ISPN-6773: treat non-MetadataAware InternalCacheEntries as not needin…

[ksobolew]

…g the entire Metadata object to be stored.

https://issues.jboss.org/browse/ISPN-6773

• Make CacheEntry not implement MetadataAware
  • Now only the Metadata_CacheEntry classes (like MetadataMortalCacheEntry)
    implement this interface, while non-Metadata_ classes (like the plain
    MortalCacheEntry) do not.
  • The getMetadata() and setMetadata() method need to be specified in
    CacheEntry interface, as they used to be inherited from MetadataAware.
• Make InternalEntryFactoryImpl.isStoreMetadata() check the type of the
  InternalCacheEntry in addition to the Metadata object and return false if it
  does not implement MetadataAware.
• If the entry is new, pass null as the InternalCacheEntry and test only the
  Metadata object.

This way the correct code path (fot Metadata* or non-Metadata*) in
EntryWrappingInterceptor is used even if the Functional API replaces the
Metadata object inside the InternalCacheEntry.

[galderz]

I'm sorry but we can't accept the change as is. CacheEntry is a public API that we expose and your change could cause issues to existing applications. I agree the interface does not make as much sense as it should (see ISPN-3460) but we can just change it like that. So, I'd suggest you find an alternative way to do what you want to do. I can certainly help you find an alternative way, but you should post a test that demonstrates why you need this change.

[ksobolew]

@galderz see 75d7d85 for the test case. In my case the EmbeddedMetadata object is created by the CacheLoader, not using the put-with-metadata API, but the result is the same.

[galderz]

@ksobolew I've sent a PR to your Infinispan repo to workaround this. Have a look and see what you think. The code in InternalEntryFactoryImpl is already making similar assumptions, so I think it should be fine.

[ksobolew]

@galderz Well, it looks like a workaround all right :) But if it's OK for you then it's OK for me

[galderz]

@ksobolew Seems like the PR needs rebasing, can you look into that?

[ksobolew]

@galderz @slaskawi Rebased.

[slaskawi]

Looking

[slaskawi]

Integrated thanks!

[danberindei]

It's no longer a marker interface :)

[ksobolew]

@danberindei well, yeah, I planned on doing a cleanup (there's also an orphaned import somewhere there) while rebasing but forgot about it :)