Wrong behaviour of array comparison when arrays contain nulls

Started by Lukas Ederabout 4 years ago3 messagesbugs
Jump to latest
#1Lukas Eder
lukas.eder@gmail.com

Hello,

A discussion has been brought to my attention about the behaviour of H2 vs
PostgreSQL when it comes to comparing arrays that contain NULL values, see:
https://github.com/h2database/h2database/issues/3476

For the following query:

SELECT array[1, NULL] = array[1, NULL]

H2 returns NULL whereas PostgreSQL returns TRUE. In my opinion and
intuition, as well as according to ISO/IEC 9075-2:2016(E) 8.2 <comparison
predicate> GR 1) b) ii), H2 is right and PostgreSQL is wrong.

On the above H2 github issue, a slack discussion was linked, to which I
have no access, but the gist of the slack discussion was that NULL is
*identical* to NULL according to the SQL standard, but identical doesn't
mean equal, and the aboe 8.2 GR 1) b) ii) clearly requires all array
elements Xi and Yi to be equal for the arrays to be equal.

Best Regards,
Lukas

#2Lukas Eder
lukas.eder@gmail.com
In reply to: Lukas Eder (#1)
Re: Wrong behaviour of array comparison when arrays contain nulls

For the record, while deviations from the standard related to ROW
expressions are documented here:
https://www.postgresql.org/docs/current/functions-comparisons.html#COMPOSITE-TYPE-COMPARISON

This particular deviation isn't documented on that page. Might be worth
adding?

On Wed, Apr 6, 2022 at 2:22 PM Lukas Eder <lukas.eder@gmail.com> wrote:

Show quoted text

Hello,

A discussion has been brought to my attention about the behaviour of H2 vs
PostgreSQL when it comes to comparing arrays that contain NULL values, see:
https://github.com/h2database/h2database/issues/3476

For the following query:

SELECT array[1, NULL] = array[1, NULL]

H2 returns NULL whereas PostgreSQL returns TRUE. In my opinion and
intuition, as well as according to ISO/IEC 9075-2:2016(E) 8.2 <comparison
predicate> GR 1) b) ii), H2 is right and PostgreSQL is wrong.

On the above H2 github issue, a slack discussion was linked, to which I
have no access, but the gist of the slack discussion was that NULL is
*identical* to NULL according to the SQL standard, but identical doesn't
mean equal, and the aboe 8.2 GR 1) b) ii) clearly requires all array
elements Xi and Yi to be equal for the arrays to be equal.

Best Regards,
Lukas

#3Tom Lane
tgl@sss.pgh.pa.us
In reply to: Lukas Eder (#1)
Re: Wrong behaviour of array comparison when arrays contain nulls

Lukas Eder <lukas.eder@gmail.com> writes:

For the following query:
SELECT array[1, NULL] = array[1, NULL]
H2 returns NULL whereas PostgreSQL returns TRUE. In my opinion and
intuition, as well as according to ISO/IEC 9075-2:2016(E) 8.2 <comparison
predicate> GR 1) b) ii), H2 is right and PostgreSQL is wrong.

If we don't impose a total order on array values, then we cannot build
btree indexes on such columns. So yes, this is a deviation from the
SQL standard, and no we are not going to change it.

As for documentation, section 9.19 says

The comparison operators compare the array contents
element-by-element, using the default B-tree comparison function for
the element data type, and sort based on the first difference.

which implies this behavior but perhaps could be more explicit.

regards, tom lane