Incorrect Decimal Division Coercion #6828
Description
Describe the bug
Currently when performing decimal division coercion_decimal_mathematics_type as called by BinaryExpr::evaluate will coerce both inputs to the wider precision type. The division kernel will then scale the left hand side by the output scale.
To see why this is an issue consider
1 DECIMAL(38, 20) / 5 DECIMAL(38, 0)
This computation shouldn't overflow as it can just perform 10e20 / 5
The issue is that the coercion logic will scale the right hand side to 5e20 requiring it to also scale the left hand side to 10e40 which will overflow
To Reproduce
❯ create table foo (a DECIMAL(38, 20), b DECIMAL(38, 0));
0 rows in set. Query took 0.001 seconds.
❯ insert into foo VALUES (1, 1000000000000);
+-------+
| count |
+-------+
| 1 |
+-------+
1 row in set. Query took 0.003 seconds.
❯ select a / b from foo;
Decimal128(38, 20) vs Decimal128(38, 20)
+------------------------------------------+
| foo.a / foo.b |
+------------------------------------------+
| 0.00000000000000000000000000000000743222 |
+------------------------------------------+
1 row in set. Query took 0.004 seconds.
Expected behavior
No response