When you write numeric values to Oracle, you might encounter precision loss.

For example, you might encounter this problem when you perform steps similar to the following:

1. Create a table in Oracle named MYDEC.
2. Add values (including 104.88) to the table.
3. Connect to SAS^® Federation Server.
4. Run code similar to the following to read a value with type NUMERIC(15,2) from an Oracle database:​​​​
   libname fedora fedsvr server="server-name" port=24141 user="user-ID"
   password="server-password"  dsn="oracle-dsn" catalog="​​​​​​​c​​​​​atalog-name"​​​​​​​ schema="​​​​​​​schema-name"​​​​​​​ ;
   proc sql ;
   drop table fedora.feddec;
   create table fedora.feddec as select * from fedora.mydec;
   select * from fedora.feddec;
   * got value 104.87 ;
   quit ;
5. Write the value to an Oracle database. The value is stored as 104.87 in the Oracle database.

This loss of precision is typical of floating-point arithmetic. Base SAS^® uses an IEEE double format (8 bytes) to save numeric values. This format does not specify exact values, and it supports precision up to 15 digits. 

For example, with this format, 104.88 in Oracle might be read into SAS as 104.879999999999995. When writing the SAS value back to Oracle with format NUMERIC(16,2), SAS incorrectly truncates the value to 104.87.

Click the Hot Fix tab in this note to access the hot fix for this issue.

Operating System and Release Information

A fix for this issue for SAS Federation Server 4.2 is available at: