On Wed, Apr 28, 2010 at 1:40 PM, Dimitri Fontaine
<dfontaine@hi-media.com> wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
>> Well, how would you define CREATE OR REPLACE TABLE? I think that
>> doesn't make much sense, which is why I think CREATE IF NOT EXISTS is
>> a reasonable approach.
>
> <hand waving time>
>
> The behavior I'd like to have would be to allow me to give a SELECT
> query to run for replacing what is there if there's something. If the
> query can not be run on the existing data set, error out of course.
>
> So you know the state for sure after the command, but it depends on your
> query being correct. And you can (de)normalize existing data using joins.
>
> The REPLACE keyword would here mean that there's a CTAS going under the
> hood, then we add the constraints and indexes and triggers etc. That
> would mean being able to express those entities changes too, but it
> seems important.
>
> Well, that may be not precise enough as a spec, but at least that's food
> for though I hope.
This type of hand-waving convinces me more than ever that we should
just implement CINE, and it should just C if it doesn't already E.
This is what has been requested multiple times, by multiple people,
including various people who don't normally poke their head into
-hackers. I think the resistance to a straightforward implementation
with easy-to-understand behavior is completely unjustifiable. It's
completely unobvious to me that all of the above will work at all and,
if it did, whether it would actually solve the problems that I care
about, like being able to write schema-upgrade scripts that would work
in a simple and predictable fashion.
...Robert