On 26.02.2013 18:40, Tom Lane wrote:
> Heikki Linnakangas<hlinnakangas@vmware.com> writes:
>> On 26.02.2013 18:23, Tom Lane wrote:
>>> (I assume
>>> the backend will bounce the other cases at some post-grammar stage.)
>
>> No. All four combinations of FROM/TO and STDIN/STDOUT are accepted:
>
> Huh. That seems like an odd decision. If we agree that that behavior
> is desirable, then your patch is ok as-is, though I do question whether
> this should be tested in the grammar at all rather than at runtime.
>
> I wonder whether this is just an oversight, or if we did it
> intentionally because people were confused about which combinations
> to use. It seems like maybe "TO STDIN" could be justified if you
> thought about stdin of the recipient rather than stdout of the server,
> but it still seems a bit sloppy thinking.
Yeah, I'd guess that it was an oversight. But it goes back all the way
to Postgres95, so it's a bit too late to change that.
- Heikki