Re: Covering Indexes - Mailing list pgsql-hackers

From Jeff Davis
Subject Re: Covering Indexes
Date
Msg-id 1343408377.18860.3.camel@jdavis
Whole thread Raw
In response to Re: Covering Indexes  (Bruce Momjian <bruce@momjian.us>)
List pgsql-hackers
On Thu, 2012-07-26 at 12:13 -0400, Bruce Momjian wrote: 
> So, do we want a TODO item about adding columns to a unique index that
> will not be used for uniqueness checks?

-1 from me, at least in its current form.

At it's heart, this is about separating the constraint from the index
that enforces it -- you'd like the columns to be available for querying
(for index only scans or otherwise), but not to take part in the
constraint.

And when you look at it from that perspective, this proposal is and
extremely limited form. You can't, for example, decide that an existing
index can be used for a new unique constraint. That's a lot more
plausible than the use cases mentioned in this thread as far as I can
see, but this proposal can't do that.

I tried proposing a more general use case when developing exclusion
constraints:

http://archives.postgresql.org/message-id/1253466074.6983.22.camel@jdavis

(allow user to specify multiple constraints enforced by one existing
index). But, at least at the time, my proposal didn't pass the
usefulness test:

http://archives.postgresql.org/pgsql-hackers/2009-09/msg01355.php

even though my proposal was strictly more powerful than this one is.

Also, this proposal extends the weird differences between CREATE UNIQUE
INDEX and a the declaration of a UNIQUE constraint. For example, if you
want DEFERRABLE you need to declare the constraint, but if you want to
use an expression (rather than a simple column reference) you need to
create the index. This problem does not exist with exclusion
constraints.

In my opinion, new innovations in unique constraints would be better
served as part of exclusion constraints, and we should keep unique
constraints simple. If we make an improvement to UNIQUE, then we will
want to do similar things for exclusion constraints anyway, so it just
seems duplicative.

Regards,Jeff Davis






pgsql-hackers by date:

Previous
From: Fujii Masao
Date:
Subject: Re: several problems in pg_receivexlog
Next
From: "Anderson C. Carniel"
Date:
Subject: Re: PostgreSQLs Extension