Adrian Klaver writes:
No I am saying that in the ALTER data_type case the column is not being
created and USING is working on data(assuming data had actually been
entered already) that exists. What you propose is a two step process,
create a column and then fill it with a default value that goes away
after the ALTER TABLE ADD COLUMN statement. In fact what you are already
doing.
I do see a use-case that's not covered by ADD COLUMN ... DEFAULT
but could be covered with USING: when you want to initialize the new
column with data taken from some other existing column(s).

Whether this comes up often enough to justify a new feature isn't
clear. You could get the same effect, for pretty much the same cost,
with
  1. ADD COLUMN new_col, not specifying any default;
  2. UPDATE ... SET new_col = some expression of other columns;
  3. ALTER COLUMN new_col SET DEFAULT, if needed.

If you need to make the column NOT NULL, that could be done after step 3,
but then you'd incur another table scan to verify this constraint.
So a USING clause could save you that extra scan.

But if you add another quantum of complication, namely that the new
column's data has to come from some other table, USING would fail at that;
you're back to having to do it with UPDATE. So it seems like there's
only a pretty narrow window of applicability for this proposed feature.
I'm having a hard time getting excited about it.

    regards, tom lane

Search Discussions

Discussion Posts

Previous

Follow ups

Related Discussions

Discussion Navigation
viewthread | post
posts ‹ prev | 10 of 11 | next ›
Discussion Overview
grouppgsql-general @
categoriespostgresql
postedAug 1, '13 at 10:50p
activeAug 3, '13 at 5:35a
posts11
users6
websitepostgresql.org
irc#postgresql

People

Translate

site design / logo © 2018 Grokbase