Re: couple bugs with schema

On Fri, Jan 13, 2006 at 08:49:11AM -0700, Jonathan Ellis wrote:
> That's probably the right thing to do.  Cleaner would be to add "AND
> a.attnum > 0" to the query itself, though, which is what the other
> query on pg_attribute does right now.

That only deals with the case where the first element of indkey is
negative; but sometimes only an element farther down in the array is
negative (browse through "select indkey from pg_catalog.indkey" --
excellent summer reading).  So you'd need something like "AND 0 <
ANY(indkey)", except that indkey is an int2vector and doesn't support
the ANY operator; there is probably a way to do it in sql, but I don't
know what it is.

My patch just stripped out the negative ids from the constraints, only
tossing out the constraint entirely if the constraint didn't contain
any nonnegative values.  That may well be wrong, and result in
constraints being misreported; maybe the whole constraint should be
skipped in there is a negative id in it (which would be equivalent to
the proposed but unworkable SQL above).  It would be better to find
fewer constraints than find ones that don't actually exist.  I don't
have time to research this today, but if no one else knows how these
catalogs actually work in regard to indices and system columns I'll
look into it soon.

j


-- 
Jacob Smullyan

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)

iD8DBQFDx9nKuqamFyFXXLIRAs7yAJ9BTfi6Orwfw2n74/vPmP6T4+BsQgCfQBAH
41xj0esJxVodS9YbYvw5xXM=
=uuTA
-----END PGP SIGNATURE-----