9.4.2b1 -> 9.4.2rc1 libisc.so interface change

Adam Tkac atkac at redhat.com
Mon Oct 15 13:51:01 UTC 2007


On Mon, Oct 15, 2007 at 09:43:19AM -0400, Danny Mayer wrote:
> Mark Andrews wrote:
> >> On Fri, Oct 12, 2007 at 02:19:42PM -0400, Danny Mayer wrote:
> >>> Adam Tkac wrote:
> >>>> Hi,
> >>>>
> >>>> I've got report that libisc.so has interface revision number 33 in
> >>>> 9.4.2rc1 (had 32 in 9.4.2b1). I think maintenance releases should
> >>>> keep API/ABI because changes like this will break other applications.
> >>>> Better should be keep old interface for compatibility and remove it
> >>>> in next major release. Btw change of socket_eventlist_delete function
> >>>> doesn't worry me but that all programs linked to libisc.so have to be
> >>>> relinked is my problem.
> >>>>
> >>> For at least that function you shouldn't have to worry since it's an
> >>> internal-only function. Applications (including BIND9) should never be
> >>> calling it.
> >> Yes, you're right, noone should use functions from libisc library. But when y
> >> ou're shipping OS you have to keep API/ABI compatibility during OS life. So i
> >> f We have BIND 9.4 in distro We should have libisc.so.32 during distro life. 
> >> This is main point why I've started this thread - keeping API/ABI compatibili
> >> ty in maintenance releases.
> >>
> >> Adam
> > 
> > 	In general we try very hard not to change the major api number
> > 	in maintenance releases.
> > 
> > 	socket_eventlist_delete() really should be static.  It's
> > 	not actually made public (not in libisc.def).
> 
> Well that's only on Windows! Adam is talking about Linux.
> 
> Danny
>

I'm talking about API interface number change which affects all systems. And due this change all programs on all systems linked to libisc has to be relinked (it doesn't matter that change doesn't affect Linux).

Adam


More information about the bind-workers mailing list