On 2017/08/04 2:13, Robert Haas wrote:
> On Thu, Aug 3, 2017 at 12:35 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> Robert Haas <robertmhaas@gmail.com> writes:
>>> On Wed, Aug 2, 2017 at 9:41 PM, Amit Langote
>>> <Langote_Amit_f8@lab.ntt.co.jp> wrote:
>>>> Attached is a patch. I think this could be considered a bug-fix,
>>>> backpatchable to 9.6 which introduced this behavior change [1].
>>
>>> I could go either way on that. It's not inconceivable somebody could
>>> be unhappy about seeing this behavior change in a minor release.
>>
>> FWIW, I vote with the camp that this is a clear bug and needs to be
>> fixed. 9.6 broke a behavior that could be relied on before that.
>> We do not normally hesitate to fix regressions in minor releases.
>>
>> (That's not a vote for the patch as submitted; I haven't reviewed it.
>> But we need to fix this.)
>
> OK. I'm going to commit and back-patch the substantive fix with a
> comment change, but I'm not going to include Amit's documentation
> changes for now because I'm not sure they are going to be sufficiently
> clear. There's not a lot of context for them where he put them.
Thanks for committing the code changes.
About the documentation changes, it seems that the only places where any
description of NOT VALID appears is ALTER TABLE, ALTER FOREIGN TABLE, and
ALTER DOMAIN references pages. Even if the CREATE (FOREIGN) TABLE syntax
allows it, NOT VALID does not appear in the syntax synopsis, so it seems
kind of a hidden feature. Maybe, we should fix that first (if at all).
Thanks,
Amit