On Wed, Sep 9, 2009 at 10:40 PM, Andrew Dunstan<andrew@dunslane.net> wrote:
> Robert Haas wrote:
>>
>> I wonder whether it would be appropriate to do something like
>> implement a method by which copy could return text[] and then one
>> could write wrappers around that functionality to do this as well as
>> other things.
>
> It's not a function ... writing a copy() function returning text[] (or setof
> text[], which would be better) in plperlu would be trivial. But it couldn't
> read from stdin or be able to be fed data from a client in copy-in mode.
Yes, setof text[] is what I meant. Making it be able to read from
stdin or be fed data from a client in copy-in mode would be the whole
point of the feature, I suppose.
> And even if we have it what does the sql look like to insert the values into
> some table with n columns?
INSERT INTO table
SELECT t[1], t[2], t[3], t[4], ..., t[n]
FROM (...however we get the copy results...)
...although I'm not entirely sure that's compatible with the
client/server protocol.
> Anything along these lines in C is likely to be far larger than what I had
> in mind, which was a fairly minor enhancement.
It wouldn't be the first time that someone was told that a particular
enhancement didn't have enough value and that they had to do something
larger if they wanted it in core, but on the other hand, I am only
expressing my opinion, which is not binding on you or anyone else, nor
even a fully-fleshed-out proposal. All I'm saying is that it seems to
me that there is value in being able to use the CSV (or other) parsing
code, but have some way to make modifications to how/where the data is
actually inserted. However, I'm skeptical about whether the specific
thing you want to do after parsing (namely, drop excess columns,
null-fill missing ones) is sufficiently common to warrant a feature to
do only that. YMMV, of course.
...Robert
