Thread: indices and casts

I have a number of timestamp columns on various tables I'd like also
to be able to query against as dates.  I can easily enough say "SELECT
* FROM foo WHERE bar::date = '2004-10-11'" and the like, but such
queries are inevitably sequential scans, and the tables are rather
large for that to be as performant as I'd like.

Is it possible to do something functionally equivalent to "CREATE
INDEX name ON table USING btree(column::date)"?  That statement yields
a syntax error on the "::".  Trying to create the index using the
date_ops opclass yields an error to the effect that date_ops don't
work on timestamp values.

The only alternatives I can see are: to issue queries of the form
"SELECT foo FROM bar WHERE baz BETWEEN '2004-10-11' AND '2004-10-12'",
which is do-able, since most of our SQL is generated, but awkward and
prone to being forgotten in the odd case where it's not generated; or
adding another column of type date, which seems awkward and prone to
causing confusion between the date and timestamp columns.  What am I
missing (alternatively, what am I smoking)?  7.4.2 on RH9, if it
helps.

/rls

--
:wq

Rosser Schwarz <rosser.schwarz@gmail.com> writes:
> Is it possible to do something functionally equivalent to "CREATE
> INDEX name ON table USING btree(column::date)"?  That statement yields
> a syntax error on the "::".

CREATE INDEX name ON table USING btree((column::date))

See the docs concerning indexes on expressions (formerly called
functional indexes).  The parens are needed to avoid syntactic ambiguity
associated with the optional opclass name ...

            regards, tom lane

while you weren't looking, Tom Lane wrote:

> CREATE INDEX name ON table USING btree((column::date))

Nested parens.  D'oh.

It actually occurred to me to wonder if that might be the fix.  But I
thought, nah, too easy.

Thanks once again, Tom.

/rls

--
:wq