[infinispan-dev] Raison de etre of TestObjectStreamMarshaller

[Galder Zamarreno]

Galder Zamarreno wrote:
> Hi guys,
> 
> I've spotted that JdbcMixedCacheStoreTest and others currently use 
> TestObjectStreamMarshaller rather than VersionAwareMarshaller. Why is 
> this? Easies of debugging with XStream based marshaller?
> 
> To increase test coverage, I'd suggest using VAM rather 
> TestObjectStreamMarshaller in tests as well.
> 
> In the case of JdbcMixedCacheStoreTest, the extended test coverage comes 
> from the fact that fromStream and toStream use start/multiple 
> writes/finish marshalling pattern, which is different to RPC calls, 
> where it's basically, start/one write/finish.
> 
> I can see the point of dummy cache stores cos you avoid needing a 
> database, files...etc, but I'm not sure whether about a dummy/test 
> marshaller. The more the production marshaller is used, the bigger test 
> coverage we have.

One thing I've noted when testing JBoss Marshaller is that during 
payload size comparisons, old Marshaller was broken for example when 
marshalling/unmarshalling:

- 3 out of 4 InternalCacheEntries impls: MortalCacheEntry, 
TransientCacheEntry and TransientMortalCacheEntry
- Bucket
- 3 out of 4 InternalCacheValue impls: MortalCacheValue, 
TransientCacheValue and TransientMortalCacheValue

These classes seem to be primarily oriented at CacheStore 
marshalling/unmarshalling and I believe my email above would explain why 
tests passed in spite of old Marshaller being broken.

> 
> Regards,

-- 
Galder Zamarreño
Sr. Software Maintenance Engineer
JBoss, a division of Red Hat