Describe the bug
When comparing two Timestamp values with different TimeUnits, DataFusion coerces both sides to the coarser time unit. This may result in a loss of precision and lead to incorrect comparison semantics.
And other timestamp operations, such as arithmetic operations and type-unification contexts like UNION, also choose the coarser time unit.
To Reproduce
- compare ts with different time unit.
SELECT
CAST(TIMESTAMP '2024-01-01 00:00:00' AS TIMESTAMP(0)) =
CAST(TIMESTAMP '2024-01-01 00:00:00.123' AS TIMESTAMP(3)) AS eq;
Expected:
+---------------------------------------------------------------+
| Utf8("2024-01-01 00:00:00") = Utf8("2024-01-01 00:00:00.123") |
+---------------------------------------------------------------+
| false |
+---------------------------------------------------------------+
Actual:
+---------------------------------------------------------------+
| Utf8("2024-01-01 00:00:00") = Utf8("2024-01-01 00:00:00.123") |
+---------------------------------------------------------------+
| true |
+---------------------------------------------------------------+
- ts subtraction
SELECT
CAST(TIMESTAMP '2024-01-01 00:00:00.123' AS TIMESTAMP(3)) -
CAST(TIMESTAMP '2024-01-01 00:00:00' AS TIMESTAMP(0)) AS diff;
Expected:
+----------+
| diff |
+----------+
| PT0.123S |
+----------+
Actual:
+------+
| diff |
+------+
| P0D |
+------+
- ts union
SELECT CAST(TIMESTAMP '2024-01-01 00:00:00' AS TIMESTAMP(0))
UNION ALL
SELECT CAST(TIMESTAMP '2024-01-01 00:00:00.123' AS TIMESTAMP(3));
Expected:
+-----------------------------+
| Utf8("2024-01-01 00:00:00") |
+-----------------------------+
| 2024-01-01T00:00:00.123 |
| 2024-01-01T00:00:00 |
+-----------------------------+
Actual:
+-----------------------------+
| Utf8("2024-01-01 00:00:00") |
+-----------------------------+
| 2024-01-01T00:00:00 |
| 2024-01-01T00:00:00 |
+-----------------------------+
Expected behavior
No response
Additional context
I also ran a few similar queries in ClickHouse. It appears that ClickHouse
prefers the finer precision.
Describe the bug
When comparing two Timestamp values with different TimeUnits, DataFusion coerces both sides to the coarser time unit. This may result in a loss of precision and lead to incorrect comparison semantics.
And other timestamp operations, such as arithmetic operations and type-unification contexts like
UNION, also choose the coarser time unit.To Reproduce
Expected:
Actual:
Expected:
Actual:
Expected:
Actual:
Expected behavior
No response
Additional context
I also ran a few similar queries in ClickHouse. It appears that ClickHouse
prefers the finer precision.