SQL: groupby/slt_good_12/q6123 returns no result in sequential runtime

[philrz]

The following query returns no output in sequential runtime, but should return a NULL as it has in the past.

$ super -f parquet -o tab0 -c "values {col0:83,col1:0,col2:38},{col0:26,col1:0,col2:79},{col0:43,col1:81,col2:24}" &&
super -c "
SELECT CAST(NULL AS INTEGER) AS col0 
FROM tab0 
GROUP BY col0 
HAVING -MAX(-col0) NOT IN (19);"

[no output]

Details

Repro is with super commit 1699162. The query above is a simplification of the one from sqllogic ztest random/groupby/slt_good_12/q6123.fail.

While the fix in #6425 addressed most sqllogictest queries that started failing at commit d7e8921, this one continues to fail. But like the others, it was returning the expected result before the merge of #6417. Here it was returning the expected NULL result back before #6417 merged.

$ super -version
Version: 8ebde7627

$ super -c "
SELECT CAST(NULL AS INTEGER) AS col0 
FROM tab0 
GROUP BY col0 
HAVING -MAX(-col0) NOT IN (19);"

{col0:null::int32}

Also, even at current tip of main (commit 1699162) the correct result does come back if the query is executed in vector runtime.

$ super -version
Version: 1699162e1

$ SUPER_VAM=1 super -c "
SELECT CAST(NULL AS INTEGER) AS col0 
FROM tab0 
GROUP BY col0 
HAVING -MAX(-col0) NOT IN (19);"

{col0:null::int32}

[philrz]

The changes in #6477 have made it such that this query is now returning no result in vector runtime as well.

$ super -version
Version: 1201d205e

$ SUPER_VAM=1 super -c "
SELECT CAST(NULL AS INTEGER) AS col0 
FROM tab0 
GROUP BY col0 
HAVING -MAX(-col0) NOT IN (19);"

[no output]

That said, I now see that this simplified query should have been returning three NULL values all along

$ psql postgres
psql (17.7 (Homebrew))
Type "help" for help.

postgres=# \pset null 'NULL'
Null display is "NULL".

postgres=# SELECT * FROM tab0;
 col0 | col1 | col2 
------+------+------
   83 |    0 |   38
   26 |    0 |   79
   43 |   81 |   24
(3 rows)

postgres=# SELECT CAST(NULL AS INTEGER) AS col0 
FROM tab0 
GROUP BY col0 
HAVING -MAX(-col0) NOT IN (19);
 col0 
------
 NULL
 NULL
 NULL
(3 rows)

So it seems there's more to it than a simple regression. Rather than treating it as a special case I'll open a new issue when I have a better sense for a general pattern to the problem.