Thank you very much for this explanation/reply. It precisely answers
my question.
Unfortunately, it prompts a new question. I am using 8.4.2 which I
assume is new enough to trigger a "yes" response to "If you have a
version new enough to have synchronize_seqscans...". I have
absolutely no idea how to turn that off. Perhaps the best thing would
be to direct me to the documentation where turning it off is described
so that I can become more autonomous. However, accompanying that with
explicit directions would be welcome too.
I am in Greenwich +1 timezone, but I fear you are in the 2AM time
zone. Thank you again,
John
On Feb 24, 2010, at 8:06 AM, Tom Lane wrote:
> John Gage <jsmgage@numericable.fr> writes:
>> This is a two-part question:
>> 1) I have a source_text that I want to divide into smaller subunits
>> that will be contained in rows in a column in a new table. Is it
>> absolutely certain that the initial order of the rows in the
>> resultant
>> table after this operation:
>
>> CREATE TABLE new_table AS SELECT regexp_split_to_table(source_text,
>> E'regexp') as subunits FROM source_table;
>
>> will be the same as the order of these subunits in the original text?
>
> If you have a version new enough to have synchronize_seqscans, you'd
> need to turn that off. Otherwise should be OK.
>
>> 2) I would like to be able to create a serial-type column during
>> CREATE TABLE AS in the new table that "memorizes" this order so
>> that I
>> can reconstruct the original text using ORDER BY on that serial
>> column. However, I am stumped how to do that.
>
> I think the trick is to get the SRF to be expanded before the serial
> values are assigned. There's more than one way to do it, but I think
> (too tired to experiment) this would work:
>
> CREATE TABLE new_table (id serial, subunits text);
>
> INSERT INTO new_table(subunits) SELECT
> regexp_split_to_table(source_text,
> E'regexp') FROM source_table;
>
> regards, tom lane