On Thu, Apr 15, 2010 at 07:57:18PM +0200, Lars Marowsky-Bree wrote: > On 2010-04-15T16:11:55, Dejan Muhamedagic <[email protected]> wrote: > > > Hello, > > > > The cluster glue release 1.0.4 contains a change which breaks > > ABI. That would break all Heartbeat and Pacemaker installations > > which weren't built against that release. We could discuss more > > about what and how, but let's keep that for one of the future > > releases. There is now a new cluster-glue release 1.0.5 which > > keeps the same ABI. > > I think this is worth discussing. > > Personally, I prefered the previous approach (just adding the two > fields) even if it modified the ABI; I don't want to see a new > connection type. > > Arguably, the soname should have been incremented. > > The package managers then would have taken care of this and noticed that > the user needs newer versions of their dependencies. > > Can we add the patch back and increment the soname for 1.0.6?
We may add the functionality back, in case anyone actually needs it. But certainly not breaking backwards ABI compatibility, if not absolutely necessary. Which it is not for this thing. Also, users of new features should have an easy way to determine availability of new features, and should have the option to decide at runtime if they are going to live without them or error out with a meaningful message. Just failing hard because the .so version does not match is not good enough, IMO. Question is: is this functionality actually going to be used, or was it an experimental proof of concept but the CIB ACLs will be implemented without them anyways? If it's not going to be used anyways, no need to discuss this any further. -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT, Austria. _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
