Subject: Re: abi compatibility policy
To: Jason Thorpe <email@example.com>
From: David Laight <firstname.lastname@example.org>
Date: 02/27/2006 21:36:47
On Sun, Feb 26, 2006 at 06:38:34PM -0800, Jason Thorpe wrote:
> On Feb 26, 2006, at 4:30 AM, YAMAMOTO Takashi wrote:
> >recently, elad made a change to libutil which breaks abi compatibility
> >without bumping shlib major number.
> >while there are no affected releases, the interface has been there for
> >5 months. i don't see any good reason to break it here.
> >i believe that the change should be reverted or major should be
> I'm not a big fan of bumping libutil -- the last time we did it, it
> was painful -- we had to hunt down a smattering of other shlibs that
> linked against it.
> My inclination is to let this one slide and just make sure a snafu
> like this doesn't happen again.
In this case it would probably have been safe to rename the function
as well as changing it's arguments. Then programs running with the
wrong library would fail 'nicely' (or at least slightly less-nastily).
But changing an ABI really does have to be very carefully thought about,
and the then not done (speakes the man who it about to completely
rewrite a library interface at work).
David Laight: email@example.com