Bug#791173: libstxxl: library transition may be needed when GCC 5 is the default

[Simon McVittie]

On Sat, 22 Aug 2015 at 18:57:37 +0200, Sebastiaan Couwenberg wrote:
> It looks like libstxxl needs a transition after al.
> 
> At least osrm is a reverse dependency, it cannot be built due to
> undefined references to stxxl::print_msg() and others.

Indeed. When I closed the bug with "no rdeps, don't bother", osrm was
still in the NEW queue, so my check didn't find that package.

On Sun, 23 Aug 2015 at 14:08:16 +0100, D Haley wrote:
> I have contacted upstream to see which of the two options they would prefer.
> My preference is for a soname bump, but this risks being out of sync with
> upstream. I don’t think that is a major problem, as we can fix that on a
> future stxxl release, but if someone is willing to correct me here, let me
> know.

The "SONAME bump" option was only really meant to be taken if the library
had an upstream SONAME bump pending anyway (for instance icu and boost
went this route). If there is not a SONAME change already in the pipeline,
you should do the "v5" rename instead. My NMUs of gtkmm2.4, gtkmm3.0,
atlas-cpp, bullet etc. should make a reasonable template for how this works.

This ABI change is entirely about how the package is compiled *in Debian*.
Upstream should not change the SONAME for this, because there is no
guarantee that the change will be synchronized with the point at which
other distributions make the same libstdc++ ABI transition (for instance
I think Fedora may have already done it, with a mass-rebuild; and there
are probably many distros that have not started yet).

    S