Re: Any way to insert rows with ID used in another column

On 04/03/2014 07:41 AM, Ben Hoyt wrote:
> Hi folks,
>
> We have a table images in our db with id (serial primary key) and
> filename columns, where the filename is a unique text column that looks
> something like "pool-1234.jpg".
>
> The catch is that the "1234" in the filename is the image ID. We want
> the filename to include the image ID because it's a nice way of making
> it unique and gives the benefit of being able to easily map from the
> filename back to the ID for debugging and the like.
>
> Currently I insert new image rows in multiple steps:
>
> 1) begin transaction
> 2) insert a whole bunch of image rows in one multiple-row INSERT, using
> a temporary random filename
> 3) use the RETURNING clause on the above insert to get a mapping between
> the database IDs and filenames just inserted
> 4) loop through all images just inserted, and for each image, execute
> UPDATE to set the filename to the real filename which includes the new
> image ID
> 5) commit
>
> This works, but it's pretty cumbersome, and requires N UPDATE statements
> which is also slow.
>
> Is there some way to do something like this:
>
> INSERT INTO images (filename) VALUES
> ('foo' || image_id_about_to_used_for_this_row::text || '.jpg')
> ('bar' || image_id_about_to_used_for_this_row::text || '.jpg')
>
> I tried using currval() to see if that'd work, but it gave an error, I
> guess because I was using it multiple times per session.

Write a BEFORE INSERT trigger function?

>
> Thanks,
> Ben
>


--
Adrian Klaver
adrian.klaver@aklaver.com