Toast bug in CVS HEAD

A bug was introduced a while ago by this patch:

> commit 447f7364dd7227a32b58a2aff24f587dd7d7051a
> Author: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
> Date: Sat Apr 12 23:14:21 2008 +0000
>
> Create new routines systable_beginscan_ordered,
systable_getnext_ordered,
> systable_endscan_ordered that have API similar to
systable_beginscan etc
> (in particular, the passed-in scankeys have heap not index attnums),
> but guarantee ordered output, unlike the existing functions. For
the moment
> these are just very thin wrappers around
index_beginscan/index_getnext/etc.
> Someday they might need to get smarter; but for now this is just
a code
> refactoring exercise to reduce the number of direct callers of
index_getnext
> in preparation for changing that function's API.
>
> In passing, remove index_getnext_indexitem, which has been dead
code for
> quite some time, and will have even less use than that in the
presence
> of run-time-lossy indexes.

You get this assertion failure:

TRAP: FailedAssertion("!(key[i].sk_attno ==
indexRelation->rd_index->indkey.values[i])", File: "genam.c", Line: 363)

with this test case:

CREATE DATABASE vdb WITH ENCODING = 'SQL_ASCII';
\c vdb
CREATE TABLE xtable(x text);

-- we must do this because otherwise the strings are too compressible
and don't
-- get toasted externally

ALTER TABLE xtable ALTER x SET STORAGE EXTERNAL;

---retrieving a substr(toasteddatum,x,y) where where x .. x+y spans two
chunks.
INSERT INTO xtable (SELECT REPEAT('ABCDEFGHIJ',400));
SELECT substr(x,1000,2000) from xtable ;

Basically, this comment and code in genam.c:
> ! /*
> ! * Change attribute numbers to be index column numbers.
> ! *
> ! * This code could be generalized to search for the index key numbers
> ! * to substitute, but for now there's no need.
> ! */
> for (i = 0; i < nkeys; i++)
> {
> ! Assert(key[i].sk_attno == irel->rd_index->indkey.values[i]);
> ! key[i].sk_attno = i + 1;
> }

is wrong, because it assumes that there's only one scankey per index
column, but that's not true for toast_fetch_datum_slice(), which uses
two scankeys for the chunkid, to fetch a range. Attached is a patch to
fix that, as suggested in the comment. Comments? I'll apply if not..

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com

Heikki Linnakangas <heikki(dot)linnakangas(at)enterprisedb(dot)com> writes:
> Basically, this comment and code in genam.c:
> ...
> is wrong, because it assumes that there's only one scankey per index
> column, but that's not true for toast_fetch_datum_slice(), which uses
> two scankeys for the chunkid, to fetch a range. Attached is a patch to
> fix that, as suggested in the comment. Comments? I'll apply if not..

Huh, can't believe I missed that that caller might use non-sequential
column numbers.

It's kind of annoying to introduce a search when it's so seldom needed,
though. How about something like

/* fast path for common case */
if (key[i].sk_attno == irel->rd_index->indkey.values[i])
key[i].sk_attno = i + 1;
else
... search as you have it ...

regards, tom lane

Tom Lane wrote:
> It's kind of annoying to introduce a search when it's so seldom needed,
> though. How about something like
>
> /* fast path for common case */
> if (key[i].sk_attno == irel->rd_index->indkey.values[i])
> key[i].sk_attno = i + 1;
> else
> ... search as you have it ...

I doubt it's worth it, given that there's only a couple of columns in
the index and in the scan key anyway.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com