Thread: BUG #17575: row(NULL) IS NULL inconsistent with IS NOT DISTINCT FROM NULL
BUG #17575: row(NULL) IS NULL inconsistent with IS NOT DISTINCT FROM NULL
From
PG Bug reporting form
Date:
The following bug has been logged on the website: Bug reference: 17575 Logged by: Alexey Borschev Email address: aborschev@gmail.com PostgreSQL version: 14.4 Operating system: Ubuntu Description: Hi, PG hackers! I noticed strange behavior of ROW(NULL): I expect, that IS NULL operator should give the same result as IS NOT DISTINCT FROM NULL similarly, IS NOT NULL operator should give the same result as IS DISTINCT FROM NULL: SELECT row(NULL::int) = row(NULL::int) AS "test= " , row(NULL::int) IS NULL AS IS_NULL , row(NULL::int) IS NOT NULL AS NOT_NULL , row(NULL::int) IS DISTINCT FROM NULL AS IS_DISTINCT_FROM_NULL , row(NULL::int) IS NOT DISTINCT FROM NULL AS NOT_DISTINCT_FROM_NULL test= | is_null | not_null | is_distinct_from_null | not_distinct_from_null --------+---------+----------+-----------------------+------------------------ | t | f | t | f But here row(NULL::int) IS NULL -> true, and row(NULL::int) IS NOT DISTINCT FROM NULL -> false ! Functions num_nulls and num_nonnulls consider row(...) as non-nulls: SELECT num_nulls(row(NULL::int), row(NULL::int, NULL::TEXT), row(NULL::int, 'Bob'::TEXT)) , num_nonnulls(row(NULL::int), row(NULL::int, NULL::TEXT), row(NULL::int, 'Bob'::TEXT)) ; num_nulls | num_nonnulls -----------+-------------- 0 | 3 Next point: I expected that IS NULL and IS NOT NULL operators must always return opposite results, but: SELECT row(NULL::int, 'Bob'::TEXT) IS NULL AS Row_IsNULL , row(NULL::int, 'Bob'::TEXT) IS NOT NULL AS Row_NotNULL ; row_isnull | row_notnull ------------+------------- f | f - They both return False on same input! Can we fix or document this PG issue? These tests was done on fresh installation of PG 14 vanilla, Ubuntu, no additional configuration: postgres=# select version(); version ----------------------------------------------------------------------------------------------------------------------------------- PostgreSQL 14.4 (Ubuntu 14.4-1.pgdg20.04+1) on x86_64-pc-linux-gnu, compiled by gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0, 64-bit P.S. It would be nice to have an abbreviation for IS NOT DISTINCT FROM operator, for example == , and have this operator supported in == ANY(...) and JOINs (hash, merge, nested loops)
Re: BUG #17575: row(NULL) IS NULL inconsistent with IS NOT DISTINCT FROM NULL
From
"David G. Johnston"
Date:
On Friday, August 5, 2022, PG Bug reporting form <noreply@postgresql.org> wrote:
The following bug has been logged on the website:
Bug reference: 17575
Logged by: Alexey Borschev
Email address: aborschev@gmail.com
PostgreSQL version: 14.4
Operating system: Ubuntu
Description:
Hi, PG hackers!
I noticed strange behavior of ROW(NULL):
I expect, that IS NULL operator should give the same result as IS NOT
DISTINCT FROM NULL
similarly, IS NOT NULL operator should give the same result as IS DISTINCT
FROM NULL:
SELECT row(NULL::int) = row(NULL::int) AS "test= "
, row(NULL::int) IS NULL AS IS_NULL
, row(NULL::int) IS NOT NULL AS NOT_NULL
, row(NULL::int) IS DISTINCT FROM NULL AS IS_DISTINCT_FROM_NULL
, row(NULL::int) IS NOT DISTINCT FROM NULL AS
NOT_DISTINCT_FROM_NULL
test= | is_null | not_null | is_distinct_from_null |
not_distinct_from_null
--------+---------+----------+-----------------------+------ ------------------
| t | f | t | f
But here row(NULL::int) IS NULL -> true, and row(NULL::int) IS NOT DISTINCT
FROM NULL -> false !
Next point:
I expected that IS NULL and IS NOT NULL operators must always return
opposite results, but:
SELECT row(NULL::int, 'Bob'::TEXT) IS NULL AS
Row_IsNULL
, row(NULL::int, 'Bob'::TEXT) IS NOT NULL AS
Row_NotNULL ;
row_isnull | row_notnull
------------+-------------
f | f
- They both return False on same input!
Can we fix or document this PG issue?
It is documented. The paragraph just below the tip.
David J.
PG Bug reporting form <noreply@postgresql.org> writes: > I noticed strange behavior of ROW(NULL): > I expect, that IS NULL operator should give the same result as IS NOT > DISTINCT FROM NULL > similarly, IS NOT NULL operator should give the same result as IS DISTINCT > FROM NULL: I don't see any particular reason to expect that. row(NULL) is a row object containing one null field, which is in fact not identical to a null composite value. The SQL spec dictates that IS NULL should return true for both cases, but that doesn't mean that no other operator is allowed to distinguish them. I'd say that this is a wart of IS NULL rather than desirable behavior we should copy elsewhere. > Next point: > I expected that IS NULL and IS NOT NULL operators must always return > opposite results, but: You have not read the SQL standard, then. It's quite clear about that. > Can we fix or document this PG issue? ... nor our documentation. See https://www.postgresql.org/docs/current/functions-comparison.html para beginning "If the expression is row-valued,", near the bottom of the page. regards, tom lane
Re: BUG #17575: row(NULL) IS NULL inconsistent with IS NOT DISTINCT FROM NULL
From
Алексей Борщёв
Date:
Thank You, David and Tom! Now these tricks around row(null) are much clearer to me! I think this issue should be closed. пт, 5 авг. 2022 г. в 17:23, Tom Lane <tgl@sss.pgh.pa.us>: > > PG Bug reporting form <noreply@postgresql.org> writes: > > I noticed strange behavior of ROW(NULL): > > I expect, that IS NULL operator should give the same result as IS NOT > > DISTINCT FROM NULL > > similarly, IS NOT NULL operator should give the same result as IS DISTINCT > > FROM NULL: > > I don't see any particular reason to expect that. row(NULL) > is a row object containing one null field, which is in fact not > identical to a null composite value. The SQL spec dictates that > IS NULL should return true for both cases, but that doesn't mean > that no other operator is allowed to distinguish them. I'd say > that this is a wart of IS NULL rather than desirable behavior > we should copy elsewhere. > > > Next point: > > I expected that IS NULL and IS NOT NULL operators must always return > > opposite results, but: > > You have not read the SQL standard, then. It's quite clear > about that. > > > Can we fix or document this PG issue? > > ... nor our documentation. See > > https://www.postgresql.org/docs/current/functions-comparison.html > > para beginning "If the expression is row-valued,", near the > bottom of the page. > > regards, tom lane