On 08/21/2014 02:09 PM, Marko Tiikkaja wrote:
> On 8/21/14, 1:19 PM, Heikki Linnakangas wrote:
>> On 08/07/2014 01:11 AM, Marko Tiikkaja wrote:
>>> On 7/21/14, 10:56 PM, I wrote:
>>>> Here's a patch which allows you to notice those annoying bugs with INTO
>>>> slightly more quickly. Adding to the next commit phest.
>>>
>>> New version, fixed bugs with set operations and VALUES lists.
>>
>> Looks good.
>>
>> There's probably more checking like that that you could add, but that
>> can be done as add-on patches, if ever. The INTO mistake happens a lot
>> more easily.
>
> Yeah, I think the mistake is at least as easy to do in "FOR .. IN
> <query>", and I'm planning to add checks for that as well. But I
> haven't found the time to look at it amongst all the other patches and
> projects I have going
Ok.
> (and also, unfortunately, I'm on vacation right now).
Oh, have fun!
>> It seems weird to pass a PLpgSQL_row struct to check_sql_expr.
>> check_sql_expr only needs to know how many attributes is expected to be
>> in the target list, so it would be more natural to just pass an "int
>> expected_natts".
>
> I'm not sure about this, though. AFAICT all the interesting cases are
> already holding a PLpgSQL_row, and in that case it seems easier to just
> pass that in to check_sql_expr() without making the callers worry about
> extracting the expected_natts from the row.
Hmm. The integer FOR syntax I used in my example does not, it always
expects 1 output column.
> And we can always change
> the interface should such a case come up, since the interface is
> completely internal. Just my 0.02EUR, of course.
You might want to add a helper function to count the number of
attributes in a PLpgSQL_row. Then the check_sql_expr call would be
almost as simple: check_sql_expr(..., get_row_natts(row)).
- Heikki
