On 2018/04/22 2:29, Tom Lane wrote:
> Amit Langote <Langote_Amit_f8@lab.ntt.co.jp> writes:
>> I think if this bug/open item can be resolved by adopting the minimal
>> patch, then we should use it for that. Maybe, we can discuss the rest of
>> the changes independently. If they make things better overall, we should
>> definitely try to adopt them.
>
> Yeah. While I think that getting rid of the grammar restrictions on what
> a partbound can be is a good idea, it seems like this is not the sort of
> improvement to be making post-feature-freeze. And it's certainly not
> something to back-patch to v10.
Agreed.
> I propose the attached slightly-less-invasive version of Amit's original
> patch as what we should do in v10 and v11, and push the patch currently
> under discussion out to v12.
Here too.
>> About the changes in transformPartitionBoundValue() to check for collation
>> conflict, I think that seems unnecessary.
>
> I agree. We can document that the partbound expression is reduced to a
> simple constant and leave it at that. Nobody has yet been confused by
> the possibility of putting COLLATE in a default expression, and I don't
> believe that anybody will be confused here.
Yes, I think so.
> (Speaking of documentation, nobody seems to have noticed that
> partition_bound_spec appears in alter_table.sgml too.)
Oops, thanks for fixing that. Actually, partition_bound_spec wasn't
expanded like it is now in the synopsis of alter_table.sgml at the time
the original patch was written. Commit a2a22057617 (dated Feb 2) added
it, whereas the patch was posted on Jan 29.
Thanks,
Amit