db-jdo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andy Jefferson <a...@jpox.org>
Subject Re: JDO TCK Conference Call Friday, Aug 12, 9 am PST
Date Fri, 12 Aug 2005 19:16:57 GMT
> Issue 106 is the biggest single problem. It is about mapping Object
> fields, which are covered in the specification. Section 6.4.3 says
> the implementation must support Object fields as First Class Objects.

OK, but the wording of that part of the spec was discussed in the EG back on 
02 April 2005 (thread titled "updates to tck11 project", for those with EG 
access) and the general conclusion was that it was totally vague, and maybe a 
JDO implementation should be able to map "some PC type" to that Object field 
but how it can be testable by the TCK is open to debate. How does the JDO 
impl know that this Object field can only be a "SimpleClass" PC object 
maybe ? There's no info that I can see in the metadata to define this.

It's the same problem space as interfaces. With those we have the "implements" 
attribute which tells us which actual types we can store there. Consequently 
JPOX models an interface field as being one FK for each implementation type 
(so if there are 3 possible implementations, then the schema has 3 FKs). With 
"Object" we have nothing to go on, so at the time of mapping the schema we 
know that we may have to map something. Clearly the JDO implementation could 
define an extension tag (akin to the JPOX "implementation-classes" that we 
had for interfaces) that defines that this Object field can be represented by 
these 2 PC types, but that becomes not testable by the TCK at that point.

If there's some action on JPOX from this, just let us know so we can work out 
what is needed.


View raw message