Michael Paquier <michael@paquier.xyz> writes:
> On Wed, Mar 14, 2018 at 05:36:26PM +1300, Thomas Munro wrote:
>> Hmm. I suppose we could have invented a new extended hook with a
>> different name and back-patched it so that PG10 would support both.
>> Then binary compatibility with existing compiled extensions wouldn't
>> be affected AFAICS, but you could use the new extended hook in (say)
>> 10.4 or higher. Then for PG11 (or later) we could remove the old hook
>> and just keep the new one. I suppose that option is still technically
>> open to us, though I'm not sure of the committers' appetite for messing
>> with back branches like that.
> The interactions between both hooks would not be difficult to define: if
> the original hook is not defined, just do not trigger the second. Still
> that's too late for v10, so I would rather let it go. New features are
> not backpatched.
Yeah. There would be no good way for a v10 extension to know whether the
additional hook is available --- it would have to know that at compile
time, and it can't --- so I don't see that this is practical.
Ideally we'd have noticed the problem before v10 got out ... so my own
takeaway here is that this is a reminder to extension authors that they
ought to test their stuff during beta period.
regards, tom lane
