Re: postgres_fdw bug in 9.6

On Sat, Apr 8, 2017 at 12:03 AM, Robert Haas <robertmhaas(at)gmail(dot)com> wrote:
> On Mon, Apr 3, 2017 at 2:12 AM, Etsuro Fujita
> <fujita(dot)etsuro(at)lab(dot)ntt(dot)co(dot)jp> wrote:
>> On 2017/04/01 1:32, Jeff Janes wrote:
>>> On Thu, Mar 23, 2017 at 5:20 AM, Etsuro Fujita
>>> <fujita(dot)etsuro(at)lab(dot)ntt(dot)co(dot)jp <mailto:fujita(dot)etsuro(at)lab(dot)ntt(dot)co(dot)jp>> wrote:
>>> Done. Attached is a new version of the patch.
>>> Is the fix for 9.6.3 going to be just a back port of this, or will it
>>> look different?
>>
>> +1 for backporting; although that requires that GetForeignJoinPaths be
>> updated so that the FDW uses a new function to create an alternative local
>> join path (ie, CreateLocalJoinPath), that would make maintenance of the code
>> easy.
>
> Well, the problem here is that this breaks ABI compatibility. If we
> applied this to 9.6, and somebody tried to use a previously-compiled
> FDW .so against a new server version, it would fail after the upgrade,
> because the new server wouldn't have GetExistingLocalJoinPath and also
> possibly because of the change to the structure of JoinPathExtraData.
> Maybe there's no better alternative, and maybe nothing outside of
> postgres_fdw is using this stuff anyway, but it seems like a concern.

I had submitted a patch in [1]. We thought that that patch is good to
fix the issue on the backbranches. But it got berried in the thread.
If you think that's a feasible solution for backbranches, I will work
on the comments.

--
Best Wishes,
Ashutosh Bapat
EnterpriseDB Corporation
The Postgres Database Company