Re: GIN - Generalized Inverted iNdex. Try 2.

> I just built this, and noticed that the regression test for "opr_sanity"
> fails with your patch. I attached the regression.diffs.

Sorry, this part isn't done yet, because we are waiting of community decision..
We don't add regression test yet.

If community don't like to include GIN in core, we make a contrib/gin, but in
this case GIN can't use WAL feature because of WAL interface can't call
user-defined function.

The reason for first diff is a hardcoded 'gist' index:
-- We have to exclude GiST, unfortunately, since it hasn't got any fixed
-- requirements about strategy operators.

SELECT p1.oid, p1.amname, p2.oid, p2.opcname, p3.amopsubtype
FROM pg_am AS p1, pg_opclass AS p2, pg_amop AS p3
WHERE p2.opcamid = p1.oid AND p3.amopclaid = p2.oid AND
p1.amname != 'gist' AND
p1.amstrategies != (SELECT count(*) FROM pg_amop AS p4
WHERE p4.amopclaid = p2.oid AND
p4.amopsubtype = p3.amopsubtype);

Second is right diff.

For the thread one reason is that operations &&, ~, @ defined for anyarray, but
used for particular types.

>
>
> ------------------------------------------------------------------------
>
> *** ./expected/opr_sanity.out Wed Jan 25 18:35:51 2006
> --- ./results/opr_sanity.out Wed Apr 26 08:31:13 2006
> ***************
> *** 778,785 ****
> WHERE p4.amopclaid = p2.oid AND
> p4.amopsubtype = p3.amopsubtype);
> oid | amname | oid | opcname | amopsubtype
> ! -----+--------+-----+---------+-------------
> ! (0 rows)
>
> -- Check that amopopr points at a reasonable-looking operator, ie a binary
> -- operator yielding boolean.
> --- 778,791 ----
> WHERE p4.amopclaid = p2.oid AND
> p4.amopsubtype = p3.amopsubtype);
> oid | amname | oid | opcname | amopsubtype
> ! ------+--------+------+-----------+-------------
> ! 2742 | gin | 2745 | _int4_ops | 0
> ! 2742 | gin | 2745 | _int4_ops | 0
> ! 2742 | gin | 2745 | _int4_ops | 0
> ! 2742 | gin | 2746 | _text_ops | 0
> ! 2742 | gin | 2746 | _text_ops | 0
> ! 2742 | gin | 2746 | _text_ops | 0
> ! (6 rows)
>
> -- Check that amopopr points at a reasonable-looking operator, ie a binary
> -- operator yielding boolean.
> ***************
> *** 825,831 ****
> 783 | 10 | <<|
> 783 | 11 | |>>
> 783 | 12 | |&>
> ! (24 rows)
>
> -- Check that all operators linked to by opclass entries have selectivity
> -- estimators. This is not absolutely required, but it seems a reasonable
> --- 831,840 ----
> 783 | 10 | <<|
> 783 | 11 | |>>
> 783 | 12 | |&>
> ! 2742 | 1 | &&
> ! 2742 | 2 | @
> ! 2742 | 3 | ~
> ! (27 rows)
>
> -- Check that all operators linked to by opclass entries have selectivity
> -- estimators. This is not absolutely required, but it seems a reasonable
> ***************
> *** 847,854 ****
> WHERE p1.amopopr = p2.oid AND p1.amopclaid = p3.oid AND
> NOT binary_coercible(p3.opcintype, p2.oprleft);
> amopclaid | amopopr | oid | oprname | opcname
> ! -----------+---------+-----+---------+---------
> ! (0 rows)
>
> SELECT p1.amopclaid, p1.amopopr, p2.oid, p2.oprname, p3.opcname
> FROM pg_amop AS p1, pg_operator AS p2, pg_opclass AS p3
> --- 856,869 ----
> WHERE p1.amopopr = p2.oid AND p1.amopclaid = p3.oid AND
> NOT binary_coercible(p3.opcintype, p2.oprleft);
> amopclaid | amopopr | oid | oprname | opcname
> ! -----------+---------+------+---------+-----------
> ! 2746 | 2750 | 2750 | && | _text_ops
> ! 2745 | 2750 | 2750 | && | _int4_ops
> ! 2746 | 2751 | 2751 | @ | _text_ops
> ! 2745 | 2751 | 2751 | @ | _int4_ops
> ! 2746 | 2752 | 2752 | ~ | _text_ops
> ! 2745 | 2752 | 2752 | ~ | _int4_ops
> ! (6 rows)
>
> SELECT p1.amopclaid, p1.amopopr, p2.oid, p2.oprname, p3.opcname
> FROM pg_amop AS p1, pg_operator AS p2, pg_opclass AS p3
>
> ======================================================================
>
>
>
> ------------------------------------------------------------------------
>
>
> ---------------------------(end of broadcast)---------------------------
> TIP 5: don't forget to increase your free space map settings

--
Teodor Sigaev E-mail: teodor(at)sigaev(dot)ru
WWW: http://www.sigaev.ru/