[hibernate-dev] Oracle12cDialect identity support

[Gail Badner]

Oracle12cDialect was added in 5.0 by HHH-9044. Unfortunately there are
problems with the new identity support that Oracle12cDialect introduced
(HHH-9983). The fix for HHH-9983 involved SPI changes, so it was applied to
master for 5.1 (only).

A couple of weeks ago, Andrea found that identities seemed to work using
5.0 when running BulkManipulationTest using Oracle JDBC 12.1.0.1.0. At the
same time it failed for me in the same way as HHH-9983 using 12.1.0.2.0.
Then Andrea and I switched the JDBC versions (Andrea tried 12.1.0.1.0; I
tried 12.1.0.2.0) and we were both able to run BulkManipulationTest
successfully. This was very strange, especially since we were using the
same Oracle instance, each using a different user. I didn't see any
difference in how we were setting Hibernate properties. Several days later,
I tried Oracle JDBC 12.1.0.2.0 again, but BulkManipulationTest was failing
again in the same way as HHH-9983.

Does anyone know how to get Oracle12cDialect's identity support to work for
5.0 reliably? Is there some version of Oracle JDBC or DB configuration that
is required?

If so, please let me know. The rest of this email assumes it is not
possible. If I'm missing some configuration detail, then just ignore the
rest of this message.

When the unit tests are run using Oracle12cDialect on 5.0, there are many
failures:
* tests with entities explicitly mapped to use an identity fail; these
failures did not happen in 4.3 using Oracle10gDialect because these tests
where skipped (because Oracle10cDialect#supportsIdentityColumns returns
false)
* use @GeneratedValue with default strategy or use hbm mapping for ID
<generator class="native"/>; these failures did not happen in 4.3 because
the native generator for Oracle10gDialect uses a sequence; the native
generator for Oracle12cDialect uses an identity.

Also, starting in 5.0, when using Oracle 12c, StandardDialectResolver will
automatically resolve to Oracle12cDialect if the dialect is not explicitly
mapped. Previously it would have resolved ot Oracle10gDialect.

I would be OK with changing the native generator in 5.0 from sequence to
identity if identities were working properly. I think it will cause
confusion when applications break. To get things working with 5.0,
applications will need to explicitly set the dialect to Oracle10gDialect,
extend Oracle12cDialect to disable identity support, or change ID mappings
to specify sequences.

Here are some alternatives to get things working.

Alternative A:
1) for 5.0 through 5.x override Dialect#getNativeIdentifierGeneratorClass
to return SequenceStyleGenerator.class; I suggest the change for 5.x
because I don't think it would be an acceptable to change the native
generator for Oracle12cDialect from using a sequence to using an identity
in  5.x;
2) for 5.0 only, remove support for identities;
3) for 6.0, if desireable, extend Oracle12gDialect (or whatever is current
at that time) to change the native generator to use an identity.

Alternative B:
1) for 5.0 through 5.x, leave Oracle12cDialect as is, and add a new dialect
with native generator that uses a sequence (i.e.,
Oracle12cNativeSequenceGeneratorDialect or something else that is not so
ugly);
2) for 5.0 through 5.x, change StandardDialectResolver to automatically
resolve to Oracle12cNativeSequenceGeneratorDialect for Oracle 12g server.

Alternative C:
1) for 5.0 through 5.x, change native generator for Oracle12cDialect to use
sequence; leave identity support as is (broken in 5.0; fixed in 5.1);
2) for 5.0 through 5.x, add a new dialect with native generator that uses
an identity (Oracle12cNativeIdentityGeneratorDialect or something else that
is not so ugly).
3) change StandardDialectResolver to automatically resolve to
Oracle12cDialect for Oracle 12g server.

I prefer alternative A. Alternatives B and C involve maintaining 2 dialects
for the same Oracle version; I would think that one of them would be
deprecated moving forward.

Thoughts?

Thanks,
Gail