Re: WIP: System Versioned Temporal Table

On Fri, Jan 8, 2021 at 7:34 AM Simon Riggs <simon(dot)riggs(at)enterprisedb(dot)com> wrote:
>
> On Fri, Jan 8, 2021 at 7:13 AM Simon Riggs <simon(dot)riggs(at)enterprisedb(dot)com> wrote:
>
> > I've minimally rebased the patch to current head so that it compiles
> > and passes current make check.
>
> Full version attached

New version attached with improved error messages, some additional
docs and a review of tests.

* UPDATE doesn't set EndTime correctly, so err... the patch doesn't
work on this aspect.
Everything else does actually work, AFAICS, so we "just" need a way to
update the END_TIME column in place...
So input from other Hackers/Committers needed on this point to see
what is acceptable.

* Anomalies around use of CURRENT_TIMESTAMP are not discussed or resolved

* No discussion, comments or tests around freezing and whether that
causes issues here

* What happens if you ask for a future time?
It will give an inconsistent result as it scans, so we should refuse a
query for time > current_timestamp.

* ALTER TABLE needs some work, it's a bit klugey at the moment and
needs extra tests.
Should refuse DROP COLUMN on a system time column

* Do StartTime and EndTime show in SELECT *? Currently, yes. Would
guess we wouldn't want them to, not sure what standard says.

* The syntax changes in gram.y probably need some coralling

Overall, it's a pretty good patch and worth working on more. I will
consider a recommendation on what to do with this.

--
Simon Riggs http://www.EnterpriseDB.com/