问题
I have a table (Oracle 12.1.0.2.0) where when I sum up the values I get
ORA-01426: numeric overflow
01426. 00000 - "numeric overflow"
*Cause: Evaluation of an value expression causes an overflow/underflow.
*Action: Reduce the operands.
When I check the values in SQL-Developer I see in the column "NULL" which should be no issue for the function sum.
When I use dump(col1) for this table the result is
Typ=2 Len=2: 0,3
shouldn't this be NULL for columns with NULL?
Also checks 'is not null' didn't filter rows.
Exporting the data only exported a part of the first row and stopped after the first occurence of this 0,3.
SQL Navigator showed "error" as value for this column.
Table was created as select. Now it is truncated and refilled and NULL rows are NULL and summing up works.
What was the content of that column? And why SQL-Developer shows NULL instead of an error?
回答1:
The column value wasn't null, it was corrupt. The internal representation used for numbers is described in the documentation, or various other places like this.
The first byte is the exponent, and it can - just - be zero, but not with only a 3 following it. The closest I think you can get is 0,3,102 for -9.8*x10^125
From the docs:
The NUMBER data type stores zero as well as positive and negative fixed numbers with absolute values from 1.0 x 10^-130 to but not including 1.0 x 10^126.
So lets look at how some at the extreme ends are stored:
with t (n) as (
select 1 * power(10, -130) from dual
union all select 1 * power(10, 125) from dual
union all select -1 * power(10, -130) from dual
union all select -1 * power(10, 125) from dual
union all select -9.7 * power(10, 125) from dual
union all select -9.8 * power(10, 125) from dual
union all select -9.85 * power(10, 125) from dual
union all select -9.9 * power(10, 125) from dual
)
select n, dump(n) d1, dump(n, 1016) d2 from t
N D1 D2
----------- ------------------------------ ------------------------------
1.000E-130 Typ=2 Len=2: 128,2 Typ=2 Len=2: 80,2
1.000E+125 Typ=2 Len=2: 255,11 Typ=2 Len=2: ff,b
-1.000E-130 Typ=2 Len=3: 127,100,102 Typ=2 Len=3: 7f,64,66
-1.000E+125 Typ=2 Len=3: 0,91,102 Typ=2 Len=3: 0,5b,66
-9.700E+125 Typ=2 Len=3: 0,4,102 Typ=2 Len=3: 0,4,66
-9.800E+125 Typ=2 Len=3: 0,3,102 Typ=2 Len=3: 0,3,66
-9.850E+125 Typ=2 Len=4: 0,3,51,102 Typ=2 Len=4: 0,3,33,66
-9.900E+125 Typ=2 Len=3: 0,2,102 Typ=2 Len=3: 0,2,66
If you specify an arithmetic expression whose value has an absolute value greater than or equal to 1.0 x 10126, then Oracle returns an error.
select 1 * power(10, 126) from dual;
ORA-01426: numeric overflow
Your dumped value of 0,3 doesn't have the 102 at the end denoting a negative number, but if it was positive the first byte would be at least 128.
There have been instances of numbers being corrupted by OCI programs mishandling them, and even legacy import doing the same. Without knowing how the data was originally created you'll probably never know exactly what went wrong, or when, or what the value was originally supposed to be.
It's odd that SQL Developer shows null in the results grid (it seems abort if you query as a script); in SQL*Plus it shows no value even if you set null to a fixed string. SQL Developer, or the JDBC driver, may just be silently swallowing the inability to convert from internal representation.
来源：https://stackoverflow.com/questions/51364254/numeric-overflow-with-null-column-0-3