[CAUSEWAY-812] Isis 1.5 blob mapping broken for PostgreSQL (when set to null)

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: archetype-quickstart-wrj-1.5.0, objectstore-jdo-1.5.0
• Fix Version/s: archetype-todoapp-1.6.0, core-1.6.0
• Component/s: Persistence JDO
• Assignee: Daniel Keir Haywood
• Reporter: Daniel Keir Haywood
• Created: 17/Jun/14 21:53
• Updated: 20/Feb/15 14:06
• Resolved: 17/Jun/14 22:07

Description

As reported on mailing list: http://markmail.org/thread/zoiu2v4sklo7mevq

~~~
Looking into this, I've learnt that PostgreSQL does not support BLOB datatype, but instead supports "bytea" (byte array) and "oid". The first is in-row storage, where the data is always returned in the result set, the second is a pointer to off-row storage; PostgreSQL provides its own API for handling oid types).

For the Isis Blob class (org.apache.isis.applib.value.Blob) we provide our own IsisBlobMapping; this is automatically registered through metadata held in the JDO applib jar. It is the job of this mapping class to read/set values on the underlying ResultSet / PreparedStatement. DataNucleus provides some additional wrapping infrastructure.

The warning mentioned in the mailing list post indicates that DN is treating the BLOB datatype as a "bytea". This almost works fine; our IsisBlobMapping can persist non-null values ok, but attempting to set a null value breaks PostgreSQL; it treats the null as a null pointer of type "oid" (offline storage), rather than a null of type "bytea".

After some experimentation, I've found a different way to set null that works for PostgreSQL and also works fine for HSQLDB, and MS SQL Server.

This ticket is to apply that change.