[DERBY-4062] Remove single-argument getDataValue overrides from DataValueFactory interface

• Type: Sub-task
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 10.6.1.0
• Bug behavior facts: Data corruption, Wrong query result
• Assignee: Bryan Pendleton
• Reporter: Bryan Pendleton
• Created: 17/Feb/09 01:53
• Updated: 21/Jan/11 17:52
• Resolved: 14/May/09 02:36

Description

[ Issue title edited to reflect the discussion about how to clarify the use of this interface. ]

I believe the problem involves o.a.d.iapi.types.DataValueFactory.
This interface defines dozens and dozens of overloads of the method
getDataValue(), for lots of different combinations of datatypes.

For most of the Java "boxed" types (Short, Long, Float, Double, etc.),
DataValueFactory defines a pair of getDataValue() methods. For example,
here are the method pair that the interface defines for Short:

/**

• Get a SQL smallint with the given value. A null argument means get
• a SQL null value. The second form uses the previous value (if non-null)
• to hold the return value.
  *
  */
  NumberDataValue getDataValue(Short value);
  NumberDataValue getDataValue(Short value, NumberDataValue previous)
  throws StandardException;

HOWEVER, for the Integer type, DataValueFactory doesn't define both overloads,
but only defines the 'previous'-style overload:

/**

• Get a SQL int with the given value. A null argument means get
• a SQL null value. Uses the previous value (if non-null)
• to hold the return value.
  *
  */
  NumberDataValue getDataValue(Integer value, NumberDataValue previous)
  throws StandardException;

The actual implementation, in o.a.d.iapi.types.DataValueFactoryImpl, though,
does implement both the Integer overloads. But this method is NOT present
in the DataValueFactory interface:

NumberDataValue getDataValue(Integer value);

Because this method is not present in the interface, code such as

row.setColumn(SYSXPLAIN_RESULTSET_NO_OPENS, dvf.getDataValue(no_opens));

which the code anticipates will invoke the above method, instead calls the method

public UserDataValue getDataValue(Object value);

which has a very different behavior (instead of returning a SQLInteger, it returns a UserType).

This accidental invocation of the wrong implementation method was causing data corruption
errors in regression tests for the DERBY-2487 patch, which uses the above setColumn call.
Instead of inserting SQLInteger values into the system table, the code was inserting
java.lang.Integer UserType values; since those values don't match the defined type of
the column(s) in the system catalog, the table appeared to be corrupt.

I believe that this problem never affects external Derby applications, but only internal Derby code,
as the DataValueFactory interface is an internal interface only. Still, since it appeared to
cause data corruption and invalid query results, it is potentially a quite serious problem.

See this thread in the derby-dev archives for a bit more discussion:
http://mail-archives.apache.org/mod_mbox/db-derby-dev/200902.mbox/[e-mail]%3E