Repository navigation
make_array should support pure null array #7144
Description
Describe the bug
DataFusion supports now heterogeneous arrays (i.e. containing NULL with another data type), but does not support pure null array.
To Reproduce
DataFusion:
❯ select make_array(NULL, NULL, NULL);
+----------------------------+
| make_array(NULL,NULL,NULL) |
+----------------------------+
| [] |
+----------------------------+
❯ select make_array(NULL, 1, NULL, 2, NULL);
+----------------------------------------------+
| make_array(NULL,Int64(1),NULL,Int64(2),NULL) |
+----------------------------------------------+
| [, 1, , 2, ] |
+----------------------------------------------+
Expected behavior
The output should be equal like in other databases (DuckDB, ClickHouse, PostgreSQL):
DuckDB:
D select list_value(NULL, NULL, NULL);
┌──────────────────────────────┐
│ list_value(NULL, NULL, NULL) │
│ int32[] │
├──────────────────────────────┤
│ [NULL, NULL, NULL] │
└──────────────────────────────┘
D select list_value(NULL, 1, NULL, 2, NULL);
┌────────────────────────────────────┐
│ list_value(NULL, 1, NULL, 2, NULL) │
│ int32[] │
├────────────────────────────────────┤
│ [NULL, 1, NULL, 2, NULL] │
└────────────────────────────────────┘
ClickHouse:
SELECT [NULL, NULL, NULL]
Query id: fe6008c8-7d4c-4c1e-96c4-852c900e5530
┌─[NULL, NULL, NULL]─┐
│ [NULL,NULL,NULL] │
└────────────────────┘
1 row in set. Elapsed: 0.001 sec.
SELECT [NULL, 1, NULL, 2, NULL]
Query id: 24391d20-dc13-4624-9df1-befb01edab4a
┌─[NULL, 1, NULL, 2, NULL]─┐
│ [NULL,1,NULL,2,NULL] │
└──────────────────────────┘
1 row in set. Elapsed: 0.001 sec.
PostgreSQL:
postgres=# select array[NULL, NULL, NULL];
array
------------------
{NULL,NULL,NULL}
(1 row)
postgres=# select array[NULL, 1, NULL, 2, NULL];
array
----------------------
{NULL,1,NULL,2,NULL}
(1 row)
Additional context
No response
Activity
alamb commentedon Jul 31, 2023
I can't think of any practical usecase for making making pure null arrays , though maybe I am missing something. Adding lots of code for a case no one is likely to actually use seems unecessary to me.
Thus, perhaps following DuckDB's behavior and picking some arbitrary type like
int32[]is reasonable. Or maybe even generating a "not yet supported" error and waiting for someone to report on their actual usecase before we spend engineering time on sorting out how to make it behave 🤔izveigor commentedon Jul 31, 2023
@alamb In my opinion, we should not pay attention to pure null arrays now, so I agree with you that it is better to wait until another user raises this issue.
jayzhan211 commentedon Aug 6, 2023
How about differentiating empty and null? Is the priority of this high?
select make_array(NULL, 1, NULL, 2, NULL);Expected to get[null,1,null,2,null]instead of[,1,,2,]