Differences between Oracle 10G Client ODBC from 10.1.x to 10.2.x (datatype)

Question

What differences where implemented to Oracle 10G Clients ODBC connectivity drivers between 10G Release 1 and Release 2 concerning datatypes?
I have an application that performs a qualified read returning a group of records (this is successful) and then performs a delete on each record read. No errors occur on completion of the application. When using release 1, the records are physically removed. Using release 2 a portion of the same group of records are not removed. The key for the table is a numeric field with six decimal places (not my design). My applications C code defines the field as a DOUBLE.

Answer 1

I certainly agree that something changed. The question is whether it was a change in function or not. With a lot of these sorts of errors, tracking down the "source" of the error involves finding a change in a piece of code that allowed what was always possible to actually occur. That is, you may have two pieces of code that are functionally identical but that end up allowing the sorts of "round off" errors that are always possible when using doubles. The driver might, for example, now be going down a code path where it shifts a number to the right or left rather than dividing or multiplying by 2, which may make this sort of thing more likely. Functionally, given that doubles don't guarantee infinite precision, these operations are identical, but one may be significantly more likely to cause problems.
Is your application using the ODBC API directly? Or is it going through a higher-level API? If you are going through ODBC directly, you might look to see whether the data types that are being returned by the driver to your code are the same across driver versions.
Justin