Re: [HACKERS] Performance testing of COPY (SELECT)

Bruce Momjian írta:
> Tom Lane wrote:
>
>> Alvaro Herrera <alvherre@commandprompt.com> writes:
>>
>>> Zoltan Boszormenyi wrote:
>>>
>>>> My v8 had the syntax support for
>>>>     COPY (SELECT ...) (col1, col2, ...) TO
>>>> and it was actually working. In your v9
>>>> you rewrote the syntax parsing so that
>>>> feature was lost in translation.
>>>>
>>> Interesting.  I didn't realize this was possible -- obviously I didn't
>>> test it (did you have a test for it in the regression tests?  I may have
>>> missed it).  In fact, I deliberately removed the column list from the
>>> grammar, because it can certainly be controlled inside the SELECT, so I
>>> thought there was no reason the support controlling it in the COPY
>>> column list.
>>>
>> I would vote against allowing a column list here, because it's useless
>> and it strikes me as likely to result in strange syntax error messages
>> if the user makes any little mistake.  What worries me is that the
>> above looks way too nearly like a function call, which means that for
>> instance if you omit a right paren somewhere in the SELECT part, you're
>> likely to get a syntax error that points far to the right of the actual
>> mistake.  The parser could also mistake the column list for a table-alias
>> column list.
>>
>> Specifying a column list with a view name is useful, of course, but
>> what is the point when you are writing out a SELECT anyway?
>>
>
> If you don't support COPY view TO, at least return an error messsage
> that suggests using COPY (SELECT * FROM view).  And if you support COPY
> VIEW, you are going to have to support a column list for that.  Is that
> additional complexity in COPY?  If so, it might be a reason to just
> throw an error on views and do use COPY SELECT.
>

No, it oes not have any additional complexity,
it uses the same code COPY tablename TO uses.

> Seeing that COPY VIEW only supports TO, not FROM, and COPY SELECT
> support only TO, not FROM, it seems logical for COPY to just support
> relations, and COPY SELECT to be used for views, if we can throw an
> error on COPY VIEW to tell people to use COPY SELECT.
>

The additional hint would be enough if the VIEW case is
not supported.