ResultClassROF can fail to set fields/properties when JDBC driver returns unassignable type about datanucleus-rdbms HOT 9 CLOSED

Created pull request #458

from datanucleus-rdbms.

Please provide a simple testcase that demonstrates the problem. Test suites need to be able to recreate such things.

from datanucleus-rdbms.

How can I write a test case that shows how Datanucleus fetch data from the database (in this case which ResultSet getter has been called)?

I have problems with the mapping of SQL queries to ResultObjects (with Oracle 11g).
I have seen in debugger, that there was a call of ResultSet.getObject() for all my number columns.
In Oracle ResultSet.getObject() deliveres a BigDecimal for all number types.
This leads to error messages like

Query needs to return objects of type "..." but it was impossible to set the field "..." type "java.math.BigDecimal". The field should have either a public set/put method, or be public.

I have to add an setter with a BigDecimal parameter to my result class for all number fields (of type short, int, Integer, Long, ...).

from datanucleus-rdbms.

Using ResultSet.getInt()/getLong()/... hides a problem with setting the values in the result object fields.
For example the BigDecimal got from Oracle has to be converted to an int.
This works on the first row of the ResultSet because of the conversion in the catch in ResultClassROF.getObject() line 456 .
For the following rows ResultClassFieldSetter.set will be called directely with the BigDecimal. At this point, there will be no conversion of the BigDecimal to Integer, because Integer is not assignable from BigDecimal (ResultClassROF:889).
This results in the 021204 NucleusUserException in ResultClassROF:343 mentioned above.

from datanucleus-rdbms.

My test case can reproduce the problem but I was able to reproduce the problem with Oracle only.
Strangely it does get no error with 'mvn clean compile test' but if I call SimpleTestMain directly.

test-jdo-sql-ResultSetROF-getObject.zip

from datanucleus-rdbms.

Thx. The testcase reproduces a problem using Junit on Oracle (11g) (no need for any "main"), due to how Oracle JDBC maps types

from datanucleus-rdbms.

Suggest that you try with the change made above, since that is way simpler than your PR and works.for.me

from datanucleus-rdbms.

It looks like your commit solves the conversions problem for Oracle. Thx.

My commit would be an additional enhancement for the way the data is fetched from the ResultSet.
ResultClassROF.getResultObject(ResultSet, int) has an optimization to use the suitable getter (instead of getObject()), if the query has a simple return type. My commit expands this optimization for the fields of the class in case of a class return type.

from datanucleus-rdbms.

Marking as fixed, but retaining PR since would be nice to rework the whole class and could use as part of that

from datanucleus-rdbms.