On Wed, Feb 26, 2020 at 06:38:57PM +0800, John Naylor wrote:
> On Wed, Feb 26, 2020 at 3:51 PM David Fetter <david@fetter.org> wrote:
> >
> > I believe the following should error out, but doesn't.
> >
> > # SELECT date_trunc_interval('1 year 1 ms', TIMESTAMP '2001-02-16 20:38:40');
> > date_trunc_interval
> > ═════════════════════
> > 2001-01-01 00:00:00
> > (1 row)
>
> You're quite right. I forgot to add error checking for
> second-and-below units. I've added your example to the tests. (I
> neglected to mention in my first email that because I chose to convert
> the interval to the pg_tm struct (seemed easiest), it's not
> straightforward how to allow multiple unit types, and I imagine the
> use case is small, so I had it throw an error.)
I suspect that this could be sanely expanded to span some sets of
adjacent types in a future patch, e.g. year + month or hour + minute.
> > Please find attached an update that I believe fixes the bug I found in
> > a principled way.
>
> Thanks for that! I made a couple adjustments and incorporated your fix
> into v3: While working on v1, I noticed the DTK_FOO macros already had
> an idiom for bitmasking (see utils/datetime.h),
Oops! Sorry I missed that.
Best,
David.
--
David Fetter <david(at)fetter(dot)org> http://fetter.org/
Phone: +1 415 235 3778
Remember to vote!
Consider donating to Postgres: http://www.postgresql.org/about/donate