[libvirt] [PATCH] libvirt_public.syms: Fix virDomainSetBlockThreshold placement

[Daniel P. Berrange]

On Tue, Apr 04, 2017 at 11:59:20AM +0200, Michal Privoznik wrote:
> On 04/04/2017 11:56 AM, Daniel P. Berrange wrote:
> > On Tue, Apr 04, 2017 at 09:42:47AM +0200, Michal Privoznik wrote:
> > > On 04/04/2017 09:27 AM, Jiri Denemark wrote:
> > > > On Tue, Apr 04, 2017 at 09:24:41 +0200, Michal Privoznik wrote:
> > > > > The symbol was introduced in the 3.2.0 release (bb09798fbeb5f)
> > > > > and not 3.1.0 release as its current placement in the file
> > > > > suggests.
> > > > > 
> > > > > Signed-off-by: Michal Privoznik <mprivozn at redhat.com>
> > > > > ---
> > > > >  src/libvirt_public.syms | 6 +++++-
> > > > >  1 file changed, 5 insertions(+), 1 deletion(-)
> > > > > 
> > > > > diff --git a/src/libvirt_public.syms b/src/libvirt_public.syms
> > > > > index 428cf2e..37a8a18 100644
> > > > > --- a/src/libvirt_public.syms
> > > > > +++ b/src/libvirt_public.syms
> > > > > @@ -755,8 +755,12 @@ LIBVIRT_3.0.0 {
> > > > > 
> > > > >  LIBVIRT_3.1.0 {
> > > > >      global:
> > > > > -        virDomainSetBlockThreshold;
> > > > >          virDomainSetVcpu;
> > > > >  } LIBVIRT_3.0.0;
> > > > > 
> > > > > +LIBVIRT_3.2.0 {
> > > > > +     global:
> > > > > +        virDomainSetBlockThreshold;
> > > > > +} LIBVIRT_3.1.0;
> > > > > +
> > > > >  # .... define new API here using predicted next version number ....
> > > > 
> > > > Oh, this is unfortunate. But doesn't this change break ABI?
> > > 
> > > 
> > > Well, the change I'm referring to in the commit message broke ABI in the
> > > first place. This will break ABI only for apps linked with 3.2.0 release and
> > > un-break apps linked with every other release.
> > 
> > No, the original change doesn't break ABI at all. It merely added a new
> > symbol to the existing ABI. This means that applications which used the
> > new API would build, link & run just fine - they'll simply not get an
> > automatic dependancy on libvirt 3.2.0 added to the RPM deps. By contrast,
> > changing the version now, will definitely break ABI, since any apps which
> > linked to libvirt 3.2.0 using this API will have unresolveable symbols at
> > link load time.
> 
> The apps linked with 3.2.0 and using the SetBlockThreshold API would get a
> false dependency on 3.1.0 so downgrading libvirt from 3.2.0 to 3.1.0 would
> break for them: rpm would allow it but in the end the app fails to load. But
> this is much narrower use case. I agree. So lets keep the code as is.

What we could do is list the symbol in *both* versions. That way next time
an app is rebuilt, the linker should pick the 3.2.0 tagged symbol, ensuring
it gets the stronger RPM dep, but we still heve the 3.1.0 tagged symbol for
compat

Regards,
Daniel
-- 
|: http://berrange.com      -o-    http://www.flickr.com/photos/dberrange/ :|
|: http://libvirt.org              -o-             http://virt-manager.org :|
|: http://entangle-photo.org       -o-    http://search.cpan.org/~danberr/ :|