16:57:30 -0500, Robert Edmonds wrote:
> >
> > > I will upload protobuf 2.5.0-5 to unstable shortly. Is there anything I
> > > need to do to schedule binNMUs of the reverse deps or is that handled by
> > > the release team?
> > >
> > Scheduled now.
> >
> And they started failing. At least ia64 and sparc look like protobuf
> itself being broken.
Ugh, sorry! I see the problem now: the architecture-dependent
primitives upstream added in the new version is exported into the
protobuf library's public header files *and pulled in by code generated
by the protobuf compiler*, which means it has to work with the C++
compiler used to build the packages depending on protobuf, not just
protobuf itself.
I've prepared a new protobuf source package which reverts upstream's
weird architecture-dependent reimplementation of pthread_once() to the
portable version that was used in protobuf 2.4.1. The changes since
2.5.0-5 can be seen on the master branch of:
git+ssh://git.debian.org/git/collab-maint/protobuf.git
This successfully builds for me on amd64, i386, powerpc, and sparc, and
I've used the resulting packages to rebuild mosh, mumble, and protobuf-c
by hand on amd64. I don't have any reason to think this will cause
architecture-specific FTBFS's because all the architecture-specific
stuff in libprotobuf-dev's public header files is now gone.
Would you like me to upload this to unstable or do you think it should
go via experimental first?
--
Robert Edmonds
edmonds@debian.org
Reply to: