Re: Accessing composite type columns in indexes

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Michael Glaesemann <grzm(at)myrealbox(dot)com>
Cc: pgsql-general <pgsql-general(at)postgresql(dot)org>
Subject: Re: Accessing composite type columns in indexes
Date: 2006-03-04 04:31:52
Message-ID: 10962.1141446712@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-general

Michael Glaesemann <grzm(at)myrealbox(dot)com> writes:
> ... it appears that one can't directly access the columns of a
> composite type when creating an index, i.e., neither UNIQUE (foo.bar)
> nor UNIQUE ((foo).bar) work.

You need both, ie something like

create table foo (bar date_co_interval);
create unique index fooi on foo (((bar).from_date));

The outer set of parens is required for any index expression. Basically
that's to fix a grammar conflict against the possible presence of an
index opclass, that is given

create index fooi on foo (x ! y)

is that an infix operator expression "x ! y", or a postfix operator
expression "x !" followed by an opclass name?

The inner set of parens is because "a.b" is always interpreted as a
table and column name. To refer to a column, and then qualify it with
a composite-type field, we require you to write "(b).c" or "(a.b).c".
It'd be legal to write the same index as
create unique index fooi on foo (((foo.bar).from_date));

Make sense now?

regards, tom lane

In response to

Browse pgsql-general by date

  From Date Subject
Next Message Michael Fuhr 2006-03-04 04:34:04 Re: Accessing composite type columns in indexes
Previous Message Michael Fuhr 2006-03-04 04:01:53 Re: record OID to table