Greetings, a somewhat length summary of tonight's detective work on OJB. :)
Originally the story started when I was debugging Oracle test-case failures in OJB and homed in on ClassDescriptor and the use of java.sql.Timestamp.
In the "locked-by-timestamp" locking tests, the code in ClassDescriptor would set a "new java.sql.Timestamp()" as object value before writing the lock out to DB. When the value was read back from Oracle the .getNanos() always resulted 0 and would cause locking exceptions (since nanos in the object written out was not always, almost never in fact, set to 0).
I misinterpreted "setNanos", didn't check the JDBC doco properly, and submitted a patch that Armin applied with setNanos(0) in the ClassDescriptor-code (which will in fact truncate nanos, micros and millis *doh*).
This did solve the Oracle failures in the test-cases, but later on a user realized that this created problems on other platforms. Armin then accepted a patch from this user that did not truncate the nanos completely, but folded it down to microsecond precision.
This still failed with Oracle though, but today for the first time I really checked the DDL and saw that Torque was generating a DATE-column for LOCKED_BY_TIMESTAMP.TIMESTAMP_ - argh! How silly.
A bit of googling reveals that Oracle in pre-9 versions does not have TIMESTAMP, only DATE - with second precision. That is why my old patch worked on Oracle but stuffed up locking precision to such a low granularity to cause problems on other platforms.
Summary: there was nothing wrong in ClassDescriptor in the first place, just the Torque-generated DDL for the test cases.
I have rolled out both patches in order to raise lock precision back to nanosecond for all platforms.
The locking-test will only work for Oracle9i or higher and only if we get Torque to generate correct DDL. I interpret this Torque issue:
http://issues.apache.org/scarab/issues/id/TRQS263
as that they are working on it. It might actually already be possible to switch between Oracle v<=8 or 9+ in Toqrue 3.1.1 - I will look into this later.
For now, I think I will submit a change to the timestamp locking testcases that will treat Oracle* platforms as a "known issue" and skip on -DOJB.skip.issues=true, using ojbSkipKnownIssueProblem. At least in 1.0-branch.
Regards, Martin
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
