问题
I am working in c# .net 2 (Visual Studio 2005 SP1) attempting to fill a dataset with the results from a select * from table from an Oracle10g database. The .net framework, IDE and database cannot be changed at this client site.
I'm connecting using the ODP.net provider the dll version is 2.102.2.20
When I run the fill command I get an Exception:
Arithmetic operation resulted in an overflow
Also if I attempt to view the offending column in the Visual Studio designer (Show Table Data) I get for every row for this column in the table. The code works perfectly if my query selects other columns with integers for example omitting this column.
The column in question looks fine when I view it in the database from Toad, data looks like:
919.742866695572
I need the precision as it's required for a monte carlo simulation.
If instead of using a data adapter to fill the datatable I use a datareader and call dataReader.getValue(columnIndex) I get the same error but if I call dataReader.GetOracleDecimal(columnIndex) then I get the result I am looking for, no error.
I would rather use data adapter and filling a dataset (note these are untyped datasets as I couldn't get autogenerated strongly typed datasets to work from an oracle db). I don't want to use datareader and walk through the results (pick out the column values) as I am trying to write this as a generic method to work for many scenarios regardless of number of columns, index of decimal columns that would require specific get calls by datatype.
Can anyone help? Can I use new versions of the ODP.net dlls to connect to the older Oracle10g database? am wondering if this will help.
Thanks
回答1:
The problem is that the precision of the result value is too high to convert to a System.Decimal without some data loss. I forget the exact number of digits allowed, but it is around 18 or so. Is it acceptable to round() the result value to that many digits? In the example you gave, a round(MyColumn, 15) or so should be sufficient...
回答2:
You can try the latest version of ODP.Net (11g). It is backwards compatible. I use it to connect to a 10g database just fine. I think it should work with VS 2005 too. For clickonce deployment, just add the dlls referenced by this question: What is the minimum client footprint required to connect C# to an Oracle database? An important note is that if you have the latest version of odp.net, all of the dlls are included in the install directory. You don't need to download instant client separately. Just search for them.
回答3:
If you move to a higher version of ODP.NET like 12.x, the error just changes to "Invalid Cast Exception" from "Arithmetic overflow". You would have to CAST numeric values to cut-down insane precision values like 29 or 30 decimal places to make it more practical like 2 to 4 decimals.
To identify the columns and rows which have excessively big decimal values, you can run the SQL below after substituting the MY_SCHEMA, MY_TABLE and the number 10 with say 25 to identify columns that have values over 25 decimal places. This SQL will generate a SQL that should be run to identify the problem columns.
SELECT 'SELECT ' || LISTAGG('MAX(LENGTH(TO_CHAR(ABS(' || column_name || ') - FLOOR(ABS(' || column_name || '))))) - 1 AS decimals_' || column_name || CHR(13)
, CHR(9)|| ', ') WITHIN GROUP (ORDER BY rn) ||
' FROM ' || owner || '.' || table_name || CHR(13) ||
' WHERE ' || CHR(13) ||
LISTAGG('(LENGTH(TO_CHAR(ABS(' || column_name || ') - FLOOR(ABS(' || column_name || ')))) - 1) > 10 ' || CHR(13)
, CHR(9)|| ' OR ')
WITHIN GROUP (ORDER BY rn) AS Nasty_Numbers_Finder_Query
FROM
(
SELECT owner, table_name, column_name,
row_number() OVER ( PARTITION BY table_name ORDER BY rownum) rn
FROM dba_tab_columns
WHERE
OWNER = 'MY_SCHEMA'
AND table_name = 'MY_TABLE'
AND (data_type LIKE '%FLOAT%'
OR data_type LIKE '%NUMERIC%')
) a
GROUP BY owner, table_name
For more information, I have blogged about it here.
来源：https://stackoverflow.com/questions/1043626/odp-net-oracle-decimal-number-precision-problem-when-filling-a-dataset-exceptio