[hibernate-dev] OGM test refactoring

[Jonathan Halliday]

In the course of developing the cassandra backend for OGM I've hit an 
issue with some of the backendtck tests.

These tests are in core and reused by each backend. It's possible to 
exclude tests at class or method level per backend, which prevents 
irrelevant tests from running for a given backend. So far, so good.

The problem I'm seeing stems from the fact that even when a test is 
excluded, the annotated POJOs it uses are still parsed and passed 
through schema creation, since that occurs in a setup step before the 
individual tests are iterated and (not) run.

There are certain POJOs that contains mappings cassandra can't support. 
Even with the corresponding test disabled, when these are passed to the 
schema creator it blows up, effectively preventing the other legitimate 
tests in the same class from running.

In an ideal world I'd like a class/method level annotation listing the 
required POJOs to replace the getAnnotatedClasses method, allowing POJOs 
to be parsed or skipped in accordance with the corresponding annotation 
on the same test method.

@RequiredPOJO(Foo.class)
@Test
@SkipByGridDialect(...)
public void testFoo()

Alternatively I can (I think) subclass the test class in the backend and 
override the getAnnotatedClasses method to exclude the problem POJOs, 
whilst leaving the @SkipByGridDialect on the shared parent for clarity. 
  Or simply copy and paste the whole thing and nuke the offending 
POJOs/methods in my backend's copy.

On a related note, many of the offending tests could be altered in such 
a way that the resulting schema is acceptable to cassandra, which would 
largely sidestep the issue but alter the test semantics. In many 
instances I suspect the semantic in question is one that's not critical 
to the test's intent, but I may be wrong.

Essentially cassandra uses what a relational db would call 'index 
organized tables', i.e. the on-disk layout of data is a function of the 
table's key. As a result, it's not possible to declare tables without a 
primary key. Such mappings result from collections with bag semantics, 
such as @OneToMany List<Object>, since the list may have dups. Removing 
the bag semantics such as by adding an index column, avoids this issue. 
But then you're testing something different...


Any preference on the approach to changing the test suite?

Jonathan.

-- 
Registered in England and Wales under Company Registration No. 03798903 
Directors: Michael Cunningham (USA), Matt Parson
(USA), Charlie Peters (USA), Michael O'Neill(Ireland)